Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Transient flow and transport in unsaturated heterogeneous media : field experiments in mine waste rock Nichol, Craig Ferguson 2002

Your browser doesn't seem to have a PDF viewer, please download the PDF to view this item.

Item Metadata

Download

Media
831-ubc_2002-732207.pdf [ 26.22MB ]
Metadata
JSON: 831-1.0052471.json
JSON-LD: 831-1.0052471-ld.json
RDF/XML (Pretty): 831-1.0052471-rdf.xml
RDF/JSON: 831-1.0052471-rdf.json
Turtle: 831-1.0052471-turtle.txt
N-Triples: 831-1.0052471-rdf-ntriples.txt
Original Record: 831-1.0052471-source.json
Full Text
831-1.0052471-fulltext.txt
Citation
831-1.0052471.ris

Full Text

TRANSIENT FLOW A N D TRANSPORT IN U N S A T U R A T E D HETEROGENEOUS MEDIA: FIELD EXPERIMENTS IN MINE W A S T E R O C K B Y CRAIG FERGUSON NICHOL B.A.(Hons.), Cambridge University, 1990 M . S c , University of Birmingham, 1992 A THESIS SUBMITTED IN PARTIAL F U L F I L L M E N T OF THE REQUIREMENTS FOR THE D E G R E E OF DOCTOR OF PHILOSOPHY in THE F A C U L T Y OF G R A D U A T E STUDIES Department of Earth and Ocean Sciences We accept this thesis as conforming to the required standard THE UNIVERSITY OF BRITISH C O L U M B I A March 2002 © Craig Nichol 2002 In presenting this thesis in partial fulfilment of the requirements for an advanced degree at the University of British Columbia, I agree that the Library shall make it freely available for reference and study. I further agree that permission for extensive copying of this thesis for scholarly purposes may be granted by the head of my department or by his or her representatives. It is understood that copying or publication of this thesis for financial gain shall not be allowed without my written permission. Department of CwQt % Qc&w &se<*C<?<, The University of British Columbia Vancouver, Canada Date 2.lh/o L. DE-6 (2/88) ABSTRACT It is essential to understand how water is transported and stored in mine waste rock to predict the rates and quantities of mineral weathering products released to the environment. The lack of understanding of flow mechanisms is the largest of the difficulties facing efforts to predict waste rock leaching. A large scale (8m x 8m x 5m high) constructed waste rock pile experiment (CPE) has been built upon a contiguous grid of 16 drainage lysimeters to better characterize the flow of water in unsaturated waste rock. The measurement of water flow parameters in a waste rock experiment required the development and testing of new instrument methods to determine water content, using time domain reflectrometry (TDR), and matric suction, using thermal conductivity sensors. The use of a resistive coating on the TDR probe conductors is found to be a successful strategy for obtaining measurable TDR signals and hence water contents. The thermal conductivity sensors are successful in certain soil conditions, but not those present in the CPE. Water flow data from the first two and a half years of the experiment are examined. Average net infiltration to an uncovered waste rock pile is 55% of precipitation. Large rainfall events have net infiltration of 55% to 85%. Water flow is spatially heterogeneous at scales less than 2m. In-situ instrumentation is found to be a poor (non-conservative) predictor of water flow. Data are analysed from the first year of a tracer test carried out under transient infiltration. Physical mechanisms of water flow are identified including non-capillary flow, flow in macropores, and water flow in the granular matrix. Water chemistry is different between different pore sizes, and between spatially distinct areas of the pile. The residence time distribution has an estimated mean of 3.0 to 3.9 years under average infiltration conditions. i i TABLE OF CONTENTS ABSTRACT ii TABLE OF CONTENTS iii LIST OF TABLES x LIST OF FIGURES xi ACKNOWLEDGEMENTS xxi DEDICATION xxii CHAPTER 1: INTRODUCTION 1 1.1 PURPOSE 1 1.2 STUDY OBJECTIVES 2 1.3 S U M M A R Y OF R E S E A R C H P R O G R A M 3 1.4 INSTRUMENTATION 3 1.5 EXPERIMENTS 4 1.6 STRUCTURE OF THESIS A N D PRESENTATION OF R E S E A R C H P R O G R A M 5 CHAPTER 2: INTRODUCTION TO THE CONSTRUCTED PILE EXPERIMENT.... 0-2.1 INTRODUCTION 9 2.2 E X P E R I M E N T A L DESIGN 9 2.2.1 Waste Rock Composition and Internal Pile Structure 10 2.2.2 Instrumentation 10 2.3 INITIAL RESULTS 11 2.4 CONCLUSIONS 12 i i i CHAPTER 3: EVALUATION OF UNCOATED AND COATED TDR PROBES FOR HIGH ELECTRICAL CONDUCTIVITY SYSTEMS 20 3.1 A B S T R A C T 20 3.2 INTRODUCTION 21 3.3 THEORY 23 3.3.1 Effect of solution electrical conductivity 25 3.3.2 Waveform Differencing and Remote diode shorting 26 3.3.3 Resistive Probe Coatings 27 3.4 M A T E R I A L S A N D METHODS 29 3.5 RESULTS A N D DISCUSSIONS 32 3.5.1 Uncoated Probes 32 3.5.2 Coated Probes in Water 36 3.5.3 Coated Probes in Saturated Sand 39 3.5.4 Effect of probe coating 41 3.5.5 Recommendations 43 CHAPTER 4: TIME DOMAIN REFLECTROMETRY MEASUREMENTS OF WATER CONTENT IN COARSE WASTE ROCK 56 4.1 A B S T R A C T 56 4.2 INTRODUCTION 57 4.3 THEORY 58 4.3.1 Dielectric Properties of Soil-Air-Water Mixtures 58 iv 4.3.2 Measurement of Dielectric Permittivity using Time Domain Reflectrometry...60 4.3.3 CPE Probe Design 63 4.3.4 Temperature Dependence 64 4.4 E X P E R I M E N T A L METHODS 65 4.4.1 TDR Instruments and Waveform Collection 66 4.4.2 Probe Head/End Offset 66 4.4.3 Performance in Electrically Conductive Media 67 4.4.4 Laboratory Calibration Measurements for CPE waste rock 68 4.4.5 Field Measurements 69 4.5 RESULTS A N D DISCUSSION 70 4.5.1 Probe Head/End Offset 71 4.5.2 Effect of Electrical Conductivity .'. 72 4.5.3 CPE Waste Rock Calibration 74 4.5.3.1 Probe Variability and Grain Size 74 4.5.3.2 Soil Water Electrical Conductivity 76 4.5.4 Field Measurements 77 4.5.4.1 Field Temperature Correction 77 4.5.4.2 Corrected Field Data 80 4.6 CONCLUSIONS 81 CHAPTER 5: FIELD EVALUATION OF THERMAL CONDUCTIVITY SENSORS FOR THE MEASUREMENT OF MATRIC SUCTION 94 5.1 A B S T R A C T 94 v 5.2 INTRODUCTION 95 5.3 T H E R M A L CONDUCTIVITY SENSORS 97 5.3.1 Sensor hysteresis 98 5.3.2 Correction Factor for Ambient Temperature 99 5.4 FIELD INSTRUMENTATION 103 5.5 RESULTS 106 5.5.1 Uncorrected Measurements 106 5.5.2 Sensor Hysteresis and Ambient Temperature 109 5.5.3 Comparison to Tensiometer Data I l l 5.6 DISCUSSION 113 5.7 RECOMMENDATIONS 116 CHAPTER 6: WATER FLOW IN UNSATURATED HETEROGENOUS POROUS MEDIA: A CONSTRUCTED WASTE ROCK PILE 126 6.1 A B S T R A C T 126 6.2 INTRODUCTION 127 6.3 E X P E R I M E N T A L DESIGN A N D CONSTRUCTION 131 6.3.1 Waste Rock Composition And Internal Pile Structure 133 6.3.2 Instrumentation 135 6.4 FIELD ACTIVITIES A N D D A T A A N A L Y S I S 137 6.4.1 Analysis Methods 138 6.5 RESULTS A N D DISCUSSION 139 6.5.1 Material Characterization 139 vi 6.5.2 Whole Pile Water Balance 144 6.5.3 Individual Lysimeter Data 146 6.5.4 Outflow Hydrographs 148 6.5.5 Event Based Outflow Volume 152 6.5.6 In-situ measurements and direct measurements of outflow 156 6.6 CONCLUSIONS 159 CHAPTER 7: SOLUTE TRANSPORT IN AN UNSATURATED HETEROGENEOUS MEDIUM: A CONSTRUCTED WASTE ROCK PILE 183 7.1 A B S T R A C T 183 7.2 INTRODUCTION 184 7.3 SOLUTE TRANSPORT IN U N S A T U R A T E D M E D I A 185 7.3.1 Homogeneous Porous Media 186 7.3.2 Transport 187 7.3.3 Preferential Flow, Heterogeneity, Macropores and Non-capillary Flow 188 7.3.4 Conceptual Model of Effects of Heterogeneity During Transient Infiltration. 191 7.3.5 Heterogeneity and Transient Infiltration 194 7.3.6 Monitoring Methods at the CPE 195 7.4 FIELD METHODS A N D D A T A A N A L Y S I S 197 7.4.1 Rainfall Record and Tracer Application 197 7.4.2 In-situ and Outflow Water Sampling 197 7.4.3 Data Analysis 199 7.5 RESULTS A N D DISCUSSION 200 vii 7.5.1 Whole Pile Tracer BTC and Mass Recovery 200 7.5.2 BTC's and Tracer Mass Recovery in Individual Lysimeters 203 7.5.3 Summary of the Character of Individual BTC's 205 7.5.4 Flow and Tracer Spikes 206 7.5.5 Tracer Arrival with the Wetting Front 209 7.5.6 No Tracer Arrival with the Wetting Front 210 7.5.7 Hydrographs and BTCs from Summer 2000 212 7.5.8 Summary of Observations from Lysimeter Measurements 213 7.5.9 In-Situ Tracer Measurements 214 7.5.10 Residence Time Estimates 219 7.6 CONCLUSIONS 221 CHAPTER 8: SUMMARY AND CONCLUSIONS 239 8.1 DURATION OF INITIAL WETTING PHASE 239 8.2 HOW A R E PHYSICAL P A R A M E T E R S BEST M E A S U R E D ? 240 8.3 W H A T P A R A M E T E R S A R E CRITICAL? W H A T IS THE D E G R E E OF SPATIAL VARIABILITY OF FLOW? 242 8.4 THE F A T E OF PRECIPITATION, FLOW MECHANISMS, SPATIAL V A R I A B I L I T Y , RESIDENCE TIME 242 8.5 CHANGES IN PILE CONSTRUCTION PRACTICES 243 REFERENCES 245 V l l l APPENDIXES: APPENDIX A : PAPER SUBMITTED TO 5 t h I.C.A.R.D. C O N F E R E N C E 256 APPENDIX B: CPE CONSTRUCTION DETAILS 267 APPENDIX C: CPE SUCTION LYSIMETERS 293 APPENDIX D: CPE TENSIOMETERS 299 APPENDIX E: CPE TEMPERAT URE PROBES 311 APPENDIX F: CPE D A T A L O G G I N G 313 APPENDIX G: D E V E L O P M E N T OF R A I N F A L L SIMULATORS FOR L A B O R A T O R Y C O L U M N A N D CPE EXPERIMENTS 327 APPENDIX H: INSTRUCTIONS FOR OPERATING CPE R A I N F A L L SIMULATOR A N D CONDUCTING A N ARTIFICIAL R A I N F A L L EVENT 338 APPENDIX I: CPE TDR PROBE CALIBRATION DETIALS 342 APPENDIX J: CPE T H E R M A L CONDUCTIVITY SENSOR CALIBRATION A N D M E A S U R E M E N T DETAILS 345 APPENDIX K : CPE TIPPING B U C K E T FLOW G A U G E C A L I B R A T I O N A N D M A I N T E N A N C E 350 APPENDIX L: M E A S U R E M E N T OF DISSOLVED CHLORIDE IN W A T E R SAMPLES F R O M THE CPE 368 APPENDIX M : L A B O R A T O R Y C O L U M N EXPERIMENT DETAILS 384 ix LIST OF TABLES Table 6.1 Summary of monthly rainfall and precipitation statistics, measured monthly precipitation and measured monthly outflow of the whole CPE 180 Table 6.2 Summary of artificial rainfall events 181 Table 6.3 Estimated outflow volumes for individual rainfall events and individual lysimeters 182 x LIST O F FIGURES Figure 2.1 Site of the constructed pile experiment 14 Figure 2.2 Simplified plan view of the constructed pile experiment 15 Figure 2.3 Simplified cross section A - A ' of constructed pile 16 Figure 2.4 Grain size of waste rock in constructed pile 17 Figure 2.5 Daily rainfall volume and outflow rate summary 18 Figure 2.6 Summary of volumetric water content and matric suction profile at pile center 19 Figure 3.1 Schematic of TDR probe design 45 Figure 3.2 Raw time domain reflectrometry waveforms with manually determined tangent-lines to the probe end reflection: 160 mm length uncoated probe 46 Figure 3.3 Remote diode shorting method waveforms for diodes located: (AB) at the probe head and (CD) at the probe base: 160 mm length uncoated probe in water solutions of varying electrical conductivity 47 Figure 3.4 Manual probe shorting method waveforms for short circuits using copper wire located: (AB) at the probe head end of the exposed conductors and (CD) at the probe base end of the exposed conductors: 160 mm length uncoated probe in water solutions of varying electrical conductivity 48 Figure 3.5 Automated remote diode shorted method measured two-way travel time, travel time measured using manual probe shorting and waveform differencing and travel time calculated from Hasted (1973) data: 160 mm length uncoated probe in water solutions of varying electrical conductivity 49 Figure 3.6 Raw time domain reflectrometry waveforms with manually determined xi tangent lines to the probe end reflection: 281 mm length coated probe in water solutions of varying conductivity 50 Figure 3.7 Remote diode shorting method waveforms for diodes located: (AJ3) at the probe head and (CD) at the probe base: 281 mm length coated probe in water solutions of varying electrical conductivity 51 Figure 3.8 Automated remote diode shorted method measured two way travel time: 281 mm length coated probe in water solutions of varying electrical conductivity 52 Figure 3.9 Remote diode shorting method waveforms for diodes located: (AB) at the probe head and (CD) at the probe base: 281 mm length coated probe in silica sand saturated with water solutions of varying electrical conductivity 53 Figure 3.10 Automated remote diode shorted method measured two way travel time: 281 mm length coated probe in silica sand saturated with water solutions of varying electrical conductivity 54 Figure 3.11 Relationship of coated probe measured apparent dielectric to apparent dielectric. A) Linear correction B) Equation 6 using known probe dimensions: r0= 12.5 mm., Kc=2.8 C) Equation 6 using r 0 as fitting parameter: r0= 53.5 mm., Kc=2.8 55 Figure 4.1 TDR waveform collected using uncoated TDR probe in water with 1 dSm"1 electrical conductivity. Tangent-line methods are used to determine the timing of the probe head (t0) and probe end (ti) reflection 84 Figure 4.2 Design of TDR probes installed in field experiment 85 xn Figure 4.3 Variation of probe head/end offset (t0ff) with sample dielectric constant 86 Figure 4.4 Corrected travel time (t*/tcair) measured using a coated probe in water solutions of varying electrical conductivity (A), and silica sand saturated with electrically conductive water (B,C). Probe head travel time determined using remote diode method. Corrected travel time determined using the remote diode method at probe end (A,B) the flat-tangent method at the probe end (C), or manually determined dual-tangent method at probe end 87 Figure 4.5 Effects of probe variability and grain size on the relationship of corrected travel time (t*/tcair) to volumetric water content. 88 Figure 4.6 Calibration curves relating corrected travel time (t*/tcair) to volumetric water content for waste rock with high soil water electrical conductivity (A) and low soil water electrical conductivity (B) 89 Figure 4.7 Daily variation in measured corrected travel time (t*/tcair) for a probe at 20 cm depth (A). Comparison of corrected travel time variation with temperature variation (B) 90 Figure 4.8 Field determined ambient temperature correction, (At*/t c aj r )/AT, as a function of corrected travel time, (t*/tcajr), for a probe at 20 cm depth 91 Figure 4.9 Application of field determined ambient temperature correction to field data: (A) corrected travel time (t*/tcair); (B) 24 hour moving window average of corrected travel time (t*/tc air)24a v g; and (C) temperature corrected travel time (t*/t c a i r)T 92 xiii Figure 4.10 Effect of ambient temperature correction and remote diode shorting bias on measured water contents: (A) volumetric water content using appropriate calibration for soil water electrical conductivity; (B) temperature corrected travel time (t*/t C A; R)T; (C) ambient soil temperature; and (D) volumetric water content using wrong calibration for soil water electrical conductivity 93 Figure 5.1 Schematic of TC sensor design (after Feng, 1999) 119 Figure 5.2 Effects of hysteresis on: (A) TC sensor volumetric water content and (B) measured sensor core temperature rise. Simplified from data in Feng (1999)... 120 Figure 5.3 Correction factor for ambient soil temperature. The field measured TC sensor core temperature rise is multiplied by the correction factor to obtain the core temperature rise that would have been measured at 23 °C 121 Figure 5.4 Preliminary estimate of matric suction based upon laboratory calibration curve for a drying sensor 122 Figure 5.5 Daily fluctuation of matric suction for a TC sensor installed at 4.5 m depth: matric suction calculated using laboratory calibration curve for a drying sensor (A, kPa); 24-hour moving-window average of estimated suction (B, kPa); air temperature (C, °C); ambient soil temperature (D, °C) and variation of estimated matric suction from the 24-hour moving-window average of estimated suction (E) 123 Figure 5.6 Demonstration of the effects of 24-hour moving window averaging, sensor hysteresis, and ambient temperature correction for a sensor at 0.2 m depth. Matric suction is calculated from individual readings using the laboratory calibration curve for a drying sensor (1) and a wetting sensor (2). xiv Corrected matric suction is calculated using a 24-hour moving-window average of TC sensor output and using a drying calibration curve (3) and wetting calibration curve (4) and ambient soil temperature (5) 124 Figure 5.7 Assessment of TC sensor accuracy, precision and response time for two sensors (X and Y) located at 0.2 m depth: estimated matric suction suction based upon a 24-hour moving-window average of sensor output and a drying sensor calibration curve (1) and a wetting sensor calibration curve (2); matric suction measured manually using a tensiometer (3); matric suction measured using a tensiometer and pressure transducer (4); and rainfall amount and timing 125 Figure 6.1 Simplified cross section of constructed pile experiment including details of basal lysimeters (Inset A), instrument profiles (Inset B) and a plan view of experiment core (Inset C) 164 Figure 6.2 Side view photograph of constructed pile experiment 165 Figure 6.3 Grain size analyses of samples collected during pile construction 166 Figure 6.4 Laboratory and field derived soil water characteristic curves and hydraulic conductivity curves. A) Laboratory data for coarsest (squares), medium (triangles) and finest (circles) of 5 L grab samples for drying (closed symbols) and wetting (open symbols) B) Typical field measured data and estimated boundary drying curve C) Estimated boundary drying curves from 8 instrument locations D) Estimated hydraulic conductivity curves derived from field estimates of unsaturated hydraulic conductivity and the SWCC presented in (C) using the method of Fredlund and Xing (1994) 167 xv Figure 6.5 Daily precipitation totals and whole pile outflow rate for March 1999 to March 2001 168 Figure 6.6 Variability in total accumulated precipitation (A) and total outflow volume (B) between lysimeters from September 1998 to March 2001 169 Figure 6.7 Monthly total outflow volume for the period of August 1999 to March 2001 expressed as percentage of total for A) all lysimeters and B) with lysimeters 6 and 13 removed 170 Figure 6.8 Topographic survey of pile surface conducted August 2001 using 0.3 m x 0.3 m survey resolution. Drainage catchments are outlined in heavy shaded lines. Lysimeter numbers are indicated 171 Figure 6.9 Outflow hydrographs of lysimeters 6, 9 and 10 in response to the July 18, 2000 artificial rainfall event 172 Figure 6.10 Variation of peak flow rates during large rainfall events with the event magnitude expressed as the average maximum flow rate of all lysimeters. Highest flow rate divided by lowest flow rate (triangles) and average of the four highest rates divided by the average of the four lowest rates (circles)... 173 Figure 6.11 Outflow hydrographs for individual lysimeters combined into quarters (A) and halves (B), and all of the pile (A and B) in response to the July 18, 2000 artificial rainfall event 174 Figure 6.12 Outflow hydrographs for A) lysimeter 6 and B) lysimeter 9 in response to the September24, 1999 and October 20, 1999 artificial rainfall events. Hydrographs are translated laterally such that the first increases in flow are coincident 175 xvi Figure 6.13 A) Distribution of outflow volume between lysimeters expressed as percent deviation of outflow volume from the average outflow volume for different rainfall events (light lines) and the whole experimental period (heavy line) 176 Figure 6.14 Variation of spatial distribution of outflow volume for varying rainfall event magnitude. Ratio of volume of highest volume lysimeter to lowest volume lysimeter (triangles) and total volume of highest four volume lysimeters to lowest four volume lysimeters (circles) against the average maximum flow rate of the rainfall events 177 Figure 6.15 Flow rate recession curves of whole pile, slowest flowing lysimeter and fastest flowing lysimeter for A) winter 1999-2000 and B) 2000-2001. Comparison of winter 1999-2000 and 2000-2001 data for whole pile outflow (C) 178 Figure 6.16 Volumetric water content in instrument profile A and outflow hydrograph of lysimeter 10 in response to the September 20, 1999 artificial rainfall event 179 Figure 7.1 Conceptual framework of transient water and tracer movement in unsaturated heterogeneous porous media: (A) In-situ tracer profiles; (B) Outflow hydrograph and (C) flux-averaged tracer concentration in outflow 224 Figure 7.2 Precipitation (A), outflow flux rate (m3s"1m"2) (B) and flux-average tracer concentration (C) plotted against time for the experimental period from February 1999 to March 2001 225 Figure 7.3 Outflow flux rate (m3s"1m"2) (A) and flux-average tracer concentration xvii (B) plotted against cumulative outflow volume normalized to cross sectional area, and zeroed at the tracer application event. The experimental period from February 1999 to March 2001 is shown 226 Figure 7.4 Cumulative tracer recovery in outflow of the whole CPE experiment, normalized to the mass of applied tracer 227 Figure 7.5 Chloride concentration in outflow plotted against time for the grid of sixteen contiguous lysimeters at the pile base 228 Figure 7.6 Spatial variation of tracer transport. Spatial variation in tracer mass recovery (A) and normalized outflow volume (m m") (B) by lysimeter. Fraction of tracer mass recovered recovered plotted against total outflow volume (C). Spatial variation in the recovery ratio: the fractional tracer 3 2 recovery divided by the normalized volume per lysimeter (m m" )(D) 229 Figure 7.7 Breakthrough curves (A, C) and outflow hydrographs (B, D) for lysimeters 2 (A, B) and 12 (C, D) for the tracer application event, and two following events. Data from the tracer application event is expanded in the inset graphs 230 Figure 7.8 Breakthrough curve (A) and outflow hydrograph (B) for lysimeter 13 or the period of for the tracer application event and two following rainfall events. Data from the tracer application event is expanded in the inset graph 231 Figure 7.9 Breakthrough curve (A) and outflow hydrograph (B) for lysimeter 9 for the period of August 28 to November 1, 1999. Data from the tracer application event is expanded in the inset graph 232 xviii Figure 7.10 Breakthrough curves (A, C) and outflow hydrographs (B, D) for lysimeters 9 (A, B) and 6 (C, D) for the period of June 6 to October 26, 2000 233 Figure 7.11: Tracer data from in-situ suction lysimeters located in instrument profile B: plotted as profiles (A); breakthrough curves (B); tracer concentration in outflow (C); and cumulative tracer recovery as the ratio to tracer applied (D). Cumulative recovery of outflow volume (D - dashed line) and cumulative recovery of tracer in outflow (D- heavy line) 234 Figure 7.12: Tracer data from in-situ suction lysimeters located in instrument profile A : plotted as profiles (A); breakthrough curves (B); tracer concentration in outflow (C); and cumulative tracer recovery as the ratio to tracer applied (D). Cumulative recovery of outflow volume (D - dashed line) and cumulative recovery of tracer in outflow (D- heavy line) 235 Figure 7.13: Tracer data from in-situ suction lysimeters located in instrument profile C: plotted as profiles (A); breakthrough curves (B); tracer concentration in outflow (C); and cumulative tracer recovery as the ratio to tracer applied D). Cumulative recovery of outflow volume (D - dashed line) and cumulative recovery of tracer in outflow (D- heavy line) 236 Figure 7.14: Comparison of tracer mass estimates from suction lysimeters and lysimeters. Ratio of total tracer mass remaining in CPE estimated from suction lysimeter samples, divided by tracer mass remaining determined from tracer recovered in outflow, plotted against cumulative outflow volume (m3m~2) 237 xix Figure 7.15: Estimated mean residence times from suction lysimeter data. Estimated residence time (years) derived from the velocity of the first moment, and estimated annual net infiltration. Plotted as a function of suction lysimeter depth 238 xx ACKNOWLEDGEMENTS This project is part of the Waste Rock Hydrology Research Program, a joint research program initiated between the University of British Columbia, the University of Saskatchewan, Cogema Resources Incorporated, Cameco Corporation, and the Natural Sciences and Engineering Research Council of Canada. Additional funding was provided through an NSERC Postgraduate Fellowship, and a University of British Columbia Graduate Fellowship. I would like to thank the many people who assisted with the construction and operation of the experiments at Cluff Lake and at U B C : Lloyd Daigneault, Leonard Mineault, Ray Kirkland, Tracy Bellehumeur, Nakib Ahmed, Sara Harrison, Keely Bright, Julie-Ann Moore, Pamela Fines, Bonnie Sjoberg, and Joe Marcoline. The assistance of the site staff at the Cogema Resources Cluff Mine operation, and in particular the Environment group, made this work possible. I am grateful to Del Fredlund and Fangsheng Shuai for the opportunity to use the thermal conductivity sensors. I am grateful to Leslie Smith and Roger Beckie for their support, guidance and for having doors that are always open (before 4:30). Thanks to Ward Wilson, Lee Barbour and Jim Hendry for their help from Saskatchewan and Vancouver. Thanks to all the members of the hydrogroup over the years for their friendship and good humour. xxi II "Yes, Christopher Robin?" "I'm not going to do Nothing any more." "Never again?" "Well, not so much. They don't let you." Pooh waited for him to go on, but he was silent again. "Yes, Christopher Robin?" said Pooh helpfully. "Pooh, when I'm - you know - when I'm not doing Nothing, wil l you come up here sometimes?" "Just Me?" "Yes, Pooh." "Wil l you be here too?" "Yes, Pooh, I will be, really. I promise I wil l be, Pooh." " A . A . Milne The House at Pooh Corner For Christine, Hamish, Alan and Heather xxii CHAPTER 1: INTRODUCTION 1.1 PURPOSE This thesis describes research undertaken to investigate the infiltration and flow of water within unsaturated mine waste rock. Mine waste rock is barren or low-grade material removed from an open pit or underground mine to gain access to the ore-grade material. Typically, waste rock is dumped in large unsaturated piles immediately adjacent to the mine workings. Fresh waste rock weathers after excavation, which may lead to the release of poor quality drainage water from the base of the dump into the receiving environment. The most common weathering reaction is the oxidation of metal sulphide minerals to produce sulphate, acidity and free metal ions. It is essential to understand how water is transported and stored in waste rock to predict the rates and quantities of mineral weathering products released to the environment. Waste rock is a more complex granular material than most natural soils. The grain size varies from clay to boulders, and the textures range from cobbles and boulders supported in a granular matrix to areas which are clast-supported with little or no granular matrix. This extreme variation in grain size, combined with a range of large-scale structures created within waste dumps by the haulage process, make water flow in waste rock very difficult to monitor experimentally, or simulate numerically. Water flow in waste rock is not well understood and represents the largest source of uncertainty in current efforts to predict the loadings of weathering products from existing piles, or to make predictions of potential future loadings prior to pile construction. Examination of the previously conducted field experiments on mine waste rock and data sets from operating mines (Chapter 6, Appendix A) indicated that existing data sets of 1 observations of flow and tracer were insufficient to allow a detailed study of flow mechanisms. This work therefore focussed on the collection of detailed observations of the flow of water in waste rock. When later combined with information on primary and secondary geochemical reactions, this will permit insight to the coupling between fluid flow and the release of metals from waste rock. The experimental work in this thesis is focussed on waste rock, but the mechanisms of water flow and solute transport discussed are also relevant to any unsaturated heterogeneous porous media. 1.2 STUDY OBJECTIVES The key questions that are addressed by this research are: • What is the duration of the initial wetting of a waste rock pile before any release of water is to be expected? • Which physical measurements of waste rock properties are critical to the characterization of flow in waste rock, and how are these best achieved? • What is the fate of precipitation onto a waste rock pile surface and how is the flow of water initiated? • What mechanisms control the flow of water within a waste rock pile, and how does water outflow at the base relate to water stored in the granular matrix within the pile? • What is the degree of spatial variability in unsaturated flow through waste rock? • What is the variation in residence times of water in waste rock? • What changes in pile construction practices may be beneficial in reducing the generation of acid rock drainage, or improving the ability to predict acid rock drainage. 2 1.3 SUMMARY OF RESEARCH PROGRAM The research program undertaken had two broad themes. The first was the development of instrumentation for the measurement of the hydrogeological parameters necessary to study water movement in waste rock. The second major focus was two experiments on waste rock: a specially constructed waste rock pile (CPE) at an active mine site and a laboratory column experiment. 1.4 INSTRUMENTATION The time domain reflectrometery method (TDR) of water content measurement was selected as the best means of making automated measurements of water content. The method measures the velocity of an electromagnetic pulse propagating in a probe placed in contact with the soil. The pulse velocity is related to the dielectric properties of the soil, and hence to the water content of the soil. Initial testing revealed that existing TDR instrumentation technology was not adequate in mine waste rock due to the high electrical conductivity of the soil water solutions present. This leads to degradation of the signal used by the TDR instrument, causing both inaccuracy and eventual failure of conventional TDR methods. Several previously proposed methods of improving signal quality were combined to design a new TDR probe. This probe was successful in obtaining interpretable measurements of the pulse velocity (Chapter 3). Further work was undertaken to calibrate the measurements of TDR pulse velocity to the actual water content of waste rock (Chapter 4). Both field and laboratory experiments were undertaken to calibrate TDR velocity measurements to water content. The final calibration work undertaken was to derive a correction for changes in ambient temperature 3 based upon the data collected during the constructed pile experiment. TDR probes were installed in both the laboratory column and constructed pile experiments. The measurement of water suction was undertaken using tensiometers (Appendix D) and thermal conductivity sensors (Chapter 5). Tensiometers and manometer tubes were used in the laboratory column experiment. Tensiometers equipped with pressure transducers were used in the CPE. Further automated measurements of soil matric suction in the CPE were undertaken using thermal conductivity (TC) sensors. These sensors are composed of a porous ceramic material which is heated using a controlled heat pulse. The thermal properties of the ceramic are measured, and hence the water content of the ceramic is estimated. Using calibrated measurements, this is related to the matric suction of the soil surrounding the probes. A new theoretical correction for changes in the ambient temperature of the soil is derived and applied to the field data. The performance of the thermal conductivity sensors is compared to that of tensiometers to determine the long term accuracy. 1.5 EXPERIMENTS Two experiments were conducted using waste rock obtained from the Cluff Lake mine located in Northern Saskatchewan. The main experiment is a large scale constructed waste rock pile experiment (CPE), undertaken at the Cluff Lake mine site. The experiment core is 8m x 8m x 5m in height. The CPE was constructed in the fall of 1997 and the summer of 1998. The experiment was operational from September 1998. Flow experiments were conducted in 1999 and 2000, and the results of analysis of water flow include data up to March 2001. A conservative tracer test was started in September 1999, and the results of monitoring up to October, 2000 are presented in this thesis. The CPE experiment is ongoing, and additional monitoring of water flow and tracer transport have been conducted but are not 4 presented in this thesis. Analysis of the results of flow measurements are presented in Chapter 6, and results of a conservative tracer test are presented in Chapter 7. A small scale laboratory column experiment ( lm diameter, 2m height) was constructed at the University of British Columbia, using material transported from the mine. The experiment was built in the spring of 1998, and several flow experiments were conducted. The experiment was then allowed to drain between the summer of 1998 and the spring of 1999. Another series of flow experiments were conducted in the spring of 1999, leading up to a conservative tracer test started in June, 1999. This experiment was used primarily as a testing ground for technologies and ideas prior to the construction and testing of the CPE. Detailed results from these experiments are not included in the main body of this thesis, but a summary of construction details, experiments and results are presented in Appendix M . 1.6 S T R U C T U R E O F THESIS AND P R E S E N T A T I O N O F R E S E A R C H P R O G R A M The results to date of this research are presented in the five main chapters. Chapters 3 to 7 are in the form of manuscripts submitted or prepared for submission to journals. A l l references between manuscripts have been updated to reflect the thesis structure. Chapter 2 presents a brief overview of the constructed pile experiment and its construction, and serves as background for instrument development detailed in Chapters 3 to 5. Some of the introductory material is repeated in the introduction to Chapter 6. Chapters 3 and 4 present the work undertaken to develop instrumentation to monitor water content in mine waste rock using the time domain reflectrometry (TDR) method. Chapter 3 describes the new TDR probe design, and the overall success of the strategies adopted to improve the quality of TDR measurements in high conductivity systems. Chapter 5 4 presents the details of the specific calibration of our waste rock material, and the derivation of field-based temperature corrections. There is some repetition of the background of the TDR method in the introductory sections of these two manuscripts. Chapter 5 presents the results of study of thermal conductivity sensors for the measurement of soil matric suction. The thermal conductivity sensors used for this research were prototype sensors purchased from Dr. Del Fredlund and his research group at the University of Saskatchewan. The sensors were calibrated by Dr. Fredlund's group prior to their installation in the CPE in 1998. During data collection, Dr. Fredlund's group provided additional information on sensor hysteresis and a preliminary temperature compensation equation. In return, several suggestions were made to improve the data collection and interpretation. Chapter 6 presents the first part of the results obtained from the constructed pile experiment. This includes a summary of the design and construction of the constructed pile experiment similar to Chapter 2. The results of water flow in the constructed pile are discussed, including: the overall water balance and calculation of net infiltration; spatial heterogeneity of water flow; temporal changes in water flow; evidence for preferential flow in waste rock; water balances from individual flow events; conditions which increase of decrease preferential flow; and the style and spatial scale of flow behaviour observed during winter drainage. Chapter 7 presents the results of the conservative tracer test carried out on the constructed pile. Data from September 1999 to October 2000 have been analysed. The results include an overall summary of the progress of the tracer test. This is followed by the presentation of a generic conceptual model of flow and transport in unsaturated porous 6 media. This framework is used as context to interpret the breakthrough curves and hydrographs of individual lysimeters. This includes discussion of the spatial and temporal variability of tracer transport mechanisms. The results from in-situ tracer measurements are then used to demonstrate the manner in which the generic transport model presented is characterized by the measurement methods used. Chapter 8 presents the conclusions of this thesis and a summary of the major contributions of this research. A series of thirteen Appendices are attached to this thesis. Appendix A presents the full text of the Nichol et al. (2000) paper which is referenced in several chapters and is summarized for Chapter 2. Appendix M provides the construction details, experimental outline and brief results summary from the laboratory column experiment. These two appendices provide additional information about the research program that directly contribute to the research presented in the main body of the thesis. Appendices B to L provide a permanent record of the details necessary for continued operation of the experiment. The constructed pile experiment has been taken over by other researchers. Many of the details of the construction of the CPE and the methods used to calibrate and operate equipment at the CPE are required for its continued operation. These Appendices provide: (B) detailed notes on the construction of the CPE, including photographs; (C) details of the preparation and operation of the suction lysimeters; (D) details of the installation and operation of the tensiometers and pressure transducers; (E) calibration details of the temperature probes installed in the CPE; (F) details of the datalogging equipment and programming for the CPE; (G) details of the construction and calibration of the rainfall simulators developed for both the laboratory column and CPE; (H) instruction for operation of the rainfall simulator at the CPE; (I) calibration measurements 7 taken for the TDR probes installed in the CPE; (J) calibration measurements taken for the thermal conductivity sensor installed in the CPE; (K) calibration details for the tipping bucket gauges used to monitor flow rate at the CPE; and (L) details of the method development and analysis protocol for the analysis of dissolved chloride in water from the CPE. Appendices D, G, J, K and L include more in-depth analysis of experiments required to develop and calibrate the methods contained in these appendices. 8 C H A P T E R 2: I N T R O D U C T I O N T O T H E C O N S T R U C T E D P I L E E X P E R I M E N T 2.1 I N T R O D U C T I O N A brief description of the constructed pile experiment is presented in this chapter to provide an introduction to the experiment sufficient for understanding the information on instrument development included in Chapters 3 through 5. These chapters refer to the paper by Nichol et al. (2000) for background information. The full text of this paper is included in Appendix A. A brief summary of the pile construction and early results is presented here. A more comprehensive review of waste rock issues, and further details of the experiment construction are included in Chapter 6. 2.2 E X P E R I M E N T A L D E S I G N The CPE was built at the Cluff Lake Mine in northern Saskatchewan, owned and operated by Cogema Resources Inc. of Saskatoon. The location of the Cluff Lake site is shown in Figure 2.1. Mean annual air temperature at Cluff Lake is 0°C. Average annual precipitation is 455 mm, with 305 falling as rainfall. The CPE foundations were constructed in 1997, and the remainder of the pile built from June to September, 1998. Automated and manual monitoring of the pile commenced in September 1998. A conservative tracer test was started in September, 1999, and is ongoing. Simplified plan views and cross sections of the pile are shown in Figures 2.2 and 2.3. The instrumented core of the pile has a footprint of 8 m by 8 m. Outflow from the base of the pile is collected in a contiguous grid of 16 gravity drainage lysimeters (Figure 2.3). Maintaining the lysimeters under suction was not considered to be technically feasible. Accordingly, a design leading to the formation of a water table at the base of the pile was chosen. Outflow 9 from each 2 m by 2 m lysimeter is separately piped to an instrumentation hut where outflow is monitored using tipping-bucket rain gauges. The top surface of the pile was finished to resemble other existing piles on site which were free-dumped and then contoured to a smooth surface. No attempt was made to create a trafficked dump surface nor any type of compaction or artificial cover, although these may be options for later study. Detailed notes, drawings and photographs detailing the construction of the pile are provided in Appendix B. 2.2.1 Waste Rock Composition And Internal Pile Structure The waste rock for this study was mined from the DJ Extension open pit at the Cluff Lake mine in the fall of 1996, and placed in the CPE in 1998. The grain size distribution of the 0.6m minus fraction of the waste rock is shown in Figure 2.4. Further details of the methods of grain size analysis and the material characteristics are discussed in Chapter 6. Grain size is highly heterogeneous, ranging from 1.5 m diameter boulders to clay, with areas that are matrix supported and areas with matrix-free cobbles and boulders. 2.2.2 Instrumentation The instrumentation used to measure water content, matric suction, temperature and matrix soil water chemistry were placed in three instrument profiles (Figures 2.2, 2.3). Water content is monitored continuously using time domain reflectrometry (Chapters 3 and 4). Matric suction is monitored year round using thermal conductivity sensors (also known as heat dissipation sensors) with laboratory-derived temperature and sensor hysteresis corrections (Fredlund and Shuai 1999, Feng 1999, Chapter 5). Tensiometers with pressure transducers were installed in 1999 to provide independent measurements of matric suction 10 when air temperatures are above freezing. Temperatures are measured using commercial pre-calibrated thermistors. Outflow water samples are collected in a cascade of mixing cells, where the water is directed through containers of increasing volume (0.2L, 2.0L and 25L) before final overflow to waste. Both instantaneous grab and longer term composite samples can be obtained without post-sampling compositing. Matrix water chemistry is monitored through the periodic manual extraction of water samples from suction lysimeters (also called soil water samplers) placed within the pile during construction. Approximately 12,000 samples have been collected during this research. Selected samples of outflow and soil-water were analysed for dissolved chloride to monitor the progress of a conservative tracer test (Chapter 7). Pile surface evaporation will be calculated using data from a full weather station and surface mini-lysimeters by other researchers involved in this project. Processed and calibrated data are not yet available. 2.3 I N I T I A L R E S U L T S A summary of rainfall and outflow measured during the initial wetting-up period of the pile from September, 1998 to July, 1999 is presented in Figure 2.5. Snowfall over the winter is generally blown off the top of the pile. Snow depth was measured to be 80 to 100 mm, which was included as 9 mm of rainfall in mid-March. The CPE was initially very dry due to intense evaporation from the soil surface during construction. Between September 1998 and July 1999, approximately 7000 litres of water fell as precipitation on the pile, and 220 litres of water exited as outflow. Details of the outflow records and water balance after the wetting up period are given in Chapter 6. 11 Preliminary estimates of water content by TDR and matric suction by thermal dissipation sensor from instrument profile A are shown in Figure 2.6. The TDR response down to a depth of 1 m was affected by ground freezing, and low measurements from November to March indicate the conversion of free water to ice, which does not register as water content. During freezing conditions the matric suction sensors do not operate, and data are absent, or erratic. The progress of the first wetting front in the capillary-dominated matrix material can be followed by observing increases in water content and changes in matric suction. The TDR data show that the initial wetting front from fall and spring precipitation events took 9 months to travel the 5 m depth to the base of the pile. A n increase in water content at 4.5 m depth immediately preceded the first lysimeter outflow. The relationship of measured water contents and outflow behaviour are discussed further in Chapter 6. Initial matric suction measurements were high due to intense evaporation from the pile surface at each instrument level during the time required to install each set of instruments. Matric suction measurements from the thermal dissipation sensors show declines at all levels. The matric suction values reported in this figure are based on raw measurements to demonstrate the initial trends. Further details of calibrations, hysteresis effects and temperature corrections are included in Chapter 5. 2.4 C O N C L U S I O N The CPE experiment was designed and built to provide a detailed, long-term data set of water flow and solute transport and geochemistry in a mine waste rock pile. The early experience presented here indicates the five meter high experiment required a wetting-up period of 11 months until the disturbances created by construction were overcome. This re-12 equilibration time implies short term results from large lysimeters installed to monitor net infiltration to waste rock piles cannot be relied upon. At least one full season would be required in the northern Saskatchewan climate to equilibrate such an experiment. 13 Cluff Lake F igure 2.1: Site of the constructed pile experiment. Location of the Cluff Lake Mine (Lat. 58 22', Long. 109 32'), owned and operated by Cogema Resources Incorporated of Saskatoon. 14 15 16 17 (%) lua^uoo 3jn;sjO|/\| oujawnioA (e<j>|) uoipons oujeifli 19 CHAPTER THREE: EVALUATION OF UNCOATED AND COATED TDR PROBES FOR HIGH ELECTRICAL CONDUCTIVITY SYSTEMS 3.1 ABSTRACT The effects of high sample electrical conductivity upon time domain reflectrometry (TDR) measurements are investigated. High sample electrical conductivity reduces the quality of a TDR waveform through the loss of signal amplitude. Two strategies to obtain higher signal-to-noise waveforms are examined: waveform differencing by remote diode shorting and covering probe conductors with resistive coatings. Experiments using electrically conductive water solutions show conventional dual-tangent line waveform analysis is accurate for solution electrical conductivities up to 5 dS m"1. Waveform differencing using manual short circuits is accurate, but the remote diode shorting method is systematically biased for three-rod probes in electrically conductive water solutions. The bias is related to the electrical properties of the diodes used. Three-rod Zegelin-type probes with a high resistance coating on the central rod are capable of returning analyzable waveforms for solutions with electrical conductivities at least as high as 70 dS m"1. Travel time in a coated probe determined by the remote diode shorting method was constant in solutions up to 5 dS m"1, but varied in an inconsistent manner at higher electrical conductivities. Travel time estimates using a coated probe and remote diode shorting in silica sand saturated with electrically conductive water solutions are less accurate in comparison to conventional waveform analysis. Current theoretical descriptions of the relationship between the apparent dielectric permittivity estimated using a coated probe to the actual sample apparent dielectric permittivity for three-rod probes do not lead to a simple 20 analytical expression to account for the presence of a coating. Samples of known dielectric permittivity were measured using a coated probe. An empirical fit of the experimental data is obtained using an equation of the form for a coated coaxial cell. A three rod coated probe with a single diode at the probe head is a practical means to collect interpretable waveforms in media with high electrical conductivity. 3.2 INTRODUCTION Soil water content can be measured in the field using gravimetric methods, neutron scattering, or techniques based upon the thermal or electrical properties of soil-air-water mixtures. Electrical methods are advantageous because they are easily automated to conduct unattended measurements with high sampling rates at multiple locations. Time domain reflectrometry (TDR) is a commonly used electrical technique. With knowledge of the electrical properties of soil-air-water mixtures, and appropriate experimental calibrations, the bulk soil dielectric permittivity can be related to the volumetric water content and the bulk soil electrical conductivity can be related to the soil water electrical conductance. While TDR methods have gained wide acceptance and usage in soils, relatively less attention has been given to TDR's application to high electrical conductivity materials such as mining waste. In these materials, conventional TDR probe designs and signal analysis techniques fail to provide sufficiently accurate estimates of the sample dielectric properties. This chapter describes an approach to design an automated TDR system for the measurement of water content in unsaturated mine waste rock. The TDR system is used in both a laboratory column experiment and a field-scale experiment (Nichol et al., 2000, Appendix A). In this application, a high soil water electrical conductivity (5 to 20 dS m"1) caused by oxidizing sulphide minerals in the mine waste lead to poor signal quality using conventional 21 TDR techniques. Sufficient signal amplitude could not be reliably obtained using a short probe of conventional design. Two techniques for obtaining improved signals were selected: waveform differencing using remote diode shorting (Hook and Livingstone, 1992), and the use of a resistive probe coating (Kelly et a l , 1995; Ferre et al., 1996, Mojid et al., 1998). The aim is to derive a method to reliably collect waveforms that can be interpreted by automated waveform fitting to derive an estimate of travel time and hence apparent dielectric permittivity. In this chapter, both techniques are tested for the collection of interpretable waveforms in high electrical conductivity media. A brief summary of the TDR method is provided which discusses how it is affected by bulk soil electrical conductivity. The relevant theory of waveform differencing by remote diode shorting and the use of probes covered with a high-resistance coating are described. A series of experiments are then presented. The first examines the performance of a short (160mm) uncoated probe using water solutions of variable electrical conductivity. The second experiments analyse the performance of the waveform differencing method using the remote diode shorting method and manual short circuits under the same conditions as the first experiments. This chapter then examines the performance of a coated probe under the same conditions for both conventional TDR waveform collection and analysis, and using the remote-diode shorting method. The performance of a coated probe in silica sand saturated with water solutions of varying electrical conductivity is presented. A last experiment demonstrates a practical means of compensating for the effect of a probe coating on the estimation of sample dielectric permittivity. 22 3.3 T H E O R Y TDR can be used to determine the apparent relative bulk dielectric permittivity (e a p p), the bulk electrical conductivity (o Dc), and in certain circumstances, the frequency-dependent real and imaginary parts of the complex sample dielectric permittivity (Hoekstra and Delaney, 1974, Dalton et al., 1984, Topp et al., 1988, Heimovaara et al., 1996, Friel and Or, 1999). TDR measures the propagation of a fast rise time, step voltage pulse through a coaxial cable to a waveguide (probe) in contact with the sample. Part of the incident pulse energy reflects back to the TDR instrument from each impedance change in the transmission line and probe, with a full reflection of the remaining pulse energy at the end of the probe. The sum of the incident voltage and reflected pulse voltage measured at the TDR instrument is plotted in time and presented as a waveform. The arrival of reflected energy at the TDR instrument causes a change in voltage over time, and thus a deflection of the waveform. The apparent velocity (v a p p) of the pulse along the probe is determined from the probe length (L) and the arrival times of the reflections from the head and end of the probe (t0, ti). The dielectric permittivity of the medium surrounding the probe is related to the velocity of an electromagnetic wave propagating in the transverse electric and magnetic mode (TEM) along a waveguide by: e = s app c vf/0 v V "PP J [3.1] where s 0 is the permittivity of free space and c is the velocity of light in a vacuum. It is assumed that the magnetic permeability of soils and soil-air-water solutions (u) equals that of 23 free space (p0) and therefore (p.0/|f) is unity (Topp et al., 1980). In this thesis, all dielectric permittivities are expressed relative to the permittivity of free space, and the term relative is taken as understood. Successful TDR measurement of the sample dielectric permittivity requires collecting a waveform in which the pulse travel time at the probe head and probe end can be accurately determined. In conventional waveform analysis, the timing of a reflection is determined using two lines fitted to the deflection of the waveform caused by the arrival of reflected pulse energy. The first is a tangent-line fit to the steepest slope of the rising or falling limb of the deflection. The second line is either a horizontal line fit to the pre-deflection waveform (flat-tangent method), or a tangent fit to the slope of the pre-deflection waveform (dual-tangent method) (Hoekstra and Delaney, 1974, Topp et al., 1982). The time of the intercept of the pre- and post- deflection lines is the arrival time of the reflected pulse energy. The dual-tangent method is more accurate for high electrical conductivity solutions (Wraith and Or, 1999) and is used as the reference method for the alternate methods examined. The dielectric permittivity estimated by tangent-line analysis of a TDR waveform is termed the apparent dielectric permittivity (e a p p) because the estimate of dielectric permittivity is not obtained for a single frequency. Frequency domain analyses of TDR waveforms using Fourier transform techniques (Hoekstra and Delaney, 1974, Heimovaara et al., 1996, Friel and Or, 1999) demonstrate that the incident TDR pulse generated by common TDR instruments contains energy in a range of frequencies up to 1.5 GHz. The deflection of the waveform is the integration of the arrivals of all of the frequencies reflected. Tangent-line methods of waveform analysis are biased towards the higher frequencies as these create the sharpest waveform deflections. The frequencies which dominate travel time estimates 24 from tangent-line waveform analysis are in the range of 700 M H z to 1.0 GHz. (Hoekstra and Delaney, 1974, Heimovaara et al., 1996, Friel and Or, 1999). 3.3.1 Effect of solution electrical conductivity Ions in solution affect both the waveform quality and the pulse velocity. DC conductance and ohmic losses due to current between the probe conductors dissipate the pulse energy within the waveguide. The energy reflecting from the end of the probe is reduced, the waveform deflection becomes less distinct, contributing to greater uncertainty in tangent-line fitting and travel time determination. In samples with high electrical conductivity, the amplitude of the signal reflected from the probe end may be fully dissipated within the exposed probe conductors and no travel time can be determined. The effect of electrical conductivity on the signal amplitude decreases with a shorter length of probe. The practical lower limit for probe length with common TDR instruments is 0.1 to 0.15 m (Kelly etal., 1995). Dissolved ions directly decrease the dielectric permittivity of a water solution by reducing the number of highly polarizable water molecules per unit volume and indirectly by the interaction of ions with the electrostatic bonding structure of liquid phase water molecules (Hasted, 1973). In contrast, the DC conductance resulting from ions in solution contributes to the imaginary component of the complex valued dielectric permittivity, leading to an increase in dielectric permittivity. However, at the frequencies which contribute most to tangent-line analysis of a TDR waveform, the DC conductance of water has negligible effect on the pulse velocity (Or and Wraith, 1999). Tabulated experimental measurements of E(co)soiution for common electrolytes are available in Hasted (1973) and Robinson and Stokes (1959). 25 The dielectric properties of the water phase in soil differs from that of water in a water solution. Both polar water molecules and dissolved ions interact with mineral surfaces. The dielectric and conductive properties of these diffuse, bonded layers are different from the rest of the water solution (Or and Wraith, 1999). Experiments in saturated and unsaturated soils have shown that increasing the concentration of dissolved ions in soil water increases the measured travel time when soil water content is constant (Wyseure et al., 1997; Sun et al., 2000; Topp et al., 2000). The measured increases in travel time are larger than can be explained by the direct effect of the increased bulk electrical conductivity on the real and imaginary parts of the complex dielectric permittivity (Topp et al., 2000). This indicates that increasing the concentration of dissolved ions in the soil water solution affects other components of the complex-valued dielectric permittivity of the soil/air/water/bound water/dissolved ions mixture in a manner that is not fully understood. 3.3.2 Waveform Differencing and Remote diode shorting Hook et al. (1992) describe a method to improve TDR signal quality using waveform differencing. A raw waveform is first collected from the probe. A short circuit then is created between the signal and ground conductors at the probe head. A second waveform is then collected, termed the shorted waveform. The incident pulse energy wil l leave the signal conductor and return to the TDR instrument along the ground conductor. This causes a sharp downwards deflection in the shorted waveform. The difference waveform is the difference between the raw waveform and shorted waveform. The timing of the deflection in the difference waveform is determined by tangent-line fitting to derive the travel time for the pulse through the short circuit and back to the instrument. The same process is then repeated with a short circuit created at the probe end. 26 Positive-intrinsic-negative diodes (PIN) can be used to create the short circuits remotely, and thus allow the waveform differencing method to be automated. PIN diodes have a low resistance under a forward-bias DC current and a high resistance under no current or a reverse-bias DC current. Hook et al. (1992) do not provide details or specifications for the diodes used. MPN3404 diodes were used, as suggested by the manufacturers of the MoisturePoint TDR system (Environmental Sensors Inc, Victoria, B.C.) which incorporates the remote diode method (Young, 1998a). The impedance of these diodes, measured at 100 MHz, switches from greater than 5MQ to less than 0.8Q under a forward bias current (On Semiconductor, 2000). Our experimental waste rock pile allowed for direct burial of the TDR probes on an open soil surface. This removed the requirement for an open-ended probe design that can be inserted into in-situ soil materials. The probe construction is based upon a three rod Zegelin-type probe design (Young, 1998a), with the inclusion of a single PIN diode at the probe head, and two PIN diodes at the probe base. A schematic of the probes is shown in Figure 3.1. The velocity of the TDR pulse within in the sample is determined from the difference between the time for the TDR pulse pass through the probe head diode, and the probe end diode. Calibration measurements in materials of known dielectric permittivity allow the measured time to be corrected for the distance between the diodes located within the probe head and end termination structures and the start of the exposed conductors. 3.3.3 Resistive Probe Coatings The ohmic loss of signal voltage between the signal conductor and the ground conductors can be reduced by introducing a high resistance coating (Kelly et al., 1995). In this work, a polyolefm heat shrink resistive coating is applied on the center rod (Livingstone, 27 1997, pers. comm., Mojid et a l , 1998). The energy of the travelling T E M wave wil l still extend outside the coating. The inclusion of an additional dielectric material complicates the relationship between the measured electrical properties and the electrical properties of the sample (Annan, 1977a, Knight et a l , 1997). The polyolefin heat shrink used has an apparent dielectric permittivity of approximately 3, lower than most soils materials, and therefore the estimate of apparent dielectric permittivity using a coated probe (ecp) wil l always be less than the apparent dielectric permittivity that would have been estimated using an uncoated probe ( S a p p ) -Three-rod or multi-rod probes are designed to emulate a coaxial cell (Zegelin et al., 1989). A simple analytical solution for the effect of a dielectric material of uniform thickness around the central conductor of a coaxial cell can be derived from Annan (1977a). Kcp = Ks K, In r: In + ^ l n | K. [3.2] K c p = Apparent dielectric determined from travel time in coated probe Kc = Dielectric constant of coating Ks = Dielectric constant of sample rc = Outer radius of coating rt = Outer radius of inner conductor r0 = Radius to outer conductor Equations have been derived which exactly describe the effects of coatings applied to three-rod or multi-rod probes, but these equations do not have analytical solutions and are 28 solved numerically (Zegelin et al.,1989; Knight et al., 1994, 1997). A simpler analytical solution is therefore tested. A l l probe designs with a fixed dielectric material within the sampling zone have a non-linear relationship between the apparent dielectric permittivity estimated directly from the measured travel time and the actual sample dielectric permittivity. This non-linear relationship means that the sample dielectric permittivity must be uniform along the probe length to properly determine the sample dielectric permittivity (Ferre et al., 1996). 3.4 M A T E R I A L S AND M E T H O D S The primary objective of this research is to obtain reliable TDR signals and therefore estimates of the apparent dielectric permittivity in samples with high electrical conductivity. A series of experiments were conducted to compare the results of traditional methods of TDR signal collection and analysis with alternative methods that have been proposed for increasing signal quality. The probes used for the experiments (Figure 3.1) are unbalanced, Zegelin-type three-conductor probes constructed using 3.2 mm diameter 316 stainless steel rods of varying length. Coated probes include a polyolefin heat shrink, carefully applied to the center conductor to ensure no air gaps between rod and coating. The average thickness was 0.40 mm. Probe designs are quoted in the remainder of the chapter as three numbers a/b/c where a is the length of exposed rods in mm, b is the rod diameter in mm, and c is the center rod to outer rod spacing in mm. Reference raw waveforms with 5000 points in time were collected using a Tektronics 1502C instrument (TTDR) connected to a PC running W A T TDR Version 3.11 (Waterloo Center for Groundwater Research, Waterloo, Ontario). A l l waveform analysis of TTDR waveforms was performed manually using the dual-tangent analysis method. A l l waveforms 29 are presented as the reflection coefficient displayed against the two-way travel time. This approach represents the currently accepted practice of TDR measurement against which the alternate methods are compared. Raw waveforms and diode-shorted waveforms with 256 points in time were collected using a MoisturePoint™ TDR (MTDR) instrument (Environmental Sensors Inc., Victoria, B.C., Canada) with firmware Version 1.27. The M T D R instrument was purchased with proprietary software and hardware specifically designed for conducting automated, diode-shorted TDR measurements. Travel times were measured using the automated waveform fitting routines which are included in the instrument firmware. Additional waveforms for presentation and for detailed waveform analysis were collected manually using a PC connected to the M T D R and running the instrument control software ViewPoint™ Version 1.34. For all measurements, all instrument settings and waveform fitting parameters in the instrument firmware and software were optimized for maximum accuracy and stability of tangent-line fitting. The M T D R instrument warm up time was set to 15 seconds and a minimum of 5 measurements were discarded prior to all recorded measurements to ensure the instrument returned a stable value of travel time. The first experiments determine the effect of solution electrical conductivity on conventional travel time analysis and on the performance of the remote diode shorting method. Waveform and travel time analysis was carried out for a 160/3.2/25 uncoated probe immersed in isothermal water of variable solution electrical conductivity. The probe was connected to the TTDR and M T D R using 6 metres of Commscope F660BEF 75f2 RG6 coaxial cable, a male F-connection termination, and an F-to-BNC adapter. Solution electrical 30 conductivity was altered from 0 to 5 dS m" using potassium chloride. Raw waveforms and probe head and end diode-shorted waveforms were manually collected for each solution. The effectiveness of the MPN3404 diodes as short circuits was assessed by taking waveform measurements using manually-created short circuits. For distilled water and for a solution electrical conductivity of 2 dS m"1, 6 gauge copper wire was used to create a short circuit at the start of the exposed rods and then at the end of the exposed rods. Shorted waveforms of each short circuit were collected using both TTDR and MTDR. The second set of experiments address the effect of a resistive coating; waveform and travel time analysis was carried out for a 281/3.2/25 coated probe using the same experimental setup as the uncoated probe. The electrical conductivity of the solution was altered from 1 to 70 dS m"1 using sodium chloride. The measurements for 30 to 70.8 dS m"1 solutions have some conductive amplitude loss of signal, and different overall probe impedance due to failure of the probe head seal and invasion of electrically conductive water into the probe head. Conclusions can still be drawn from these results and they are included in the following analysis. No manually created short circuits were tested in these probes due to the presence of the probe coating. The performance of a 281/3.2/25 coated probe and remote diode method was then assessed for silica sand saturated with electrically conductive water solutions. Water electrical conductivity was altered from 0 to 20 dS m"1 using sodium chloride. Raw and diode shorted waveforms were collected using the MTDR. Travel time was determined using automated remote diode shorting at the probe head and end. Water solution and bulk soil electrical conductivity were measured independently using a four electrode conductivity cell and conductivity meter. 31 The effect of the resistive coating on the estimation of sample dielectric permittivity was determined in the final set of experiments. Travel time measurements were made using a coated probe and materials of known dielectric with low electrical conductivity, the latter determined using a calibrated 295/3.2/25 uncoated probe. Measurements were conducted using the M T D R instrument and averaging a minimum of 15 measurements. For the purposes of these experiments, the best dielectric materials are liquids, which allow simple probe immersion, uniformity of sample and a controlled range of dielectric properties. Organic liquids can be used to obtain a range of dielectric properties. However, the 281/3.2/25 coated probe used in this study was required for other calibration work and the compatibility of the probe construction materials with various organic liquids was not known. To prevent damage to this probe, a more limited range of inert materials was used. Air, oven-dried silica sand, wetted silica sand, methanol and water were used as dielectric media. Liquids were placed in a 15.2 cm diameter P V C container. Granular materials were packed within a 12 cm x 12 cm x 63 cm watertight container. 3.5 R E S U L T S A N D DISCUSSIONS 3.5.1 Uncoated Probes The conventional method of TDR raw waveform analysis is first demonstrated for an uncoated probe in water. Raw waveforms are presented in Figure 3.2 for solution electrical conductivities from 0 to 5 dS m"1. The reflection from the probe head is virtually identical for all solution electrical conductivities. The negative slope on the probe segment of the waveform decreases and the probe end reflection shows a decrease in the final amplitude as solution electrical conductivity increases. Manually-determined tangent-lines to the end reflections are presented for 0 and 2 dS m"1. The signal quality is significantly degraded at 32 5.0 dS m"1 making manual waveform fitting challenging. In the range of 0 to 5.0 dS m"1, no changes in travel time greater than the method uncertainty were detected. By definition, the transition time of a reflection is half the time for the signal to rise from the pre-reflection tangent to the maximum amplitude and is related to the highest frequencies contained within the reflected signal (Hook and Livingstone, 1992). The transition time of the probe end reflection was consistently 1.2 nanoseconds (ns) from 0 to 5 dS m"1. These water solution electrical conductivities do not contribute to either increased transition time or decreased amplitude of the high frequency components of the TDR pulse. Signals obtained using remote diode shorting are shown in Figure 3.3. The gain- and offset-corrected M T D R raw waveforms, diode-shorted waveforms, and diode difference waveforms are shown for distilled water (panels A , B) and a solution electrical conductivity of 2 dS m"1 (panels C, D). Waveforms are separated for when the diode is located at the probe head (panels A , C) and at the probe base (panels B , D). The probe head and probe end diode-shorted waveforms for distilled water both drop rapidly due to the short circuit created by the diodes. The resulting diode difference waveforms have distinct signals, with sharp waveform rises and short transition times, making tangent-line fitting both simple and accurate. Note that the vertical position of the difference waveforms in this figure, and following figures, has been shifted vertically for greater clarity of presentation. However, at 2 dS m"1, both the probe head (panel C) and probe end (panel D) diode-shorted waveforms more closely follow the raw waveform. As a result, the diode-difference waveforms demonstrate reduced total amplitude, increased time to maximum amplitude, and a less distinct shape than those in distilled water. The 2 dS m"1 probe head and probe end diode-33 difference waveforms are more difficult to analyse by tangent-line fitting than the raw waveforms presented in Figure 3.2. The performance of the MPN3404 diodes can be understood by examining the data for near-perfect short circuits created with copper wire. Figure 3.4 presents the waveforms obtained using manually created short circuits in distilled water (panels A , B) and in a 2 dS m"1 water solution (panels C, D) for the probe head (panels A , C) and probe end (panels B , D). The shorted waveforms at the probe head and end are similar for both solution electrical conductivities. The difference waveforms have different amplitudes due to the difference in amplitude of the raw waveform, but similar transition times. The poor signal quality of the difference waveforms in Figure 3.3 is therefore not the result of the waveform differencing method itself, but results from the MPN3404 diodes not creating adequate short circuits. At the probe head, the multi-frequency incident TDR pulse encounters two possible transmission paths when a short circuited is present. Wave energy wil l be reflected and transmitted along the waveguide or into the short circuit in proportions determined by the frequency dependent complex impedance of each path. The impedance of the manual short circuit is near zero, and thus all wave energy diverts from the signal conductor to the ground conductor. After 85 ns, the manually shorted waveforms in Figure 3.4A and 3.4C are both are flat, illustrating that there are no reflections present in the shorted waveform from any energy that passed the short circuit and entered the probe. In contrast, the pathway created by the remote diode short has a low, but finite impedance. Pulse energy does divide between the diode pathway and the probe pathway when it encounters the two impedances in parallel. The diode-shorted waveform collected at 2 dS m"1 at the probe head closely follows the raw waveform (Figure 3.3C), indicating that 34 significant signal energy continued past the location of the diode in the probe head, entered the probe and returned reflections from impedance changes within the probe. The transition time of the probe-end reflections in the diode-shorted and the raw waveforms indicates that the frequency spectra of the energy pulse transmitted into the probe was similar. The difference between the diode-shorted waveform and the raw waveform represents the energy that returned via the diode. The gradual departure of the diode-shorted waveform from the raw waveform indicates the energy passing through the diode was dominated by lower frequencies. The probe impedance at low frequency (Zoo) was calculated from the reflection coefficient at long times using the method of Mallants et al. (1996). The probe impedance is 4000 Q in distilled water and 20 ft in 2 dS m"1 water. The MPN3404 PIN diodes used have a rated impedance of 0.8ft at 100 MHz. It is clear from Figure 3.3 that the MPN3404 diode and the impedance of the probe design used are more similar at higher frequencies, and thus shorting becomes less effective. At the probe end, a similar division of pulse energy occurs as at the probe head. Connecting diodes across the end of the probe conductors requires the presence of a probe termination structure, similar to the probe head, to house the diodes. Energy is reflected and transmitted at both the end of the exposed waveguides, and again at the diodes, which connect the end of the center conductor to ground within the probe end termination structure. At 2 dSm"1 (Figure 3.3D), the higher final amplitude of the diode-shorted waveform than in distilled water (Figure 3.3B) indicates a greater proportion of the energy was reflected back along the signal conductor, instead of through the diode short. In high electrical conductivity systems, the combination of the effects of imperfect diode shorting at the probe head and end leads to large errors in travel-time estimation, and 35 hence water content. The automated, instrument-determined travel time for varying solution electrical conductivity is shown in Figure 3.5, with the theoretical travel time calculated from Equation 3.1, and the travel time estimated from the manually created shorts in Figure 3.4. The travel time estimated using manually created shorts does not include travel time within the probe head and end termination structures from the diodes to the ends of the exposed conductors. The automated determination of travel time increased by 6 ns. Remote-diode shorting hinders travel time estimation under the conditions examined. In contrast, manually fitted tangent-line analysis of the manual short circuit data presented in Figure 3.4 determined no change in travel time greater than the uncertainty in the analysis, similar to the results of manual tangent-line fitting to the raw waveforms in Figure 3.2. The method of waveform differencing does not significantly extend the range of waveform analysis to media with higher electrical conductivities. At the probe end, waveform differencing by either manual short circuits, or diode short circuits relies upon the collection of a raw waveform with non-zero reflected signal amplitude. This method will therefore only be successful up to similar solution electrical conductivities as conventional raw waveform analysis, or approximately 5 dS m"1 for a 0.16 m probe. 3.5.2 Coated Probes in Water The performance of coated probes is first assessed using conventional waveform analysis. Raw waveforms for solution electrical conductivities between 1 to 70.8 dS m"1 collected using the TTDR are shown in Figure 3.6. The raw waveforms from 1 to 20 dS m"1 indicate little signal amplitude loss and a consistent trend in probe impedance. The transition time of the probe head reflections is similar for 1 to 70.8 dS m"1. No attempt was made to precisely match the impedances of the cable, connectors and probe head. Multiple 36 reflections are seen between 67 and 73 ns caused by these impedance changes and the start of the exposed conductors. The accidental penetration of high electrical conductance water into the probe head causes the sharp drop in probe-head impedance observed in the lowest three waveforms. The multiple internal reflections originating from within the probe base are smoothed out at higher electrical conductivities. The dual-tangent method of waveform analysis was assessed by manual tangent-line fitting. The better preservation of probe head internal reflections in coated probes does not affect manual tangent-line fitting, but would make automated tangent-line analysis of the probe head reflection more difficult. The signal from the end reflection of a coated probe has lower transition times than the signal from the uncoated probe, more signal amplitude and therefore tangent-lines are easier to fit to the waveform. Manually-determined tangent-lines to the probe end reflection are presented on Figure 3.6. Between 1 dS m"1 and 70.8 dS m"1 the travel time determined by the tangent-line method decreases by 0.4 ns. This decrease in travel time will be discussed in final section. The signal attenuation encountered in uncoated probes, caused by high sample electrical conductivity, is reduced in coated probes and raw waveforms are interpretable to >70 dS m"1. The waveform differencing method was assessed for coated probes using only the remote diode shorting method as no manual short circuits could be created due to the presence of the probe coating. The diode waveforms collected at 1 (panels A , B) and 20 dS m"1 (panels C, D) are presented in Figure 3.7, with the probe head waveforms on the left (panels A , C) and the probe end waveforms on the right (panels B , D). In contrast to the uncoated probes, the overall probe impedance calculated from the final reflection coefficient remains high throughout the range of solution electrical conductivities presented. The probe 37 impedance at low frequency changes from 6000 Q. to 1000 Q from distilled water to 20 dS m"1. The impedance contrast between the shorted diode and the probe remains high and therefore the diode acts as a better short circuit than in uncoated probes. The probe head diode waveforms indicate negligible changes in the slope and intercept of the diode waveforms between 1 and 20 dS m"1. The probe head diode-difference waveforms eliminate internal reflections within the probe head, and are smoother and more easily interpreted using tangent-line analysis than the raw waveforms in Figure 3.6. The diode-shorted waveform for the probe end diode in a 1 dS m"1 solution shows a very small reflection from the conductor to probe end transition; none is visible at 20 dS m"1. The probe end diode-shorted waveform at 20 dS m"1 has a 4 nS longer transition time than the 1 dS m"1 waveform due to frequency-dependent transmission by the diode and frequency filtering by the solution. The automated measurements of travel time using remote diode shorting from 1 to 70 dS m"1 are presented in Figure 3.8. The travel time calculated using the tangent-line method remains constant up to 5 dS m"1. Between 5 dS m"!and 20 dS m"1 the travel time rises by 0.5 ns, beyond which it decreases. The combined effects of the longer diode waveform transition times and the raw waveform smoothing contribute to the inconsistent estimation of travel time. The decrease after 20 dS m"1 may be the result of the invasion of water into the probe head. The probe coating therefore allows collection of a waveform interpretable by travel time analysis in samples with electrical conductivities that are much higher than uncoated probes. Waveform differencing improves accuracy of travel time determination at the probe head, even with the MPN3404 diodes. The success of the waveform differencing method with manual short circuits in uncoated probes implies that either manual short circuits or perfect remote diode short circuits created would be similarly 38 successful in a coated probe, but the well preserved end reflection removes the necessity for waveform differencing at the probe end. 3.5.3 Coated Probes in Saturated Sand The first experiments presented used electrically conductive water solutions to isolate the effects of DC conductance on signal quality from any other effects arising from the interaction of water and soil particles. The data collected in a saturated silica sand are now examined to determine the quality of the waveforms collected by a coated probe using a sample material closer to the expected field conditions. Figure 3.9 presents waveforms collected in silica sand saturated with distilled water (panels A , B), and saturated with a water solution with an electrical conductivity of 19.6 dS m"1 (panels C, D). Waveforms are displayed for the probe head (panels A, C) and probe end (panels B, D). When the saturating solution was 19.6 dS m"1, the bulk electrical conductivity of the saturated silica sand, measured with a four electrode conductivity probe, was 8.3 dS m"1. The diode-difference waveforms at the probe head are good quality, and again, probe head travel time is more easily estimated from the difference waveform than the raw waveform. The difference waveforms have the same rise time and waveform fitting derives the same travel time. At the probe end, the waveforms in the silica sand saturated with 19.6 dS m"1 water (panel D) are significantly more rounded, with longer transition times. The transition time of the probe end reflection is the same as in 20 dS m"1 water (Figure 3.7D), but the final amplitude is actually lower. Figure 3.10 presents the two way travel times estimated from the waveforms in Figure 3.9 using automated remote diode shorting at both the probe head and end (A) and manual dual-tangent fitting to the raw waveforms (B, open circles). The measured travel 39 times for both the remote diode method, and the manual method both increase, but the remote diode method overestimates the increase. Curve C presents the travel time measured in water of the same bulk electrical conductivity, taken from Figure 3.8. This indicates that the increase in travel time determined by curves A and B is not directly the result of increased electrical conductivity of the water solution, but arises from the interaction of that water solution with the mineral particles. The measured apparent dielectric permittivity increases in both the manual and remote diode waveform analysis methods and it can be therefore inferred this is a real increase in apparent dielectric permittivity similar to observations by Wyseure et al. (1997), Sun et al. (2000) and Topp et al. (2000). As discussed in the theory section, the apparent dielectric permittivity is not an actual physical property of the sample media , but is an average measurement over a non-specific range of frequencies defined by the tangent-line method of waveform analysis. An increase in apparent dielectric permittivity may therefore be the result of either changes in the frequency range used to define the apparent dielectric permittivity or actual changes in the dielectric properties of the media. It is clear from the waveform smoothing in Figure 3.7 and 3.9 that changes in the frequency content of the TDR pulse reflections are occurring with increasing electrical conductivity. Changes in the actual dielectric permittivity of the soil-water mixture may also be occurring. Detailed investigation of changes in frequency content of the TDR pulse, and changes in the dielectric permittivity of soil/water/air/bound water/dissolved ion mixtures are best determined using frequency domain techniques where it is possible to determine both the real and imaginary parts of the complex dielectric permittivity across all the frequencies in the TDR pulse (Hoekstra and Delaney, 1974; Heimovaara et al., 1996; Friel and Or, 1999)'. This 40 type of analysis, using both uncoated and coated probes, would increase our understanding of the effects of high concentrations of dissolved ions on dielectric permittivity, TDR signal frequency content and on the apparent dielectric permittivity derived from tangent-line analysis. These methods are not currently suited to automation in a field situation, and therefore this study has focussed the analysis on conventional travel time measurements. The primary concern is with signal quality, and obtaining a waveform interpretable by automated waveform fitting. The results obtained in saturated silica sand indicate that a coated probe is sufficient to determine a measurement of travel time and hence apparent dielectric permittivity, but that apparent dielectric permittivity alone is not sufficient to determine the water content of a soil containing a high electrical conductivity water solution. Calibration of the apparent dielectric permittivity to water content for mine waste material is presented further in Chapter 3. 3.5.4 Effect of probe coating The apparent dielectric permittivity estimated using a coated probe must be related to the apparent dielectric permittivity of the soil to accurately estimate water contents. The third set of experiments determined the measured travel time for both a coated probe and an uncoated probe in materials with different dielectric permittivity. Apparent dielectric permittivity for coated and uncoated probes were estimated from these measured travel times using Equation 3.1. The apparent dielectric permittivity (ecp) derived from a coated probe is plotted against the apparent dielectric permittivity (e a p p) determined with an uncoated probe in Figure 3.11. 41 The dielectric permittivity of the probe coating can be estimated from where the apparent dielectric permittivity from the coated probe and the actual apparent dielectric permittivity coincide. This is estimated to be at a dielectric permittivity of 2.8. Three fitted curves are shown. Curve A represents a two-point linear correction for the presence of a coating material using the air and water measurements as suggested by Young (1998b). The maximum overestimation of the apparent dielectric permittivity using curve A occurs at a dielectric permittivity of 30, which corresponds to the apparent dielectric permittivity of saturated silica sand. The estimate of apparent soil dielectric permittivity using curve A would be double the true value. Curve B is derived from Equation 3.2 using the known probe dimensions and the estimated probe coating dielectric permittivity. Experiments by Friel and Or (1999) demonstrate that three-rod probes may emulate coaxial cells up to frequencies of 0.7 GHz. Equation 3.2 may be appropriate up to that frequency, but based upon the poor fit between the measured data and curve B, it is not applicable at the 0.7 to 1GHz that is estimated to dominate the travel time analysis of TDR waveforms. Curve C is the best-fit equation to the measured data obtained by modifying Equation 3.2. The rod and coating dimensions are retained as the known values, but r 0 is used as a fitting parameter. Using the form of Equation 3.2, the correct curve shape is captured while using a relatively simple fitting equation. It is now possible fo assess the accuracy of a coated probe for the determination of the apparent dielectric permittivity of a high electrical conductivity sample. A 0.4 ns change in the measured travel time was noted in the previous discussion of the performance of coated probes in solutions between 1 and 70 dS m"1 (Figure 3.6). The highest solution electrical conductance, 70.8 dS m"1, corresponds to a sodium chloride concentration of 0.77 mol/L 42 (Mojid et al., 1998). Using the data tabulated in Hasted (1973), the dielectric constant would be expected to decrease from 80.7 to 69. Using Equation 3.1, the two-way travel time in an uncoated probe of the same length as the coated probe should decrease by 1.3 ns. The actual decrease in measured travel time for the coated probe is affected by the probe coating. Using the best fit curve (Figure 3.9, Curve C), this calculated change of dielectric permittivity would correspond to a change in measured apparent dielectric permittivity from 29.3 to 27.3 and a travel time difference of 0.38 ns. This corresponds to the measured 0.4 ns change in travel time between 1 and 70 dS m"1. The coated probes are therefore sensitive enough to determine small changes in the dielectric permittivity. 3.5.5 Recommendations The method of waveform differencing for determination of the travel time is accurate for high electrical conductivity samples (Figure 3.4) when the waveguides are perfectly shorted. As this method requires a raw waveform with an end reflection, it has the same upper limit of solution electrical conductivity as does the interpretation of raw waveforms. The remote diode method of creating short circuits using the MPN3404 diodes recommended by Young (1998) is inaccurate for media with high electrical conductivities. The MPN3404 diodes have frequency-dependent impedances that are incompatible with the probes design used. The results obtained for the remote diode method may be improved by the use of diodes other than the MPN3404 that have lower impedance characteristics at frequencies of 700 M H z to 1GHz. Alternate diodes have not been investigated , but can recommend that any diodes selected can be easily checked for performance by measuring diode waveforms of an uncoated probe in an electrically conductive water solution. The presence of significant 43 signal amplitude in the diode-shorted waveform is an indication of unsuitable diode characteristics. Measurement of apparent dielectric permittivity in high electrical conductivity systems requires the use of a probe coating to obtain an interpretable waveform. The three-rod Zeglin type probe with a coating on the center rod obtained interpretable waveforms in both water solutions and saturated silica sand. The accuracy of the probe head measurements for coated probes (Figure 3.7) demonstrates that probe head travel time for coated probes is more easily determined using remote diode shorting, even with the MPN3404 diode. The accuracy of locating the probe end reflection using conventional waveform analysis (Figure 3.6) indicates that probe base travel time should be determined from the raw waveform using a dual-tangent method. No probe end diodes or terminal block (Figure 3.1) are therefore required. A n empirical correction (Figure 3.9) for the probe coating should be derived by TDR measurements in liquids of known dielectric properties. The calibration media should have dielectric permittivities close to the range of expected dielectric permittivities in the soil and calibration of individual probes is required. The use of a resistive coating permits the determination of travel time and hence apparent dielectric permittivity in high electrical conductivity systems. Further work using frequency domain techniques is required to investigate the changes in apparent dielectric permittivity with both soil water content and soil water electrical conductivity. 44 1. RG6 coaxial cable: F660 BEF CommScope 75 ohm cable 2. Female F-connector 3. Male F-Connector to plug 4. Polyethylene terminal block 5. Outer rod: unshielded 3.2 mm dia. 316 stainless steel. 6. Center rod: uncoated 3.2 mm dia. 316 stainless steel. Optionally,coated with 0.4 mm thick polyolefin heat shrink 7. Motorola MPN3404 PIN diode 8. Terminal block sealed with silicon sealant Figure 3.1: Schematic of TDR probe design. 45 Figure 3.2: Raw time domain reflectrometry waveforms with manually determined tangent-lines to the probe end reflection: 160 mm length uncoated probe. 46 Two Way Travel Time (ns) R- Raw S- Diode-Shorted D- Diode-Difference Figure 3.3: Remote diode shorting method waveforms for diodes located: (AB) at the probe head and (CD) at the probe base: 160 mm length uncoated probe in water solutions of varying conductivity 47 Probe Head Distilled Water Probe End Distilled Water] Probe Head 2 d S m 1 Water Solution Probe End 2 d S m 1 Water Solution 1.6 1.4 S 1 . 2 ~ 1 . 0 1 0 . 8 | . 0 .61 0.4 "J 0.2 £ o.o-S 1.6 1.4 1.2 1.0 0.8 0.6 0.4 0.2 0.0 0) T3 3 "5. E < a> .> re a> 90 100 75 80 85 90 95 100 70 80 Two Way Travel Time (ns) R- Raw S- Manual-Shorted D- Difference 110 120 Figure 3.4: Manual probe shorting method waveforms for short circuits using copper wire located: (AB) at the probe head end of the exposed conductors and (CD) at the probe base end of the exposed conductors: 160 mm length uncoated probe in water solutions of varying conductivity 48 20-18-16-14-12-10; 8-Manual short and waveform difference: Measured Theoretical 1 2 3 4 5 Water Solution Electrical Conductivity (dS m1) Figure 3.5: Automated remote diode shorted method measured two way travel time, travel time measured using manual probe shorting and waveform differencing and travel time calculated from Hasted (1973) data: 160 mm length uncoated probe in water solutions of varying electrical conductivity. 49 Two Way Travel Time (ns) Figure 3.6: Raw time domain reflectrometry waveforms with manually determined tangent lines to the probe end reflection: 281 mm length coated probe in water solutions of varying electrical conductivity 50 I I ' ' I 1 , 1 , I l l 70 80 90 100 110 70 80 90 100 110 120 Two Way Travel Time (ns) R- Raw S- D iode-Shor ted D-Difference F igure 3.7: Remote diode shorting method waveforms for diodes located: (AB) at the probe head and (CD) at the probe base: 281 mm length coated probe in water solutions of varying electrical conductivity. 51 11.0 1 10.5-> ns 10.0-I : o £ 9.5-9.0-0 10 20 30 40 50 60 70 Water So lut ion E lectr ica l Conduc t i v i t y (dS m 1 ) F igure 3.8: Automated remote diode shorted method measured two way travel time: 281 mm length coated probe in water solutions of varying electrical conductivity 52 70 80 90 100 110 70 80 90 100 110 Two Way Travel Time (ns) R- Raw S-Diode-Shorted D- Diode-Difference Figure 3.9: Remote diode shorting method waveforms for diodes located: (AB) at the probe head and (CD) at the probe base: 281 mm length coated probe in silica sand saturated with water solutions of varying electrical conductivity 5 3 12 a> 6 T 1 1 1 1 1 1 i i 1 1 0 2 4 6 8 10 Sample electrical conductivity (dS m'1) Figure 3.10: Automated remote diode shorted method measured two way travel time: 281 mm length coated probe in silica sand saturated with water solutions of varying electrical conductivity. 54 Apparent Dielectric Permittivity Figure 3.11: Relationship of coated probe measured apparent dielectric permittivity to apparent dielectric permittivity. A) Linear correction B) Equation 3.2 using known probe dimensions: r0= 12.5 mm., Kc=2.8 C) Equation 3.2 using r 0as fitting parameter: r0= 53.5 mm., Kc=2.8 55 CHAPTER 4: TIME DOMAIN REFLECTROMETRY MEASUREMENTS OF WATER CONTENT IN COARSE WASTE ROCK 4.1 ABSTRACT The TDR method is calibrated and applied to determine the water content of mine waste rock in a large scale field experiment. High electrical conductivity of the soil water in waste rock reduces the quality of a TDR waveform through the loss of signal amplitude. Two strategies are used to obtain higher signal-to-noise waveforms: waveform differencing by remote diode shorting and covering probe conductors with a resistive coating. The TDR pulse travel-time measured in soils containing electrically conductive water solutions increases systematically, partly due to bias in the remote diode and tangent-line measurement methods. This bias affects the experimental calibration data, and is larger than the variations in the calibration data caused by variations in probe construction, grain size, or cable length. Two empirical calibrations are derived for waste rock with low and high electrical conductivity soil water, respectively. An ambient temperature correction is derived from measured field data based upon the diurnal fluctuation of measured travel time with temperature. The variation of apparent dielectric permittivity with temperature is positively correlated with temperature at low water content, and negatively correlated at high water content. This trend indicates a significant influence of water bound to mineral surfaces on the variation of apparent dielectric permittivity with temperature. Examination of the field data indicates the effect of electrical conductivity of the soil water on the accuracy of the remote diode method is larger than the ambient temperature effect. 56 4.2 INTRODUCTION Long-term measurements have been made of the water content of mine waste rock in a large scale field experiment (Nichol et al., 2000, Appendix A). Automated measurements of soil water content are most easily achieved using methods based upon the electrical properties of soils, with time domain reflectrometry (TDR) being the most common method. TDR measures the propagation of a fast rise time, step voltage pulse through a coaxial cable to a waveguide (probe) composed of parallel conductors in contact with the sample. TDR can be used to determine the dielectric properties of the sample and the bulk soil electrical conductivity. With appropriate calibrations, the bulk soil dielectric permittivity can be related to the soil water content, and the bulk soil electrical conductivity can be related to the electrical conductivity of the soil water solution. The measurement of water content in coarse, heterogeneous mine waste rock presents a challenge when using the TDR method. Oxidation of sulphide minerals within waste rock leads to soil water with high electrical conductivity. This can lead to high signal attenuation, and subsequent difficulty in the estimation of soil electrical properties. Several strategies have been proposed to overcome this limitation, including using short probes (Kelly et al., 1995), resistive probe coatings (Knight et a l , 1997, Mojid et al., 1998), and waveform differencing using remote diode shorting (Hook et al., 1992). A review of the success of these strategies is presented in Chapter 3. In this chapter, work undertaken to relate the measured electrical properties of mine waste rock to the volumetric water content is presented. These experiments demonstrate that while TDR measurement of water content is well accepted in theory, its practical application to difficult soil material can be challenging. This required three issues to be addressed. 57 Firstly, the effectiveness of the probe design deployed in the constructed pile experiment (CPE) is determined. This design differs from the design recommended in Chapter 3. Two potential sources of error are highlighted: design deficiencies in the probe head and end connections; and a systematic bias of the remote-diode method used in soils with high electrical conductivity soil water solutions. Secondly, laboratory and field calibrations are undertaken to relate the electrical properties of mine waste rock to the water content. Thirdly, the effects of ambient temperature upon the measured electrical properties are corrected for using an algorithm derived from two and a half years of field data. The processed field data are then presented to demonstrate the relative effects of probe design errors, calibration variation, and the effect of ambient temperature on the final calculated values of water content, and the overall performance of the TDR measurement system. 4.3 THEORY 4.3.1 Dielectric Properties of Soil-Air-Water Mixtures The measurement of soil water content using electrical methods is related to the change in the dielectric permittivity of soil-air-water mixtures as a result of the variation in the proportion of water. The complex dielectric permittivity of a substance, s(co), is dependant upon the angular frequency (co) of the electrical field. It has a real component, e'(co), which expresses recoverable energy storage, and an out-of-phase imaginary component, e"(co), which describes both ohmic losses arising from the electrical conductivity (ODC) and polarization losses, sp(co). The dielectric permittivity is (Topp et al., 1980): 58 e(a>) = s'{co) - ie"(co) [4.1] e"{ai) = ep{fo) + ^ - [4-2] <» The dielectric constant (K) is defined as the static dielectric permittivity relative to the permittivity of free space (e'(0)/ s0). The dielectric constant is often used, sometimes inappropriately, in place of e(co), as it allows the dielectric properties to be expressed as a single value. The dielectric constant of soil changes as water (K~80) replaces air (K~l ) in the pore space of dry soil (K~3 to 5). The relationship of the soil dielectric permittivity to the soil water content can be described using either a theoretical mixing model, or by direct calibration. Various theoretical mixing models have been proposed to relate the electrical properties of soil-air-water mixtures to the electrical properties of the individual components (Birchak et al., 1974, Dobson et al., 1985; Roth et al., 1990; Herkelrath et al., 1991; Dirksen and Dasberg, 1993; Heimovaara et al., 1994, Ferre et al., 1996, Y u et al., 1997). However, the dielectric permittivity of water within a soil environment may not be the same as that of water as a free phase. Water close to mineral surfaces, especially clays, is bound by electrostatic forces. Similarly, the presence of high concentrations of dissolved ions in water constrains liquid water molecules by partly binding them to the dissolved ions. (Robinson and Stokes, 1955; Hasted, 1973). The dielectric and electrical conductance properties of bound water are therefore different from those of free-phase water (Hoekstra and Delaney, 1974, Topp et al., 1980; Herkelrath et al., 1991; Dirksen and Dasberg, 1993; Or and Wraith, 1999). Several studies have attempted to relate the amount and influence of bound water to theoretical 59 descriptions of the water bonding processes, and measurements of clay mineralogy, surface area or hygroscopic water content (Dirksen and Dasberg, 1993; Or and Wraith, 1999). Work is progressing, but no theoretical description or experimental correlation exists to reliably account for bound water independent of actual measurements of the soil dielectric permittivity. A significant silt and clay fraction (10% of the <600mm fraction or 20% of the <5mm fraction) is present in the CPE waste rock, in addition to high total dissolved solids in the soil water. It was therefore assumed that a fraction of bound water would be present, and work proceeded directly to an empirical calibration between measurements of the apparent dielectric permittivity of the waste rock material, and the volumetric water content of the waste rock. 4.3.2 Measurement of Dielectric Permittivity using Time Domain Reflectrometry A TDR instrument propagates a fast rise time, step voltage pulse through a coaxial cable to the probe. Part of the incident pulse energy reflects back to the TDR instrument from each impedance change in the transmission line and probe, and a full reflection of the remaining pulse energy occurs at the end of the probe. The sum of the incident voltage and reflected pulse voltage measured at the TDR instrument is plotted in time and presented as a waveform. The arrival of reflected energy back at the TDR instrument causes a change in the measured voltage over time, and thus a deflection of the waveform. Figure 4.1 illustrates a representative waveform for a TDR probe immersed in an electrically conductive water solution. The simplest analysis of a TDR waveform is to determine the times that the waveform is deflected by the arrival of energy from the probe head and probe end reflections (to, ti) and use the difference to estimate an apparent pulse velocity within the probe (v a p p): 60 where L is the length of the probe. Two forms of line-fitting for wave form analysis are commonly used to determine the locations in time of the reflections within the TDR waveform (Hoekstra and Delaney, 1974, Topp et al., 1980, Wraith and Or, 1999). One line is fit to the pre-deflection waveform and this can either be a horizontal line (flat-tangent method) fit to the lowest point on the pre-deflection waveform, or a tangent to the pre-deflection waveform (dual-tangent method). The second line in both methods is a tangent fit to the falling (probe head) or rising limb (probe end) of the deflection. The time of the intercept of the pre-deflection and post-deflection fitted lines is the arrival time of the reflected pulse energy. A comparison of the flat-tangent and the dual-tangent methods by Wraith and Or (1999) indicated the dual-tangent method is more accurate for high electrical conductivity samples. The dielectric permittivity of the medium surrounding the probe is related to the velocity by: £app £c V V "PP J MJ [4.4] where e0 is the permittivity of free space and c is the velocity of light in a vacuum. It is assumed that the magnetic permeability of soils and soil-air-water solutions (u) equals that of free space (p0) and therefore (p0/|f) is unity (Topp et a l , 1980). 61 The dielectric property calculated by interpreting the arrival times of reflections in the waveform by tangent-line analysis is termed the apparent dielectric permittivity, (s a p p), not the dielectric permittivity, s(to), because the apparent dielectric permittivity is not derived for a single frequency. The TDR equipment deployed at the CPE was not capable of automated recording of full waveforms with sufficient detail to permit the interpretation of dielectric properties in the frequency domain. A l l the TDR interpretations presented are therefore based upon travel time analysis and are measurements of apparent dielectric permittivity only. Tangent-line methods of waveform analysis to determine travel time may be impossible to apply in systems with high electrical conductivity. Conduction of energy between the central signal conductor and the outer ground conductors leads to signal voltage loss, a weak reflection from the probe end, and ultimately a waveform on which tangent-lines cannot be accurately fitted. This is demonstrated in Figure 4.1, where the waveform descends between t0 and ti, and the end reflection after ti is small. The conductive loss of signal between the signal and ground conductors can be reduced by the use of resistive coatings on the conductors (Knight et al., 1997, Chapter 3). The signal energy is retained, and the amplitude of the reflection from the end of the probes is increased. Waveform differencing using remotely shorted diodes has also been proposed as a method to improve waveform interpretability and obtain improved estimates of the arrival times (t0, ti) (Hook et al., 1992). The TDR pulse is selectively diverted from the signal conductor to the ground conductors by positive-intrinsic-negative (PIN) diodes. Travel times to the probe head (t0 ) and end (ti) are estimated from the arrival of the diverted signal 62 returning along the ground conductor. Hook et al. (1992) and Chapter 3 provide further discussion of the remote diode and waveform differencing methods. 4.3.3 C P E Probe Design The probes that were installed in the CPE (Figure 4.2) are based upon a probe design suggested by Young, 1998a, for use with the MoisturePoint ™ TDR (MTDR) instrument (Environmental Sensors Inc., Victoria, B.C., Canada). The probes are Zegelin-type, three-conductor probes constructed using 3.2 mm diameter 316 stainless steel rods (Zegelin et al., 1989). The coated probes include a 0.4 mm thick polyolefm heat shrink on the center conductor. Probe designs are quoted in the remainder of the chapter as three numbers a/b/c, where a is the length of exposed rods in mm, b is the rod diameter in mm, and c is the center rod to outer rod spacing in mm. The probe design includes a single positive-intrinsic-negative (PIN) diode (On Semiconductor, MPN3404) at the probe head, and two PIN diodes at the probe end. The presence of diodes at the probe end required the use of a terminal block, a probe termination structure at the probe end, which is not normally included in a TDR probe. The TDR probes were manually buried within the waste rock pile as it was being constructed and thus the terminal block did not interfere with the TDR probe installation, as it would in a more standard soils application where the probe is pushed into the soil to avoid disturbing the natural soil structure. The PIN diodes used for the remote diode shorting method are located within the probe head and end termination structures, which are manufactured out of polyethylene terminal blocks coated with industrial silicon sealant (Figure 4.2). The diodes are located immediately prior to the start of the exposed TDR rods at the probe head, and immediately after the exposed rods at the probe end. The remote-diode method determines the time that 63 the propagated voltage signal passes through the diodes when they are short-circuited, and the signal returns along the ground conductor. To determine the travel time in the soil only, a probe head/end offset correction must be applied to account for the signal travel time within the termination structures between the diodes and the conductors exposed to soil (Young 1997). The probes were installed July of 1998. Several difficulties with the probe design and waveform interpretation method were subsequently uncovered after the probes were irretrievably installed: (1) it was determined that the two-point linear calibration of Young (1998b) was insufficient to describe the effects of the coating on the calculated sample apparent dielectric permittivity; (2) the MPN3404 diodes do not create short circuits of sufficient quality for the remote diode method to be successful; and (3) the probe head and end structures were found to be inadequate. Problems 1 and 2 are discussed in Chapter 3. Problems 2 and 3 are further investigated in this chapter. 4.3.4 Temperature Dependence The waste rock in our experiment undergoes variations in ambient temperature from -20 to 25 °C. Measurements below 0 °C are not considered in this work due to ice formation. The effect of ambient temperature upon the measured soil apparent dielectric permittivity must be determined prior to the final calculation of soil water content. The dielectric permittivity of a wetted soil changes with ambient temperature as a function of two competing processes. First, the dielectric of liquid water in the free phase decreases with increasing ambient temperature. Second, the volume fraction of water bound to mineral surfaces decreases with increasing temperature (Hoekstra and Delaney, 1974; Wraith and Or, 1999; Or and Wraith, 1999). The dielectric constant of the bound fraction (~3) is lower than 64 that of free water (-80) and thus the conversion of bound water to unbound water with increasing temperature leads to a higher dielectric permittivity at the same water content. Both the dielectric permittivity and conductive properties of water that remains bound to the surface of clay minerals and of water bound in hydration layers of dissolved ions can also be expected to vary with temperature (Or and Wraith, 1999). Soils with a significant bound water fraction can therefore be expected to have a positive trend of dielectric permittivity with temperature at low water contents and a negative trend of dielectric permittivity with temperature at high water contents, when the relative contribution of the bound water fraction is lower. The cross-over point will be a function of the clay fraction and porosity of the sample. The effect of temperature can be measured empirically by controlling the water content of a sample, then progressively changing the ambient temperature of the sample (Pepin et al., 1995; Halbertsma et al., 1995; Wraith and Or, 1999). The calibration is repeated for different water contents, and over the full range of ambient temperatures expected in the field. A n alternative method is presented based upon the field data, using the measured daily variations of the apparent dielectric permittivity and ambient temperature at the same location. 4.4 E X P E R I M E N T A L M E T H O D S Experiments were first conducted to investigate the ability of the TDR system used to collect valid measurements of the apparent dielectric permittivity, secondly to relate the measured apparent dielectric permittivity to the water content of the waste rock, and thirdly to determine a correction for ambient soil temperature. 65 4.4.1 TDR Instruments and Waveform Collection For the laboratory experiments, raw waveforms and diode-shorted waveforms with 256 points in time were collected using a M T D R instrument with firmware Version 1.27. Travel times were measured using the automated waveform fitting routines which are included in the instrument firmware. Additional waveforms for detailed waveform analysis were collected manually using a PC connected to the M T D R and running the instrument control software ViewPoint™ Version 1.34. For all measurements, all instrument settings and waveform fitting parameters in the instrument firmware and software were optimized for maximum accuracy and stability of tangent-line fitting. For all laboratory calibration work, a minimum of twenty measurements of travel time were obtained in each media, and averaged. The CPE experiment required the use of 34 m coaxial cables (Commscope F660BEF RG6) and up to two Campbell Scientific SDMX-50 coaxial multiplexers in series to connect multiple TDR probes to the TDR instrument. Long cables and multiplexers may increase the measured travel time slightly by changing the frequency content of the TDR pulse. This effect was checked during our calibration experiments, and was found to be small enough to be ignored. Laboratory calibration work was conducted using 6 m cables (Commscope F660BEF RG6) to connect the TDR instrument to the TDR probe. 4.4.2 Probe Head/End Offset To determine the probe head/end offset, automated travel time measurements were taken using a 291/3.2/25 uncoated probe in air and in methanol and water contained in a 15 cm diameter P V C chamber. In addition, prior to installation of the TDR probes into the CPE, travel time measurements were conducted for each probe in oven dried silica sand, and de-ionized water contained within 30 cm diameter P V C containers. These measurement were 66 taken in accordance with the two-point probe calibration procedures described in Young (1998b). These measurements were indicated to be sufficient to parameterize the probe head/end offset and a linear correction equation to compensate for the presence of the probe coating. 4.4.3 Performance in electrically conductive media Measurements of travel time using the automated curve fitting routines of the MoisturePoint instrument were carried out to determine the performance of a 281/3.2/25 coated probe and the remote diode shorting method using MPN3404 diodes in electrically conductive media. Measurements were taken in a water solution in which the electrical conductivity was varied from 0 to 70 dSm"1 using NaCl. Measurements were also conducted in silica sand (K~27) saturated with water solutions of with varying electrical conductivity. The electrical conductivity of the water solutions saturating the silica sand was varied from 0.24 to 17.5 dSm"1 using NaCl. This caused the bulk soil electrical conductivity to vary from 0.029 to 8.3 dSm"1 which represents the variation of soil-water conductivities expected in the field experiment. Travel time in the water solutions (Curve A, K~80) was determined using the remote diode shorting method for both the probe head and end (Chapter 3). In the saturated silica sand, three methods were used to measure ti at the probe end. The travel time was measured using remote diode shorting for both the probe head and probe end. The travel time was also measured using the remote diode method to determine the probe head travel time (t0) and using the automated curve fitting routine of the MoisturePoint TDR system to perform flat-tangent fitting to the raw waveform collected at the probe end. The travel time was also measured by manual determination using the dual-tangent method. 67 4.4.4 Laboratory Calibration Measurements for CPE waste rock An empirical calibration of the travel time-water content relation was obtained by packing waste rock into a box at known water contents, and measuring the travel time using coated probes. Three sets of calibrations were performed. First, calibrations were performed using two different probes, 281/3.2/25 and 376/3.2/25, using waste rock collected from the CPE during construction, to assess the magnitude of variation between probes. Secondly, two calibrations were conducted to determine the effect of variable grain size on the measured travel time using a 281/3.2/25 probe. CPE waste rock samples were sieved to obtain two fractions, <5mm and <25mm. Thirdly, calibrations were performed on two different waste rock materials to encompass the range of soil water electrical conductivities expected in the field. Measurements were conducted using both 281/3.2/25 and 376/3.2/25 probes in washed waste rock material. This waste rock was flushed with many pore volumes to remove stored weathering products. Upon saturation, the measured electrical conductivity of the soil water solution was <0.8 dSm"1. The second material was waste rock sampled as the waste rock pile was constructed, which included a reserve of stored weathering products. Measurements were obtained using a 281/3.2/25 probe. Waste rock was first oven dried for 24 + hours at 110 °C. It was progressively wetted using a hand spray bottle and thoroughly mixed prior to packing. At each water content, the soil was repacked three times and the measurements replicated to determine variability caused by the packing arrangement. Repeated calibrations from oven-dried soil to wetted soil were made. The travel time was measured with automated remote diode shorting for both the probe head and end. Volumetric water content was determined for each packing 68 from the difference of total mass and dry mass, and the volume of the packed sample. The dry density of the material was kept similar to that obtained with waste rock at the water contents measured during construction of the experimental waste rock pile. The container used was 12 cm x 12 cm in cross section, representing 2.4 times the outer rod to outer rod spacing. The size of the container was checked prior to use. Measurements of the travel time for an uncoated probe in water in a 35 cm diameter cylindrical container and the 12cm x 12cm container were found to be within 0.7%, confirming that the container was adequate to contain the majority of the wave energy of the TDR pulse. Coated probes do not correctly average water content i f the water is not evenly distributed along the probe length (Ferre et al., 1996, Knight et al., 1997). A l l measurements were therefore taken with the TDR probe and soil container horizontal to minimize vertical gradient in water content. Wetting was stopped when it was clear that water was redistributing within the soil container during the time taken to measure the travel time. Measurements were also conducted for saturated waste rock. A coated TDR probe was placed into a 15 cm diameter P V C pipe, and soil was packed around the probe at residual water content. The pipe was slowly flooded from the bottom upwards over 24 hours by equilibration of the water level in the pipe with a large external water reservoir. 4.4.5 Field Measurements Two and a half years of measured TDR travel times are available from 21 probes installed within the constructed pile at depths from 0.1 m to 4.5 m. Instrumentation was placed within the CPE during construction of the waste rock pile. The TDR probes were manually placed upon a leveled waste rock surface and buried by hand to 20 cm depth. Further waste rock was placed upon the probe using a tracked excavator with a i m bucket. 69 It was necessary to remove larger grain size particles (>25mm) from the vicinity of the TDR probes to prevent damage to the TDR probes during burial. Automated TDR measurements are conducted using an M T D R instrument connected to a Campbell Scientific CR10X datalogger. The TDR probes are connected using 34 metres of cable, and either one or two Campbell Scientific SDMX-50 multiplexers connected in series. Measurement intervals range from 20 minutes to 60 minutes with the time of year. Travel time is measured using the automated remote diode measurement routine of the M T D R at both the probe head and end. Five measurements of travel time are discarded at the start of each round of measurements to ensure the instrument is properly warmed up and stable. During a single measurement round, each probe is measured three times in succession and the average and standard deviation of the three measurements are recorded. Ambient soil temperature is recorded every 15 minutes at each instrument location using commercially available thermistors. Separate dataloggers are used to collect TDR and temperature data, and the measurements are not synchronized. The soil ambient temperature corresponding to each TDR measurement was determined by linear interpolation between the nearest temperature measurements. 4.5 R E S U L T S A N D DISCUSSION For all measurements of travel time, the raw measured travel time (t) is converted to an adjusted travel time (t*) by subtraction of the probe head/end offset (t 0ff). The adjusted travel time (t*) is then divided by the calculated travel time in air ( t c a i r ) , calculated from the probe length and the dielectric constant of air. Division by t c a j r allows comparison of raw travel time measurements of probes of varying length. A l l measurements are presented as t*/t c air, referred to in the remainder of this chapter as the corrected travel time. 70 4.5.1 Probe Head/End Offset The results of the measurements of travel time in air, methanol and water are presented in Figure 4.3. The dielectric properties of these substances are known (CRC Press, 1994), and a theoretical travel time can be estimated from Equations 4.3 and 4.4 using the length of the exposed conductors. The difference between the calculated travel time and the measured travel time is the probe head/end offset (t0ff). The results show that t 0ff increased with increasing sample medium dielectric constant. This behaviour indicates that the electromagnetic field of the TDR signal traveling within the probe head and end termination structures was not contained within the termination structures and its velocity is affected by the dielectric permittivity of the surrounding medium. This difficulty with the probe design was not discovered until after probes were installed in the field. To correct for this effect, it was necessary to derive an empirical equation, relating t 0ff to measured travel time based upon the data in Figure 4.3 and measurements from the probe calibrations performed before installing the probes (Appendix I). t 0ff can be expected to vary between probes depending upon the exact dimensions of the probe head and end, and upon the thickness and nature of the silicon coating. The apparent dielectric permittivity of the oven-dried silica sand (2.7) used in the two-point calibration of the probe coatings is very close to the dielectric permittivity of the probe coatings (2.8) (Chapter 3). The travel time measured by a coated probe in oven-dried silica sand will therefore be equal to the travel time that would be measured in the absence of the coating. Using the probe length, a theoretical travel time in oven-dried silica sand was calculated, and a value of t 0ff determined for each probe. t 0 fr in oven dried silica sand ranged from 0.464 to 0.696 nanoseconds (ns), with an average of 0.571 for 21 probes. The empirical correction 71 uses the measurement of t0ff in silica sand for each individual probe, and increases t0ff from that value based upon the slope of data in Figure 4.3. The connection of the coaxial signal conductor to the central probe rod is relatively direct, leading to minimal opportunity for interaction of the propagating signal with the medium surrounding the probe head or end. The remote diode method short circuits the signal propagating along the signal conductor at the probe end. The signal must then travel through the un-shielded wire connecting the central rod to the ground conductors at the probe end (Figure 4.2, Item 8), then from the ground conductors to the coaxial cable shield at the probe head (Figure 4.2, Item 9). This gives increased opportunity for interaction of the signal with the surrounding medium. Had the wiring connecting the inner and outer conductors in the probe head and end structures been properly shielded, the t0ff correction would not have been dependant upon the dielectric permittivity of the sample material. 4.5.2 Effect of Electrical Conductivity Figure 4.4 presents the variation of corrected travel time (t*/t c aj r) for the measurements conducted in electrically conductive water solutions (Curve A), and in silica sand saturated with electrically conductive water solutions (Curves B,C, Open Circles D). Curve B presents the travel time measured using remote diode shorting for both the probe head and probe end. Curve C presents measurements taken using the remote diode method to determine the probe head travel time (t0) and using the automated curve fitting routine of the MoisturePoint TDR system to perform flat-tangent fitting to the raw waveform collected at the probe end. Points at D presents the manually determined measurements of travel time using the dual-tangent method. 72 There should be no measureable change in the measured travel time over this range of electrical conductivities for water solutions (Or and Wraith, 1999). Chapter 3 determined that travel times determined using remote diode shorting and MPN3404 diodes in coated probes was biased in electrically conductive water solutions. This systematic bias was introduced because the MPN3404 diodes did not create short circuits of sufficient quality for the remote diode shorting method to be successful. This bias is small over the electrical conductivity range presented in Figure 4.4 (Curve A). Curves B,C and D indicate that a systematic change occurs in the measured travel time in saturated silica sand. Similar results were obtained by Wyseure et al. (1997). Both the automated and manual measurements of travel time increase with increasing electrical conductivity of the saturating solution. However, both the automated measurements methods overestimate the magnitude of this change. Figure 4.4 indicates the flat-tangent method is no better than the remote diode method for the probe design used. The increase in measured travel time determined by manual measurements may relate to a real change in the complex dielectric permittivity of the soil-air-water-bound water mixture present in electrically conductive soil. Changing ionic strength may alter both the dielectric and conductive properties of water bound in mineral and ionic hydration layers (Or and Wraith, 1999). It may also relate to the changes in frequency content of the TDR pulse during passage through the soil. As discussed in the theory, the apparent dielectric permittivity is an average measurement which integrates dielectric permittivity over an unspecified range of frequencies. Changes in the frequency content of the TDR pulse brought about by the interaction of the probe design with the electrically conductive soil will therefore cause variations in the measured apparent dielectric unrelated to changes in the soil 73 itself. The precise reasons for the increase in travel time cannot be investigated without using frequency domain analysis of the TDR waveforms, which is not the purpose of this chapter. For the practical measurement of water content in the field, commercially available equipment is based upon travel time measurement. This result indicates that the measurement of travel time alone may not be sufficient to determine water content in soils with both bound water, and highly electrically conductive soil-water solutions. Figures 4.3 and 4.4 indicate the travel time measured using the probe design installed in the CPE and the remote diode method using MPN3404 diodes varies with both sample dielectric permittivity and electrical conductivity of the soil water solution. To compensate for and eliminate these errors empirically would require unwieldy calibrations. In addition, the use of coated probes means that the probes cannot be used for the measurement of in-situ bulk soil electrical conductivity by analysis of the raw waveform. It is not possible, with the system used, to perform simultaneous measurement of travel time and electrical conductivity, and thus it is not possible to deterministically correct the field travel time measurements for the bias introduced by the electrical conductivity of the soil water. These results provided the justification for measuring two end-member empirical calibrations for a washed (low soil water electrical conductivity) and unwashed (high soil water electrical conductivity) waste rock. 4.5.3 C P E Waste Rock Calibration 4.5.3.1 Probe Variability and Grain Size The effects of probe -variability and grain size on the measured calibrations using unwashed waste rock are presented in Figure 4.5. The use of a probe coating implies a unique relationship between measured travel time and the true sample apparent dielectric 74 permittivity that was not fully characterized by the two-point calibration performed for each probe prior to installation (Chapter 3). Combined with the previously noted variation in the probe head/end offset, it was expected that there would be variation between probes. Figure 4.5A indicates a slight shift in the calibration curve upwards to higher travel times between the 281/3.2/25 probe and the 376/3.2/25 probe. The magnitude of the shift in the calibration between these two probes was smaller than the range of variability of measurements between different replicates at the same water content with a single probe. From these results, it appears that variations in the texture or nature of the waste rock are larger than the variations in the probe characteristics caused by the coatings and probe head and end termination structures. The unaccounted for inter-probe variation will lead to some uncertainty in the absolute value of calculated water content within the CPE experiment. The results shown in Figure 4.5B indicate that the difference between the measured travel times for <5mm and <25mm fractions was small compared to the variability between replicates of the same soil at the same water content. This indicates that the water content measured by TDR wil l not be affected by the grain size of the material around the probe. However, it should be noted that while the TDR probes wil l correctly measure the water content of the waste rock around the probe, that waste rock is not representative of the bulk waste rock contained within the entire experiment. The TDR probes only measure the water content in the finer grain size fraction of the pile as a result of the grain size selection during placement of the probes. The water content measurements derived for the field installed probes wil l represent water in the finer grained (< 5mm) fraction over a scale of approximately 40 cm. The actual waste rock contains particles up to 1.5 m in diameter. A n investigation of the conversion of water contents to account for larger grain size fractions can 75 be found in Yazdani et al. (2000). This work indicates that particles > 5mm occupy volume with impermeable material, but do not contribute to water retention under capillary tension. In unsaturated material, they act to lower volumetric water contents under the same matric suctions. The actual value of volumetric water content of the bulk waste rock that includes all grain size fractions, will thus be some lower fraction of the TDR measured water content to account for the presence of particle sizes greater than 5mm. 4.5.3.2 Soil Water Electrical Conductivity The results of the calibrations conducted in waste rock with varying soil water electrical conductivity are presented in Figure 4.6. The empirical Topp equation, reformulated as a square root mixing model by Ferre et al. (1996), is presented as a reference comparison only. The Topp equation has been calculated using the non-linear calibration of the effect of the probe coating on the measured apparent dielectric permittivity presented in Chapter 3. The data for unwashed (Figure 4.6A) and washed (Figure 4.6B) soil are similar between 10 to 20 % water content. Near saturation, the effects of soil water electrical conductivity on the measurement of travel time is stronger, and the curve for the unwashed soil trends upwards. In contrast, the washed soil calibration converges to the Topp equation as higher water contents (>15%). At low water contents (<10 to 15%), both calibrations fall above the Topp equation, with the washed material at a slightly higher travel time. The differences between washed and unwashed waste rock calibrations are significant enough to justify separate calibrations. In-situ water samples extracted from the field experiment indicated that water to 1 m depth reaches electrical conductivities of 2 to 3 dSm"1 (Nichol et a l , 2000, Appendix A). Figure 4.4 indicates that the systematic increase in travel time did not commence until after 76 the bulk soil electrical conductivity exceeded 1.5 dSm m"1, which corresponds to a soil water solution of 4 dSm"1 in saturated material. It is expected therefore that the measurements in washed material will be representative of probes installed near surface (<lm) where the electrical conductivity of the soil water is low. Measurements of soil water electrical conductivity during the waste rock calibration experiments using saturated, unwashed waste rock indicated soil water electrical conductivities of 10 to 20 dSm"1. This range of electrical conductivities is similar to the range of electrical conductivities measured for in-situ soil water samples extracted from the experimental pile using suction lysimeters between 1.75 m and 4.5 metres below the surface of the pile. Travel time measurements conducted in unwashed waste rock matches the most electrically conductive soil water solutions near the base of the constructed pile and are expected to be more representative of probes installed at 1.75 m and below. 4.5.4 Field Measurements 4.5.4.1 Field Temperature Correction Data from three TDR probes installed at 20 cm depth, located immediately adjacent to temperature probes, were analysed to derive a field based temperature correction. A maximum daily temperature variation of approximately 7 °C was recorded. A l l travel times were corrected for probe head/end offset, and t*/tcair was determined. A 24 hour moving window average of t*/tcajr was calculated, and the difference between each individual measurement and the average (t*/tcajr - (t*/tcajr) 24 a vg = A t*/tcai r) was calculated. Figure 4.7A presents individual measurements of t*/tcajr and the calculated (t*/t c aj r)24 a Vg- This figure indicates significant daily fluctuation in measured travel time. 77 A 24 hour moving window average was calculated of ambient in-situ temperature (T"24avg) and the deviation of each individual temperature measurement from the average (T-T24avg =AT) was calculated. Figure 4.7B presents the variation in measured travel time, A t*/tcair with the variation in measured temperature, AT. For this water content range, the systematic daily variation of the measured travel time is in phase with the in-situ temperature. This behaviour is consistent with the presence of a significant water fraction bound to mineral surfaces, or within the hydration layers of dissolved ions. This variation would not be correctly described by a temperature correction based upon the properties of free phase water only. For each of the three probes, the ratio of (At*/t c ai r)/( AT) was calculated, which represents the change in travel time with temperature. Figure 4.8 presents a plot of the temperature correction factor, (At*/t c air)/( AT) against t*/t c ajr for a probe at 20 cm depth. The data have been filtered to remove all data where the soil water content was changing rapidly, and the 24-hour moving window average is a poor representation of water content. Only those measurements where ATM.5 °C are shown in Figure 4.8. This temperature filter was varied, and 1.5 °C was found to be optimal for the field data. Figure 4.8 indicates that the correction factor is positive at low travel times (low water content) and negative at higher travel times (high water contents). The highest corrected travel time on Figure 4.8 is -2.6. Analysis of daily variation of measured travel times at higher water contents was not possible as these water contents only exist during short-lived rainfall events, where determination of a 24 hour average of travel time is not possible. A linear equation has been fitted to the data describing the change in the temperature factor with travel time. The positive correction factor at low water contents 78 indicates that the apparent dielectric permittivity of the water increases with temperature due to the unbinding of bound water. The influence of bound water decreases with increasing water content as the relative fraction of bound water to free water decreases. The crossover from a positive to negative correction factor for this probe location occurs at approximately t*/t c air~2.65, or approximately 15% water content. At higher water contents, the correction factor becomes negative, implying the higher temperature lowers the free water dielectric permittivity, and this trend is now larger than the unbinding of bound water The slope of the temperature correction factor (At*/t c a i r ) / AT) against t*/t c ai r was similar for all three probes, and an average value of 0.006 was chosen. The cross-over point from positive to negative was different, and varied from 2.6 to 2.9. This behaviour indicates the influence of bound water varies from location to location due to variations in the grain size and clay content of the material immediately surrounding the probe. The correction factor will therefore vary for each probe location. A n average empirical equation was derived to be applied to the other probes: (At * I tcair) / AT = -0.006 * f t ^ + (2.65*0.006) [4.5] This equation is applied to t*/t c ai r as shown in Figure 4.9, for a probe at 20 cm depth, which presents the measured travel time (A,t*/t c aj r), the 24 hour moving window average of the measured travel time (B, (t*/ t c a i r )24avg), and the temperature corrected travel time (C). A l l three time periods have similar magnitudes of daily temperature fluctuation. The data from June and April, 1999, indicate the correction is sufficient to remove the diurnal variation of measured travel time to below the measurement precision. The data in 79 September are at a higher water content, near the cross-over point, and demonstrate little diurnal variation in both the raw (A) and corrected (C) data. 4.5.4.2 Corrected Field Data The magnitude of the differences between the effects of electrical conductivity and temperature on the final calculated value of water content can be assessed by examination of the corrected field data. The raw time is corrected for probe head/end offset, then (At*/t c aj r)/ AT) is calculated using Equation 4.5. AT is determined as the difference in ambient temperature from 21 °C, the temperature of the laboratory during calibration. t*/t c ai ris adjusted based upon Equation 4.5, or upon the probe specific calibrations for the three probes used. The water content is then calculated using either the washed or unwashed waste rock calibration equation appropriate for the depth of the probe. Figure 4.10 presents calculated water contents for a probe located at 20 cm depth following a large rainfall event in September, 1999, and for a probe located at 3.0 m depth as it underwent the cooling portion of a yearly temperature cycle in the winter of 1999/2000. Three curves are presented for each probe: the calculated water content with no temperature correction using the washed (0.2 m depth) or unwashed (3.0 m depth) calibration (A); the same waste rock calibration including the temperature correction (B); and the ambient temperature (C). The difference between the washed and unwashed calibrations is shown by curve D, which represents the water content calculated using the opposite (washed or unwashed) calibration to that used for curves A and B. The data for both shallow and deep probes indicate the effect of temperature (A,B) is smaller than the variation in the calculated water content caused by the effect of electrical conductivity on the remote diode method (D). For the probe at 20 cm depth, using no 80 temperature correction (A) underestimates water content at low water contents, but only overestimates water content for very short periods during short-lived peak water flow. The difference between calibrations (A,D) for electrical conductivity is more pronounced at low water contents where the two calibrations diverge. Near the peak water contents measured in the field, 17 to 20%, the two calibrations are quite similar. At 3 m depth, the water content is generally higher. By coincidence, the water content is close to the cross-over range where the ambient temperature correction is very small and the uncorrected and temperature corrected data are indistinguishable (A, B). The two calibrations for different electrical conductivities also coincide at these measured travel times, and the two calibrations are very close (A, D). 4.6 C O N C L U S I O N S The TDR method using travel time analysis has been empirically calibrated and applied to determine the water content of coarse grained waste rock. This required consideration of the effects of poor probe design, electrical conductivity of the soil water solution, and grain size of the waste rock. The effects of grain size and inter-probe variation (Figure 4.5) alter the relative position of the measured calibration curves, but have little effect on the slope of the either the unwashed or washed calibration curves. These effects are averaged by including all these data in the calibrations derived in Figure 4.6. An estimate of the total error in absolute water content is difficult because the variability of the probe coatings and the probe head/end offsets were not properly measured prior to installation, but a single measurement of t*/t c ai r in Figure 4.6 corresponds to a spread of +/- 2% water content in the range of 0 to 15% water content. The approximate calibrations required to compensate for electrical conductivity will increase the error estimation. At saturation, the same 81 measured travel time corresponds to a 12% difference in volumetric water content due to the effects of electrical conductivity on the remote diode method. The timing of the arrival of wetting fronts and the transition time for measurements relate directly to the measured travel time, and thus wil l be accurate. In addition, the change in water content at any give probe location will be more accurately described. The measurement of water content using TDR is promoted as a simple and effective means of water content estimation using commercially available, automated equipment. This work has demonstrated that the practical application of TDR can be more difficult. Based upon this experience with probe design, calibration and field installation, and the knowledge gained from the literature, the following recommendations can be made for the practical usage of coated probes in materials with high electrical conductivity soil water solutions: 1) Three-rod Zegelin-type probes with a resistive coating are a reliable means of determining the travel time in media with high sample electrical conductivity. A successful design is recommended in Chapter 3. A check of the performance of any design should be undertaken in soil saturated with solutions of differing electrical conductivity prior to probe deployment. 2) Measurement of travel time alone may not be sufficient to characterize the dielectric properties of water-soil-air mixtures where the soil water has a high electrical conductivity. 3) The construction of the probe head/end should be such that the propagating TDR pulse is fully contained within the probe head. A l l wiring should be properly shielded. Travel times should be measured in both air and water. The theoretical travel time can be 82 determined, and the probe head offset determined. Only those probes that show no variation in probe head/end offset should be used. 4) Remote diode shorting can be used to determine the timing of the signal arrival at the probe head (Chapter 3) 5) Remote diode shorting should not be used at the probe end. The dual-tangent analysis method should be used to determine the timing of the probe end reflection with no probe end termination structure or diodes present (Chapter 3) 6) Coated probes installed in the field should be matched with short uncoated probes which are used solely for the determination of bulk soil electrical conductivity. This wil l enhance the capability for soil calibration and provide information on the electrical conductivity of the soil water solution. 7) A l l coated probe calibrations and field installations should ensure uniform water distribution along the probe length. Coated probes should not be installed vertically. 8) A l l field measurements should include the simultaneous measurement of ambient in-situ temperature to permit correction for the effects of changes in temperature. 9) The effects of ambient temperature can be compensated for by the examination of field data or by laboratory calibration. The temperature compensation is dependent upon the clay fraction and varies with water content. 10) The measurement of the difference in water content is more accurate than that of the absolute water content in coarse, heterogeneous waste rock. The measured water content is representative of the water content in the finer (<5 mm) fraction. 83 Two Way Travel Time (ns) Figure 4.1: TDR waveform collected using uncoated TDR probe in water with 1 dSm' 1 electrical conductivity. Tangent-line methods are used to determine the timing of the probe head (t0) and probe end (t,) reflection. 84 8 ¥ 7 8 1. RG6 coaxial cable: F660 BEF CommScope 75 ohm cable 2. Female F-connector 3. Male F-Connector to plug 4. Polyethylene terminal block 5. Outer rod: unshielded 3.2 mm dia. 316 stainless steel. 6. Center rod: 3.2 mm dia. 316 stainless steel coated with 0.4 mm thick polyolefin heat shrink 7. On Semiconductor MPN3404 PIN diode 8. 22 gauge un-insulated bus wire 9. 22 gauge un-insulated bus wire 10. Terminal block sealed with silicon sealant Figure 4.2: Design of TDR probes installed in field experiment. 8 5 1.2n +£ +J O 1.0-tn i t o 0.8-•D C 0) 0) 0.6-n a) 0.4-n o 0.2-H 1 1 1 1 1 • 1 r~ 0 20 40 60 80 S a m p l e Dielectr ic Cons tan t F igure 4.3: Variation of probe head/end offset (toff) with sample dielectric constant. 86 6 E o o 3 -I 1 1 1 1 1 1 1 r 0 2 4 6 8 Figure 4.4: Corrected travel time (fVt^) measured using a coated probe in water solutions of varying electrical conductivity (A), and silica sand saturated with electrically conductive water (B,C,D). Probe head travel time determined using remote diode method. Corrected travel time determined using the remote diode method at probe end (A,B), the flat-tangent method at the probe end (C), or manually determined dual-tangent method at probe end (D). 87 0) E o > TJ £ O Q) L-L-o O A) Different Probes: Unwashed Waste Rock ft* 281 mm probe 376 mm probe 5 10 -1— 15 — i — 20 — i — 25 — i — 30 — i — 35 40 at E > L _ to TJ £; o 0) l _ l _ o o 4H 2 1 B) Grain Size: Unwashed Waste Rock . A 5 281 mm probe in 5mm' soil 281 mm probe in 25 mm' soil - i — 10 - i — 15 — i — 20 - i — 25 — i — 30 — i — 35 40 Volumetric Water Content (%) Figure 4.5: Effects of probe variability and grain size on the relationship of corrected travel time (P/t^J to volumetric water content. 88 6 o E a> > TJ *, O CD o O U n w a s h e d waste rock: best fit equat ion T o p p equat ion (Ferre et al, 1996) —r-10 - r — 20 - r ~ 30 40 o E > T5 £ O 0) l _ o o W a s h e d waste rock: best fit equat ion T o p p equat ion (Ferre et al, 1996) B 10 20 30 Volumetr ic water content (%) 40 F igure 4.6: Cal ibrat ion curves relating corrected travel t ime ( tVt^) to volumetr ic water content for waste rock with high soil water electr ical conductivity (A) and low soil water electrical conductivity (B). 89 2.45 2.40 2.35H 2.30H 2.25 T J C o 4  H D ) < C '—' 3 > CJ) w O 3 2 E +J as i_ i_ 3 CD O Q . 1 J= E +J CM 0 i c <D 0) -1 l_ 3 rati am 0) >•--2 a o E 0) 0) O) -3 4-1 •*-> ro >_ c o > -4 Ambi ns J u n e 1999 F igure 4.7: Daily variation in measured corrected travel time (tVt^) for a probe at 20 cm depth (A). Comparison of corrected travel time variation with temperature variation (B). 90 o c o o Is 0) 1- '5 3 +^ +J -~; 2 4-o < a. ~" E a> c 5 E < 0.020 0.015 -0.010 -0.005 -0.000 -0.005 --0.010 --0.015 --0.020 (At*/tcalr )/AT = -0.006 * ( t * / U + 0.0159 Data fi ltered: AT>1.5 °C T 2.2 2.3 2.4 2.5 2.6 Cor rec ted travel t ime t*/t 2.7 F igure 4.8: Field determined ambient temperature correction, (At*/trair )/AT, as function of corrected travel time, (fVt^J, for a probe at 20 cm depth. 91 2.45 June 1999 Figure 4.9: Application of field determined ambient temperature correction to field data: (A) corrected travel time (tVt^); (B) 24 hour moving window average of corrected travel time (F/UW a n d (c) temperature corrected travel time ( t V U r 92 20-Depth 20 c m •20 H 5 M O O g 0) i _ 3 4-1 ra i _ CD Q. E 23 25 27 29 Sept . 1999 01 03 05 Oct . 1999 M O Q . 28 A u g 06 D e c 15 Mar 23 J u n 1999 1999 2000 2000 Date F igure 4.10: Effect of ambient temperature correction and remote diode shorting bias on measured water contents: (A) volumetric water content using appropriate calibration for soil water electrical conductivity; (B) temperature corrected travel time (fYt^)^ (C) ambient soil temperature; and (D) volumetric water content using wrong calibration for soil water electrical conductivity. 93 CHAPTER 5: FIELD EVALUATION OF THERMAL CONDUCTIVITY SENSORS FOR THE MEASUREMENT OF MATRIC SUCTION. 5.1 ABSTRACT Thermal conductivity (TC) sensors indirectly measure matric suction in a material through its effect upon the water content and hence thermal properties of an engineered ceramic probe. TC sensor output must be corrected for sensor hysteresis and for ambient temperature. The effects of hysteresis are described and an equation presented to correct for ambient probe temperature. Two years of monitoring data from a field site are assessed to determine long-term sensor performance and accuracy. Air entrapped in the sensors during initial wetting has led to unpredictable long-term drift in those sensors continually exposed to matric suctions less than 20 kPa. Comparison of uncorrected field measurements to values corrected for both sensor hysteresis and ambient sensor temperature indicates the magnitude of these corrections can be similar. Corrected TC sensor measurements of matric suction are compared to tensiometer measurements. The field results demonstrate that TC sensor accuracy is highly variable in the low matric suction range. Initial sensor response, lags 1 to 4 days behind tensiometer measurements and the transition time of TC sensor measurements is significantly greater than tensiometers. It is recommended that TC sensors only be installed where the matric suctions are expected to exceed a limiting value of matric suction determined from the hysteresis characteristics of the sensor. The expected rates of change of matric suction in a field application should be estimated prior to sensor installation, and TC sensors installed only i f the expected matric suction changes are slower than the response time of the TC sensors. 94 5.2 I N T R O D U C T I O N Nichol et al. (2000) (Appendix A) describe a field experiment to characterize unsaturated flow within waste rock at a mine site in northern Saskatchewan. As part of that experiment, matric suction has been monitored over a two year period. The experimental pile, which is 5 m high, is composed of waste rock from an open pit operation. The material is highly heterogeneous, with boulders up to 1.5 m in diameter, 20-30% by mass less than 2mm in size, and a silt size fraction less than 5% by mass. Spatially and temporally detailed measurements of volumetric water content and matric suction are being collected on an ongoing basis to understand the transport of sulfide mineral weathering products from the pile. Measurements of matric suction can be obtained using tensiometers, gypsum blocks, granular matrix sensors, filter paper methods, psychrometers, and thermal conductivity (TC) sensors. Fredlund and Rahardjo (1993) provide an excellent review of most methods. Filter papers and psychrometers are not appropriate for long term in-situ field monitoring. Granular matrix sensors and gypsum blocks rely upon measuring the electrical conductivity of a water solution within the probe and both sensors contain probe components constructed of sulphates. Both of these factors are incompatible with the waste rock at the field study site. In addition, granular matrix sensors are not sufficiently accurate or reliable for long-term monitoring (Thomson and Armstrong, 1987, McCann et al., 1992 and Spaans and Baker, 1992). Thermal conductivity sensors and tensiometers were therefore chosen to measure matric suction in this study. The use of tensiometers is reviewed elsewhere (Cassel and Klute, 1986, Fredlund and Rahardjo, 1993) and their accuracy and limitations are 95 understood within their range of operation (matric suctions between 0 to -85 kPa). In contrast, thermal conductivity sensors are a developing technology. TC sensors measure the thermal conductivity of a ceramic probe in hydraulic contact with the soil water. The water content of the ceramic changes in response to the matric suction and therefore the thermal properties of the ceramic change. A laboratory calibration determines the relationship between matric suction, the water content of the ceramic, the thermal properties of the ceramic, and the sensor measurements recorded by a datalogger. The modern design of a thermal conductivity sensor was first proposed by Phene et al. (1971a; 1971b). Histories of the development of TC sensors and discussions of previously available commercial sensors are provided by Fredlund (1992), Reece (1996) and Feng (1999). The experimental waste rock pile provides the opportunity to examine long-term TC sensor performance under field conditions. In this chapter, TC sensor operation and calibration is reviewed. The raw sensor output must be corrected for both hysteresis in the relationship between matric suction and water content of the ceramic (Feng, 1999; Feng and Fredlund, 1999), and for the effect of ambient temperature on the thermal properties of water. A new correction factor for the ambient temperature effect is derived. Methods used to collect TC sensor and tensiometer measurements at the experimental site are then presented. The TC sensors are used to conduct year-round, automated measurements of matric suction. Manual tensiometer measurements and automated measurements using pressure transducers were conducted to provide independent estimates of matric suction. The raw and corrected TC sensor data are presented to demonstrate the effects of hysteresis and ambient in-situ 96 temperature. The corrected data are then compared with the measurements of matric suction by tensiometer to assess the performance of the TC sensors. 5.3 T H E R M A L C O N D U C T I V I T Y SENSORS A TC sensor is composed of a heating element and a temperature measurement device encased in an engineered ceramic body. A diagram of the sensors used in this study is shown in Figure 5.1. These prototype sensors (Beta-98 Sensors) were made at the University of Saskatchewan (U. of S.) by D.G. Fredlund and F. Shuai. The thermal conductivity of the ceramic body is determined by measuring the dissipation of a known heat pulse. The core of the ceramic is heated by controlled activation of the heating element. The temperature at the core is measured before, during and after heating. The temperature rise at the core changes as the rate of dissipation of heat away from the probe center or its thermal diffusivity changes. The temperature rise is used to determine the thermal conductivity of the ceramic probe. The thermal conductivity of the ceramic skeleton is constant while the thermal conductivity of a wetted sensor will change as water and air exchange within the pore spaces. The thermal conductivity of water (-0.6 W m^K" 1) is much greater than that of air (0.026 W m^K" 1) and thus thermal conductivity increases from a dry sensor to a wet sensor. The thermal conductivity is primarily affected by volumetric water content in conjunction with the degree of interconnection of the water phase in the engineered ceramic. The water content of the ceramic, and the distribution of the water phase, is in turn controlled by the matric suction applied to the outside of the sensor and, as in any porous material, is a function of the pore size distribution of the ceramic. Laboratory-based calibration curves relating the temperature rise at the sensor core to an applied matric suction are derived by equilibrating the sensors to known matric suctions 97 using a pressure plate apparatus and measuring the temperature in the probe over the heating cycle (Feng and Fredlund, 1999). The temperature rise-matric suction relationship is not unique due to hysteresis of the ceramic's water content to matric suction relationship (Feng, 1999; Feng and Fredlund, 1999) and is also a function of changes in ambient temperature (Shuai, 1998, pers. comm.). 5.3.1 Sensor hysteresis The hysteretic nature of the relationship between water content and matric suction in the ceramic TC sensor leads to a hysteretic relationship between measured thermal conductivity and matric suction. To interpret the sensor output, it is necessary to know the hysteresis behaviour of the ceramic, and the wetting and drying history of the sensor. To follow is a review of hysteresis corrections that will later be applied to the field data Feng (1999) and Feng and Fredlund (1999) conducted experiments to record the water content and thermal properties of the sensors under different wetting and drying regimes. The curves shown in Figure 5.2 are based upon the measured curves in Feng (1999) but they have been simplified here for clarity. Four curves define the primary wetting and drying characteristics of the sensors. The ceramic has an initial drying curve derived by saturation under vacuum, followed by drying. The investigated range of sensor calibration was restricted to a matric suction of 400 kPa for practical purposes. Once the sensor equilibrates at 400 kPa matric suction, the sensor is re-wetted to 0.1 kPa to produce the main wetting curve, then re-dried to 400 kPa to produce the main drying curve. A further curve can be measured by air drying the sensor, then wetting the sensors to 0.1 kPa. This curve is termed the boundary wetting curve and is not shown, but it lies slightly below the main wetting curve. 98 The scanning curves shown on Figure 5.2A describe the water content-matric suction relationship that is followed between the main drying and main wetting curves. The scanning curves are also hysteretic. To correct for hysteresis it is therefore necessary to know the main wetting and drying curves as well as the shape of the scanning hysteresis curves. The wetting/drying history of the sensor is then used to determine whether any given measurement lies on a main curve or a scanning curve. No measurements were conducted of the main wetting curves for the Beta-98 sensors used in this study. The ceramic of the Beta-97 and the Beta-98 sensors is identical and therefore the main wetting curves are estimated from the Beta-97 data of Feng (1999) and Feng and Fredlund (1999). They experimentally determined the scanning curves and provided empirical functions to fit the measured curves. Two points should be noted from Figure 5.2. First, the core temperature rise is a non-linear function of matric suction, with the smallest changes recorded at low matric suctions. The sensors will therefore be least sensitive to changes at low matric suctions and high water contents. The second important feature is the large difference in saturation between the initial drying curve and the main wetting curve at 1 kPa. This difference is due to air trapped within the sensors during re-wetting on the short time scale of the measurement of the main drying and wetting curves during laboratory calibrations (Feng, 1999). Experiments by Feng (1999) indicate this entrapped air is gradually released through diffusion and dissolution, a . process called relaxation, and the sensor water content steadily increases with time. Relaxation is slow; a submerged sensor achieved 60% saturation within 1 hour, but only 70 % saturation at the conclusion of the one month test. To achieve 100% saturation would take years. The relaxation process raises concerns for the long term stability of sensor 99 measurements at low matric suctions. This point will be addressed in conjunction with the presentation of field results. 5.3.2 Correction Factor for Ambient Temperature Sensors deployed in the field experiment undergo variations in ambient temperature from -10 to +25 °C, with the operational range restricted to temperatures above freezing. The thermal conductivity of water and hence the thermal conductivity and core temperature rise of a wetted sensor are functions of temperature. The aim is to correct the field-measured core temperature rise to the temperature rise that would have been measured at the ambient temperature of the laboratory during calibration (23 °C). Phene (1971a) and Xing and Fredlund (1994) present an analysis of the core temperature increase in the sensor, based upon the representation of the sensor ceramic and the heater element core as spherical bodies. In practice, the sensor ceramic is not spherical, but i f it is large enough, the heat pulse does not reach the outer boundary during the heating cycle and the actual sensor shape does not affect the core temperature rise. They solved for the temperature rise (AT(t)) inside a spherical sensor as a function of the time since heating began (t). They conclude that at long times the effects of the finite heat capacity of the heating element, wire and epoxy become small and the full equation can be reduced to its first two terms (Phene, 1971a; Xing and Fredlund, 1994): AT(t)= Q fnDt [5.1] where Q is the heating rate (W), r is the radius of the heating element core (m), X is the thermal conductivity (Wnf'K" 1) and D is the thermal diffusivity (mV1). The thermal 100 diffusivity is the thermal conductivity divided by the density and specific heat capacity (Phene 1971a). Equation 5.1 can simplified: Where: AT(t,r) = A-B^=, [5.2] Q A = B = 4-nXr Q AnX^nD From these relations: Q x = -4mA D = ' Q * \An/2XBj A plot of AT(t) vs 1/Vt can be used to derive A and B. A correction for ambient temperature can be derived by taking the ratio of Equation 5.1 at two temperatures and considering the thermal conductivity and diffusivity as functions of temperature, T. If: XSens0r CT, ¥) = XDry_Semor + F(W)XW (T) • [5.3] where F(\|/) is the fractional contribution to the total thermal conductivity from the water phase as a function of the matric suction (\|/), then the temperature rise A T at time t and calibration temperature T 0 can be expressed as a function of the temperature rise ( A T ) at time t at ambient temperature T i : 101 c ,11 1-[5.4] AT(t,T0) = AT(t,Ti) * XDry+F(¥)^(Tx) ADry+F(¥)Aw(T0) * 1- c [5.4] where c" = r Values for A, and D for the TC sensors at 23 °C were estimated using Equation 5.1 and the values of A and B collected during the calibration of Beta 97 sensors (Shuai et al., 1998). Thermal conductivity and diffusivity were calculated using values of 2 watts for the rate of heating (Shuai, 1998, pers. comm.) and estimating the radius of the heater element and temperature sensing integrated circuit at the sensor core (ie: r = 0.002 m). No laboratory calibrations exist at other ambient temperatures to directly determine the changes in thermal diffusivity. Therefore, the effect of temperature upon the thermal diffusivity was estimated by making the conservative assumption that the density and heat capacity of the sensor are that of water and calculating thermal diffusivity from the thermal conductivity divided by the density and specific heat capacity. The magnitude of the third term of Equation 5.4 ranges from 1.000002 to 0.999999 over the range of ambient probe temperatures measured in the field. The third term in Equation 5.4 can therefore be set to a value of 1.0. The correction for ambient temperature depends upon the F(\j/) function in Equation 5.4. It can be derived from the laboratory calibration data by recording the full heating curve and using Equations 5.1 and 5.2. If X,dry is measured, then at each subsequent calibration measurement, the values of X(T0,\|/), A , B and AT(t,T0) are all known and Equations 5.3 and 5.4 are fully determined for both wetting and drying. The second term of Equation 5.4 can 102 then be calculated. This term, which can be termed the correction factor, represents the ratio by which the measured core temperature rise must be multiplied to correct for ambient sensor temperature. F(v|/) is a non-linear function that is dependent upon the distribution and continuity of the water phase within the sensor. The function F(\|/) is expected to be hysteretic as the relationship between water content and matric suction is hysteretic. The parameters required to derive F(i|/) and the correction factor from the laboratory calibration are not available for the sensors used in this study. The estimates of derived from Shuai et al. (1998) and the equation for A,(t|/) from Reece (1996) were used to estimate F(\|/) and hence to calculate the temperature correction for different matric suctions. The ambient temperature correction factor derived as a function of matric suction during drying is shown in Figure 5.3. The Shuai et al. (1998) data required using an estimated X,dry of 0.15 Wm^K" 1 . Estimates of matric suction corrected for ambient temperature are determined using recursion as the correction factor is a function of v|/. The ambient temperature correction is applied directly to the measured change in sensor core temperature during the heating cycle (AT(t,Ti)). Figure 5.3 indicates the correction is largest near 0 kPa when the sensor has the highest water saturation. It was previously noted that the calibration curves are least sensitive at matric suctions near 0 kPa. Ambient temperature corrections will have the largest effect on the calculated matric suction between 0 and 15 kPa. 5.4 F I E L D I N S T R U M E N T A T I O N This section describes the details of the measurements that were conducted to assess the performance of the TC sensors. The experimental pile was constructed in the summer of 103 1998 and measurements between September 1998 and October 2000 are analysed in this study. Eighteen TC sensors were calibrated by the U . of S. to determine the main drying curve. The sensors were installed in three instrument profiles at depths from 0.2 to 4.5 metres. Probe bodies were briefly wetted immediately prior to installation to improve initial equilibration time (Fredlund, 1992). Each sensor ceramic was coated with approximately 5mm of saturated silica flour paste to improve sensor-soil contact. The controlled heat pulse is delivered using a 10 volt constant voltage supply. This power supply was used in the laboratory calibration. More recent versions of the U . of S. TC sensor are calibrated and deployed using a constant current supply (Shuai, 2000, pers. comm.). Sensor internal temperature is measured using the integrated circuit (IC) temperature-sensing chip installed in each probe. A l l IC output voltage measurements and heating power control is performed using a Campbell Scientific CR-10X datalogger. The datalogger and voltage supply are located in an insulated and heated instrumentation hut. The datalogger is located approximately 20 metres from the sensors, and the lead wires are located in an above-ground cable duct. Measurements are collected from the sensors at intervals of four hours. This provides sufficient time for the previous heat pulse to dissipate, and the sensor body to re-equilibrate to the ambient in-situ temperature. The average ambient temperature within the center of the ceramic body is measured before the heat pulse. The regulated power is supplied to the heater element within the sensor core for a fixed time period (50 seconds). The temperature is recorded each second during the heating cycle. A second-order polynomial is fit by least-squares regression to the data for the last twenty seconds of the heating cycle (Devore, 1987). 104 The final heating voltage reading is calculated at a fixed time equal to the time used in laboratory calibration (50.225 seconds) using the polynomial equation. The temperature rise is calculated by subtracting the ambient voltage measurement from the calculated voltage at the end of the heating cycle. Nine Jetfill tensiometers (SoilMoisture Equipment Corp., Model 2725) were installed in three profiles in August 1999 at depths of 20, 50 and 100 centimeters. Tensiometers were installed within 0.5 m of the TC sensors. Each tensiometer ceramic was coated with approximately 5mm of saturated silica flour paste to improve sensor-waste rock contact. Tensiometers were monitored during August to October, 1999. The tensiometers were measured manually using a digital vacuum gauge with the double-puncture technique of Greenwood and Daniel (1996). Additional manual measurements were obtained using direct measurement with a mechanical vacuum dial gauge. Tensiometers were also measured using automated pressure transducers (Motorola 5100DP) at time scales of every 15 minutes. The pressure transducers were individually calibrated prior to installation using a variable hanging water column to apply an equal vacuum to all sensors. Three sets of readings were taken with increasing, decreasing and increasing vacuum applied to the sensors. The maximum vacuum applied was 23kPa. A correlation between tranducer output signal and ambient temperature was observed despite the on-chip temperature compensation (Motorola, 1998). The transducers were allowed to collect data for 15 days at 0 kPa and variable air temperature to determine the temperature dependent drift in output voltage for each sensor. The recorded output signal from each transducer was corrected for ambient temperature based upon the individual calibrations. A l l manual or automated measurements of pressure were corrected for the height of the water 105 column within the tensiometer to determine the matric suction in the material surrounding the tensiometer tip. 5.5 R E S U L T S The main features of the TC sensor measurements are first described using the uncorrected field measured data prior to hysteresis and temperature corrections. The effects of hysteresis and ambient temperature upon the measured data are then assessed. The corrected data are then compared to the tensiometer measurements to make an assessment of sensor accuracy under field conditions. 5.5.1 Uncorrected Measurements Uncorrected matric suctions for three sensors are presented in Figure 5.4. These matric suction estimates are calculated using the laboratory calibration based upon the main drying curve with no hysteresis or ambient temperature correction. Three dominant features can be seen: near surface measurements are absent or erratic in winter; estimated matric suctions follow clear trends, but the individual measurements appear "noisy"; and estimated matric suctions at deeper sensors have steadily declined over time and estimates are no longer possible. Examination of the ambient in-situ temperature data indicates that the erratic measurements in early winter correspond to freezing temperatures. Once freezing conditions start, the sensors are not meant to provide reliable results, as the heat pulse is absorbed by the latent heat of fusion of water frozen in the sensor. A l l of the near surface sensors have frozen for two winters, up to 6 months each, but all have recovered and operated normally when in-situ temperatures recovered above zero in the spring. Closer examination of the "noise" in the raw data for a probe buried at 4.5 m depth indicates systematic daily variation of sensor output. Figure 5.5 presents the uncorrected 106 matric suction calculated using the main drying curve (A), a 24-hour moving window average of the values of calculated matric suction (B), air temperature (C), the in-situ temperature (D), and the variation of matric suction defined as the difference between an individual measurements and the 24-hour moving window average (E). The probe temperature at this location is isothermal and air pressure measurements within the pile indicate no daily variation during the 10 day interval. The daily variation of calculated matric suction is systematic, and is much larger than the measurement resolution. The variations appear to be opposite phase of the measured air temperature. Several components of the TC measurement system experience daily temperature fluctuation. The observed variation in measurement may result from air temperature effects on the resistance of the heater element wiring and the resistance of the sensor output wiring within the above-ground cable duct, or the effect of temperature within the instrumentation hut on the performance of the datalogger and voltage supply. Attempts were made to compensate for this daily variation. The variation of measured core temperature rise for each individual measurement from a 24-hour moving window average of the core temperature rise was plotted against the variation in individual measurements of air temperature from the 24 hour moving window average air temperature. A weak negative correlation was found, but when this correction was applied it did not remove the daily variation seen in Figure 5.5. The variation of individual core temperature rise measurements from the 24-hour moving window average was plotted against the absolute air temperature, but no correlation was found. Similarly, no clear correlation was found to the temperature of the instrument enclosure. It is assumed the daily variation is related to the use of a constant voltage source and includes effects of both air temperature 107 and datalogger enclosure temperature. This method of heating is not recommended. U . of S. TC sensors manufactured after the Beta-98 sensors were all calibrated and deployed with a constant current source. Data from these sensors do not exhibit the same daily variations (Shuai, 2001, pers. comm.). The field readings are averaged using a 24-hour moving window average of the measured core temperature rise to smooth the daily fluctuations. This does not address the possibility of a systematic bias due to the absolute values of average air temperature or whichever factors lead to the daily variation. The calculated matric suction values may therefore be systematically too high or low. The long-term decline in calculated matric suction (Figure 5.4, 4.5 m depth) can be related to the gradual saturation of the sensors as previously noted. The sensors located deep within the pile are constantly exposed to matric suctions less than 20 kPa. Air entrapped in the sensors during initial wetting causes the difference between the main wetting curve and the initial drying curve (Figure 5.2) measured in the laboratory. Over time, the air trapped within the sensors is replaced by water. The increased degree of saturation of the sensors while exposed to a constant matric suction is calculated incorrectly as a decrease in matric suction. The calculated matric suction drifts towards zero as the degree of saturation approaches the highest value reached during the laboratory calibration on the main drying curve. The laboratory calibrations can no longer be used to calculate a matric suction value from the sensor measurements once the degree of saturation exceeds the main drying curve. The relationship between sensor water content and the external matric suction was slowly changing during the entire installation period for those sensors exposed to low matric suctions. The laboratory calibration curves are only valid immediately following installation. 108 After two years, 15 of the 18 sensors installed have saturated to water contents above the main drying curve and calculation of matric suction is not possible. 5.5.2 Sensor Hysteresis and Ambient Temperature The data can now be examined to determine the effect of sensor hysteresis and ambient waste rock temperature on the measured data. Post-processing of the data was carried out in the following sequence. A 24-hour moving window average of measured core temperature rise was calculated. Approximations for the matric suction using the drying calibration curve were derived from this average. The appropriate F(\|/) ambient temperature correction curve was chosen from Figure 5.3 based upon this preliminary estimate of matric suction. The individual raw measurements of core temperature rise were then recalculated using the ambient temperature correction factor and the ambient probe temperature. The individual, corrected, core temperature rise measurements were used to calculate a 24-hour moving-window average of core temperature rise to remove the daily variation. The averaged core temperature rise was then used to calculate a matric suction using both the sensor drying calibration curve and the sensor wetting calibration curve. Comparison of these two calculated values gives an indication of the magnitude of hysteresis effects. Figure 5.6 presents the non-temperature corrected matric suction calculated using both the main drying (1) and wetting (3) calibration curves. The estimates shown are the individual measurements, prior to the application of a 24-hour moving-window average. There is considerable daily variation that masks the underlying changes in matric suction. The effect of daily averaging can be seen in the matric suctions calculated after application of the ambient temperature correction, and using the drying (2) and wetting (4) calibration curves. The data have been smoothed by the application of the 24 hour moving window 109 average. The data presented are for the 0.2 m depth probe shown in Figure 5.4. The probes at 0.2 m depth experience the largest changes in both matric suction and ambient waste rock temperature. The importance of sensor hysteresis can be illustrated by comparing the preliminary matric suctions calculated using the main drying calibration curve (1) and the main wetting calibration curve (3). The initial estimate of the matric suction calculated using the drying calibration curve ranges from 35 to 50 kPa, while the matric suction calculated from the wetting calibration curve ranges from 20 to 30 kPa. This demonstrates that sensor hysteresis must be addressed, or significant errors will result in the estimate of matric suction. The magnitude of the effect of ambient temperature is also seen in Figure 5.6 by examining the data in curves 1, 3 and 5. As shown in curve 5, the ambient probe temperature at 0.2 m depth drops from -20 °C to -3 °C over the period presented. The uncorrected matric suction calculated with the drying calibration curve (1) increases over this period from 35 to 50 kPa, while the uncorrected matric suction calculated with the wetting calibration curve (2) increases from 20 to 30 kPa. The matric suctions calculated by correcting for ambient temperature and averaging to remove the daily variation (2, 4) rise only slightly during the same period. There are several large rainfall events in August, 1999, which is the wettest month of the year at the site. There are rainfall events of similar magnitude on both October 10 th and 20 t h, 1999, and therefore the matric suctions near surface at these times are expected to be similar to the August values. The increase in the uncorrected matric suctions (1,2) is therefore due to the long-term trend in ambient temperature. The 10 to 15 kPa differences in uncorrected matric suctions between August and October are similar in magnitude to the 20 kPa difference between the wetting and 110 drying curves. The application of an ambient temperature correction is necessary, particularly at lower matric suctions. The combination of hysteresis and ambient temperature corrections would correct a drying calibration curve (1) estimate of matric suction during the October 20, 1999 rainfall from 43 kPa to an actual value of 18 kPa (Curve 4). 5.5.3 Comparison to Tensiometer Data The corrected matric suctions calculated from the main wetting curve and main drying curve are now compared to tensiometer data in Figure 5.7. Data from two of the three sensors installed at a depth of 20 cm depth within the pile and the rainfall records for the same time period are shown in Figure 5.7. The upper graph shows the sensor response presented in Figures 5.4 and 5.6. The middle graph represents the TC sensor data which most closely matched the tensiometer data. The total rainfall volume (mm), and the timing of the rainfall are shown by the vertical lines in the bottom graph. These graphs indicate the matric suction calculated from the drying (1) and wetting (2) laboratory calibration curves, with correction applied for ambient temperature, and averaged using a 24-hour moving window average. The use of a 24-hour moving window average causes some smearing of changes in the measurements, but this is a small effect compared to the scattering in the unaveraged data seen in Figure 5.6. The manual tensiometer measurements (3, solid squares) and the automated pressure transducer tensiometer measurements (4, solid diamonds) are also shown. The data indicate TC sensor accuracy, time to first response, transition time and response to similar rainfall events are all highly variable. Sensor accuracy is easily determined by comparison of the middle and upper graphs. In the middle graph, the matric suction calculated using the wetting calibration curve approximates the tensiometer data 111 during periods of waste rock wetting, such as August 23 r . The matric suction calculated using the drying curve follows the trend in tensiometer data during the drying period prior to September 12 t h. This demonstrates the validity of using wetting and drying curves to describe sensor hysteresis. However, this TC sensor generally overestimates matric suction, and underestimates the magnitude of the changes in matric suction. The upper graph represents a more typical degree of agreement between TC sensor and tensiometer data observed at the other TC sensors. The upper graph indicates poor agreement in both the magnitude of matric suction, and the magnitude of changes in matric suction between the TC sensor and tensiometer measured matric suction. TC sensor response time and transition time can be assessed by examining the rainfall events labeled as A , B and C. Response time refers to the time to the first change in the measurement following a rainfall event. By definition, the transition time of a change in the sensor signal is the time required for the sensor measurement to have changed by 50% of the overall difference from the initial value to a final measured matric suction. The three rainfall events marked by (A) indicate rapid response of both TC sensors to the rainfall events. The apparent early response of the TC sensor readings is an artifact of the 24-hour averaging. The non-averaged readings indicate the TC sensors responded within hours to the onset of rainfall and their first response occurs at the same time as the tensiometers. In comparison, neither TC sensors responded to the September 9 t h rainfall event (B) for 12 and 48 hours after the rainfall event. This compares to 2 hours for the tensiometers. The TC sensors transition time was approximately 48 hours for event B , compared to 2 hours for the tensiometers. There was a strong drying trend after September 9 t h, as evidenced by the increase in matric suction recorded by the tensiometers. The matric suction estimated by both TC sensors 112 continued to fall, even when the matric suction measured by tensiometer had been rising for 96 hours. The unpredictable response to similar events can be assessed by examining rainfall event C on September 24 t h, in comparison to three events in August, 1999 marked as A . This event, C, represents the.largest rainfall event monitored during this period. The matric suction measured with the tensiometers indicate similar matric suctions prior to the rainfall events for both the A and C events. The changes in matric suction during these rainfall events, as measured by the tensiometers, are also similar. The data in the top graph indicate three clear responses to the August rainfall events (A). However, a TC sensor response is not discernable in the top graph during the September 24 t h rainfall (C). The estimated matric suction actually increases. In the middle graph, the TC sensor clearly responds to each of the three August events (A). The TC sensor response on September 24 t h (C) appears to lead the rainfall event. Examination of the unaveraged data on the 24 t h indicates no changes in TC sensor output greater than the systematic daily variations noted earlier. The apparent decrease in matric suction in the averaged data shown in Figure 5.7 is unrelated to the rainfall event. 5.6 D I S C U S S I O N The results presented imply that TC sensors will provide accurate and responsive results only for certain materials and conditions. As described below, properties of the TC sensor ceramic must be matched to the properties of the surrounding material. Improper matching will result in problems with long term saturation of the sensors, and slow or erratic sensor response. 113 The observed long-term saturation of the sensors will affect any of the TC sensors designs currently available, although the timing and magnitude of the effect will vary for each ceramic. The point of significant departure of the initial drying curve from the main drying curve (Figure 5.2) should be determined and represents the limiting matric suction for long-term operation, even i f short-term laboratory calibration is possible at lower suctions. The limiting value for long-term operation of the sensors used in this study is 15-20 kPa. Long-term results wil l be possible i f sensors are installed in materials in which the matric suction is expected to always exceed the limiting matric suction. The magnitude of the drift in sensor measurement is represented by the difference between the initial and main drying curves. Above the limiting matric suction, this drift is small compared to the differences caused by hysteresis, which is the difference between the main drying and wetting curves. Sensors may provide interpretable results i f the sensors are placed in materials where the matric suction exceeds the limiting value on a weekly or perhaps monthly basis. A sensor will drain to a water content close to the main drying curve i f it is dried out to matric suctions greater than the limiting value, it will . The majority of the pores which entrap and exchange air are emptied and the re-wetting process will then more closely follow the main wetting curve. If this occurs regularly, then sensor drift will be removed sufficiently often to allow the data to be interpreted. Sensors should not be installed in materials where the matric suction is constantly below the limiting value. Sensors will over-saturate, and interpretation of the output will be difficult. A method to restore the sensors used in this study, which are currently over-saturated, by over-heating the sensors and drying them out beyond the limiting matric suction value in-situ has been proposed to the U . of S. and wil l be tested in the laboratory in the future. 114 The field data indicate that matric suction calculated from the TC sensors is a damped, delayed, and sometimes unpredictable response to the matric suction in the CPE waste rock. This can be partly explained by the differences in the time scales of changes in matric suction, and the response time of the matric suction sensors. Matric suction in the CPE waste rock changes rapidly, as demonstrated by the tensiometer measurements, and wetting fronts propagate to five metres depth in as little as three hours. The waste rock behaves similarly to other coarse materials in which hydraulic conductivity drops rapidly with increasing matric suction. The hydraulic conductivity is -10" 3 to -10"4 ms"1 near saturation and drops to -10" 1 0 ms"1 at 20 to 30 kPa with an air entry value of <1 kPa. The rapid decrease of hydraulic conductivity with increasing matric suction means increases in matric suction are self-limiting as water phase movement of water downwards slows dramatically. The response time of a TC sensor is determined by the rate of water movement into or out of the ceramic. This is controlled by the hydraulic conductivity of the ceramic and the matric suction contrast between the sensor core and the surrounding material. The sensor 6 1 8 1 hydraulic conductivity has been estimated to range between 2x10" ms" at 1 kPa to 10" ms" at 30 kPa and lO'^ms" 1 at 400 kPa (Shuai et al., 1998). In the CPE waste rock, the matric suction changes are small, leading to low driving forces in or out of the ceramic. Experiments on sensor response time by Feng (1999) indicate 30 to 70 hours for sensor equilibration to a change in matric suction between 0 and 7 kPa, 20 to 100 hours for a change between 55 and 103 kPa change, and 100 to 200 hours for a change between 200 and 400 kPa. The shorter time corresponds to drying, and the longer times to wetting. In the CPE experiment, some wetting fronts arrive and recede faster than the equilibration time and the 115 sensor ceramic may be imprinted with first a wetting, then a drying trend. The distribution of water within the sensor ceramic is therefore a complex integration of the matric suction over the equilibration time. TC sensors will work best in materials where the sensor response time is always faster than the changes in matric suction brought about by the arrival and recession of wetting fronts in the material. It is advisable, but not sufficient to require that the permeability of the sensor always exceed the permeability of the surrounding material because both the geometry and driving forces of flow in the sensor, and the downward propagation of a wetting front in the surrounding material are different. A n estimate of the hydraulic properties of the material at the intended installation should be acquired, and this used to conduct preliminary modeling of the anticipated rate of change in matric suction. The slow response time of the TC sensors has one further implication. Matric suction is often monitored in conjunction with in-situ water content. Water content measurement methods such as time domain reflectrometry (TDR) respond immediately to changes in the water content of the surrounding material. TC sensor matric suction data should not be matched to water content data to estimate field soil water characteristic curves unless the response times and transition times of the two measurements are similar. 5.7 RECOMMENDATIONS The theoretical analysis of the effects of ambient temperature and the experience from two years of field application suggests the following: 1) Laboratory calibrations and field deployments should avoid the use of a constant voltage for powering the heating pulse. 116 2) The full heating and cooling curves for all laboratory calibrations should be recorded and kept for later use. 3) A n ambient temperature correction factor can be calculated from thermal conductivity measurements during calibration. 4) Air temperature and datalogger / power supply enclosure temperature should be recorded during field measurements. 5) A second order polynomial is used to fit the last twenty data points of the heating curve, and to calculate the final temperature during field measurements. The three coefficients of the fitted curve should be recoreded to subsequently determine field thermal conductivity and diffusivity values. 6) TC sensor measurements should be corrected for both sensor hysteresis and ambient temperature. 7) The initial drying curve and main drying curve of the sensors should be determined. TC sensors should not be installed in material where the matric suction is expected to be always less than the point where the initial drying curve and main drying curve diverge at low matric suctions. 8) Preliminary estimates of the range in the rate of change of matric suction should be obtained for the material at an intended installation site either by modeling or experiment. Sensors wil l operate best in materials where the changes in matric suction occur on longer time scales than the sensor response time. 9) Volumetric water content data and TC sensor matric suction data should not be matched to create field soil water characteristic curves unless the response times and transition times of the two measurements are similar. 117 10) Automated, unattended TC sensor measurements should be periodically confirmed by independent measurement of the matric suction using an alternate method. 118 Lead Wires Putty Cap Epoxy Seal I.C. temperature sensor Porous ceramic body 38.5 mm 28.5 mm \ H e a t e r resistor F igure 5.1: Schematic of TC sensor design (after Feng, 1999). 119 0) CO p i I I 11 i i | i i i i i i 11 j 1 10 100 1000 J I I M i l l 1 10 100 1000 Matric Suc t ion (kPa) F igure 5.2: Effects of hysteresis on: (A) TC sensor volumetric water content and (B) measured TC sensor core temperature rise. Simplified from data in Feng (1999). 120 Figure 5 .3: Correction factor for ambient soil temperature. The field measured TC sensor core temperature rise is multiplied by the correction factor to obtain the core temperature rise that would have been measured at 23 °C. 121 0.2 m Depth y-i 1 1 1 1 1 1 1 1/8/99 1/10/99 1/12/99 1/2/00 1/4/00 1/6/00 1/8/00 1/10/00 Calender Date (d/m/y) Figure 5.4: Preliminary estimate of matric suction based upon laboratory calibration curve for a drying TC sensor. 122 35 LU 30 00 Q < o 25 "nT a. r r 20 C 0) 15 = ? 10 •c E 5 -t d> re I— S 0 29/06/99 1/07/99 3/07/99 5/07/99 7/07/99 9/07/99 Calender Date (d/m/y) Figure 5.5: Daily fluctuation of estimated matric suction for a TC sensor installed at 4.5 m depth: matric suction calculated using laboratory calibration curve for a drying sensor (A, kPa); 24-hour moving-window average of estimated suction (B, kPa); air temperature (C, °C); ambient waste rock temperature (D, °C) and variation of estimated matric suction from the 24-hour moving-window average of estimated suction (E). 123 60 0 H 1 1 1 1—1 08/08/99 28/08/99 17/09/99 07/10/99 27/10/99 Calender Date (d/m/y) Figure 5.6: Demonstration of the effects of 24-hour moving window averaging, sensor hysteresis, and ambient temperature correction for a sensor at 0.2 m depth. Matric suction is calculated from individual readings using the laboratory calibration curve for a drying sensor (1) and a wetting sensor (2). Corrected matric suction is calculated using a 24-hour moving-window average of TC sensor output and using a drying calibration curve (3) and wetting calibration curve (4) and ambient soil temperature (5). 124 40 • 35 ro Q. 30 c o 25 o 20 3 CO o 15 10 ro S 5 0 40 • "ro 35 Q_ 30 c o 25 u 3 20 (/) 15 o +J ro 10 • 5 0 50 40 Sensor X: 20 cm depth Sensor Y: 20 cm depth E E. 30 jo 12 0 "Jo 10 03/08/99 17/08/99 31/08/99 14/09/99 28/09/99 12/10/99 Calender Date (d/m/y) Figure 5.7: Assessment of TC sensor accuracy, precision and response time for two sensors (X and Y) located at 0.2 m depth: estimated matric suction suction based upon a 24-hour moving-window average of sensor output and a drying sensor calibration curve (1) and a wetting sensor calibration curve (2); matric suction measured manually using a tensiometer (3, closed squares); matric suction measured using a tensiometer and pressure transducer (4, closed diamonds); and rainfall amount and timing. 125 CHAPTER 6: WATER FLOW IN UNSATURATED HETEROGENOUS POROUS MEDIA: A CONSTRUCTED WASTE ROCK PILE 6.1 ABSTRACT Weathering of minerals in mine waste rock can lead to mobile, water soluble weathering products. The flow of water through waste piles is one of the primary controls on the timing and rate of environmental loading. Preferential flow in an unsaturated heterogeneous porous medium such as waste rock leads to the potential for large spatial variation in leaching rates which prevents accurate prediction of loading rates. A large scale (8m x 8m x 5m high) experiment has been built upon a contiguous grid of 16 lysimeters to better characterize the flow of water in waste rock. The experiment has been operated for two and a half years. Material characterization from laboratory and in-situ data indicate strong hysteresis effects, and that measurement of representative unsaturated material properties is difficult. In-situ measurements of wetting front propagation are found to be a poor predictor of outflow timing and volume. The whole pile water balance indicates an average of 55% of precipitation is measured as outflow. Analyses of hydrographs and outflow volumes from individual lysimeters indicate the number of preferential flow paths and the total water outflow volume within preferential flow paths increases with increasing rainfall event magnitude but the total water outflow volume is more evenly spatially distributed. The scale of spatial averaging required to determine effective unsaturated flow parameters increases with increasing flow rate. 126 6.2 I N T R O D U C T I O N The exposure of minerals in mine waste materials to atmospheric oxygen leads to the weathering of primary minerals and the formation of secondary minerals and dissolved weathering products. The most common process is the oxidation of metal sulfides to produce sulphate, acidity and free metal ions. The mineralogy and geochemistry of the weathering process have been studied in detail (Jambor and Blowes, 1994; Malstrom et al., 2000), as have the processes of gas and heat movement in waste rock piles (eg: Pantielis and Ritchie, 1990, 1992, 1993; B-ennet et al., 1995, Birkham et al., 2001, Lefebvre et al., 2001 a/b). In comparison, water flow in waste rock is less well understood and represents the largest source of uncertainty in current efforts to predict environmental loadings of weathering products (Smith et al., 1995; Ericksson and Destouni, 1997; Malmstrom et al., 2000). Waste rock piles are typically placed and stored under unsaturated conditions. It is essential to understand how water is transported and stored in waste rock to predict the rates at which weathering products are released to the environment. The focus of this study is on obtaining detailed observations of the flow of water in waste rock. When later combined with information on primary and secondary geochemical reactions and solute transport properties of waste rock, this should permit insight to the coupling between fluid flow and the release of metals. A study of water movement in waste rock is also relevant to the general question of flow and transport in any heterogeneous porous medium. Mine waste rock is uneconomic or barren material moved from an underground or open-pit mining operation and typically deposited in piles adjacent to the mine workings. It is composed of broadly graded material from clay up to boulders several metres in diameter (Herasymuik et al., 1995; Smith et al., 1995; McKeown et al., 2000). Waste rock textures are 127 highly heterogeneous, from cobbles and boulders supported in a granular matrix, to clast-supported cobbles and boulders completely free of matrix (Herasymuik et al., 1995; Wilson et al., 2000). Piles may be constructed in lifts, and hence may also include internal compacted layers from haul truck traffic. Waste rock may be dumped on a flat surface (free-dumped) leading to random grain sizes and textures for each dump load. Alternately, waste rock may be dumped or pushed off the edge of the pile (end- or push- dumping) leading to cross-bedded structures down the face of the pile with interbedded layers of varying grain size. End- or push-dumping creates a grain size gradation from finer material near the lip of the lift to coarser material at the base (Herasymuik et al., 1995; Smith et al., 1995). Waste rock piles typically range from ten's of metres to several hundred metres high. Previous studies of water flow in coarse rock include small scale lab experiments (El Boushi, 1975; Dexter, 1993), large column experiments (Ward et al., 1983a,b; Frostad, 1999; L i , 2000), field studies of full scale piles (Gelinas et al., 1994; Smith et al., 1995; Ericksson et al., 1997), and numerical studies (Davis et al., 1986a/b; Ericksson and Destouni, 1997; Lopez et al., 1997; Newman et al., 1997; Gerke et al., 1998). Closely related to the problem of environmental leaching of waste rock is heap leaching, where ore minerals are purposefully leached from granular materials as the method of resource recovery (eg: Murr et al., 1981; Dixon and Hendrix, 1993ab; and Decker and Tyler, 1999). A review of these studies indicates there is no comprehensive data set with sufficient temporal and spatial detail to build and test mechanistic models for describing infiltration and metal transport through coarse-grained waste rock. The non-specific term preferential flow is typically applied to describe either any mechanism leading to flow or tracer movement faster than expected by an observer, or the 128 concentration of flow into spatially distinct areas. Preferential flow has been confirmed in waste rock by quick responses after the start of rainfall events: rapid changes of water flow rates in pile toe-drains; increases in water table elevations beneath piles; and changes in water pressure or temperature deep within waste piles (Smith et al., 1995). Channeling of flow to spatially distinct areas has been observed in small scale lab experiments on coarse materials by Elboushi (1975) and Dexter (1993), in larger scale column experiments by Murr et al.(1981) and L i (2000), and in dye tracer tests on waste rock piles by Bellehumeur (2001). The presence of preferential flow implies the potential for a significant spatial variation of leaching rates for different areas of a pile. The nature of this preferential flow is poorly characterized and creates uncertainty in efforts to predict leaching from existing piles or to make predictions prior to pile construction. Some degree of preferential flow in unsaturated porous media is common enough to be the rule rather than the exception (Flury et al., 1994). The mechanisms leading to preferential flow in natural soils include: variable infiltration at the soil surface; flow in macropore features such as large pores, cracks, wormholes and root holes; film flow in non-filled pore spaces; surface flow over large particles; instability of infiltrating wetting fronts and fingering of flow in homogeneous materials; soil hydrophobicity; spatial heterogeneity of soil properties; hysteresis effects; and the focussing of flow through the presence of large boulders or zones of fine grained materials. In this chapter, the non-specific term flow path is used to describe a location or mechanism of water flow when the exact nature is not known. A range of experimental methods have been used to study preferential flow. Small columns, or isolated soil blocks, have been used both for quantitative studies of flow and 129 transport (eg: Seyfried and Rao, 1987; Jardine et al., 1993; Sassner et al., 1994; Wildenschild and Jensen 1999a) and for qualitative visualisation of preferential flow mechanisms (eg: Booltink and Bouma, 1991). Some studies have investigated the spatial variability of flow using grids of contiguous collection lysimeters installed at the base of columns or soil blocks (eg: Wildenschild et al., 1994; Phillips et al., 1995; Boll et al., 1997; Kranz et a l , 1998). Small scale studies, however, do not encompass the patterns and mechanisms of preferential flow that are present in larger field plots (Schulin et al., 1987). Large scale studies of unsaturated flow in natural materials have been carried out in agricultural fields and hill slopes. In some studies, water movement is deduced from changes in water content and pressure monitored by intensive in-situ instrumentation, and tracer movement from in-situ methods or destructive sampling (eg: Jury et al., 1982; Biggar and Nielsen, 1976, Schulin et al., 1987; Butters et a l , 1989; Wierenga et al., 1991; and Rudolph et al., 1996) Other studies have been performed in fields with tile drains that capture outflow (Mohanty et al.1998; Villholth et al., 1998; Lennartz et al., 1999) or in hillslopes using cross-slope trenches to intercept flow (Feyen et al., 1999). The outflow at a tile drain or a hill slope base is a spatial average of all fluxes passing through the vadose zone to the base of the experiment and a correction must also be made for lateral transport at the base of the experiment (eg: Jury, 1975). It was decided to construct a large scale waste rock dump on a collection system composed of contiguous lysimeters. This design combines a large area experiment to investigate flow processes with direct measurement of flow and flux-averaged water chemistry at the pile base. By constructing the pile, both the difficulties associated with attempting to install instrumentation into an existing waste rock pile and trying to estimate 130 outflow at the pile base are avoided. Few conventional methods of drilling, coring or instrument insertion are successful in coarse, rocky waste rock. As the natural state of waste rock is a deposit formed randomly by large haulage equipment, this experiment mimics the textures found in existing waste rock. This chapter describes the design and construction of the waste rock pile. The pile has been monitored for a period of 2.5 years, during which time data were collected from natural infiltration events, a series of artificial rainfall experiments, and from a tracer test in which a conservative tracer was released during one of the rainfall experiments. After presenting the experimental design and instrumentation, the material properties are reviewed. The water balance for the pile is calculated and net infiltration is estimated. The changes in the patterns of infiltration over time are assessed, as is the flow behaviour observed at different spatial scales. Analysis of the outflow response to varying infiltration rates and the conditions that increase or decrease both flow rate and flow volume is undertaken. In Chapter Seven, the results of the tracer experiment are presented which permits a detailed analysis of the time scales of interaction of transient flow through the different hydrologic regimes of an unsaturated waste rock pile. 6.3 E X P E R I M E N T A L D E S I G N A N D C O N S T R U C T I O N The constructed pile experiment (CPE) was built at the Cluff Lake Mine in northern Saskatchewan. Air temperature at Cluff Lake ranges from -40 °C to +35 °C with a mean annual temperature of 0 °C and an average annual precipitation of 455 mm, of which 305 mm occurs as rainfall. The site is semi-arid, with precipitation typically occurring as intense summer thundershowers or spring and fall frontal systems of lower intensity but longer 131 duration. Snow fall on the top of the waste rock piles is typically removed by wind and thus accumulations are low (<0.1m ). The CPE is intended to mimic the behaviour of the upper 5m of a much larger unsaturated waste rock pile. A simplified cross sections of the pile is shown in Figure 6.1 with a partial plan view of the experiment core in Inset C. The photograph in Figure 6.2 presents a side view of the CPE, including the instrument hut. The instrumented core of the pile has a footprint of 8 m by 8 m and is 5 m high. A height of 5 m is typical for lifts in existing waste rock piles at Cluff Lake. Outflow from the base of the pile is collected in a contiguous grid of 16 lysimeters each 2m x 2m (Insets A and C). The lysimeter base represents a compromise between the need to collect outflow from both the granular matrix material and from coarse grained areas with no matrix material. Water moves through the finer material under the influence of capillary forces and gravity. In contrast, the experiments by E l Boushi (1975), Prazak et al. (1992), Dexter (1993) and Bellehumeur (2001) all demonstrate water flow by surface flow over large particles. In regions of the pile where capillary forces dominate, water cannot flow freely from unsaturated material unless either a water table is formed or water is removed by suction. Maintaining the pile base under controlled suction was not technically feasible. Accordingly, a design leading to the formation of a water table at the base of the CPE was chosen (Figure 6.1, inset A) . The pile is built on a contoured cement pad lined with a P V C geomembrane. Within each lysimeter is a layer of 9.5 to 25 mm washed gravel. Outflow from each lysimeter is separately piped to an instrumentation hut. The finer waste rock immediately above the gravel will wet to a higher saturation than would normally be encountered in a thicker unsaturated pile. Those flow paths dominated by capillary forces wil l therefore be affected by the base of the pile, 132 and these flow paths may vary from those in a higher pile. The gravel is coarse enough to not retain or delay water which flows from any free-draining flow pathways where capillary forces do not control the rate or direction of flow. The presence of the lysimeter base will have no effect on these non-capillary flow paths, and these flow paths wil l be the same as i f the pile was higher. The CPE includes impermeable sides from the base of the lysimeter grid to the pile surface to ensure a complete water balance. Only by limiting lateral diversion using these walls will a zero-pressure lysimeter successfully capture capillary dominated water flow (Bews et al., 1997). Plywood lined with a 60 mil HDPE geomembrane isolates the central core of the CPE. The walls may limit the lateral range of spatially-distinct flow pathways. Closed cell polyethylene foam strips were attached to the geomembrane to reduce gaps at the waste rock-wall contact and re-direct any water flowing down the wall back into the waste rock. A skirt on the exposed wall prevents any water flowing down the wall from above the pile surface. Although these measures cannot guarantee the absence of wall flow, they act to minimize it. The top surface of the CPE was finished to resemble other piles on site which were free-dumped and then contoured to a smooth, level surface. No attempt was made to create a sloped or compacted surface, no vegetative growth was allowed, nor was any cover material used to alter infiltration. 6.3.1 Waste Rock Composition And Internal Pile Structure The waste rock was mined from the DJ Extension open pit at the Cluff Lake mine in the fall of 1996. It is composed of aluminous gneisses and granitoids from the Precambrian Earl River and Peter River Gneiss Formations. Associated with the target uranium 133 mineralization are sulphides of iron, copper, lead, zinc and molybdenum, with localized concentrations of nickel and arsenic minerals. Despite low overall sulphide content, < 0.64%, the total dissolved solids concentrations of in-situ water samples are as high as 50,000 mg/L. Further information on the composition and weathering of the material is available in Hollings et al. (1999,2001). The waste rock was exposed to natural weather conditions from mining, until placement in the CPE. Waste rock was placed within the CPE during July and August of 1998. Delicate instrumentation within the pile, and the construction of the impermeable walls around the lysimeter grid required reworked placement of previously dumped waste rock. Final placement of the rock within the experimental core was undertaken using a tracked excavator (1 m 3 capacity). The waste rock was randomly placed and the texture approximates the range of textures seen in excavations of existing waste rock piles on site (Bellehumeur, 2001). The pile contains regions that are matrix-supported and regions with matrix-free cobbles and boulders. The internal structure of the CPE represents the simplest grain arrangement possible for randomly dumped waste rock. No attempt was made to reconstruct haul truck traffic surfaces, or the cross-bedding structures and the extreme separation of grain sizes down dump faces that result from end- or push-dumping. Prior to pile construction, a backhoe was used to make a 15 m blended composite from thirty 0.5 m 3 sub-samples of the 2600 m 3 of waste rock used to build the CPE. A i m 3 sample of this composite was collected using the backhoe bucket (0.1 m 3), stored in five 200 L drums and analysed for grain size (Rowlett et al., 1999). The maximum grain size of this sample, ~0.6 m., was determined by the diameter of the drum opening. Sixty 5 litre samples were collected on a regular grid during pile construction using a hand shovel. Maximum 134 grain size in these samples was -0.1 m. The range of grain size distributions of the grab samples is shown in Figure 6.3, along with the full grain size curve of the composite sample. The l m 3 composite appears coarser as it has a larger sample support volume and grain size cut-off and therefore includes greater numbers of particles greater than 0.1m. When the data are re-plotted for the <5mm fraction only, the composite sample plots in the middle of the grab samples. The waste rock contains boulders up to 1.5 metres in diameter, and thus any practical measurement of samples for grain size necessitates the selection of sub samples. A large number of small volume grab samples better defines the envelope of grain sizes in the <5mm size fraction, but poorly represents the larger fractions. Any comparison of grain size curves for waste rock requires knowledge of the sampling method, support volume, and the maximum grain size sampled. According to the grain size classification system adopted by Dawson and Morgenstern (1995), this pile falls at the boundary between a soil-like and rock-like pile. 6.3.2 Instrumentation The CPE was instrumented to permit the determination of water movement across the soil-atmosphere boundary, within the waste rock, and out of the waste rock at the experiment base. An automated weather station is located on top of the pile to determine wind speed and direction, air temperature, relative humidity, net radiation and rainfall. Nine pairs of mini-lysimeters were installed across the surface to permit periodic direct measurement of actual and potential evaporation to complement and calibrate evaporation calculated from the weather station. A rainfall simulator was deployed to create artificial rainfall events of controlled rate and duration, and the later application of a conservative tracer. An oscillating lawn sprinkler 135 was found that supplied rainfall rates with a uniformity better than that obtained with a spray nozzle simulator (eg: Holden et al., 1995). The interruption in application rate caused by the sprinkler oscillation is acceptable i f the interruption interval is small (Barnett and Dooly, 1972; Young and Burwell, 1972; Holden et al., 1995). Artificial rainfall events were constrained to periods of very low wind speed to minimize systematic drift of the spray pattern. Water was supplied from a nearby lake. Instrumentation within the pile was installed during waste rock placement within three profiles during waste rock placement (Figure 6.1). Water content, matric suction, in-situ temperature, matrix water chemistry, pore gas pressure and pore gas chemistry are monitored. The instrumentation used to measure water content, matric suction, temperature and matrix soil water chemistry (Figure 6.1, inset B) either require direct contact with finer waste rock as part of their principle of operation, or would be damaged during construction by cobbles or boulders. Some grain size selection was required to remove particles >25 mm immediately adjacent to instrument locations and thus these instruments monitor conditions in the matrix-supported waste rock. There are no existing instrument technologies for clast-supported, matrix-free areas. Instruments were only placed on one side of the CPE to permit unrestricted placement of waste rock on the other side. Automated measurements of volumetric water content are conducted using time domain reflectrometry (TDR). Monitoring is carried out at a 20 to 60 minute interval depending on the time of year. The high concentrations of dissolved solids in the soil water necessitated the use of three rod TDR probes employing a high resistance coating to overcome high signal attenuation. Details of the TDR design, calibration and temperature corrections are presented in Chapter 3 and 4. Water content is monitored manually using a 136 neutron probe through aluminum access tubes placed during pile construction. Matric suction is monitored year round using thermal dissipation sensors (Feng and Fredlund, 1999; Feng 1999, Chapter 5 ). Tensiometers with pressure transducers were installed near surface to provide independent measurements of matric suction when air temperatures are above freezing. Tensiometers are measured both manually and using automated pressure transducers. Water outflow rate and water outflow chemistry are monitored within the instrumentation hut. The outflow pipes leading from the experimental base to the instrumentation hut are heat-traced to permit year-round operation. Water outflow rate is monitored using sixteen tipping bucket rain gauges. Outflow water is passed through a cascade of mixing cells, where the water is directed through containers of increasing volume (0.2 L , 2 L , 25 L). The outflows are then combined and directed through a 2 L , then 100 L container prior to discharge to waste. Both instantaneous grab and longer term composite samples can be obtained without post-sampling compositing. In-situ water chemistry is monitored through the periodic manual extraction of water samples from suction lysimeters placed within the pile during construction. 6.4 FIELD ACTIVITIES AND DATA ANALYSIS During placement of the waste rock, no rainfall was recorded at the CPE, and significant evaporation occurred from waste rock surfaces exposed during pile construction. The pile instrumentation was connected in August and September 1998, and the CPE was fully operational in September, 1998. Automated data collection has been running continuously since then. The top surface is left open to natural precipitation. Limited outflow from three lysimeters was recorded from September to December, 1998 and outflow 137 recommenced in April 1999 (Nichol et al., 2000, Appendix A). A l l lysimeters were not flowing until August 8, 1999. The CPE surface freezes in November and thaws in March with the melting of the winter snow cover. Temperatures measured deeper within the pile indicate the advance of freezing conditions to 1.75 m depth within the pile, with the deeper waste rock freezing in February and thawing by April. On September 24, 1999, a conservative tracer (LiCl) was applied to the surface of the pile as a single rainfall event of three hours duration. Intensive outflow sampling was carried out following the tracer application at times scales as short as 5 minutes. Unlabelled rainfall tri t h events were created on October 10 and 20 , 1999. Four additional unlabelled rainfall events were created in July 2000 with similar sampling intensity. The pile surface is always uncovered, and additional natural rainfall events occur. Outside of the main field work periods, the CPE was maintained by on-site staff. Outflow chemistry was sampled on a weekly or three-times-weekly depending on the flow rate variability. 6.4.1 Analysis Methods Rainfall data were compiled from the two rain gauges located on top of the pile, with additional information from another nearby automated weather station at the site airport (-500 m away), from manually determined snow depth estimates, and from artificial rainfall amounts measured using rainfall collection cups deployed on the pile surface. The uniformity of artificial rainfall events was measured calculated using the Christensen Uniformity Coefficient (CUC) (Christensen, 1942) which approaches 100 for uniform application: 138 1 Y\Xi-X n Cu =100 1- n [1] X V J where X i is the ith measurement, X is the mean rainfall and n is the number of measurements. Time domain reflectrometry measurements of water content were estimated to be accurate to within 4%, partly due to variability in the TDR probe design (Chapter 4). Relative changes in water content are more accurate, and changes < 0.5 % are detectable. The measurements of matric suction using thermal conductivity sensors were corrected for hysteresis and variation in ambient in-situ temperature using laboratory-derived correction equations (Feng and Fredlund, 1999; Feng 1999, Chapter 5). Short term sensor responses were sometimes unrepresentative of in-situ conditions and long term accuracy was poor due to sensor drift. Response times of the thermal dissipation sensors was varied from hours to days, but generally was too slow to match the rapid movement of wetting fronts through the CPE (Chapter 5). Tensiometer measurements were successful, and response times were rapid (seconds to minutes). Raw measurements of the timing of individual tips of the tipping bucket rain gauges were converted to outflow rate using moving window averaging and gauge specific calibrations. 6.5 R E S U L T S A N D DISCUSSION 6.5.1 Material Characterization Any study of water flow in unsaturated material requires the measurements of the soil water characteristic curve (SWCC) and hydraulic conductivity curves (HCC) of the material. Estimates of the SWCC and HCC of the waste rock have been made from both laboratory 139 and field data (Figure 6.4). These estimates emphasize the difficulty in measuring the properties of heterogeneous waste rock. SWCC's of the <5mm fraction of the coarsest, middle and finest grab samples of the 60 collected were measured in a large diameter (0.15m) tempe cell. The <5mm fraction was selected as the effect of particles greater than 5 mm on the SWCC can be empirically corrected for using the method of Yazdani et al. (2000). Samples were placed in the tempe cell as a saturated slurry. Figure 6.4A presents the drying curves (solid symbols) obtained during drying from saturation to 80 kPa matric suction, and the re-wetting behaviour (open symbols) obtained from 80 kPa to 2 kPa. Continuous functions were fit to the drying data using the method of Fredlund and Xing (1994). Saturated porosity ranges from 21 to 23%, and air entry values range from 1.0 to 3.3 kPa. Water contents remain at 10-12% by volume (50% saturation) at 100 kPa, indicating significant water retention by the waste rock forming the fine grained tail of the grain size curve (Figure 6.3). The wetting curves show a large displacement from the measured drying curves, indicating the potential for significant hysteresis at low matric suctions. In-situ SWCCs were derived by combining water content measurements by TDR and matric suction measurements by tensiometer for instrument pairs installed near the surface of the CPE. Figure 6.4B presents an example of typical field data recorded at a single instrument location over a three week period in August 1999. The right hand edge of the clustered data represents the conditions recorded during extended drying periods, and a boundary drying curve was manually fitted this data. This data set incorporates wetting trends from multiple wetting events, shown by the sparser data to the left of the boundary drying curve. The duration of wetting events was often short compared to the sampling interval of automated TDR and tensiometers measurements (15 to 20 minutes) and thus 140 fewer data points could be collected. Short term wetting and drying during rainfall events progressed exclusively on wetting and drying scanning hysteresis curves, and no boundary wetting curves could be determined at any of the probe locations examined. Detailed examination of the matched data collected during wetting events indicates trends in the data which are not possible in the case of a uniform wetting front. For example, in response to an infiltration event, water content increases while matric suction is increasing. These responses are created when the wetting front arrives separately at the tensiometer and TDR instruments located approximately 0.2 m apart. Wetting or drying scanning hysteresis curves can therefore not be reliably determined from in-situ instrumentation during short term infiltration given the possibility for mismatch of the data. Figure 6.4C presents the estimated boundary drying curves from eight instrument locations for the same period as Figure 6.4B. Using the right-hand boundary limit of the matched data should remove the short-term mismatches in water content and matric suction data observed during wetting events (Figure 6.4B). Only partial curves can be determined from the range of conditions occurring at each location. A l l of the in-situ drying curves are located to the left of the laboratory curves, indicating changes in water contents at lower matric suctions. This indicates the presence of larger pore spaces in the field waste rock than in the tempe cells. Water contents are also generally lower than the laboratory data which is consistent with the presence of particles greater than 5mm. These particles do not contribute to water retention at higher matric suctions but replace porous granular matrix with impermeable material and therefore lower total water content (Yazdani et al., 2000). Data are sparse below 1 kPa for all curves and the data in this region represent wetting and drying 141 scanning behaviour. Although some curves exhibit breaks in slope at suctions of 0.2 to 0.6 kPa this does not indicate the an accurate air entry value. Comparison of the laboratory and field measured SWCC's indicates both the laboratory SWCC method, and the field SWCC data fail to adequately represent near-saturated conditions. Bellehumeur (2001) measured in-situ dry densities of 1630 to 1690 kg/m 3 and porosities of 36 to 39% for a nearby pile at Cluff Lake. Measurements of 1550-1800 kg/m 3 dry density and 3 Ito 34% porosity were made in laboratory calibrations of TDR probes (Chapter 4). When the waste rock is mixed and compacted at typical field measured water contents (10-15% by volume) the smallest sized particles tend to form into 1 to 2mm diameter aggregates that resist compaction and create inter-aggregate porosity. Placement of the waste rock sample as a slurry, or vacuum saturation of the waste rock, is required to measure the boundary drying curve from saturation to ensure no entrapped air in the sample and a repeatable starting point. Placement of the laboratory samples as saturated slurries in the tempe cells led to de-flocculation of these aggregates, no clast supported textures, higher dry densities (1980 kg/m3), lower saturated porosities (22%), and higher air entry values than in-situ waste rock Vacuum saturation of samples compacted at residual water contents (10-15%) led to similar de-flocculation of the fines aggregates and the collapse of the additional porosity created by these aggregates. The saturated conditions necessary to measure the boundary drying curve destroy the waste rock texture. Further pores with larger diameters are created under field conditions by particles greater than 5mm diameter, which create pores greater than 1mm when these particles are clast-supported (Kenny et al., 1985 ). Field SWCC curves indicate highest measured water contents of 20 to 23%, implying greater than 10% air filled porosity even under the highest flow conditions. This pore space may still 142 permit water flow by film flow and flow in partially filled pores (Tuller and Or, 1999) but these flow processes are not represented in the SWCC. The hydraulic conductivity curve (HCC) relating water content or matric suction to the unsaturated hydraulic conductivity can be estimated for a granular porous medium from a SWCC and an estimate of saturated hydraulic conductivity using methods based upon the representation of the porous medium as a continuum of capillary tubes (Fredlund et al., 1994). Measurements of saturated hydraulic conductivity were conducted on waste rock from several waste piles at the Cluff Lake site with similar grain size distributions using constant head and falling head tests. The results varied over 7 orders of magnitude from 4.6 x 10"5 ms"1 at 1660 kgm"3 to 10"12 ms"1 when compacted to 2300 kgm"3. Saturated permeability measured in the laboratory is a poor representation of field conditions without detailed knowledge of in-situ densities. Therefore, estimates of field H C C curves were derived from selected SWCC field curves by matching the HCC curves to field measured unsaturated conductivities, rather than laboratory measured saturated conductivities. In-situ field measurements were examined for artificial rainfall events where the moisture and matric suction measurements reached steady state during the event. Water contents between 12% and 23%, and matric suctions between 0.1 and 1 kPa, were recorded by in-situ probes near surface under rainfall at 4.9 x 10"6 m/s. The measured partial SWCC field curves were fit with an estimated SWCC using Fredlund and Xing (1994), then an unsaturated permeability estimation was performed using the method of Fredlund et al. (1994) (Figure 6.4D). The saturated permeability used to generate the HCC was adjusted until the curves matched the in-situ field data. As these data are collected under wetting conditions, it represents scanning hysteresis behaviour, not boundary drying behaviour. These curves are 143 for the matrix supported areas of the pile and are valid for permeabilities below the match point with field data, 5 x 10"6 m/s. Near saturation, the unsaturated permeability is not well estimated by these methods of calculation and is better measured empirically. 6.5.2 Whole Pile Water Balance The water balance of the CPE, and its variation in space and time, can be used to assess flow mechanisms in waste rock, and to determine long-term estimates of net infiltration. A summary of the average rainfall and precipitation statistics and the precipitation during the experimental period is provided in Table 6.1. The periods when artificial rainfall events were created in 1999 and 2000 exceed the monthly maximums recorded for these months under natural rainfall conditions by up to 20%, but monthly rainfall is still low in comparison to more temperate climates. Yearly totals indicate both 1999 and 2000 were below average precipitation years without the added artificial rainfall and 2000 was a below average year with the artificial rainfall. Figure 6.5A presents the daily precipitation record over the experimental period. Artificial rainfall events are shown as lines in bold. A summary of the artificial rainfall events created in 1999 and 2000 is presented in Table 6.2. The sprinklers used for artificial rainfall events proved successful i f used during low wind conditions. The CUC's compare well with the CUC's obtained in other field studies using large arrays of overlapping spray nozzles (eg: Miller, 1987; Chow and Rees, 1994). The intensities and durations of artificial rainfall events were based upon the rainfall rates delivered by the sprinklers used in the rainfall simulator, rainfall return period statistics for the Cluff Lake site, and rainfall return statistics for other wetter climates. The application rates are similar to the range of rates used in heap leaching (eg: Murr et al., 1981; Pantelis and Ritchie, 1993; Decker and Tyler, 1999). 144 The return periods (Hogg and Carr, 1985) for the artificial events shown in Table 6.2 indicate these events represent large storms for the semi-arid Cluff Lake site. It is the duration of these events that makes them statistically large events, not the rate. Natural rainfall events with similar daily rainfall totals (eg: August 28 & 29, 1999, 35 mm; August 26, 2000, 31.3 mm) occur as three or four intense thundershowers of 15 to 30 minutes duration with rainfall rates up to 80 mrn/hr, spread over a period of several hours and joined by low rate (<1-2mm/hr) rainfall or no rainfall. Low rainfall rates (<l-2 rnm/hr) occur under natural conditions during spring and fall frontal rains of 12 to 24 hours duration. Figure 6.5B presents the outflow flux of the whole pile as a flux rate, equal to flow rate per unit area (m3s"1m"2). A summary of the outflow volumes for the CPE on a monthly basis is shown in Table 6.1. The outflow starts at 10"11 ms"1 during the wetting up period prior to August 1999 when not all lysimeters were flowing. Following August 1999, the flow rate varies over three orders of magnitude (10"9 to 10"6 ms"1) between peak summer flow and the lowest flow at the end of winter. The highest flow rates generally correspond to the periods during which artificial rainfall events were created but several natural rainfall events created outflow responses of similar magnitude indicating the outflow rates created by the artificial events are realistic for the Cluff Lake site. Yearly water balances are calculated for 1999 and 2000 using the cyclic period for flow behaviour. Water drains slowly from November until late March following spring thaw. To estimate net infiltration, precipitation from March to November was matched to the outflow from mid-March to the following mid-March. Measured water contents indicate no net change in water content between March 2000 and March 2001 and thus recorded outflow equals net infiltration. Outflow was 57% of precipitation in 2000. Outflow was 56% of 145 precipitation for 1999, but net infiltration is estimated to be higher as storage within the pile was being replenished following drying during construction in 1998. Net infiltration was also estimated for natural rainfall only. An estimated drainage curve (under the assumption of no subsequent infiltration) was derived from the last artificial rainfall event on July 23, 2000 to March 2001. The estimated drain down was subtracted from the measured outflow for the period from July 23, 2000 to March 15, 2001. Net infiltration was estimated as 55% of precipitation for this northern Saskatchewan climate. These estimates represent long term average net infiltration to an uncompacted waste rock surface with no vegetation cover. Net-infiltration estimates for specific rainfall events are presented in later sections. 6.5.3 Individual Lysimeter Data The division of the pile base into sixteen lysimeters allows for the investigation of smaller scale processes occurring within a larger volume of waste rock. Outflow volume data show a large degree of spatial variability at the 2m x 2m scales. Figure 6.6A illustrates the variability in rainfall and snowmelt on the surface of the CPE for the period from September 1998 to March 2001. This plot is calculated by including the measured spatial variability of artificial rainfall events, and assuming natural rainfall events are evenly distributed. The variation in total outflow volume for each lysimeter is shown in Figure 6.6B for the same period. The data indicate a variation of 8% between the lowest and highest recording lysimeters for rainfall, and of 400% for outflow. Individual estimates of net infiltration range from 30% to 121% of precipitation for the sixteen different lysimeters with a standard deviation of 23%. This wide range of net infiltration estimates between 2m x 2m areas of the pile indicates that the individual 2m x 2m lysimeters are poor predictors of net infiltration. 146 The spatial distribution of outflow was examined to determine changes in the distribution of outflow volume between lysimeters with time. Figure 6.7A presents the monthly total outflow per lysimeter as a percentage of the monthly total for all sixteen lysimeters. The wetting up period from September 1998 to August 1999 and has been excluded. Spring snowmelt and early spring rainfalls occur when the pile is under the driest conditions. Outflow responses to these events can be weeks apart for different gauges, leading to wide variability in outflow volume on a monthly basis. Between 1999 and 2000, lysimeters 6 and 13 increased their share of the cumulative outflow at the expense of the slower flowing lysimeters. Their monthly outflow volume shows greater variability at the higher water flow rates experienced during summer months than at the low flow rates recorded during the winter drain down. Figure 6.7B presents the same calculation as Figure 6.7A, with lysimeters 6 and 13 removed. The overall range of variation in outflow volume for the remaining lysimeters is similar throughout the experimental period. Within the range, individual lysimeters show greater variability between 1999 and 2000. These data indicate at least one process leading to greater outflow volume reporting to lysimeters 6 and 13 that is more active at the higher flow rates during summer months and that increased from 1999 to 2000. Outflow as 120% of precipitation at lysimeter 6 can only be caused by redistribution of precipitation falling on other areas of the pile surface to within the footprint of this lysimeter. This may occur at or below the pile surface. Surface runoff and the subsequent infiltration at low elevation points on the waste rock surface is one possible mechanism. Minor surface ponding was observed at the highest flow rates created during artificial rainfall events (16 to 19 mm/hr), indicating these rainfall rates exceeded the local infiltration capacity at points on the pile. No significant ponding or runoff was 147 observed at lower rates. No runoff was allowed to exit the isolated central block of the experiment and any runoff generated infiltrated at lower elevations on the pile surface. Intense physical weathering due to freezing and thawing conditions leads to rapid disintegration of the larger particles exposed at the waste surface. This may have led to decreased surface infiltration rates, and increased run-off between 1999 and 2000. A surface survey was conducted in August 2001, and the topography of the pile surface is presented in Figure 6.8. Boundaries between surface catchments are shown by thick grey lines. Settlement of the waste rock surface around the fixed neutron access tubes indicates approximately 0.15 m of average surface settlement has occurred over three years. The footprints of lysimeters 6 and 13 correspond to two of ten lower spots on the pile surface but this is not conclusive evidence that increased outflow volume is due to runoff. The two next highest flowing lysimeters, 10 and 11 do not have low elevations within their boundaries. Lysimeters 7 and 15 contain low elevation points with catchments of greater area than lysimeter 13, but are within the middle range of total outflow volume. The lowest volume, and slowest flowing lysimeters, 3, 5, 8 and 9 do not all correspond to higher elevations. Lysimeters 6 and 13 response faster and with greater volume to all precipitation events, even when rainfall intensity is below the infiltration capacity of the surface. Surface runoff is therefore a contributing, but not dominant, mechanism creating variation in the volume of outflow. 6.5.4 Outflow Hydrographs Outflow hydrographs for all thirteen artificial rainfall events, and the larger natural rainfall events were examined to investigate variations of flow rates and flow volumes between lysimeters in response to individual events. The time to the first hydrograph 148 response, peak flow rates, volumes produced and flow recession behaviour all vary. Figure 6.9 presents the flow response recorded within three lysimeters following the artificial rainfall event on July 18, 2000. Note that outflow rate is presented on a logarithmic scale to fit the three curves on a single graph. This event represents the CPE under high flow rate conditions but the general character of the outflow response is typical of all of the artificial rainfall events and of the larger natural rainfall events. Lysimeters 6 and 9 are included as they represent the fastest and one of the slowest flowing lysimeters, respectively. Lysimeter 10 represents the third highest volume lysimeter, and is typically within the highest five for peak flow rate. Three important features are illustrated in Figure 6.9. First, the initial changes in outflow rate, and the peak flow rate occur at different times at each lysimeter. Lysimeter 6 rises to peak flow within 4.5 hours of the start of the rainfall event, lysimeter 10 after 22.5 hours and lysimeter 9 after 33.5 hours. Those lysimeters recording the lowest flow rates and total volumes are the slowest to respond to any given event. The slowest responses to an artificial rainfall event were recorded October 20, 1999. The flow rate at lysimeter 6 rose after 9 hours to a peak after 13 hours. In comparison, lysimeter 12 rose after 43 hours to a peak at 151 hours. Second, the pre-event flow rates are also different, ranging by a factor of 3.5 from 2 to 7 x 10"8 ms"1 (Figure 6.9). This range rises to a factor of 18 at the peak flow. Figure 6.10 presents the ratio of maximum flow rate in the fastest flowing lysimeter to the slowest flowing lysimeter (open triangles) for the thirteen artificial rainfall events and several natural events. The ratio of the fastest flowing four lysimeters and the slowest flowing four lysimeters is calculated as measure of how the spread in measurements changes are a larger 149 cross sectional area of the pile is considered. This ratio is shown as open circles. These ratios are plotted against the average maximum flow rate calculated from the maximum flow rates of all sixteen lysimeters. This measure combines the effects of rainfall intensity, total rainfall volume and prior wetting conditions and is representative of the overall flow state of the pile. The ratio of maximum flow rate between the single fastest and slowest lysimeters varies between 4 and 30 at peak flows and from 4 to 6 for the four fastest and slowest lysimeters. The same ratios were calculated for the whole experimental period, which includes data at the end of long flow recessions. The ratio of the fastest and slowest lysimeter ranges from 4 at a whole pile flow rate of 2xl0" 8 ms"1 to 2 at the lowest flow rate of 10"10 ms"1 at the end of the winter period. The same ratio calculated for the fastest and slowest four lysimeters varies from 3.2 at 2xl0" 8 ms"1 to 1.3 at 10"10 ms"1. Figure 6.10 demonstrates that higher peak flowrate lysimeters are always flowing at higher flow rates and the spread of flow rates is larger during high rainfall, wetter conditions. The third distinct feature of Figure 6.9 is the presence of multiple arrivals. The non-specific term arrival refers to an identifiable increase in the outflow rate. Two distinct increases in flow rates are present in the hydrographs for lysimeters 9 and 10. Closer examination of the hydrographs reveals seven distinguishable outflow arrivals in lysimeter 6, and six in each of lysimeters 9 and 10. Multiple arrivals are present in all lysimeters and for all events examined. The number of distinguishable arrivals within individual lysimeters ranges from 3 to 12 between different lysimeters for the artificial rainfall events where influx rate is constant. During natural rainfall events, multiple arrivals may be the result of varying rainfall rate. The majority of distinguishable arrivals are prior to the peak flow, on the rising limbs of the hydrographs. In some of the early arrivals, it is possible to distinguish an arrival, 150 peak and the start of a recession curve prior to the next arrival. Changes in observed slope of the recession curve are present that indicate arrivals after the hydrograph peak. The lysimeter hydrograph is the integration of all these separate arrivals at scales below that of the lysimeter. These multiple arrivals suggest at least as many spatially distinct flow paths as there are discernable arrivals. The number of faster arriving preferential flow paths within each of the 2m x 2m lysimeters is small enough to produce an integrated response that is not a single smoothed curve. The number of distinct arrivals is higher in faster flowing lysimeters than slower flowing lysimeters, and the number increases between events with increasing maximum flow rate. In the unsaturated zone, the imposition of a higher input flux increases the geometry of the water-filled pore space into larger, more conductive pore spaces. The fastest arrivals may result from preferential flow due to spatial heterogeneity of properties, non-capillary flow or saturation on boulder surfaces, but the mechanism cannot be determined from flow data alone. The physical mechanisms leading to the early arrivals in the hydrograph are further investigated using the results of a conservative tracer test in Chapter 7. The outflow data from individual lysimeters were combined to determine the outflow response at larger spatial scales. Figure 6.11A presents the outflow hydrographs for four adjacent lysimeters added together to represent the four quarters of the pile, and Figure 6.1 IB presents the outflow for halves of the pile. Both figures include the flow rate calculated for the whole pile from all sixteen lysimeters. Note the scale on the flow axes is arithmetic. The range of pre-event and maximum flow rates decreases with increasing spatial averaging, as does the range in first arrival times. The maximum flow rates for the quarters span a range of 151 a factor of 6, and the halves span a range of a factor of 1.5. The whole pile hydrograph still contains distinguishable multiple arrivals. For the artificial rainfall events, averaging between 50 and 110 distinguishable arrivals in sixteen individual hydrographs leads to between 10 to 15 separate arrivals in the whole pile hydrograph. The 8m x 8m area of the CPE was therefore not sufficient to average enough of the preferential flow paths active at highest flow rates to produce a smooth outflow hydrograph for the whole pile. 6.5.5 Event Based Outflow Volume The data for specific rainfall events can be examined to learn more about the short-term net infiltration and distribution of outflow. The outflow volumes attributable to single rainfall events were estimated using recession curve data. Both the net infiltration rate, and the spatial variability in net infiltration are found to be functions of the rainfall event magnitude. Examination of the recession curves indicates that the hydrographs for each lysimeter converge to a similar recession curve unique to that lysimeter. Figure 6.12 presents the recession curves for lysimeters 6 and 9 for the artificial events which created the highest flow rate, September 24,1999, and the lowest flow rate, October 20,1999. The curves have been translated laterally such that the first rises in the hydrograph are coincident. Each event has a different short term behaviour, but all events converge to the same recession curve at longer times. The curve for lysimeter 6 converges to a common curve after five days, whereas in lysimeter 9 the recession curves become coincident after 10 days. These estimated recession curves were used to extend the pre-event hydrographs for selected artificial and natural rainfall events. The post-event hydrograph was extended to eliminate the volume from rainfall events after the selected event. The total volume between these curves represents the outflow from the individual event. Only those rainfall events 152 where the measured outflow rate preceding the selected rainfall event was declining along a similar recession curve to the outflow rate after the event were analysed. Hydrographs that were still rising from the previous rainfall event were excluded. For some periods, several rainfall events were combined into a single estimate when the hydrographs from adjacent events could not be separated. An estimate for outflow from the whole pile was obtained by the same method from the whole pile hydrograph. The results are presented in Table 6.3. Estimated total volumes for each event from the whole pile hydrograph, and from the sum of the sixteen lysimeters are within 4% agreement. Net infiltration for each event was calculated from the whole pile hydrograph. Estimated net infiltration ranges from 85% for the artificial rainfall event on August 16, 1999 to 51% for natural rainfalls in August 2000. The data from the spring of 2000 represent rainfall and snowmelt introduced under the driest, slowest outflow conditions where this analysis was possible. The estimated net infiltration rate from this event was 52%. The long term estimate of 55% net infiltration is composed of high net infiltration during short duration intense storms (50 - 84%), coupled with low net infiltration from low intensity, low duration rainfalls. Large rainfall events similar to those considered for this analysis account for 50% of precipitation. Events of less than 5 mm daily rainfall account for 25% percent of the total precipitation. Figure 6.13 presents the variability in total volume between lysimeters for the different events. Percentage deviation from the average value is plotted against lysimeter number, with data from individual events connected by lines The heavy line indicates the deviation of total volume over the entire experimental period from Figure 6.6. The pattern of water volume reporting to each lysimeter shows consistency between the precipitation events 153 monitored, from the smallest event in the spring of 2000 to the largest on September 24, 1999. The fastest flowing lysimeters 6 and 13 returned the highest volumes in each event and lysimeters 3,5,8 and 9 were consistently lower volume in each event. Others vary from event to event. Figure 6.13 indicates that the variation in outflow volume created by increased surface run-off to lysimeters 6 and 13 during high rainfall rate events is small in comparison to the variation between lysimeters caused by spatial heterogeneity of the waste rock below the surface. Figure 6.14 presents the ratio of the volumes produced by the lysimeters reporting the highest and lowest volumes (open triangles) and the ratio of volumes from the four lysimeters reporting the highest and lowest volumes (open circles) as a function of the average maximum flow rate of the pile. The ratio of the highest and lowest decreases from 6 at the lowest flow rates to 3.5 at the highest and the ratio of the four highest and lowest decreases from 2.5 to 2. The range in maximum flow rate increases with event intensity for large rainfall events but the range of variation in outflow volume between lysimeters decreases. Newmann et al. (1997) and Wildenschild and Jenson (1999a/b) demonstrated that preferential flow due to spatial variability in material properties occurs in different spatial locations as a function of flow rate. Coarser grained areas conduct greater flow under higher infiltration rates and wetter conditions. As previously noted, the total number of distinct arrivals in large rainfalls increases with increasing event intensity, indicating more fast preferential flow paths are active. The spatial distribution also changes. The number of distinct arrivals in the fastest four lysimeters is twice the number in the slowest four lysimeters for the smallest events. As the flow rate increases, the ratio of the number of distinct arrivals in the fastest and slowest four lysimeters is reduced to 1. With increasing 154 flow rate, more fast preferential flow paths are active and the spatial distribution of preferential flow paths becomes more even between lysimeters, the lateral redistribution of water is reduced, and lysimeter volumes become more even. The preceding observations of the changes in preferential flow behaviour and volume for individual events have been derived only for those events large enough to cause distinct flow arrivals. Preferential flow occurs in those areas of the pile with the highest unsaturated hydraulic conductivity under the given infiltration rate. Under dryer, higher matric suction conditions, it is the finer grained sized fractions that retain water and have a higher unsaturated hydraulic conductivity (Newmann et al., 1997; Wildenschild and Jensen, 1999a/b). In the CPE, the smaller infiltration events cannot be analysed because they create slow wetting fronts in the granular matrix. These are overtaken by the rarer, larger events before they reach the pile base. It is therefore not possible to directly determine the distribution of volume and the hydrograph responses from single rainfall events with low infiltration rate and duration. However, the winter drain down data can be examined for evidence that the spatial distribution of water within the pile may be different at lower flow rates. Each lysimeter has a different recession curve at higher flow rates and each lysimeter starts the winter drain down period at different flux rates due to the spatial re-distribution of water at higher summer flow rates. The outflow flux rate for the fastest flowing lysimeter, the slowest flowing lysimeter and the whole pile are shown in Figure 6.15 A and 6.15B for the winter of 1999-2000 and 2000-2001. These hydrographs have been translated laterally so that they match at the lowest flow rate recorded. For both years, all three curves are indistinguishable at flux rates lower than 2.2x10"9 ms"1. Comparison of the 1999-2000 and 2000-2001 data for the whole pile (Figure 15C) indicates the two are identical for flux rates 155 below 2.0 xlO" 9 ms"1. The drain down behaviour of the pile below 2.0 xlO" 9 ms"1 is identical between different lysimeters, and at different physical scales. The flow rate of 2.2 x 10"9 ms"1 corresponds to an evenly applied net infiltration of 63 mmyear"1 net infiltration which corresponds to 14% of average annual precipitation or 21% of rainfall. At the end of winter, flow will only be occurring in the finest grain sized portions of the granular matrix material, where capillary forces dominate. It is inferred from the convergence of the drain down curves that at low flow rates, the spatial scale of variation in flow within the granular matrix is reduced and a 2m x 2m footprint is sufficient to average these variations in flow rate. 6.5.6 In-situ measurements and direct measurements of outflow. We now compare the progress of changes in water content through the pile with changes in outflow at the base of the pile. TDR measurements of water content of the granular matrix (Chapter 4) have been recorded in three profiles in the CPE. When the field data are examined in detail, TDR measured water content is found to be a poor predictor of changes in the outflow hydrograph and the volume of water outflow. For the thirteen artificial rainfall events and selected natural rainfall events, the arrival and peaks of wetting fronts at each TDR probe location were compared to the arrival and peak of the outflow recorded from the lysimeter underlying the instrument profile. Selected results from this analysis are presented. In the majority of the artificial rainfall events, where the time to the first change in outflow rate was hours to days, the outflow rate increased before the TDR measured water content at the base of the pile increased, and the outflow rate at the base peaked before the water content at the base of the pile peaked. Figure 6.16 presents the data measured in Profile A following an artificial rainfall event on September 156 20, 1999. It shows water contents from TDR instruments located in Profile A (Figure 6.16A), and the outflow hydrograph from lysimeter 10, located beneath the TDR profile (Figure 6.16B). Water content changes are generally larger near surface (5 to 7%) in comparison to deeper probes where water content changes in response to the largest rainfall events may be 1% to less than 0.5%. At the scale presented, the change in water content at 1.75 m is barely visible, but is clearly defined when examined in detail. Increases in water content are seen to progress in order down the pile, from 0.1 m to 3 m depth. The first increase in water content at 3 m depth occurs 16 hrs after the start of rainfall, and the peak is reached after 55 hrs. The flow hydrograph increases 4.75 hours after the start of the rainfall event and reaches a first peak after 9 hrs with a second peak after 33 hrs. The flow arrival and first peak occur when increases in water content are noted to 1 m to 1.75 m depth. Delayed response of the TDR in comparison to water outflow was consistent between profiles A , B and C for all larger rainfall events. The arrival and peak of the hydrograph typically occurred when the measured wetting front had reached 1.75 or 3m depth. We can infer that the spatially distinct preferential flow paths at high flow rates did not correspond to the majority of TDR probe locations. However, the TDR probe located at 4.5m in Profile A (Figure 6.16A) records a change in water content prior to changes in water content at the probes installed at 1.0 and 1.75 metres and immediately prior to changes in outflow. This response was consistent for all examined flow events. This probe location falls within a stable and predictable preferential flow path. Other TDR probes in all three profiles recorded early responses that were observed for single events only, indicating the TDR probe was located within a preferential pathway, but the location of the preferential pathway was not consistent between different precipitation events. 157 The measured water content is a marginally better predictor of flow behaviour at the low end of the outflow flux range. In two profiles (A and C), the progress of a wetting front down the instrument profile, and the subsequent increase in outflow rate followed a smooth progression during some, but not all, of the low rate infiltration events under low water content and outflow conditions, such as spring snowmelt, or small rainstorms after long dry periods in the summer. During these events, the first change in outflow occurs days to weeks after the precipitation event. A wetting front in the granular matrix could be traced through the pile, and soon after the increase in water content at 4.5 metres was recorded, the outflow rate increased. These low flow events were the only events where in-situ water content matched the timing of outflow. In profile B, the water content response was always delayed in comparison to the outflow response, even at the lowest flow rates. The late response of the instrumentation to the observed outflow implies the estimation of water flow through waste rock based solely upon the information derived from in-situ instrumentation would underestimate both the average wetting front velocity and the total volume of outflow to the base of the pile. The progress of the TDR monitored wetting front was plotted against time, and estimates derived for the predicted wetting front arrival and peak at the base of the experiment for all the artificial rainfall events. These data were combined with the outflow recession data to derive the volume of water in outflow that had discharged prior to the predicted arrival of the wetting front from the TDR data. Between 10 to 60% of the event outflow volume, with an average of 31%, arrives prior to the predicted TDR wetting front arrival, and an average of 44% of the event volume arrives prior to the predicted peak of the TDR wetting front. It was not possible to determine quantitative estimates for the lowest infiltration rates, as these events were always overprinted by the 158 responses to larger events. However, as previously indicated, qualitative observations indicate the better agreement between the timing of the movement of the wetting front observed using TDR, and the timing of changes in outflow at the base lower infiltration events. The TDR measurements may be better quantitative predictors of outflow timing and volume at lower flow rates. The TDR probes are located only in granular matrix, and the responses outlined above indicate they monitor the slower wetting front velocity portions of the pile. It is not clear i f the TDR data represent something close to the median wetting front velocity, or i f the minor grain size selection necessary to place instrumentation may have biased the material around the probes towards consistently slower flow conditions than average waste rock. 6.6 C O N C L U S I O N S The primary aim of this experimental work is to provide a comprehensive data set of water flow, transport and geochemical observations to assist in the understanding of the leaching of mineral weathering products from mine waste rock. Only the water flow data have been presented here, which is the starting point to understand the processes of leaching. Prediction of environmental loadings from waste rock requires both an understanding of the total net infiltration which determines the overall water volume, and an understanding of the contact of net infiltrating water with reactive surface area of the waste rock. The results presented indicate net infiltration under bare surface conditions is high, 50-85% of rainfall during storms, and an average of 55% long term. The variability of net infiltration calculated between different lysimeters indicates either a single large lysimeter (larger than the CPE), or a large number of smaller scale lysimeters is required to accurately determine net infiltration to bare waste rock. Commonly constructed lysimeters for field monitoring of 159 2 2 net infiltration include large plastic tanks (~4 m area) and 200 L drums (-0.4 m area). The variation in net infiltration data between the 2m x 2m lysimeters indicates that several of the tank lysimeters, or a larger number of 200L drum lysimeters would need to be installed and the data averaged to provide a similar estimate of net infiltration as a single large lysimeter such as the CPE. The measurements of water flow demonstrate preferential flow in spatially distinct flow pathways under these infiltration conditions. At least some of these pathways exist in the matrix supported waste rock. Some appear stable in space between flow events, whereas others are active only during specific flow events. Preferential flow behaviour increases with increasing water content and rainfall event intensity. The spatial distribution of preferential flow paths is also flow rate dependent. Increased numbers of preferential flow paths at the highest flow rates are more evenly distributed and lead to a slight decrease in the spatial variability of outflow. Based on the possible mechanisms identified in the introduction, it is postulated that these flow pathways are the result of infiltration and ponding at surface, ponding on large boulders beneath surface, surface flow in clast-supported areas, heterogeneity of the granular matrix or hysteresis effects, but the exact mechanism cannot be determined from flow data alone. Two strategies have been used to describe flow and transport behaviour in heterogeneous media. Approaches such as the transfer function approach of Jury et al. (1982) start with field scale observations of flow and transport and work backwards to derive a single set of parameters that describe the cumulative effects of all possible flow mechanisms. These averaged parameters are thus descriptive of some average material over the scale of the whole experiment. The hydrograph for the whole pile shows evidence that it 160 integrates the response of individual flow mechanisms but that the 8m x 8m experiment may not have been a large enough spatial scale to average sufficient numbers of preferential pathways occurring at high flow rates. This implies experiments larger than the CPE may be required to determine a single set of averaged parameters for randomly placed waste rock. A second strategy is to start from detailed descriptions of flow mechanisms at small scales and use extensive databases of measured parameters and appropriate spatial averaging or statistics to work upwards to the field scale behaviour (eg: Butters et al., 1989; Wierenga et al., 1991; Russo and Bouton, 1992; Shouse and Mohanty, 1998). It is clear that this second approach will also be difficult in this waste rock. The material characterization results indicate that simple laboratory methods fail to preserve in-situ textures and that the currently measured in-situ measurements of water content and matric suction wil l not be sufficient to define the necessary unsaturated flow characteristics for the granular matrix. Hysteresis effects are large, and field flow events are too short lived to derive adequate data from in-situ instrumentation. Immediately adjacent instruments may record different flow behaviours due to preferential flow. The response of TDR measured water contents during large rainfall events suggests that a substantial proportion of the outflow volume propagated in areas of the pile that bypass the three profiles of TDR probes. No in-situ parameters will be measured in those areas leading to the fastest propagation of wetting fronts, and 30 to 40% of the outflow volume. A careful program of in-situ measurement during pile deconstruction and laboratory testing using in-situ textures will be required i f this strategy is to succeed. The nature of waste rock piles allows a third strategy to be applied. Waste rock is an engineered material and therefore changes in surface texture are possible by altering haulage and deposition practices during pile construction. In addition, re-vegetation and re-sloping of 161 piles is often a required component of mine closure plan. Thus both within construction and post-construction changes to the pile surface may be possible. The observed convergence of recession curves at annual average net infiltrations of less than 69 mmyear"1 at Cluff Lake presents the possibility that a surface treatment of the pile capable of reducing net infiltration to a steady infiltration below this figure may also reduce the spatial scale required to predict flow behaviour. This observation does not imply that no heterogeneity in flow exists at this scale and flow rate, just that the physical scale of preferential flow is reduced and thus experimental size and cost can be reduced. Observation of the results of a tracer test would be required to determine i f the spatial scale required to observe residence time distributions is also reduced. As flow is dominated by the finer grain sized fraction of the granular matrix, predictions derived from laboratory characterizations of small samples are also more likely to be successful. Plans for future experimentation at the CPE include the compaction of the surface and the installation of a sand cover to reduce surface infiltration. These experiments will confirm i f low net infiltration does reduce the spatial scale of flow and transport behaviour and i f the variability of net infiltration estimates at the 2m x 2m scale is reduced. While reduction of net infiltration may decrease the necessary physical size of an experiment, it wil l increase the time scale necessary to conduct experiments by lengthening of the initial wetting-up period. This chapter has presented the observed responses of the pile outflow to changing infiltration and the overall success of the CPE experiment. The mechanistic reasons for these changes have not yet been addressed. The results of the first year of a transient tracer test conducted on the CPE are presented in Chapter Seven. These data are used to determine specific information about the nature of specific flow pathways. The results of the tracer test 162 also indicate the average residence time, the residence time distribution, and the spatial variability in these measures of water-rock contact time. 163 164 M M 0 5m Approximate Scale Figure 6.2: Side view photograph of constructed pile experiment. 165 Seive s ize (mm) Figure 6.3: Grain size analyses of waste rock samples collected during pile construction 166 -* P c 2 0 -B c o o 1— 15 -B ro 1 0 -"5 E 5 -3 O > 0 -25 -—' c 20-0 c o 15-o B ro § 10-o ' i — "S E 5 -3 O > o-B i 1111111|—i 111 i I I H I | — i 1111MI|— i 11111uj l | I I I I I I I I I I lllllj 1 I Mill I I I Ulll| 1 I I lllllj 1 I I 1 I I 1 | | MM 10"2 10"1 10° 101 102 103 Matr ic suct ion (kPa) Figure 6.4: Laboratory and field derived soil water characteristic curves and hydraulic conductivity curves. A) Laboratory data for the coarsest (squares), medium (triangles) and finest (circles) of 5 L grab samples for drying (closed symbols) and wetting (open symbols). B) Typical field measured data and estimated boundary drying curve C) Estimated boundary drying curves from 8 instrument profile locations. D) Estimated unsaturated hydraulic conductivity curves derived from field estimates of unsaturated hydraulic conductivity and the S W C C s presented in (C) using the method of Fredlund and Xing (1994). 167 10" i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—i—r 2/99 7/99 1/00 7/00 1/01 Date (month/year) Figure 6.5: Daily precipitation totals and whole pile outflow rate for March 1999 to March 2001. 168 F igure 6.6: Variability in total accumulated precipitation (A) and total outflow volume (B) between lysimeters from September 1998 to March 2001. 169 Figure 6.7: Monthly total outflow volume for the period of August 1999 to March 2001 expressed as percentage of total monthly outflow volume for A) all lysimeters and B) with lysimeters 6 and 13 removed 170 Figure 6.8: Topographic survey of pile surface conducted August 2001 using 0.3 m x 0.3 m survey resolution. Drainage catchments are outlined in heavy shaded lines. Lysimeter numbers are indicated. 171 Figure 6.9: Outflow hydrographs of lysimeters 6, 9 and 10 in response to the July 18, 2000 artificial rainfall event. 172 Figure 6.10: Variation of peak flow rates during large rainfall events with the event magnitude expressed as the average maximum flow rate of all lysimeters. Highest flow rate divided by lowest flow rate (triangles) and average of the four highest rates divided by the average of the four lowest rates (circles) 173 1.5x10-— 1.0x10 7 J x 20 21 Date (July 2000) Figure 6.11: Outflow hydrographs for individual lysimeters combined into quarters (A) and halves (B), and all of the pile (A and B) in response to the July 18, 2000 artificial rainfall event. 174 10* I -(ms" 10" i X —\ -_> 5= _ _ O Utfl O 10" 10" 24/9/1999 / 20/10/1999 i r i 1 r (ms 10'7 -z X _5 -5 -B u 1 o" — O -10"! 24/9/1999 20/10/1999 B n 1 1 1 1 1 i 0 2 4 6 8 10 12 14 Elapsed time (days) F igure 6.12: Outflow hydrographs for A) lysimeter 6 and B) lysimeter 9 in response to the September 24, 1999 and October 20, 1999 artificial rainfall events. Hydrographs are translated laterally such that the first increases in flow are coincident. 175 Figure 6.13: Distribution of outflow volume between lysimeters expressed as percent deviation of outflow volume from the average outflow volume for different rainfall events (light lines) and the whole experimental period (heavy line). 176 F igure 6.14: Variation of spatial distribution of outflow volume for varying rainfall event magnitude. Ratio of volume of highest volume lysimeter to lowest volume lysimeter (triangles) and total volume of highest four volume lysimeters to lowest four volume lysimeters (circles) plotted against the average maximum flow rate of the rainfall events as a measure of rainfall event magnitude. 177 0 100 200 300 Elapsed time (days) Figure 6.15: Flow rate recession curves of whole pile, slowest flowing lysimeter and fastest flowing lysimeter for A) winter 1999-2000 and B) 2000-2001. Comparison of winter 1999-2000 and 2000-2001 data for whole pile outflow (C). 178 Figure 6.16: Volumetric water content in instrument profile A and outflow hydrograph of lysimeter 10 in response to the September 20, 1999 artificial rainfall event. 179 in OJ CM O i - CO CO LO CM K O T O d s ci 00 CO o d oi d OT CO CD d •<- in * £ o 5 in K CO °° CN o . -tf ^ CM °> 00 ™ CO is. OT in d in OT oo CM O • t f CM CD OT OT p O in CD d OT CM o d d m CM o d CO d CM in p d d £ E co CC CD OT OT OO OT CD CO § TO in co T -in OT CD •tf CD CO P CO O CO ^ CM CO 00 CO CM CO •tf m d co in g & is; CD M CO °°. CM £ s ? o •r-in CM OT oo CO IS. is. OT in d in OT OT CM CD CM CO CM CD CM ••- "tf "* P CM |s! CM CO N- OT r - CM 00 CD Q-J T - CO 0 0 S3 ~ o 3 '5. 'o CD '•S °-±i CD Q. OT CD CD CO CO i CD 3 X CO in m o d o d o co d -tf CO o m CD o o CM in • t f O o o o d d d d O CD o CO d CM d d oo o OT CO oi CO • t f in C J 00 °°. 00 ^ £ •tf 00 ^ 3 = 3 OT CO CO OT O O d o P ^ T— o m . o <*? o s ° s o ^ o ^ o 5 o 5 o o s d OT d co o ? d £ o o o o d d d d p p p p d o d o oo oo O) OT "P- OT OT OT OT C OT OT OT OT £ ^ c CD E CD Q. X LU 0) Iii "D CD A—> I o 3 to o o o o o o CM CM c co cc TO 3 cc .£ — TO — w TO TO O — O — •§ .2 — TO TO a: O — 1 C TO t o < i-CO • t f CM CO CO • t f o 7— CO OT CO ts. CN OT OO oo CO CM CO m CO • t f CO CM o in co o o o o o o d d d d d d o o o p d p d oo • t f CD co OT CM |s- 00 CM CO CM CO CO CO 00 CM OT h-co in o o CO oo <N OT ® oS £ o d o -- Tr 00 CM CO 00 CO CO CO O -tf CM I*. OT CM CM O CO O i - 1 d S d " 00 oq CM o •* o ri O (D ,-j » ° is. ° CM CM CO CO oo P ^ P CD O O O CO CD CM 00 P 00 o o OT CO O CO d ^ o o o o d d d d o o o o d d d d oo oo OT OT o o OT OT OT OT O O OT OT OT OT O O •<- 1 - T - CM CM CD E * i _ CD Q. X UJ a> . E • 73 CD T> 3 C 'co CC — TO .5. CC a TO 3 TO «2 — TO — TO .!5 CC .2 CC io " o _ o _ o o O CM CD • CM in m -tf «N T -in -tf d -tf co r-OT in CM i -•tf is. co d •<- o CM OT CO OT CO OT CO h- CM in co o « d 0 3 O 2 <M 5 CM in 1 0 co 1 0 CO o ^ in Is. CO CM 00 CM CM O CM CO c CD E 0 Q. X LU a) K •a % o O JD i l a> o OT CM 00 o 00 CO CM in O OT CO ri o in CO CM 00 OT O i -0 ) 0 ) 0 0 OT OT O O i - T - CM CM o CO CD E o c CO •4—. TO aj '5. c o a) c 'c ' T O E V .TO +-> 1 5 2 CO CD Q. JS _^  CD O " O CD 3 cr CD TO JB 5 | CD > £ E •S E Q. OT O CD o c o (0 o c CO I CD c CO CD CO CD •a _3 O c ro 4—* TO T3 O O O CM i OT OT OT TO o o >. c o E T3 9? CO CD CD E TJ C ro c o ro Q. o 0 Q. >. JZ "c o E TJ CD CO ro CD E co" o w ro to c o ro 'o. o CD Q. TJ C ro 1 c ' r o i_ >s JZ C o E o ro E UJ £ CL O CO CD O JZ co 0 0 J3 JZ ' 10 \— o 1 8 0 Date Start End Duration Average Rate Christiansen Return Time Time Uniformity Period Coefficient (yr) (hrs) (mm) (mm/hr) 7/8/99 9:39 14:45 5:06 36.8 7.2 92 1 in 12 12/8/99 10:10 14:30 4:20 35.0 8.1 84 1 in 12 16/8/99 8:33 10:53 2:20 40.0 17.1 88 1 in 30 20/8/99 16:42 16:57 0:15 5.9 23.8 56 1 in 1 24/8/99 9:49 11:00 1:11 18.6 15.7 90 1 in 4 20/9/99 9:11 10:56 1:45 28.2 16.1 89 1 in 18 24/9/99 8:00 11:00 3:00 53.0 17.7 85 1 in 100+ 10/10/99 8:17 9:47 1:30 26.3 17.6 87 1 in 10 20/10/99 8:30 10:30 2:00 17.4 8.7 82 1 in 3 8/7/00 9:00 13:20 4:20 32.0 7.4 91 1 in 7 13/7/00 9:00 11:20 2:20 36.6 15.7 91 1 in 30 18/7/00 9:05 10:35 1:30 25.1 16.7 94 1 in 10 23/7/00 8:03 10:03 2:00 16.2 8.1 83 1 in 2 Average 86 Tab le 6.2: Summary of artificial rainfall events. 181 1999 2000 Date Date Artificial Total Rainfall Whole Sum Net Average Rainfall Rainfall Pile of16 Infiltration Maximum Start End in period Volume Outflow Lysimeter From Flow Estimate Estimates Whole Pile Rate (mm) (mm) (L) (L) (L) % m/s 16/8 24/8 40.0 2561 2162 2180 84 1.9E-07 24/8 17/9 44.0 71.5 4576 3216 3195 70 1.3E-07 24/9 10/10 53.0 53.0 3390 2642 2656 78 4.7 E-07 10/10 20/10 26.3 26.3 1685 985 971 58 3.7 E-08 1/3 3/6 70.3 4499 2320 2262 52 1.4 E-08 18/7 23/7 25.1 26.9 1719 1405 1460 82 9.4 E-08 4/8 26/8 48.2 3085 1560 1557 51 3.0 E-08 26/8 17/9 47.2 3021 2222 2189 74 4.6 E-08 1/10 30/10 47.2 3021 1794 1809 59 1.5 E-08 Date Date Lysimeter Outflow Volume (L) Start End 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Whole Pile 16/8 24/8 70 138 88 107 72 272 180 103 93 202 162 150 203 128 110 101 2161 24/8 17/9 165 198 112 171 121 345 185 148 148 261 223 171 343 230 195 178 3215 24/9 10/10 169 158 106 168 114 294 161 130 113 201 176 124 274 161 151 157 2642 10/10 20/10 80 65 35 76 31 157 47 32 31 51 49 24 116 49 57 71 984 1/3 3/6 106 157 50 86 74 327 139 98 126 133 170 151 212 147 156 130 2320 18/7 23/7 72 75 56 94 44 211 101 60 48 103 105 93 119 90 90 100 1405 4/8 26/8 89 87 46 77 48 249 90 67 57 82 95 78 185 114 105 88 1560 26/8 17/9 99 118 68 . 115 69 288 135 111 90 140 146 137 223 150 151 150 2221 1/10 30/10 118 153 62 86 69 240 97 62 84 118 100 82 208 118 117 95 1793 1999 2000 Table 6.3: Estimated outflow volumes for individual rainfall events and individual lysimeters 182 CHAPTER 7: SOLUTE TRANSPORT IN AN UNSATURATED HETEROGENEOUS MEDIUM: A CONSTRUCTED WASTE ROCK PILE 7.1 ABSTRACT A long duration conservative tracer test has been carried out on a heterogenous, unsaturated waste rock pile (8m x 8m x 5m high) to determine the physical mechanisms by which water flows in waste rock, and the resulting distribution of water residence time. Tracer was applied during a single rainfall event, which was followed by unlabelled rainfall in the form of natural and artificial rainfall events. Progress of the tracer is monitored using in-situ suction lysimeters and a contiguous grid of 16 lysimeters at the experiment base. A conceptual framework of the processes of solute transport under transient infiltration are presented to clarify the components of the hydrographs and breakthrough curves presented. Data from the first year have been analysed, and the long term spatial heterogeneity of flow and transport properties are presented. Data from the lysimeters demonstrate the presence of matrix flow in a heterogeneous granular porous medium and preferential flow in macropores and non-capillary pathways. Data from the in-situ suction lysimeters are used to determine transport processes at the pore scale, and the effectiveness of these instruments for monitoring the progress of tracer. Tracer transport is dominated by water flow in a small fraction of the total water content. Peak tracer concentration was observed in the lysimeters at 0.3 pore volumes, and the arrival of the center of mass is anticipated at 0.5 to 0.65 pore volumes. Median residence time is 3.0 to 3.9 years at the average annual rainfall rate. Residence time is underestimated by data from suction lysimeters due to systematic bias in the water sampled. 183 7.2 INTRODUCTION The movement of water through unsaturated mine waste rock must be understood to better predict the rates and timing of metals and acidity loadings to the environment. The grain size and textures of a waste rock pile can vary widely; the resulting porous medium is highly heterogeneous. A long duration experiment has been conducted on a 5m high waste rock pile with an 8m x 8m footprint. The pile is constructed on a contiguous grid of sixteen lysimeters (Chapter 6). The constructed pile experiment (CPE) was designed to provide a detailed set of observations of infiltration and water flow, tracer transport, and the evolution of the geochemistry of the leaching process in a waste rock pile under varying infiltration conditions. Details of the design, construction, instrumentation and operation of the CPE and the results of water flow in the first two and a half years of its' operation are presented in Chapter 6. In-situ and laboratory measurements of waste rock properties indicated the waste rock within the CPE is spatially heterogeneous. Measurements of matric suction and water content within the CPE indicated this heterogeneity occurred at scales as small as ~0.2 m. This heterogeneity is expressed in the lysimeter hydrographs by the presence of multiple changes in the outflow hydrograph (arrivals) in response to single rainfall events. This was noted for both for the hydrographs of individual lysimeters (2m x 2m) and the whole pile. The spatial distribution of water outflow was also heterogeneous. In this chapter the data from the first year of a conservative tracer test are presented. The tracer test was carried out to resolve the mechanisms of water flow within the CPE, and determine the residence time distribution of water flowing through the pile. This information is needed to better understand the time-scales of water-rock interaction, and the manner in which the outflow geochemistry may be altered by the rate and duration of individual 184 infiltration events and subsequent changes in the outflow rate. Infiltration was not maintained at steady state as the tracer test formed only part of the CPE's purpose and transient infiltration conditions were required for the other objectives. Tracer was applied in a single rainfall event, after which the surface of the pile was subject to both natural and artificial rainfall events in which no further tracer was added. In this chapter, the aim is to achieve: (1) a description of the effects of both heterogeneity and transient infiltration on the transport of solutes in unsaturated heterogeneous porous media; (2) an evaluation of flow mechanisms active in waste rock; (3) the effectiveness of in-situ instrumentation and lysimeters for the measurement of tracer transport in waste rock under transient infiltration conditions, and (4) an estimate of the distribution of solute travel times within the CPE. 7.3 SOLUTE TRANSPORT IN AN UNSATURATED MEDIUM Chapter 6 discussed the effects of spatial heterogeneity within waste rock using the non-specific term "flow path" to describe water flow that occurred either in a location or by a mechanism of which the exact nature was not known. Water flow in coarse, heterogeneous waste rock may proceed by a variety of physical mechanisms including: capillary-dominated porous medium flow in granular matrix material; capillary or non-capillary flow in macropores; surface flow over clast-supported cobbles and boulders; and by ponding on the surface of large cobbles and boulders (Chapter 6). The analysis of flow data alone cannot distinguish between these types of flow paths. In this section, more specific descriptions are developed of the possible physical mechanisms of water flow in waste rock material, and the manner in which they can be expected to affect the transport of solute. A conceptual framework is first introduced to describe transport under transient infiltration in unsaturated 185 porous media. This conceptual model is later used to frame the discussion of the results from the CPE. 7.3.1 Homogeneous Porous Media In an ideal unsaturated homogeneous porous medium, water is held in capillary tension, and water flow can be described by the Richards equation, which relates capillary tension and gravity driving forces to the Darcy water flux by the unsaturated hydraulic conductivity function. A n increase in the infiltration rate at surface creates an increase in the water velocity within the medium, and hence a change in water content. The interface between the higher water content and water velocity, caused by the infiltration, and the lower water velocity and water content deeper within the soil, is termed a wetting front, which propagates downwards through the soil profile. The velocity of the wetting front is determined by the kinematic velocity (Smith, 1982; Yamada and Kobayashi, 1988; Rasmussen et al., 2000). To first approximation, the kinematic velocity (c) is equal to the derivative of the Darcian flux (q) with respect to the water content (0), which represents the slope of the hydraulic conductivity to water content curve i f gravity gradients dominate (Rasmussen et al., 2000). The hydraulic conductivity of a porous medium increases as water content increases, and thus the wetting front will propagate faster than the Darcian advective velocity, v, of the soil water: v = - [7.1] e c = ^ [7.2] de When the wetting front arrives at the base of the soil profile, in our case the CPE lysimeters, the outflow rate increases. Two factors compete at the wetting front. The increase in wave velocity with saturation means that any small perturbations in the wetting 186 front caused by variations in hydraulic properties tend to self-correct and keep the wetting front sharp, whereas capillary pressure gradients act to widen the wetting front (Smith, 1982). The wetting front represents equilibrium between these two processes (Rasmussen et al., 2000). The end of the infiltration event starts a drying front, which also propagates faster than the Darcy water velocity. When this arrives at the base of the pile, the outflow rate recedes. Both variations in hydraulic properties and capillary gradients act to widen the drying front into an elongate trailing wave, rather than a sharp front (Smith, 1982), thus creating the faster rise and long recession of a typical hydrograph. The wetting and drying front velocities determine the advective velocity of the water within storage. The advective velocity field under transient infiltration is therefore a highly non-linear, time-varying function of the infiltration rate and duration, and the water content / matric suction / unsaturated hydraulic conductivity relationships. 7.3.2 Transport One-dimensional-solute transport in a porous medium has been described using the advection-dispersion equation: dt 8z2 dz where C is the solute concentration, z is the depth and D is the dispersion coefficient. The advective component of tracer movement is described by the second term on the right hand side. The center of mass of the solute, the first moment, travels at the mean water velocity, v. As noted above, this lags the wetting front velocity. Experiments by Rasmussen et al. (2000) noted a three orders of magnitude difference between the wetting front velocity and the average tracer velocity. 187 The first term on the right hand side describes the process of dispersion, a term which encompasses all processes by which some solute moves faster than the Darcy velocity, and some slower. Dispersion is caused by heterogeneity of the water velocity field at spatial scales smaller than the scale used to describe that average water velocity. In the advection-dispersion equation, the dispersion process is quantified using a Fickian diffusion term and the dispersion coefficient. This approximation assumes that the variation in water velocities can be described by a single mathematical term which is defined and is measurable at the same physical scale as the measurements used to define the advective components of water flow. This matter is discussed below. The dispersion coefficient must be measured experimentally for any given porous medium. For a saturated porous medium, it has been found experimentally to be linearly related to average advective velocity (Matsubayashi et al., 1997 ). In unsaturated porous media, experimental evidence shows it is a non-linear function of water saturation that is not linearly related to velocity, and that dispersion is higher in unsaturated media than saturated media. The dispersion coefficient has been found to increase with saturation and velocity to a peak near, but prior to, saturation, and then to decrease to the saturated values (Smiles and Phillip, 1978; Sharma and Tanaguchi, 1991; Maraqa et al., 1997; Matsubayashi et al., 1997). 7.3.3 Preferential Flow, Heterogeneity, Macropores and Non-capillary Flow The term preferential flow was introduced in Chapter 6 to describe water or solute movement either concentrated spatially, or faster than expected, in unsaturated media. For solute transport at the pore scale, preferential flow is a loosely defined term used to refer to water or solute movement faster than expected. Several physical mechanisms account for preferential flow. 188 Heterogeneity of soil properties at a macrosopic scale leads to spatial concentration of flow into areas with different properties. Under low infiltration and higher matric suction, water flow is concentrated into finer-grained areas where water is held under capillary tension, the water-filled pore space is more interconnected and hydraulic conductivity is higher. Under high infiltration rates, water fills more pore space and enters the larger pores. These pores have higher hydraulic conductivity, and thus carry a larger fraction of the water volume (Bews et al., 1997, Birkholzer and Tsang, 1997, Wildenschild and Jensen, 1999a). When these areas of varying properties occur at a field scale, and the variations in properties can be measured deterministically, then the process is assigned to the term heterogeneity. If water flow is concentrated at the scale of single, or several pores, then the term macropores is applied. In this case the heterogeneity is at a small scale, typically too small for the changes in properties between the macropores and the remainder of the porous medium, the matrix, to be easily measured and hence described deterministically. The loosely-defined term macropores is used to describe single pores or textures made of collections of pores within an unsaturated porous medium which are hydrologically effective at channeling water through the soil in a manner in which the water contained within them does not interact well with the water in the surrounding porous medium (Beven and Germann, 1982). Physical examples of single pores include root holes, worm holes, cracks, or larger pore spaces created by large grained particles, or open soil textures. The characterstics which describe them are (1) low or no air entry value (2) large size (3) some degree of vertical continuity within the soil profile (4) large variation in hydraulic conductivity from neighbouring pores and (5) poor mixing of water within macropores with the remainder of the porous medium. The range of physical definitions spans from -0.1 to -189 10 kPa when defined by the air entry pressure or 0.03 mm to 2 mm minimum using size (Beven and Germann, 1982), but the threshold between micropores and macropores is a matter of convention rather than physical principle (Germann and Beven, 1981). They may contain water under tension, under atmospheric pressure, or under positive pressures depending on the flow regime (Beven and Germann, 1982). They need not be filled for flow to occur. Flow can occur in films and rivulets on the pore wall sides (Germann et al., 1997; Tuller and Or, 2001). It should be made clear that macropores are not necessarily distinct from the porous medium but are part of it. Macropores are an extension of the pore space continuum to those pores that are filled near or at saturation. They may be singular in space, and easily identifiable, such as a worm hole, but they may also simply be a term to describe the largest pore spaces within the continuum. The propagation of a wetting front through a porous medium occurs when water content and hydraulic conductivity in the porous medium increases at surface. Water from the infiltration event is introduced into larger pore spaces at surface. If these pore spaces terminate in a short distance in a collection of finer pores, then the pressure wave continues in other large pores, but the water is matrix water. When water is introduced to pores large enough, and vertically continuous enough to be termed macropores, the infiltrating water flows in these pores at, or faster than, the wetting front in the finer matrix. What distinguishes them for the purposes of solute transport is their increased degree of vertical connection, and the manner in which the water they carry mixes poorly with water held in other pores. 190 Conceptual models of water flow and solute transport in macropores consider that water flows within the macropores, and gradually loses or exchanges water with the rest of the porous matrix (Beven and Germann, 1982). If the macropore is continuous, and the infiltration event is long enough, then water within the macropore at the base of the soil profile is event water. In this case, the hydrograph response corresponds to an increase in the solute break through curve (BTC) in the application event. If the macropore dead ends within the soil, the water is dispersed back into the matrix, which is termed internal catchment (Hatano and Booltink, 1992). As clast size and the resulting pore spaces are increased, the same processes used to define the term macropore extend to the larger pore space resulting from clast-supported gravel, cobbles or boulders with no matrix between. In these environments, the pore space is sufficiently large that water is never under tension, and non-capillary water flow may occur as surface flow over part or all of the clast surface. Between episodes of water flow in these pathways, small volumes of water are held in tension only at grain contacts (El Boushi, 1975, ElBoushi and Davis, 1969). This latter behaviour has been confirmed in waste rock by Bellehumeur (2001). These pathways would have even less interaction between the event water and the water resident in the granular matrix. 7.3.4 Conceptual Model of Effects of Heterogeneity During Transient Infiltration An unsaturated porous medium is described as heterogeneous i f the properties of the porous medium required to describe the Darcy velocity change between spatial locations, and measurement scales. The results presented in Chapter 6 indicate that material properties of the CPE waste rock vary in space, and water flow varies in space at scales less than 2m x 2m. The laboratory and in-situ measurements of material properties (Figure 6.4) indicated that 191 coarser waste rock had lower air-entry values and overall lower water contents than finer waste rock. Lower air entry values indicate the presence of larger pore spaces, and thus higher conductivities at lower water contents. Thus, i f the same wetting front velocity (3q/30) was recorded in both a coarse and fine sample, the coarse sample would have a lower water content, and thus a higher average advective velocity (q/0). Similarly, under the same Darcy flux, the coarser area will have both higher wetting front, and advective velocities. The effects of average advective velocity, wetting front velocity, dispersion, and spatial heterogeneity of porous medium properties can now be combined to demonstrate a typical hydrograph and BTC response to a single rainfall event. This is used later to address the results of the CPE tracer test. A conceptual explanation can be made by imagining that at least three distinct arbitrary classes of flow and transport pathways occur within the waste rock: fast, medium and slow, which refers to the velocity of the wetting fronts that propagate in each pathway. At this point, no further characterization of the actual nature or scale of these pathways is required. Figure 7.1 presents conceptual in-situ tracer profiles (panel A), the outflow hydrograph (panel B), and the BTC (panel C) that result from a single infiltration event at some time after the start of the tracer test. In our case, the plane of observation is at the base of the CPE at 5 meters depth. The time considered is after the original tracer application event, and sufficient infiltration has occurred such that the tracer has reached the lysimeters in two of the three flow pathways. The outflow hydrograph (panel B) is the integration of the wetting fronts that propagate within each pathway, and the hydrograph from each pathway. Wetting fronts will propagate to the base the experiment in the fast, medium and slow pathways with different kinematic velocity and arrival times (tf, tm, ts). 192 The average movement of water resident within each pathway is much slower (panel A), moving at the average Darcy velocity, which matches the velocity of the center of mass of the solute mass contained within each pathway. As discussed above, the observations in Chapter 6 indicate that the average advective velocity is higher in the pathways which generate the fastest moving wetting fronts. Thus, the tracer profile has advanced further in the fast pathway. Each rainfall event contributes to the downwards migration of the tracer profile, shown as the solid (pre-event) and dashed (post event) profiles. The vertical distribution of tracer in each pathway is subject to dispersion, which is here represented by a Gaussian distribution for simplicity. The B T C (panel C) measures the flux-averaged concentration in outflow. This is the result of the outflow rate and the concentration of each pathway. Prior to the event, the flux-averaged tracer concentration is a mix of the outflow from the three paths. The tracer concentration starts to rise at tf with the arrival of a wetting front within the fast flow path. Tracer in this pathway has moved further downward, and has a greater concentration at the outflow. The flux-averaged tracer concentration in outflow will rise as the proportion of the outflow composed of water from the fast flow path increases. The BTC tracer peaks at a concentration below that contained within the fast flow path, and then starts to drop with the arrival of lower tracer concentration outflow contained within the medium pathway. The BTC drops further as the flux from the fast and medium pathways decreases, and the proportion of water from the slow pathway increases. At some long time after the infiltration event, the water flux from each pathway will be similar to the proportions prior to the event. The final concentration in the outflow is higher than the pre-event concentration, as the peak concentration within each pathway has moved downward (panel B). 193 7.3.5 Heterogeneity and Transient Infiltration In a saturated porous medium, or an unsaturated porous medium under constant infiltration, the advective velocity field, the geometry of the saturated pore space, and the processes causing dispersion are constant in time. Under transient infiltration, water content, average water velocity, the geometry of the saturated pore space and dispersive processes are all time-variant and non-linear. It is not clear what the long-term average effect of a given transient infiltration regime will be in comparison to the same net infiltration applied at steady state. Several long-term, large-scale field experiments of solute transport under transient infiltration conditions have been conducted in unsaturated heterogeneous soils. These experiments have produced conflicting results. Experiments by Wierenga (1977) and Destouni (1991) found good agreement between the average water velocity, and the velocity of the first moment of the tracer mass. The BTC could be modeled using methods based upon steady state infiltration, by re-formulating the transient infiltration into a time-averaged measure. For example, the steady-state advection-dispersion equation can be re-written using either cumulative net infiltration or outflow instead of time (Wierenga, 1977). In these experiments, the effects on advection and dispersion from the changing soil conditions under transient infiltration averaged out in a manner that could be represented by the methods used to describe transport and solute dispersion under steady infiltration. Other experiments have noted differences in the velocity of the first moment of tracer mass, and the time-averaged water velocity under transient infiltration. Experiments have recorded average solute velocity both faster (Bowman and Rice, 1986; Roth et al., 1991) and slower (Jury et al., 1982; Butters et al., 1989) than the time-averaged pore water velocity. 194 The results of several experiments demonstrate that under transient infiltration conditions, the timing of the initial infiltration of the pulse is also a stochastic variable (Destouni, 1991). In comparison, the timing of the tracer application is not important in steady state experiments. Schulin et al. (1987) determined that the tracer velocity under a given flow regime was a function of the application date. They followed two tracers applied separately during summer and winter, and observed greater transport (as a function of cumulative net infiltration) for the tracer applied at the start of a wet season. Jury et al. (1982) observed slower than expected tracer transport for a tracer applied in which a 25 to 35 day delay occurred prior to rainfall. Both authors attribute the slower transport to the tracer being more mixed between large and small pores by evaporative cycling near surface. 7.3.6 Monitoring Methods at the C P E Data to describe the movement of tracer through unsaturated soils, and ultimately to parameterize predictive unsaturated transport models, are derived from measurements of in-situ tracer concentrations (eg: Rudolph et al., 1996), or from BTC's measured at tile-drains or lysimeters underlying the field plot (eg: Mohanty et al., 1998). Any observations of tracer concentrations or movement using instruments or lysimeters implies that the measurement is made over a particular physical scale, matching the sampling area of the instrument. Description of the movement of solute cannot be made independently of the scale of measurement (Butters et al., 1989a). At the CPE, the progress of tracer movement is monitored by two methods. Within the pile, tracer is monitored using in-situ suction lysimeters. The arrival of the tracer at 5 m depth is monitored in the lysimeter grid. Suction lysimeters were installed during construction of the pile. These samplers collect soil water from centimeters to tens of centimeters around the tip of the porous cup. 195 Destructive sampling followed by soil water extraction (Jury et al., 1982) was not an option. Nor was the use of TDR to monitor bulk soil conductivity, and hence ionic tracer concentration (Rudolph et al., 1996, Caron et a l , 1999), due to the high background conductivity of the soil water (5 to 20 dSm"1) caused by weathering of sulphide minerals. (Chapters 3 and 4). It is known that suction lysimeters sample a biased fraction of the water present in a soil and they measure neither volume-averaged nor flux-averaged concentration (Parker and van Genuchten, 1984, Litaor, 1988 ). They extract water from the soil around the ceramic tip by the application of an artificial pressure gradient. The water obtained is therefore a flux-averaged sample of the soil water immediately around the suction lysimeter, based not upon the flux resulting from natural matric suction and gravity gradients, but upon the flux generated by the applied suction and the geometry of the suction lysimeter tip. The exact nature of the suction lysimeter samples is therefore not well defined. The lysimeters at the base of the CPE allow direct measurement of flux-averaged concentration (Parker and van Genuchten, 1984) at the spatial scale of a single lysimeter (2m x 2m) or data can be combined to derive flux-averaged tracer concentration at larger scales up to 8m x 8m. As noted in Chapter 6, these lysimeters are at zero tension, and thus create a zone of increased water saturation in the waste rock immediately above the lysimeters. This may lead to mixing of water from different flow paths within the zone of increased water saturation that may not have occurred in a pile without the lysimeters. 196 7.4 FIELD METHODS AND DATA ANALYSIS 7.4.1 Rainfall Record and Tracer Application The CPE was operational in September 1998 and has been exposed to both natural rainfall and artificial rainfall events. A series of artificial rainfall events was conducted between August and September, 1999. On September 24 t h, a single rainfall event of 53 mm depth and 3 hours duration was labeled with 2100 mg/L chloride as L i C l . At the time of the tracer test, the pile water content and outflow rate was similar to that recorded during typical wet season (July/August) conditions. In our experiment, the initial application of tracer was during a period of relatively high water flow and resident water contents. The application event was also a large precipitation event for the northern Saskatchewan climate. This choice was made to maximize the possibility of activating non-capillary flow paths (Shipitalo and Edwards, 1996). This initial condition likely increased the participation of high flow rate pathways in the initial application event. Chloride ion was introduced with lithium to reduce disturbance to the geochemistry of the pile due to cation exchange or other geochemical reactions with the weathering sulfide minerals. Data from the tracer test have been analysed up to October 24, 2000. In this period, 154 mm of artificial rainfall has been added in six additional artificial rainfall events and 313 mm of natural rainfall has occurred. Average yearly rainfall at the site is 305 mm with a standard deviation of 76 mm and a maximum of 518 mm (1981-1997 site data). 7.4.2 In-situ and Outflow Water Sampling In-situ water samples are collected for geochemical and tracer analysis from suction lysimeters located in three instrument profiles (Figure 6.1). These were installed during construction in matrix supported areas of the pile, with a thin paste of silica flour to ensure 197 good hydraulic contact between the ceramic cup and the granular waste rock. Suction lysimeters installed at 0.2, 0.5 and 1.0 m depth are composed of a 0.022 m diameter x 0.05 m length ceramic cup connected to a 100 mL reservoir. The deep samplers are a 0.05 m diameter x 0.05 length ceramic cup connected to a 500 mL reservoir. During low flow rate periods, suction lysimeters were typically placed under 50 kPa suction, and left overnight. For samples collected during high flow rate conditions, the suction lysimeters were emptied within three hours. The timing of samples was assigned to the mid-point of the sample collection interval. Typical sample recoveries were 20 to 60 mL in the small suction lysimeters and 100 to 300 mL in the large suction lysimeters. Samples could not be collected during winter months from November to March when air temperatures are consistently below 0°C. Sampling restarted in April when near surface waste rock had thawed. Outflow chemistry in the lysimeter grid was sampled on a variable time scale. Immediately following artificial rainfall events, high frequency sampling of outflow was carried out by directly collecting samples from the sixteen tipping bucket gauges used to the monitor flow rate. Sampling of outflow during lower flow rate periods ranged from daily to weekly depending on flow rate. Between extensive field programs, long-term sampling carried out by site staff was performed tri-weekly or weekly. These samples were collected from a 0.2 L container placed in line after the tipping bucket gauge and therefore represent composite samples of varying duration depending on flow rate. At the lowest flow rates recorded at the end of the five month winter periods (0.1 to 0.2 Lday"1), this volume represented 1 to 2 days of outflow. 198 7.4.3 Data Analysis Water samples were analysed for dissolved chloride using a modification of the mercuric thiocyanate / ferric nitrate spectrophotometric method for chloride (Standard Methods, 1995) using a Hach 2100 portable spectrophotometer. The high and variable concentrations of other dissolved ions led to background interferences that varied between suction lysimeters and outflow gauges. These interferences were partly removed by adjusting the ratio of reagent to sample to 30:1, to optimize performance over the range of 0 to 400 mgL"1. Interferences were further minimized using calibration standards created by known addition of chloride to samples collected prior to tracer application. Outflow samples from the detailed sampling periods were selected for analysis based upon the rate of change of flow rate and tracer concentration. A total of 4420 outflow samples were analysed, with 900 duplicates. Between 216 and 312 samples were analysed per lysimeter, with an average of 275. Tracer concentration measurements have been processed up to October 24, 2000. Measurements of outflow rate were calculated from individual tipping bucket outflow tips using gauge specific calibration. Data were linearly interpolated to determine flow rates and cumulative outflow volumes at fixed intervals, every 30 seconds during summer, and every 15 minutes during winter. Water flow data have been processed up to March, 2001. Chloride concentration in the outflow was linearly interpolated between measurements to derive chloride concentration estimates at the same fixed time intervals as the flow rate estimates. Chloride mass flux and cumulative mass were determined using flow rate and chloride data from the individual lysimeters. Tracer mass flux and cumulative tracer mass were estimated by subtracting the background chloride concentrations estimated from samples collected prior to tracer application. Background chloride in outflow was assumed 199 to be constant. The values for flow rate and cumulative volume in individual lysimeters were combined to form the outflow record of the whole pile. Total chloride mass flux rates from individual lysimeters were combined with cumulative outflow volume data to derive flux-averaged chloride concentration for the whole pile. 7.5 R E S U L T S A N D DISCUSSION 7.5.1 Whole Pile Tracer B T C and Mass Recovery Figure 7.2 presents the record of precipitation (panel A), the outflow hydrograph for the CPE (panel B) and the flux-averaged chloride concentration in outflow for the whole pile (panel C). Note that the vertical scale in panel B is logarithmic. Artificial rainfall events are shown as darker lines and the date of the tracer application event is indicated. Outflow rates are presented as fluxes (mVW2) to allow direct comparisons between measurements at different physical scales. BTC's under transient infiltration conditions can also be presented plotted against the cumulative net infiltration, or outflow volume (Wierenga, 1977 ). In Figure 7.3, the outflow hydrograph (panel A), and the flux-averaged chloride concentration (panel B) are presented plotted against cumulative outflow volume, normalized by cross sectional area and calculated relative to the start of the tracer application. Key events on Figure 7.3 have been labeled with their dates. In this figure it is possible to see multiple arrivals in each hydrograph response that indicate that the hydrograph responses to individual events are composed of the integrated arrivals from several flow paths (Chapter 6). The outflow rate climbed over five orders of magnitude during the wetting up of the pile from April to August, 1999. During this time, water storage depleted by evaporation during construction was replenished. Subsequently, flow rate varies over three orders of magnitude. Background chloride concentration ranged between 24 to 27 mgL"1 prior to August, 1999. 200 Several key features of the BTC can be noted in Figures 7.2 and 7.3. Firstly, tracer reaches the base of the CPE during the application event. The flux-averaged chloride concentration rises quickly to 350 mgL"1 within hours of the tracer application, beginning with the first wetting front arrival. At this peak, the discharge is composed of 17% by volume of water applied in the infiltration event (event water). Secondly, the duration of this high tracer concentrations in the first event is short-lived. The flux-averaged chloride concentration peaks early, on the rising limb of the outflow hydrograph (Figure 7.3). The BTC falls much more rapidly than the flow recession curve, reaching an inter-event concentration of ~50 mg L" 1 , or 1% event water by volume. This result indicates that the outflow in this first event is dominated by water already in storage within the pile, termed resident or pre-event water. Thirdly, during subsequent artificial rainfall events on October 10 and 20 (no tracer input), the flux-averaged chloride concentration again rises sharply with each flow peak, and falls to a progressively higher inter-event concentration during flow recession. These early events indicate that the flux-averaged BTC is composed of outflows from different pathways within the pile that contain tracer in different concentrations. At the start of the experiment, those pathways that contribute a greater proportion of the outflow volume in the earlier parts of the hydrograph response to each infiltration contain higher tracer concentrations. The pattern of the hydrograph and BTC is similar to the conceptual framework discussed in the preceding section. Lastly, the early pattern of short-lived increases in flux-averaged concentration during peak flow and a gradually increasing inter-event concentration continues until July 18, 2000 at which time the flux-averaged chloride concentration reaches a second experiment peak of 332 mgL"1, or 16% of the applied tracer concentration. After this time, the pattern changes. 201 A small drop in flux-averaged chloride concentration occurs at the start of each event response, followed by a rise to a peak concentration, and a more gradual fall to the inter-event flux-averaged chloride concentration. The inter-event flux-averaged chloride concentration also begins to decline. Later in the experiment, those pathways that contributed a greater proportion of the outflow volume during earlier parts of response to each infiltration now contain lower tracer concentrations. These issues are discussed in sections 7.5.3 to 7.5.7 using the BTC's from individual lysimeters, where the relationships of tracer concentration and water outflow are more easily demonstrated. Comparison of Figures 7.2 and 7.3 indicates that several features of the outflow BTC in Figure 7.2 are distorted by the presence of the long winter period from November to May in which there is little infiltration and low outflow flux. This period represents 40% of the time period presented in Figure 7.2, but only 10% of the cumulative outflow in Figure 7.3. Over the winter, the BTC rises from November to January, then declines prior to two increases in April and May, 2000. In Figure 7.3, the changes over the winter period can be seen to be minor contributions to the overall BTC, and the changes in flux-averaged concentration from the small volume rainfall events in the April 2000 become less prominent. In general, plotting the BTC against cumulative outflow (Figure 7.3) better represents the relative magnitude of infiltration events. There were four larger precipitation events after July, 2000 that led to significant flow responses and rises in the BTC. It should be noted that these events appear to have smaller changes in chloride concentration over the event, but this may be a sampling artifact. These events occurred during tri-weekly sampling of outflow by site staff. This sampling frequency 202 was not adequate to measure the peak chloride concentrations, which are typically only hours in duration. The cumulative fraction of applied tracer recovered in the CPE outflow is plotted against cumulative outflow volume in Figure 7.4, reaching 34% by October 24, 2000. The first moment of the tracer pulse has not reached the base of the pile. Also shown along the x-axis are the approximate pore volumes of outflow. This is calculated from the resident pore water volume and the cumulative outflow. Resident pore water volume is not constant as in-situ water contents change rapidly following infiltration events (Chapter 6). Changes of 6 to 7% in water content occur near surface, which reduce to approximately 1% below l m depth. The pore volumes were calculated by assuming a long-term average water content of 16%, based on monitoring of water contents by TDR. By this method, approximately 0.5 pore volumes of water have flowed through the pile. The peak slope in the cumulative mass recovery curve is between 0.15 and 0.2 m3m"2, or 0.3 pore volumes, which corresponds to the peak in chloride concentration in July 2000 (Figure 7.2C). After this time, the cumulative tracer curve flattens out as the inter-event chloride concentration decreases after July 2000. The peak chloride concentration arrives well in advance of one pore volume, and prior to the first moment. The declining concentration of chloride in the outflow after July 2000 implies a gradual decrease in the mass flux that will lead to a long tail on the cumulative breakthrough curve. 7.5.2 BTCs and Tracer Mass Recovery in Individual Lysimeters The division of the pile base into sixteen contiguous lysimeters allows for investigation of transport at smaller scales (2m x 2m). The BTC's of the sixteen lysimeters are presented in Figure 7.5, plotted as chloride concentration against time. In four lysimeters, 203 (2,3,7,8), single concentration peaks exceed the maximum scale of 600 mgL"1, and the peak concentration is noted. The cumulative outflow volume varies by a factor of 4 between lysimeters, and plotting the BTC's against cumulative volume makes discussion of specific rainfall events difficult. The differences in BTC's indicate a wide range of spatial and temporal responses at the 2m scale within the CPE. Varying transport between each lysimeter footprint results from the spatial variability of surface infiltration and the spatial variability of medium properties. The effects of spatial variability on the long-term transport between lysimeters is presented in Figure 7.6. Figure 7.6A presents tracer mass recovery to October 24, 2000, expressed as the fraction of the total tracer mass applied to each lysimeter footprint. Tracer recovery ranges from 11% to 117%, with an average of 34% and a standard deviation of 26%. For comparison, the spatial distribution of cumulative outflow volume is presented in Figure 7.6B. The spatial variation in tracer recovery is much greater than that of outflow volume, but those lysimeters recording the highest outflow volumes generally recover greater tracer mass (Figure 7.6C). For example, lysimeter 6 recorded the highest tracer recovery (117%) and recorded 120% net infiltration, compared to the average of 55%. This indicates substantial lateral transfer of water flow and tracer mass within the CPE. Heterogeneity of material properties at the macroscale (2m) leads to variations in both outflow volume and residence time. The relative transport in each lysimeter can be illustrated by introducing a dimensionless recovery ratio, Figure 7.6D, which expresses the mass of tracer recovered in a lysimeter as a fraction of the tracer mass applied at surface, divided by the total outflow flux per lysimeter (m3m"2). Differences in total water volume represents the variation in advective 204 movement of tracer mass among the lysimeters and is the first-order control of tracer recovery (Figure 7.6B). Variation in the recovery ratio represents the second-order variations in tracer movement caused by differences in dispersive properties between lysimeters. In general, faster flowing, higher volume lysimeters such as 6 and 13 have higher recovery ratios, thus higher dispersion. The effect of dispersion on the tracer recovery can be seen by comparing lysimeters 1, 5 and 9 in Figure 7.6B and Figure 7.6D. They have similar volumes of outflow, but different recovery ratios. Lysimeter 1 has a tracer recovery ratio similar to lysimeter 6. Lysimeters 5 and 9 have opposite rankings of outflow volume and tracer recovery. The mechanisms that govern flow volume, and solute dispersion are weakly linked. 7.5.3 Summary of the Character of Individual BTC's The high temporal sampling rate of the data collected from individual lysimeters in the CPE provides an opportunity to examine BTC's in greater temporal detail than typical field-scale studies. These can be used to determine the nature of flow mechanisms operating in the CPE, and their variation in space. The character of the BTC's of individual lysimeters at early times, prior to July 2000, can be grouped into three classes (Figure 7.5): (1) lysimeters 2,3,8,12 and 16 exhibit distinct spikes in chloride concentration; (2) lysimeters 1,4, 6,13 and 15 show a rise in tracer concentration to a peak with rising flow rate, which tails off to a gradual rise in inter-event chloride concentration, similar to the BTC of whole pile; and (3) lysimeters 5,9,10,11 and 14 exhibit dips in chloride concentration during the tracer application event, followed by later slow increases. The character of the BTC's after the peak in chloride concentration also involve several classes: (1) lysimeters 3 and 12 continue to exhibit higher chloride during flow events, and rising inter-event chloride 205 concentration; (2) lysimeters 2,4,5,7,9,10,11,14 and 15 exhibit higher event chloride concentrations, yet gradually decreasing inter-event chloride concentrations; and (3) lysimeters 1,6,8,13 exhibit a third character where rainfall events lead to lower chloride concentrations during peak flows, which return to a higher, but steadily decreasing inter-event chloride. Each of these classes is examined, and the flow mechanisms responsible are identified. 7.5.4 Flow and Tracer Spikes The first class of early time behaviour is seen in Figure 7.7. This presents the tracer concentration and outflow hydrographs from lysimeters 2 (panels A , B) and 12 (panels C, D), which exhibit distinct spikes in flow rate and tracer concentration at early times. Tracer concentration is presented in the top plots (A, C) and outflow hydrographs in the bottom plots (B, D). The application event and the next two rainfall events of the tracer test are presented, with the September 24, 1999 tracer application event expanded in the inset of each graph. Examination of lysimeter 2 in panels A and B indicates a distinct spike in the flow response prior to the main rise in the outflow hydrograph in each of the three rainfall events presented. The outflow rate rises rapidly within 1.5 to 2 hours of the start of the rainfall event, reaches a short lived peak, and then returns to the outflow rate recorded prior to the rainfall event within 3 to 12 hours. Further changes in the flow rate occur later with the arrival of multiple wetting fronts typically 14 to 48 hours later. This character of response has been observed following both natural and artificial rainfall events in several lysimeters. Tracer concentration during the spike in the flow rate reaches 1124 mgL"1 during the tracer application event, and 400 and 253 mgL"1 during the following two rainfall events. The tracer concentration falls to the pre-application concentration with the fall in flow rate. The 206 inter-event tracer concentration does not begin to rise above background until after the October 20 t h event. In lysimeters 2,3 and 16, the second and third events both contain tracer during the spike in flow rate. A similar spike in flow rate and tracer concentration is seen after 1 hour in lysimeter 12 but the only during the tracer application event, and not in subsequent events. This response is indicative of flow and transport through a high conductivity pathway, which has little storage capacity, and little exchange of water or tracer with the remainder of the granular matrix. The water which creates the hydrograph response is predominantly composed of event water. These conditions could be met by either water flow down the walls of the experiment or by water flow as surface flow over clast-supported cobbles and boulders. As indicated in Chapter 6, considerable effort was made to prevent wall flow by the inclusion of skirts around the edge of the experiment both above and below ground surface to re-direct water flow back into the waste rock. Flow directly down the wall from above ground is minimized. If water is flowing down the wall, it is being introduced beneath the ground surface, perhaps by ponding on the upper surface of large clasts in contact with the wall. The four lysimeters located in the center of the experiment not bounded by one or more walls did not show this type of early response. The response recorded in lysimeter 12, where tracer only appeared during the application event may indicate wall flow. The presence of high tracer concentrations in lysimeter 2 during the second and third unlabelled rainfall events indicates some interaction between the water in this pathway and the tracer introduced in the application event. This outflow and tracer response would also be characteristic of non-capillary flow over clast-supported particles with small amounts of 207 water stored at grain to grain contacts, or on the upper surface of large particles (El Boushi, 1975, E l Boushi and Davis, 1969). Non-capillary water flow over larger particles could be generated either at surface by localized ponding, or by sub-surface ponding of resident or event water on the top surface of larger clasts. This latter mechanism would re-mobilize water containing high tracer concentration from near the CPE surface into non-capillary flow paths if the large clast was located near the pile surface. This would generate BTC peaks during the second and third rainfall events which contain significant tracer chloride, as seen in Figure 7.7A. Calculation of the total outflow volume produced from this pathway indicates approximately 0.1% of the total outflow volume between September 1998 and March 2001 flowed within this pathway, and 1.3% of the total tracer applied was recovered during flow spikes. A waste rock pile with greater than 20% of a sand size fraction can be classified as soil-like (Smith et al., 1995), which indicates that the majority of larger clasts are matrix supported. The grain size analysis of the CPE indicates approximately 25% sand sized (>2mm) grains. Areas of clast-supported texture with no matrix filling voids, were observed during construction. The tracer data suggest that few of these paths reach from the pile surface to 5m depth without interacting or dead-ending within matrix material. Under different circumstances, i f surface runoff to clast-supported areas of the pile were to occur, or the fraction of finer-sized material was reduced, then this mechanism may be more important for both infiltration past the evaporative zone and the duration of water-rock contact, which is a fundamental process in the development of water chemistry deeper within the pile. 208 7.5.5 Tracer Arrival with the Wetting Front Lysimeters 1,4, 6,13 and 15 show a rise in tracer concentration with the first wetting front arrival, indicating event water was able to flow at the velocity of the wetting front (Figure 7.8). The tracer concentration increases distinctly after each infiltration event in the fall of 1999. The typical maximum tracer concentration reached 100 to 150 mgL"1, or 0.5% to 7% event water at the flow peak. Figure 7.8 presents the tracer concentration (panel A) and the outflow hydrograph (panel B) measured in lysimeter 13 after the tracer application event. The arrival of chloride is simultaneous with the first wetting front arrival (insets). The hydrograph rises sharply, and this arrival has started to recess by the time a second broader arrival can be seen. The arrival of chloride is part of the main hydrograph and chloride is present through the whole hydrograph response. The BTC starts to drop sharply with the second large wetting front arrival, and then levels out to a steadier figure (70 to 80 mgL"1). In Figure 7.8, the first three rainfall events have hydrograph and B T C patterns similar to the conceptual model presented earlier, where the hydrograph and B T C can be explained using multiple flow paths. Event water containing tracer is able to flow from surface to the base of the experiment during the application event as part of the wetting front. Unlike the spikes in flow rate and tracer concentration described in Figure 7.7, these flow paths are part of the porous medium in the granular matrix. Tracer labeled water accounts for a maximum of 6% of the outflow hydrograph, indicating the majority of the hydrograph response is resident water from within the granular matrix. This BTC response results from macropores, or spatially distinct zones of high conductivity material. As stated before, both of these flow path types are part of the continuum of the heterogeneous porous medium. At the 2m x 2m 209 scale, there is no way to distinguish the effects of many distributed large pores which would be termed macropores, from the effect of several macroscopic areas of coarser material which would be termed heterogeneity. Both are present in waste rock. What does distinguish these processes is that some of the pores filled during the infiltration event were large enough that the hydraulic conductivity matched the velocity of the wetting front. The cumulative tracer mass arriving in these lysimeters during the application event was 5% of the total mass applied to them. It can also be noted that the tracer concentration levels out early within the hydrograph response. This flattening of the BTC was found in many of the BTC's and carefully examined. It was often found to correspond to the latest recorded discernable arrival on the hydrograph. This can be explained i f tracer mass is lost from these macropores or coarse zones into the surrounding porous medium during the tracer application event. This tracer mass is carried forwards when the later slowing wetting fronts carry through in the finer zones, and leads to the leveling out of the BTC. 7.5.6 No Tracer Arrival with the Wetting Front In lysimeters (5,9,10,11 and 14), the first response during the tracer application event was a drop in the chloride concentration (Figure 7.4). Figure 7.9A presents the tracer concentration and Figure 7.9B the outflow hydrograph measured in lysimeter 11 before and after the tracer application event. During the tracer application event, the chloride concentration drops from 30 mgL"1 to 3 mgL"1. The concentration recovers to pre-event levels within 48 hours of the tracer application, then continues to rise slowly above the background. This response demonstrates that the initial response to the tracer application was composed of water already resident within the pile. 210 Of particular interest is that the pathways which contributed to the first response of the hydrographs contained resident water with a lower background chloride than the flux-average during low flow rates. As the outflow became dominated by the slower flow paths at the end of the event, the tracer concentration returned to that seen in long term outflow. These data indicate that the resident soil water chemistry is not in equilibrium between different flow pathways. The fast pathways that contributed water early in the hydrograph, macropores or spatially distinct coarse zones, contained resident water which was also contained lower concentrations of dissolved weathering products. It also implies that an assumption of uniform background chloride is not strictly valid, and that separation of background and tracer flux wil l be difficult when chloride concentrations tail off to pre-application event concentrations. The longer term behaviour of these lysimeters is slightly different (Figure 7.5). Lysimeters 9, 11 and 14 exhibit a slow rise in tracer concentration, whereas lysimeters 5 and 10 exhibit a more rapid rise in chloride. The BTC of lysimeters 5 and 10 indicate that tracer labeled water reached the base of the experiment in the subsequent two rainfall events. In these lysimeters, the resident water with lower background chloride present within the fastest pathways had only been within the pile for perhaps one or two events, a matter of days to a week, before it was expelled in the tracer application event. Once these faster pathways reached the base, the BTC resembles the conceptual model, and the BTC varies sharply with flux rate. The longer slower rise in lysimeters 9, 11 and 14 may indicate that the macropores, or coarse pathways activated during the tracer application event, are less active and less significant in subsequent events. Tracer does not break through within the faster pathways in 211 these lysimeters until the spring of 2000, after several rainfall events. The lower chloride water resident in the fastest pathways during the tracer application event may have been in the pile for a longer time period. 7.5.7 Hydrographs and BTCs from Summer 2000 A l l of the BTC's presented in Figure 7.5 were examined event by event to determine the relative timing of the changes in the hydrograph and BTC. In all lysimeters, it was observed that the peak concentration gradually occurred later on the rising limb of the hydrograph, eventually corresponding to the hydrograph peak. The peak concentrations were measured in July, 2000, 9 months after the tracer was applied. In our conceptual model (Figure 7.1) this would occur approximately when the peak concentration in the fastest flow paths has moved past the plane of observation. The peak tracer concentration is generated by the flux-average of the outflow rate and tracer concentration in the fastest and medium flow paths. Figure 7.10 presents the two types of post-peak B T C character. The behaviour typical of most of the lower flux rate lysimeters and the whole pile B T C is shown in Figure 7.1 OA which presents the BTC and hydrograph for Lysimeter 9 during July to October, 2000. Each event leads to higher chloride concentrations, with the peak of chloride concentration corresponding to the peak in flow rate. The chloride peak is once again tempered by the last arrivals after the flow peak to a steadier inter-event concentration. The overall inter-event concentration steadily declines. In this case, the contributions of declining chloride from the fastest flow paths, and increasing chloride from the medium flow paths average out to a chloride concentration increase over the concentration in the slow pathway at the hydrograph peak. The late arrivals from within the slow pathway are when this pathway contributes the 212 greatest fraction of mass, and the chloride concentrations level out briefly. The fastest pathways are past the chloride concentration peak and thus when the outflow returns to the longer duration flux-average, the concentration slowly drops. In lysimeters 1,6,8 and 13, the BTC's changed more after the concentration peak. Figure 7.1 OB presents the BTC and hydrograph for lysimeter 6 for July to October, 2000. During this period, each rainfall event leads to a decrease in tracer concentration, followed by a return to a steadier tracer concentration during inter-event times. The inter-event tracer concentration drops steadily. In these lysimeters, the peak concentrations have passed in both the fastest and medium flow paths. Their tracer concentrations are below the concentrations in the slower flow paths. Thus, the rising limb and peak of the hydrograph dominated by flux from fast pathways leads to a drop in chloride, followed by a recovery to the concentration in the slowest flow paths. 7.5.8 Summary of Observations from Lysimeter Measurements The effects of non-capillary flow paths, macropores or spatial heterogeneity and flow in the finer matrix material can all be distinguished with the BTCs from individual lysimeters. Chapter 6 indicates that there are between 3 to 12 discernable wetting front arrivals in each 2m x 2m lysimeter for each rainfall event. When the scale of observation is increased to the whole pile (8m x 8m), there are still 6-15 distinct wetting front arrivals discernable in the hydrograph from individual rainfall events. Many of the BTC features observed in individual lysimeters are also present within the whole pile BTC. The 8m x 8m scale of the whole pile is therefore insufficient to average either flow or transport properties. 213 7.5.9 In-Situ Tracer Measurements Figures 7.11, 7.12, 7.13 present the chloride concentrations measured in suction lysimeter samples obtained from instrument profiles B , A and C. These profiles are located above lysimeters 13, 11 and 12 respectively. Only selected sampling dates are presented for clarity. Data are presented as profiles versus depth (panel A). The flux-averaged tracer concentration in the lysimeter underlying the profile is plotted at 5 m depth for comparison. BTC's (panel B) are plotted as tracer concentration at a single depth against the cumulative outflow volume (m 3nf 2) from the gravity-drained lysimeter underlying the profile (Russo et al., 1989, Wierenga , 1977). The BTC of the underlying lysimeter is presented in panel C. The cumulative tracer mass passing each suction lysimeter was estimated from the measured BTC, the background chloride concentration and the cumulative water volume recorded in the underlying lysimeter. The results are presented in panel D. Cumulative recovery is plotted as a ratio to the total tracer applied per lysimeter. Also shown is the total outflow volume from the underlying lysimeter as a ratio to the average outflow, and the cumulative tracer in outflow. The tracer concentrations and the timings of changes in concentrations from all instrument profiles indicate that water flow is occurring in three dimensions in spatially distinct flow pathways, some of which bypass the suction lysimeters. In profile B , (Figure 7.11) tracer progresses from 0.2m to 1.0 m in an expected fashion, arriving later, with a lower and wider chloride peak (panel B). However, the BTC's at 1.75 and 3.0 m are sharper and higher than at 1.0 m, indicating these suction lysimeters fall within a different flow pathway, with different dispersive characteristics. The outflow BTC (panel C) shows that tracer appears at the base of the CPE during the tracer application event, and peaks when the 214 suction lysimeter at 1.75 m peaks. The BTC peak at 4.5 m is delayed, implying this suction lysimeter lies within a much slower pathway. The tracer concentration at 4.5 m just exceeds that in the outflow at the end of the analysed period. These data indicate the fastest tracer movements occur in spatially distinct pathways that are not sampled by the suction lysimeters. There is a low probability that a rare, fast pathway wil l intersect with an instrument location. Similar results were noted for the measurement of water content using TDR (Chapter 6). Figure 7.1 ID indicates that the calculated cumulative tracer flux passing some suction lysimeters exceeds the amount applied to the surface above the instrument profile. The concentration of tracer in all of the deeper suction lysimeters is still increasing at the end of the monitored period. Lysimeter 13 recorded the second highest volume of outflow, 1.6 times the average and thus a total tracer recovery greater than the applied mass on the 2m x 2m infiltration surface above the lysimeter can be expected from Figure 7.5C. The calculation of mass flux past near surface lysimeters is only approximate as the water flux past shallow lysimeters may be different from the water outflow at 5 m depth due to lateral movement of water into the footprint of adjacent lysimeters. In instrument profiles A and C (Figure 7.12 and Figure 7.13), the data more clearly indicate flow in spatially distinct pathways. In instrument profile A , the tracer appears to progress downwards in two distinct pulses (Figure 7.12A). During the tracer application event, tracer appears at 1.75 m and 3.0 m at the same time tracer is recorded at 0.2m (panel 12B). By October 15, 1999 the two pulses are distinct. Tracer concentration rises at 4.5 m, and in the outflow (panel 12C), immediately following the application event. The peak in outflow concentration occurs after the peak in concentration at 4.5 m. However, the tracer 215 concentration in the outflow is consistently lower than the corresponding concentration at 4.5 m. The cumulative recoveries of tracer mass and total outflow volume (Figure 7.12D) indicate a range of tracer recovery that brackets the ratio of outflow recorded. Tracer recovery in outflow is lower than the tracer flux past any lysimeters. In profile C (Figure 7.13), the profile of tracer concentration has two peaks (panel 13 A) . Except for the short-lived spike in tracer concentration, tracer concentrations in the outflow do not change until after tracer has been recorded at 4.5 m (panel 13 B). The peak concentration at 4.5 m and in outflow appear to match in time, but outflow is again at a lower concentration than the suction lysimeter at 4.5 m. Figure 7.13D indicates lysimeter 12 reports lower than average outflow volume, and lower than average tracer recovery by the suction lysimeters and outflow lysimeter. The cumulative mass flux is similar to instrument profile B , where the outflow recovery is much lower than the recovery at 4.5 metres. The tracer mass stored within the CPE can be calculated using the profile data in panel A of Figures 7.11, 7.12 and 7.13. Stored tracer mass was calculated from the measured chloride concentration, the estimated background chloride concentration and estimated soil water content. Water content is variable in response to infiltration events, but an approximate value of 0.1 was applied from surface to l m depth and 0.16 after 1 m, based upon long term monitoring using TDR. The qualitative character of this calculation is insensitive to allowing water content to vary by the range recorded in TDR measurements. A second estimate of tracer mass in the CPE was determined by subtracting the calculated cumulative tracer mass in outflow from the original applied mass. The ratio of these estimates is shown in Figure 7.14. The ratio initially rises up to 8 in profile A and 3.5 to 4 in profiles B and C. It then falls below 1 in profiles A and C by October, 2000. 216 When the measured concentration from the suction lysimeters is assumed to apply to all the resident water, a volume-average sample, the stored mass is overestimated. This can be compared to the estimates of tracer mass flux past the suction lysimeters in panel D of Figure 7.11, 7.12 and 7.13. Within each instrument profile, the mass flux estimates calculated for individual suction lysimeters either bracket, or overestimate the calculated mass flux in outflow. The tracer concentrations derived by the suction lysimeters are therefore closer to flux-averaged samples. The calculation of stored mass also indicates that the tracer is unevenly distributed at the pore scale within the water resident in the granular waste rock. In order to match the calculations of stored tracer mass based on the suction lysimeter data and the BTC data, the tracer concentrations determined from suction lysimeter data would only be representative of l /8 t h to 1/4 of the 16% by volume average resident water content around the suction lysimeter tip. As discussed previously in this section, the suction lysimeters collect a sample that is closer to a flux-averaged tracer concentration. The tracer mass flux is therefore dominated by a small fraction of the pore space, representing those pores with the highest hydraulic conductivity. It was also noted in Section 7.3.6, that suction lysimeters collect a biased sample. Early in the tracer test, the largest, highest conductivity pores contain higher tracer concentrations than the flux average passing the suction lysimeter elevation. The flux created by the suction lysimeter is greater than the natural flux, and therefore the suction lysimeter samples a greater water fraction from the larger pores than the flux averaged concentration under a natural gradient. The calculation of mass flux past the lysimeter therefore uses the total water flux resulting from the true head gradient, multiplied by the 217 tracer concentration in the most mobile pathway only. This leads to an overestimation of tracer mass flux. Hence, the cumulative tracer mass recoveries shown in panel D of Figures 7.11, 7.12 and 7.13 indicated lower cumulative recovery in outflow compared to the flux past the suction lysimeters at the base of the pile. Later in the event, the tracer concentration within water flowing in the larger pores has already started to decline, similar to Figure 7.10. When the estimate of stored mass falls below 1 in Figure 7.14, this indicates the most conductive pore space sampled by the suction lysimeters are now relatively free of chloride compared to the overall distribution of chloride in the remaining pores. The long time period (1 year) over which these changes in the estimates of stored mass occur indicates that the uneven tracer distribution at the pore scale persists over a long time scale. During transient infiltration, the passage of each wetting front changes the geometry of the saturated pore space as the water content increases. The mixing between the water contained within the smallest pores and the largest pores is thus slow. Suction lysimeters are commonly used to determine the geochemistry of drainage products at existing waste rock piles. These results indicate that the samples obtained from suction lysimeters must be treated with caution. In heterogeneous waste rock, the water obtained is not representative of the fastest moving pathways within the granular matrix, nor any flow over clast-supported cobbles and boulders. These pathways are unlikely to be intercepted by any single instrument. Furthermore, the samples obtained are more representative of a flux-averaged concentration, than a volume-averaged concentration. If a total dissolved contaminant inventory is being compiled to determine long term drainage from a waste rock pile, suction lysimeters are poor (non-conservative) estimators of volume averaged concentration, or total stored dissolved mass of weathering products. 218 7.5.10 Residence Time Estimates The tracer test data can be used to derive estimates of the velocity of the first moment of the tracer pulse which can be used to derive average residence time. Residence times can be estimated from: net infiltration; BTC curves from suction lysimeters; and tracer BTC's at the base of the CPE. Average residence time for the CPE can be determined from the average annual average rainfall rate (303 mm, 1981-1997 average), average net infiltration under natural rainfall conditions (55%, Chapter 6), and using an average water content of 0.1 in the upper meter, and 0.16 in the lower four meters. The average residence time calculated is 4.4 years. Using the rainfall rate recorded during the first year of the tracer test, 487 mm, the estimated residence time is 2.8 years. This estimate assumes water moves by piston flow through the whole water content, which is in conflict with the estimates of resident tracer mass (Figure 7.14), which indicates tracer movement is dominated by a limited fraction of the pore water. BTC data from suction lysimeters were examined to determine the cumulative outflow flux at the peak in concentration, and the cumulative outflow flux at 50% of the recovered mass or the first moment. The peak concentration always preceded the first moment, indicating preferential flow behaviour in the monitored pore fraction and non-Fickian dispersion. The cumulative outflow flux from the first moment was used to determine the cumulative outflow flux required for the first moment to reach 5 m depth. This was converted to a residence time by application of the average annual rainfall, and the estimated net infiltration rate. The result is presented in Figure 7.15, as estimated residence time against depth of sensor. It should be noted that the cumulative outflow used is that measured from the underlying lysimeters, which includes any lateral re-distribution of water. 219 Data from deeper sensors are more consistent, and indicate a residence time from 0.7 to 1.0 years in profiles A and C, and 4.7 years for profile B . The samples derived from suction lysimeters are biased towards the more mobile fraction of the water content, and therefore will generally underestimate residence time. The residence time estimate for profiles A and C is between 25% and 35% of the estimate from net infiltration. This is in general agreement with the discrepancy in stored tracer mass estimates which indicated 12 to 25% of the resident water content dominated tracer transport. Data from sensors near surface, 0.2 to 1.0 m depth, indicate much higher estimated residence times. This is likely due to a variety of factors. The flux past shallow lysimeters may be poorly represented by the basal outflow in the underlying lysimeter, as the degree of lateral re-distribution of water is unknown. The shallow suction lysimeters are also within the zone of greater changes in water content due to infiltration and evaporation. During the initial infiltration of the tracer application event, water infiltration is driven by both gravity, and the capillary gradient. Early time infiltration is dominated by the sorptivity of the soil, and tracer labeled water will enter into storage depleted by near surface evaporation. Later in the infiltration event, water infiltration is dominated by gravity forces, and the infiltration capacity. That water taken up by initial sorptivity may remain longer in the near surface zone where further repeated evaporation and infiltration causes greater mixing of chloride between different pore sizes during wetting and drying cycles. Similarly discrepancies between lower near surface velocities, and higher velocities at depth have been found in other studies (Schulin et al., 1987; Butters and Jury, 1989). The wider variations in water flow velocity created by varying conditions near surface are damped with depth, and more steady water flow occurs (Destouni, 1991). 220 A third preliminary estimate of residence time can be obtained from the outflow BTC's. Estimation of the arrival of the first moment is difficult for individual lysimeters because the final tally of total mass reporting to each lysimeter is not known. A n estimate can only be derived for the whole pile. Visually estimating an extension to the BTC in Figure 7.4 indicates an approximate arrival of the first moment at 0.5 to 0.65 m cumulative outflow flux, which corresponds to 1.8 to 2.4 years at the higher rainfall rate over the tracer test, or 3.0 to 3.9 years at the lower longer term average. This preliminary estimate favours an earlier arrival of the first moment under current infiltration conditions than indicated by the assumption of piston flow. The residence time estimate is quite close to the estimate of 2.8 years assuming piston displacement at the current rate. The actual transport occurring in the CPE is therefore not as rapid as would be suggested by the limited water content participating in transport suggested by Figure 7.14. This confirms that the suction lysimeters sample only a mobile fraction of the true tracer mass located within the CPE. 7.6 C O N C L U S I O N S The tracer test was conducted to determine the physical mechanisms of flow governing the movement of water, and to determine the residence time distribution of water within the CPE. The tracer test has revealed the following: • The hydrographs and BTC's of individual 2m x 2m lysimeters monitor the integrated arrivals of flow and tracer from multiple flow paths. • The hydrograph and BTC of the whole pile, 8m x 8m, still contains evidence of multiple flow paths. 221 Some of these flow paths are not dominated by capillary forces, and have little interaction with the granular matrix. These contribute a small proportion (<0.1%) of the total outflow. Some event water is able to flow 5m with a single rainfall event. This water flows within macropores or spatially distinct zones of coarser material, and proceeds at the same velocity as the wetting front. This water accounts for ~5 % of the outflow of larger rainfall events. The majority of the water (>90%) in outflow as the result of any individual rainfall event is pre-event water. During a transient tracer test, the contributions of different flow paths are temporally separated. Fast flow paths contribute more to the flux average soon after a rainfall event and slow flow paths contribute more later in the flow response. The interval between outflow samples must be flux-determined, not temporally determined. Long term monitoring of the water quality of outflow from the base of a waste rock pile (such as a in a toe-drain) on a daily or weekly basis will not represent the long term flux-average. Observations of tracer movement from in-situ samplers indicates that preferential tracer transport occurs both in spatially distinct areas of coarse material (macro-scale heterogeneity) and at the pore scale by the dominance of large pores in the water and tracer fluxes. The suction lysimeter data indicate that these samplers do not collect unbiased samples. The samples are closer to flux-averaged samples, but are biased towards the more mobile water fraction of the flow path in which the instrument is located. The fastest flow paths are likely bypassing the instrument locations. Estimates of transport properties derived 222 from suction lysimeter samples would lead to an overestimation of first breakthrough, and an underestimation of average residence time. The interpretation of the geochemistry of water samples collected from in-situ suction lysimeters must take this sample bias into account. Water contained in spatially distinct flow paths caused by porous medium heterogeneity at the field scale have different residence times, and hence different chemistry. At the pore scale, within the spatially distinct flow pathways, water in larger pores has a different chemistry than water in smaller pore spaces. The water chemistry of the outflow at the base of pile is the flux-average of the water chemistry resident within the pile. Water chemistry wil l change with flow rate as the proportion of the flux-average changes. Mean water residence time is 1.8 to 2.4 years at the rainfall rates occurring during the tracer test to date. This corresponds to 3.0 to 3.9 years at the average annual rainfall rate. The residence time distribution is expected to have a long tail. 223 C L Q "Fast" Tracer concentration • • i i "Med ium" Tracer concentration "S low" Tracer concentration Center of m a s s Solute velocity P lane of observat ion 0 t, u Infiltration Event s Wetting Front Arrival Time (t) Wetting Front Z/ts Kinematic Velocity (Z/t) Fast Figure 7.1: Conceptual framework of transient water and tracer movement in unsaturated heterogeneous porous media: (A) In-situ tracer profiles; (B) Outflow hydrograph and ( C) flux-averaged tracer concentration in outflow. 224 60 50 H 40 A E E o -4—' ro | . 30 o £ 20 10 1 cn E, 300 c o ro ~ 200 cu o c o CD 100 T3 o 0 I I I I I I I I I I I I I I I I I I I I I I I I 2/99 7/99 1/00 7/00 Date (month/year) 1/01 F igure 7.2: Precipitation (A), outflow flux rate (m s'1) (B) and flux-average tracer concentration (C) plotted against time for the experimental period from February 1999 to March 2001. 225 CO E x o ZS O CD O c o cc I — c CD o c O O 0 TJ O Tracer Application Sept 24, 1999 July 1 3 - 2 3 2000 Oct 1 0 - 2 0 1999 2 100 H -0.2 August - Oct 2000 -0.1 0 0.1 0.2 0.3 Normalized outflow volume (m3m'2) 0.4 Figure 7.3: Outflow flux rate (m s"1) (A) and flux-average tracer concentration (B) plotted against cumulative outflow volume normalized to cross sectional area, and zeroed at the tracer application event. The experimental period from February 1999 to March 2001 is shown. 226 Figure 7.4: Cumulative tracer recovery in outflow of the whole C P E experiment, normalized to the mass of applied tracer. 227 o l d 600 P e a k 1120 i i i i i i i • i J 13 14 P e a k 740 P e a k 960 • I ' T I I I 15 I • I I T * 1 P e a k . 8 700 i i i i i i i i i i 12 8/99 12/99 4/00 8/00 8/99 12/99 4/00 8/00 8/99 12/99 4/00 8/00 8/99 12/99 4/00 8/00 Date (month/year) F igure 7.5: Chloride concentration in outflow plotted against time for the grid of sixteen contiguous lysimeters at the pile base. 228 Figure 7.6: Spatial variation of tracer transport. Spatial variation in tracer mass recovery (A) and normalized outflow volume (m3m'2) (B) by lysimeter. Fraction of tracer mass recovered recovered plotted against total outflow volume (C). Spatial variation in the recovery ratio: the fractional tracer recovery divided by the normalized volume per lysimeter (m3m'2)(D). 229 Figure 7.7: Breakthrough curves (A, C) and outflow hydrographs (B, D) for l