Open Collections

UBC Theses and Dissertations

UBC Theses Logo

UBC Theses and Dissertations

Diagnosis of headwater sediment dynamics in Nepal’s middle mountains: implications for land management Carver, Martin 1997

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

Item Metadata

Download

Media
831-ubc_1997-250296.pdf [ 18.15MB ]
Metadata
JSON: 831-1.0076943.json
JSON-LD: 831-1.0076943-ld.json
RDF/XML (Pretty): 831-1.0076943-rdf.xml
RDF/JSON: 831-1.0076943-rdf.json
Turtle: 831-1.0076943-turtle.txt
N-Triples: 831-1.0076943-rdf-ntriples.txt
Original Record: 831-1.0076943-source.json
Full Text
831-1.0076943-fulltext.txt
Citation
831-1.0076943.ris

Full Text

DIAGNOSIS OF H E A D W A T E R SEDIMENT D Y N A M I C S I N N E P A L ' S M I D D L E MOUNTAINS: IMPLICATIONS F O R L A N D M A N A G E M E N T  by Martin Carver B.Sc, University of Manitoba, 1984 M.A.Sc., University of Waterloo, 1988  A THESIS SUBMITTED I N P A R T I A L F U L F I L M E N T O F THE REQUIREMENTS FOR THE DEGREE OF DOCTOR OF PHILOSOPHY IN T H E F A C U L T Y OF G R A D U A T E STUDIES  (Interdisciplinary Studies in Resource Management Science)  We accept this thesis as conforming to the required standard  The University of British Columbia June, 1997 ® Martin Carver, 1997  In  presenting  degree at the  this  thesis in  University of  partial  fulfilment  of  the  requirements  British Columbia, I agree that the  for  an advanced  Library shall make it  freely available for reference and study. I further agree that permission for extensive copying  of  department  this thesis for or  by  his  or  scholarly purposes may be granted her  representatives.  It  is  by the  understood  that  head of copying  my or  publication of this thesis for financial gain shall not be allowed without my written permission.  The University of British Columbia Vancouver, Canada  Date  DE-6 (2/88)  11  Diagnosis of Headwater Sediment Dynamics in Nepal's Middle Mountains: Implications for Land Management Abstract  An evaluation of headwater erosion and sediment dynamics was carried out to assess the health of the Middle Mountain agricultural system in Nepal. Controversial statements predicting this system's imminent demise and identifying Middle Mountain farming practices as major contributors to downstream sedimentation and flooding have long been promoted and have suggested the following research hypothesis: soil and sediment dynamics and the indigenous management techniques within headwater Middle Mountains basins do not indicate a deterioration in the health of the agricultural system. Three questions were addressed in this research. What are the main controls on normal-regime erosion? How effective is the system of indigenous management at modifying sediment dynamics? What do headwater sediment budgets (erosion, storage, and yield) reveal about the health of the agricultural system? Answers to these questions are suggested and development initiatives proposed. Intensive monitoring was carried out during 1992-1994 within nested basins ranging in size from 72 to 11 141 ha. Variation of storm-period variables in time and space was assessed using five recording rain gauges and a network of up to fifty 24-hour gauges. Surface erosion was measuredfromfive erosion plots on steep bari (rainfed cultivated land). Suspended sediment behaviour was examined through event sampling at seven hydrometric stations. Basin sediment yield was determined for three of these nested basins. Sediment storage was assessed using accumulation pins in khetfields(irrigated cultivated land), khet  canals, and bari ditches and through erosion and channel surveys. An annual average of 77 storms were identified over the three-year period with 3.5% of these  delivering more than 30 mm total rainfall and a peak 10-minute rainfall intensity of more than 50 mm/h. About 1/3 of all storms regardless of magnitude occurred during the pre-monsoon season. Pre-monsoon and monsoon storms delivered equivalent high-intensity short-term rainfall disputing the hypothesis that it is a higher rainfall intensity in the pre-monsoon season which causes an elevated sediment regime  Ill  during that season. Total storm rainfall was significantly higher during the monsoon season whereas the period without rain before a storm begins was longer for pre-monsoon storms. The source of suspended sediment was found to vary with season and spatial scale. During the pre-monsoon season, surface erosionfrombari was severe when high-intensity rain fell on bare ground. Indigenous farming practices were found to be effective at limiting surface erosion except during the premonsoon season when targeted intervention may be useful. During the pre-monsoon season, nutrient loss from headwater basins due to sediment export was at its highest. Severely degraded land remained bare throughout the rainy season, producing sediment at an elevated rate and in relation to total rainfall. The onset of the monsoon season reduced this bari source markedly due to the complete development of a vegetative cover under conventional management. The pre-monsoon-season surfaceerosion mechanism of sediment production was replaced with scale-dependent mechanisms resulting from the higher total rainfall of monsoon-season storms. Within the steep terraced hillslopes, the capacity of runoff ditches was more often exceeded resulting in episodic-regime rilling, gullying, and in some instances, terrace failure. When sufficiently heavy and widespread, monsoon storm rainfall led also to stream discharge high enough to damage riparian areas and the system of irrigation dams. The farmers alter the sediment regimes profoundly and their management activities reduce soil loss collectively over all spatial scales. Sediment budgets reveal that a significant component of the sediment produced in the study basin (5.3 km) was recaptured (35% to 50%) because of these indigenous 2  farming practices. Objective calibration of indigenous knowledge showed it to be well founded but inconsistent. Farmers practise techniques which are well adapted to this environment reflecting their stated receptiveness to innovation and outside support. The detailed measurements show that the important controls on erosion are variable temporally and spatially over scales too small to be considered by conventional monitoring programs in these environments. Spatial differences in rainfall delivery, hysteresis effects, variability in land-surface response, and management activities conspire to yield sediment dynamics which are difficult or impossible  iv to quantify with typical limited monitoring. Site-specific opportunities for investigation should be exploited and a high degree of uncertainty be anticipated. Management recommendations focus on two topics. An improved vegetative cover during the premonsoon season is required to reduce soil erosion during that period. Greater retention of these nutrientrich soils would directly benefit the upland farmer. Rehabilitation of degraded lands and the halting or reversing of further degradation would benefit all farmers by providing a greater land base for biomass production especially in light of an increasing population. Both strategies would benefit hydropower developments by limiting reservoir sedimentation. Above all, proposed changes should enhance - not undermine - indigenous management. Current soil dynamics may be sustainable but it is unlikely that they can remain so in the future under the increased landuse intensification that may be necessary with projected population increases unless support is provided strategically from outside sources. Working with the farmers to develop techniques to improve their ability to recapture previously-eroded soil is a useful area of applied research. The high degree of skill and adaptability of the farmers within this environment suggest that carefully designed intervention which targets vulnerable aspects of the agricultural system while not undermining the present methods have a reasonable likelihood for success.  V  Diagnosis of Headwater Sediment Dynamics in Nepal's Middle Mountains: Implications for Land Management TABLE OF CONTENTS  Abstract Table of Contents List of Figures List of Tables List of Abbreviations List of Symbols Acknowledgements Foreword  u v  *  x  x i v x v x  ' * * u  x  x x  x x n  PART I Biophysical Analyses  l  1. General 1.1 1.2 1.3  1 1 2 5  Introduction Problem statement Research context Goals and Objectives  2. Study Area 2.1 Jhikhu River basin 2.1.1 Location and physiography 2.1.2 Farming system and landuse 2.1.3 Regional climate 2.1.4 Local climate 2.2 Study catchments 3. Methods 3.1 Field methods 3.1.1 Climate 3.1.2 Stream measurements 3.1.3 Erosion plots 3.1.4 Soil properties 3.1.5 Soil movement 3.1.6 Mapping 3.1.7 Interviews and questionnaires 3.2 Laboratory methods 3.2.1 Stream sediment samples 3.2.2 Stream water samples 3.2.3 Soil samples 3.3 Data Synthesis 3.3.1 Stage-discharge relations 3.3.2 Automated data 3.3.3 Tipping-bucket rainfall data 3.3.4 Geographic Information System  8  8 8 10 11 12 19 2 6  26 26 31 32 33 34 36 38 39 39 40 40 4 1  41 42 42 43  vi 4. Characterising and Monitoring Monsoonal Rainfall for Studying Erosion and Sedimentation 45  4.1 Introduction 4.2 Research background 4.2.1 Patterns of rainfall delivery 4.2.2 The Asian monsoon . 4.3 Definitions 4.3.1 Storm 4.3.2 Storm-period variables 4.4 Storm-period variables 4.4.1 Distributions 4.4.2 Seasonal distributions 4.4.3 Event ranking 4.5 Spatial variation 4.5.1 Elevation 4.5.2 Storm cell 4.6 Integration 4.6.1 Classification . . 4.6.2 Storm frequency 4.7 Summary and conclusions 4.7.1 Summary of quantitative findings 4.7.2 Conclusions 5. Diagnosing Headwater Controls on Erosion and Sediment Transport  5.1 Introduction 5.2 Research background 5.2.1 Behaviour of fine-sediment erosion and transport 5.2.2 Quantitative Himalayan data 5.3 Surface erosion on cultivated rainfed uplands 5.3.1 Controlling factors 5.3.2 Erosion plots: annual regimes 5.3.3 Erosion plots: seasonal regimes 5.3.4 Erosion plots: event regimes 5.4 Stream sediment regimes 5.4.1 Controlling factors 5.4.2 Seasonal regimes 5.5 Landuse signatures 5.5.1 Surface degradation: comparison of two basins 5.5.2 Surface degradation: within Lower Andheri basin 5.5.3 Sediment storage by water diversion for irrigation 5.6 Conclusions  45 47 47 51 54 54 57 58 59 62 64 68 68 71 82 82 85 86 87 90 92  92 93 93 103 107 108 108 110 114 119 • • 120 120 134 136 139 146 151  VII  6. Signatures of Erosional Sources and Sediment-Transport Behaviour in the Physical and Chemical Character and the Patterns of Movement of Suspended Sediments  6.1 Introduction 6.2 Research background 6.2.1 Hysteresis 6.2.2 Particle-size behaviour 6.2.3 Fingerprinting 6.3 Hysteresis 6.3.1 Single events 6.3.2 Multiple events 6.3.3 Implications 6.4 Entrainment and transport behaviour by particle-size class 6.4.1 Controlling factors 6.4.2 Seasonal regimes 6.4.3 Hysteresis 6.5 Fingerprints of suspended sediment 6.5.1 Sediment properties 6.5.2 P-Q relations 6.5.3 Implications 6.6 Conclusions  PART II Management and Implications  153  153 153 154 158 164 168 169 177 180 181 181 184 202 208 208 212 224 228  231  7. The Influence of Indigenous Management on Sediment Dynamics in the Middle Mountains of Nepal 231  7.1 Introduction 7.2 Research background 7.2.1 Indigenous knowledge 7.2.2 Indigenous management 7.3 Environmental perceptions and system of soil classification 7.3.1 Farmer attitudes and perceptions 7.3.2 Soil classification 7.3.3 Significance 7.4 Techniques of water management and erosion control 7.4.1 Description of some techniques observed in the Jhikhu basin 7.4.2 Irrigated lands 7.4.3 Rainfed lands 7.4.4 Significance 7.5 Implications of quantitative study for indigenous management 7.6 Conclusions  231 231 232 235 236 237 240 245 246 246 249 253 255 255 • 256  viii 8. Sediment Budgets: Implications for Landuse Management  8.1 Introduction 8.2 Research background 8.2.1 The sediment-budget technique 8.2.2 Sediment-yield calculation methods 8.2.3 Case studies 8.3 Sediment sources and pathways 8.4 Components of sediment budget 8.4.1 Normal-regime behaviour 8.4.2 Episodic sediment production 8.5 Sediment budgets in space and time 8.5.1 Detailed basin sediment budgets 8.5.2 Basin sediment production and delivery across temporal scales 8.5.3 Three-year sediment production and yield across spatial scales 8.6 Implications for nutrient loss 8.7 Conclusions and recommendations 8.7.1 Conclusions 8.7.2 Recommendations 9. General Discussion, Conclusions, and Recommendations  9.1 Conclusions 9.2 General discussion 9.3 Recommendations 9.3.1 Farming system 9.3.2 Monitoring 9.3.3 Further research 9.4 Postscript Appendices  259  259 259 260 266 269 274 277 277 289 295 296 298 304 309 311 311 313 316  316 319 324 324 326 327 328 329  Al. Photographs of the Study Area 329 A2. Descriptive Rainfall Statistics of Study Area 331 A3. Location and Performance Information of Rain Gauges 334 A3.1 Location 334 A3.2 Performance characteristics 336 A4. Questionnaires, Interviews, and Site Description for Indigenous Knowledge 339 A4.1 Perceptions, attitudes, and approaches 339 A4.2 Soil classification (1992) 339 A4.3 A/ier-accumulation management 340 A4.4 Soil classification (1992) 341 A4.5 5ari-erosion management 342 A4.6 Irrigation-dam management 343 A4.7 Site descriptions of soils described by farmers (1992 & 193 interviews) . . 344 A5. Stage-Discharge Relations 345 A6. 1992 Survey of Mass Wasting 350 Literature Cited  353  ix LIST OF FIGURES  Figure 2.1  Location of the Jhikhu River basin within Nepal  9  Figure 2.2  Mean, maximum, and minimum monthly rainfall at (a) Kathmandu airport (HMG Records 1968-1990) and (b) Panchkhal (HMG Records 1978-1994;, means based on 1978-1985 and 1988-1994) 13  Figure 2.3  Mean, maximum, and minimum monthly rainfall at (a) Baluwa (1992-1994; means based on 1993-1994), (b) Bela (1990-1994), and (c) Dhulikhel (1990-1994) 15  Figure 2.4  Mean monthly rainfall (a) and maximum 24-hour rainfall (b) at Kathmandu Airport (1968-1990), Panchkhal (1978-1985 and 1988-1994), Baluwa (1993-1994), Bela (19901994), and Dhulikhel (1990-1994) 16  Figure 2.5  Mean-monthly and extreme-monthly maximum/minimum temperatures at (a) three lowelevation sites (Panchkhal, 1978-1994; Baluwa, 1993-1994; Bhimsenthan, 1993-1994) and (b) three high-elevation sites (Kathmandu airport 1968-1990;, Bela, 1990-1994; Bhetwaltok, 1993-1994) 18  Figure 2.6  Locations of the nested catchments, the hydrometric stations, and the erosion plots 20  Figure 2.7  Elevational profiles of the study streams (vertical scale is exaggerated by 10 times) 23  Figure 3.1  Lay-out of monitoring network for rainfall and temperature within the Jhikhu River basin 27  Figure 4.1  Number of storms in relation to (a) generalised minimum-time-without rain, and (b) specific minimum-time-without-rain at the five study locations 56  Figure 4.2  Cumulative frequency distributions of storm-period variables (1992-1994) at the five study locations: (a) maximum 10-minute intensity (1 ), (b) maximum 60-minute intensity 0«>)> (c) total rainfall (R T)> (d) duration (T ), (e) start of maximum 10-minute rainfall intensity (T ), (f) start of maximum 60-minute rainfall intensity (T ), and (g) period without rain before storm start (S) 60 10  DUR  TO  10  Figure 4.3  10  Ranking of largest monitored storms in upland and lowland of Andheri basin by (a) I and (b) R 65  10  T  Figure 4.4  Ranking of monitored floods at (a) Lower Andheri, (b) Kukhuri, and (c) Jhikhu stations 67  Figure 4.5  Effect of elevation on rainy-season rainfall at eight monitoring stations distributed across the Jhikhu River basin (100 km; 1992-1994): (a) June-September, (b) June, (c) July, (d) August, and (e) September . . 69 2  Figure 4.6  Effect of elevation on rainy-season rainfall at 20 monitoring stations located on the Andheri basin hillslope (10 km; 1992-1994): (a) June-September, (b) June, (c) July, (d) August, (e) September 70 2  X  Figure 4.7  Effect of elevation on total rainfall at 18 monitoring stations located on the south-facing hillslope (10 km ; 1992): (a) July - only 7 data available, (b) August, and (c) September 72 2  Figure 4.8  Rainfall isolines for storm rainfall over Andheri basin for (a) an upland event during the transition season which was also the heaviest event of the study period and (b) a basin event typical of the monsoon season 77  Figure 4.9  Rainfall isolines for storm rainfall over Andheri basin for upland events during the (a) pre-monsoon season and (b) monsoon season 78  Figure 4.10  Rainfall isolines for storm rainfall over Andheri basin for lowland events during the (a) pre-monsoon season and (b) monsoon season 79  Figure 4.11  Number of rainfall monitoring sites required in relation to the relative error for a range in coefficient of variation 81  Figure 5.1  Soil loss from the erosion plots on an event basis, 1992-1994  Figure 5.2  Average maximum maize height (a) and maize leaf length (b) at all erosion plots in 1994 113  Figure 5.3  The effect of maximum 10-minute rainfall intensity (J ) on soil loss at all erosion plots, seasonally stratified for all events of 1992-1994 115  Figure 5.4  The effect of maximum 10-minute rainfall intensity flio) on event runoff coefficient ( C R ) at all erosion plots, seasonally stratified for all events of 1992-1994 117  Figure 5.5  Relation between runoff coefficient ( C R ) and event soil loss at all erosion plots, seasonally stratified for all events of 1992-1994 118  Figure 5.6  Sediment rating curves for monitored hydrometric stations based on entire data set (19921994): (a) Kukhuri, (b) Upper Andheri, (c) Lower Andheri, (d) Dhap, and (e) Jhikhu 121  Figure 5.7  Seasonally-stratified sediment rating curves based on all entire data set: (a) Kukhuri, (b) Upper Andheri, (c) Lower Andheri, (d) Dhap, and (e) Jhikhu 126  Figure 5.8  Seasonal variation in a, and b with basin area  Figure 5.9  Seasonal sediment rating curves overlain for all study basins showing both the expected functional relations and envelopes representing confidence limits to these relations (at 90%): a) monsoon regime b) pre-monsoon regime 131  Figure 5.10  Suspended sediment data of the transition season and the seasonal regression regimes based on the entire data set: (a) Kukhuri, (b) Upper Andheri, (c) Lower Andheri, (d) Dhap, and (e) Jhikhu 133  Figure 5.11  Seasonally-stratified suspended-sediment data for the mid-reach Andheri River station based on 1993-1994 data: (a) mid-reach #1 (11), and (b) mid-reach #2 (12) . . . . 135  112  10  f  ( C =  A  Q  B  )  128  xi Figure 5.12  Seasonal sediment rating curves for Lower Andheri and Dhap basins overlaid for comparison 137  Figure 5.13  Sediment-phosphorus rating curves stratified by rainfall location and colour for Lower Andheri basin (2) based on (a) rainfall differences only, and (b) both rainfall differences and their relative upland-lowland sediment contributions 141  Figure 5.14  Locations of the rain gauges within and near the Andheri basin overlain on the red soil map 143  Figure 5.15  Sediment rating curves for Lower Andheri station (2) including sedimentsfromonly lowland and upland events (1992-1994) 144  Figure 5.16  Map of Andheri basin showing all irrigation diversion dams on the Kukhuri and Andheri rivers 148  Figure 6.1  C-Q relations for events within Kukhuri basin (station 10) illustrating single-value behaviour 171  Figure 6.2  C-Q Relations for events within Andheri Lower basin (2) and Dhap basin (3) illustrating CCW hysteretic behaviour 173  Figure 6.3  C-Q relations for events within Jhikhu basin (station 1) illustrating CW hysteresis behaviour 175  Figure 6.4  CO behaviours: (a)-(c) step CO C-Q behaviour at Jhikhu and Lower Andheri stations, and (d) CO behaviour at Kukhuri station 176  Figure 6.5  Functional relations for rising and falling limbs contrasted within the pre-monsoon and monsoon seasons 179  Figure 6.6  Comparison of pre-monsoon and monsoon sediment rating curves for four particle-size classes at Kukhuri station (10) based on functional analysis 187  Figure 6.7  Comparison of pre-monsoon and monsoon sediment rating curves for four particle-size classes at Upper Andheri station (9) based on functional analysis 188  Figure 6.8  Comparison of pre-monsoon and monsoon sediment rating curves for four particle-size classes at Lower Andheri station (2) based on functional analysis 189  Figure 6.9  Comparison of pre-monsoon and monsoon sediment rating curves for four particle-size classes at Dhap station (3) based on functional analysis 190  Figure 6.10  Comparison of pre-monsoon and monsoon sediment rating curves for four particle-size classes at Jhikhu station (1) based on functional analysis 191  Figure 6.11  Seasonal contrasts of functional relations (b and aj) for suspended coarse- and fine-sand fractions 195 5  XII  Figure 6.12  Seasonal contrasts of functional relations (b and af) for suspended silt and clay fractions 196  Figure 6.13  Seasonal functional relations for suspended coarse sand contrasted by basin area  Figure 6.14  Seasonal functional relations for suspendedfinesand contrasted by basin area . . 198  Figure 6.15  Seasonal functional relations for suspended silt contrasted by basin area  200  Figure 6.16  Seasonal functional relations for suspended clay contrasted by basin area  201  Figure 6.17  Seasonal functional relations for Kukhuri basin for four particle-size classes . . . 203  Figure 6.18  Seasonal functional relations for Upper Andheri basin for four particle-size classes4203  Figure 6.19  Seasonal functional relations for Lower Andheri basin for four particle-size classes 205  Figure 6.20  Seasonal functional relations for Dhap basin for four particle-size classes  Figure 6.21  Seasonal functional relations for Jhikhu basin for four particle-size classes . . . . 207  Figure 6.22  C-Q graphs by particle size for pre-monsoon and transition season events at four stations 209  Figure 6.23  C-Q graphs by particle size for monsoon-season events at four stations  Figure 6.24  Soil colour and available-phosphorus maps for Andheri basin  213  Figure 6.25  Soil colour map for Jhikhu basin  214  Figure 6.26  Sediment rating curve stratified by colour and sediment-phosphorus rating curve stratified by both colour and season for Kukhuri basin (10) 215  Figure 6.27  Sediment rating curve stratified by colour and sediment-phosphorus rating curve stratified by colour and season for Upper Andheri basin (9) 216  Figure 6.28  Sediment and sediment-phosphorus rating curves stratified by colour and season for Lower Andheri basin (2) 217  Figure 6.29  Rainfall distribution over Andheri basin for an upland and a lowland pre-monsoon event, overlaid on the soil colour map 220  Figure 6.30  C-Q (a) and P-Q (b) graphs stratified by rainfall location (upland and lowland) and sediment colour (red and brown) for Lower Andheri basin (2) 221  Figure 6.31  Sediment and sediment-phosphorus rating curves stratified by season and colour for Dhap basin (3) 223  Figure 6.32  Sediment and sediment-phosphorus rating curves stratified by season and colour for Andheri Mid #2 basin (12) 224  f  197  206  210  Xlll  Figure 6.33  Sediment and sediment-phosphorus rating curves stratified by season and colour for Andheri mid #1 basin (11) 225  Figure 6.34  Sediment and sediment-phosphorus rating curves stratified by season and colour for Jhikhu basin (1) 226  Figure 7.1  Nutrient enrichment in khet fields due to annual deposition  Figure 8.1  Sediment routing within the Andheri River basin including sources and opportunities for storage 275  Figure 8.2  Sediment yield at Lower Andheri in relation to average rainfall for lowland events and corrected for included production and storage from non-degraded land 281  Figure 8.3  Annual fine-sediment yield andfine-sedimentdelivery ratio in relation to scale during 1992-1994 308  Figure Al.l  Terraced agriculture in the study area: (a) rainfed (pari) and (b) irrigated (khet) . 329  Figure A1.2  Reach of the Lower Andheri River during mid-monsoon at (a) high flow and (b) low flow 330  Figure A3.1  Catch ratios (Custom/Tipping-bucket) at Sites 1, 2, 3, and 4  Figure A3.2  Catch ratios (Custom/Tipping-bucket & Custom/HMG) at Sites 5, 6, and 7 . . . . 336  Figure A5.1  Stage-discharge relationship at station 1, Jhikhu River at Bhendabaribesi  343  Figure A5.2  Stage-discharge relationship at station 2, Lower Andheri River  344  Figure A5.3  Stage-discharge relationship at station 3, Dhap River at Shree Rampati  345  Figure A5.4  Stage-discharge relationship at station 9, Upper Andheri River  346  Figure A5.5  Stage-discharge relationship at station 10, Kukhuri River at Andheri River . . . . 347  257  335  XIV  LIST OF TABLES  Table 2.1  Slope, aspect, and elevation of the Jhikhu River basin (from 1:20 000 mapping) . . 10  Table 2.2  Landuse of the Jhikhu River basin in 1990  Table 2.3  Descriptive temperature statistics from the six climate stations  Table 2.4  Topography of the six study catchments based on 1994 1:5 000 mapping. Data for the Dhap catchment are taken from 1990 mapping at 1:20 000 scale 21  Table 2.5  1990 landuse distribution of study sub-catchments within the Andheri basin, mapped  11 19  at 1:20 000 scale. (Total area based on 1994 1:5 000 photogrammetric map.) . . . 22 Table 2.6  Slope, order, and bankfull discharge of the study streams at each hydrometric statior22  Table 3.1  Comparison of catch ratios of the rain gauges at seven test sites  Table 3.2  Installation information of temperature measurements  31  Table 3.3  Characteristics of erosion plots in the study  33  Table 3.4  Topics examined for farmer indigenous knowledge  38  Table 3.5  Tolerances used to parse automated data  43  Table 4.1  Published storm definitions  49  Table 4.2  Frequency of occurrence of four storm-period variables in three classes  Table 4.3  Spatial variation within a storm cell and distribution of events according to lowland, upland, and basin area-events for storms of R ^ 3mm and R ^ 10mm 74 Seasonal distributions of spatial variation within a storm cell and distribution of events according to lowland, upland, and basin area-events for storms of R ^ 10 mm total rain 76  Table 4.4  T  29  61  T  T  Table 4.5  Definitions of minor, intermediate, and major storm classes in relation to total rainfall and peak 10-minute rainfall intensity 83  Table 4.6  Distribution of storm events in three storm classes (minor, intermediate, major) at five sites 83  Table 4.6  Distribution of storm events in three storm classes (minor, intermediate, major) at five sites 84  Table 4.8  Number of storms in each year (1992-1994) at each site as given by recorded data 86  Table 4.9  Expected average seasonal storm frequency by class (minor, intermediate, major) within the study area 86  Table 5.1  Surface erosion rates as determined by field studies using erosion plots within or near the Middle Mountains 104  XV  Table 5.2  Surface erosion rates as determined by field studies using check dams and hydrometric stations 105  Table 5.3  Annual rate of soil loss (tonnes/ha) from all plots, 1992-1994  Table 5.4  Surface-soil characteristics of erosion plots  Table 5.5  Percentage of each plot's annual erosion occurring in the pre-monsoon, transition, and monsoon season, 1992-1994 110  Table 5.6  The percentage of the total annual erosion at each plot which occurred in the two most-damaging events of each year, 1992-1994 114  Table 5.7  Sediment-rating-curve relations based on seasonally-stratified data using log-linear regression (1992-1994, assuming Q known without error) excluding datafromthe transition season 124  Table 5.8  Sediment rating-curve relations derived using functional analysis  Table 5.9  Annual and seasonal sediment-rating-curve relations for Lower Andheri station based on lowland/upland data using log-linear regression 145  Table 5.10  Contributing areas and number of irrigation dams for Kukhuri and Lower Andheri hydrometric stations (stations 10 and 2 respectively) 147  Table 5.11  Flow-ratio comparison of 16 individual floods for Kukhuri (station 10) and Lower  108 109  127  Andheri (station 2) hydrometric stations  149  Table 5.12  Sediment accumulation in irrigated fields measured using pegs  151  Table 6.1 Table 6.2  Theoretical models of primary hysteretic behaviour Classes of sediment behaviour observed in C-Q graphs  Table 6.3  Seasonal sediment-rating-curve relations derived using log-linear regression (19921994, Q known without error) for rising and falling hydrograph limbs 177  Table 6.4  Seasonal sediment-rating-curve relations derived using functional analysis for rising and falling hydrograph limbs 178  Table 6.5  Dominant hysteresis indicated by net behaviour of suspended sediment of rising and falling hydrograph limbs 180  Table 6.6  Number of samples analysed for particle-size distribution by station and season . 182  Table 6.7  Pre-monsoon sediment-rating-curve relations for coarse sand, fine sand, silt, and clay at stations 1, 2, 3, 9, and 10 using log-linear regression (1992-1994, Q known without error) excluding data from the transition season 185  Table 6.8  Monsoon sediment-rating-curve relations for coarse sand, fine sand, silt, and clay at stations 1, 2, 3, 9, and 10 using log-linear regression (1992-1994, Q known without error) excluding datafromthe transition season 186  155 170  XVI  Table 6.9  Pre-monsoon sediment-rating-curve relations derived using functional analysis for coarse sand,finesand, silt, and clay at stations 1, 2, 3, 9, and 10 excluding data from the transition season 192  Table 6.10  Monsoon sediment-rating-curve relations derived using functional analysis for coarse sand,finesand, silt, and clay at stations 1, 2, 3, 9, and 10 excluding data from the transition season 193  Table 6.11  Annual coarse andfine-sandratings for Kukhuri basin based on data from entire rainy season 194  Table 6.12  Number of sediment samples analyses for colour and phosphorus by station and season  211  Table 6.13  Mean sediment phosphorus content by station and season  Table 7.1  Farmer descriptions of selected soils (1993)  241  Table 7.2. Table 7.3  Definitions of primary terms of indigenous soil classification system Distribution of terms used by farmers to name specific soils (1992)  243 244  Table 7.4  Summary of quantitative results from farmer interviews about the irrigation and khet system 250  Table 7.5  iton-management data based on farmers interviews and detailed measurements . 253  Table 8.1  Annual and seasonal rates of normal-regime sediment production (t/ha) from surface erosion determined for bari, shrub, forest, and grassland 280  Table 8.2  Sediment production by surface erosion from gullied lowlands (tonnes) calculated using an empirical relation derived from lowland rainfall events (limits represent 95% confidence) 282  Table 8.3  Annual and seasonal sediment yield at Kukhuri (10), Lower Andheri (2), and Jhikhu (1) hydrometric stations (limits represent 95% confidence) 284  Table 8.4  Average annual soil accumulation measured annually (1992-1994) in khetfields. 286  Table 8.5  Estimated annual rates of accumulation (cm/yr) and total annual sediment storage by basin (tonnes) in the khetfields(ha), khet canals (m), and bari ditches (m) . . . . 287  Table 8.6  227  Seasonal storage (tonnes) by basin resulting from sediment deposition within khet, khet canals, and bari ditches  288  Table 8.7  The 10 high-flow events at Kukhuri station with the highest sediment yield during 1992-1994 292  Table 8.8  The 14 high-flow events at Andheri station with the highest sediment yield during 1992-1994 294  XVII  Table 8.9  Seasonal and annual sediment budget components (production, storage, yield) in tonnes in 1992 for Kukhuri and Lower Andheri basins 297  Table 8.10  Seasonal and annual sediment budget components (production, storage, yield) in tonnes in 1993 for Kukhuri and Lower Andheri basins 298  Table 8.11  Seasonal and annual sediment budget components (production, storage, yield) in tonnes in 1994 for Kukhuri and Lower Andheri basins 299  Table 8.12  Seasonal and annual sediment production, yield, and overall percentage delivery for the Kukhuri basin (1992-1994) 300  Table 8.13  Seasonal and annual sediment production, yield, and overall percentage delivery for the Lower Andheri basin (1992-1994)  Table 8.14  Seasonal and annual sediment yieldfromthe Jhikhu basin (1992-1994)  Table 8.15  Seasonal and annual sediment yield as percentage of seasonal and annual totals  301 303  respectively for individual and selected groups of events for all spatial scales . . . 305 Table 8.16  Average seasonal sediment budgets across all spatial scales  Table 8.17 Table A2.1  Sediment yield at nested hydrometric stations for single events 310 Average monthly, maximum monthly, minimum monthly, and maximum 24-hour rainfall (mm) at Baluwa, 900 m (1992-1994, averages based on 1993-1994) . . . 331 Average monthly, maximum monthly, minimum monthly, and maximum 24-hour rainfall (mm) at Kathmandu Airport, 1336 m (1968-1986) 332 Average monthly, maximum monthly, minimum monthly, and maximum 24-hour rainfall (mm) at Bela, 1211 m (1990-1994) 332  Table A2.2 Table A2.3  306  Table A2.4  Average monthly, maximum monthly, minimum monthly, and maximum 24-hour rainfall (mm) at Dhulikhel, 1500 m (1990-1994) 333  Table A2.5  Average monthly, maximum monthly, minimum monthly, and maximum 24-hour rainfall at Panchkhal, 865 m (1978-1994, averages based on 1978-1985 and 19881994) 333  Table A3.1  Rain-gauge summary information including gauge number, elevation, and location 334  Table A4.1  Summary of site data for 11 soils selected and described by farmers in 1993 . . . 344  Table A4.2  Summary of site data for 16 soils selected and described by farmers in 1992 . . . 344  Table A6.1  Summary of sediment production and delivery of episodic erosion attributed to inadequate runoff managementfromthe event of July 10, 1992  Table A6.2  351  Summary of sediment production and delivery of episodic erosion attributed to streambank erosion due to high-flow conditionsfromthe event of July 10, 1992 352  XV111  LIST OF ABBREVIATIONS  ASCE  American Society of Civil Engineers  DHM  Department of Hydrology and Meteorology  DSCWM  Department of Soil Conservation and Watershed Management  FAO  Food and Agricultural Organisation  HMG  His Majesty's Government  SDR  sediment delivery ratio  USLE  Universal Soil Loss Equation  xix LIST OF SYMBOLS  X  ratio of the error variances  a  coefficient of the power-law relation C=aQ  b  exponent of the power-law relation C=aQ  C  total suspended sediment concentration, g/1 (cover factor in  (E^/EQ ) 2  b  b  C  USLE)  event runoff coefficient (erosion plots)  R  = the percentage of incoming storm rainfall which runs off the plot CF  correction factor for calculating stream discharge  CV  coefficient of variation  E  realised measurement error  E  C  E  Q  2  error variance of C  2  error variance of Q  F  falling limb  GH  gauge height (cm)  I  10  maximum 10-minute storm rainfall intensity (mm/h)  J-eo  maximum 60-minute storm rainfall intensity (mm/h)  K  soil erodibility factor  L  slope length factor  M  monsoon season  n  sample size  N  number of storms  P  pre-monsoon season (management factor in  Q  stream flow rate (m/s)  R  falling limb (rainfall factor in  R  2  (USLE)  (USLE)  3  correlation coefficient  USLE)  USLE)  XX  RMHM  minimum total event rainfall to be considered a storm (mm).  R  total storm rainfall (mm),  T  s  sample standard error  S  time without rain between storms, h (slope steepness factor in USLE)  s  standard error of the estimate  r  S  minimum time without rain to declare new storm (h)  t  student's t-distribution  T  transition season  mN  T  10  timing of I relative to the storm start (h) 10  Tgo  timing of  T  total storm duration (h).  D U R  relative to the storm start (h)  V  flow velocity (m/s)  x  sample mean  COMMON SUBSCRIPTS  exp  expected value (based on marginal regression)  f  resultfromfunctional analysis  meas  measured value  min  minimum  max  maximum  r  resultfrommarginal regression  surface  measured at the surface  T  total  xxi ACKNOWLEDGEMENTS  I would like to express my thanks to the International Development Research Centre for its direct support of this research through a Young Canadian Researcher's Award to me and their ongoing support of the Mountain Resources Management Project of which this research has been a part. I would also like to thank the Natural Sciences and Engineering Research Council of Canada for supporting me financially during 1990-1992. In addition, the International Centre for Integrated Mountain Development QCIMOD, Kathmandu) and the Integrated Survey Section (His Majesty's Government, Kathmandu) both provided invaluable assistance and administrative support during my field seasons in Nepal during 1991-1994. The dedication of Dr. Hans Schreier, my research supervisor, in collaboration with Mr. Pravakar B. Shah (ICIMOD) made this work possible. Without their tenacity in keeping the larger project alive, this research would not have occurred. I would also like to thank my entire Supervisory Committee - Dr. Hans Schreier, Dr. Mike Church, Dr. Les Lavkulich, Dr. Tim Ballard, and Dr. Hamish Kimmins - for giving direction especially during the earlier years of this work. An interdisciplinary research project like this is rarely achieved without the support and participation of many people. A dedicated team of individuals in Nepal under the direction of Mr. Pravakar B. Shah assured the completion of this research. The tireless enthusiasm of Mr. Gopal Nakarmi and the calm thoroughness of Mr. Bhuban Shrestha were instrumental in motivating field staff and making thefieldwork successful. My stay in thefieldwas made much easier and more enjoyable by the collaboration of Mr. A. Raj Pathak whose good humour and capacity forfieldwork is well known throughout Nepal. Many farmers within the study area participated directly in this research through data collection. I am grateful to them and would like to mention in particular Mr. Dipak Bhetwal, Mr. Uddav Pathak, Mr. Hirinath Bhetwal, Mr. Prem Lama, and Mr. Sudarsan Accharya and their families for their partnership and the high quality of their work.  XXll  Technical support in both Canada and Nepal has been invaluable. Claire Dat, Kathi Hofmann, and SiPing Tu worked diligently with irreplaceable sediments in the UBC Soil Science and Geography laboratories. Technical supportfromSandra Brown, Yao Cui, and Wayne Tamagi in the Geographic Information System Laboratory (UBC) is also gratefully acknowledged. Andrew Faulkner's suggestions on database management improved the efficiencies of the analyses. I have benefitted greatlyfromdiscussions with many other scientists. Mr. Brian Carson kindly shared with me his experience and reflectionsfromhis years of work in Nepal. Dr. Johannes Ries and Petra Schweizer provided hospitality and inspiration on the mechanics of carrying out the field work required for this study. Similarly, Ms. Susanne Wymann and Dr. Jemuel Perino shared their experiences of carrying out research in Nepal. Discussions with Dr. Marwan Hassan, Dr. Judy Haschenburger, and Dr. Tony Kozak at UBC were always very helpful. Thanks also to my office friend, Kathy Cook, who shared with me her critical thinking. A special appreciation is extended to Dr. Mike Church who influenced profoundly the study design and data analysis in this research and whose level of involvement in this work, for me, matched that of a co-supervisor. His ability to marry the highest standard of scientific endeavour with the realities of applied research will have a long-lasting effect on my future research activities and scientific interpretations. And most of all, thank you to Kathi for accepting these years of research and for being there at their conclusion.  FOREWORD  The findings reported in this document are based on extensive field and laboratory data. The original data are available on diskette by contacting the author through the Institute for Resources and Environment, University of British Columbia, 436E-2206 East Mall, Vancouver, B.C., CANADA V6T 1Z3 (or via email at ire@unixg.ubc.ca).  1  PART I Biophysical Analyses 1. General Introduction 1.1 Problem Statement  Nepal's dramatic mountainous relief, frequently incompetent bedrock, rapid tectonic uplift, and warm monsoonal climate all indicate that weathering and erosion have long been intense and an important aspect of life in Nepal for a very long time. More recentiy, with expanding populations both in Nepal and in downstream neighbouring regions, there has been increasing alarm about steepland cultivation and other agricultural landuse practices in the mountainous regions due to the presumed acceleration of erosion. Concern about stream sediment levels and their potential both for endangering the viability of upland farming and for yielding negative consequences downstream have led to many strong statements about sediment dynamics on these southern slopes of the Himalaya (World Bank 1979; Eckholm 1975). Unfortunately, when the bases of these statements are examined, little or no quantitative scientific data are available to substantiate the statements made. A clearer picture of the fertility and landuse dynamics of these headwater catchments is emerging. Detailed measurements in the Jhikhu River basin show that although there have been significant increases in both forest cover and cultivated land from 1972 to 1990 (Schreier et al. 1994), the fertility of the agricultural land (Wymann 1993) and the quality of the forests (Schmidt 1992) are declining. We do not, however, have equivalent quantitative measurements of erosion and sediment transport even though tolerable rates of surface erosion are an essential underpinning to a sustainable agricultural system. No matter how good the economy might be, and no matter how many inputs can be brought in from the outside, if erosion is excessive, widespread site degradation will put the entire system into decline. It is clear that rates of erosion in the steepland agricultural areas of Nepal are high. What is far less clear is what these rates are over various spatial and temporal scales, whether these rates are  acceptable within the framework of a sustainable agricultural system, and how effective the indigenous population is at controlling erosion. By looking in detail at sediment dynamics in loworder Middle Mountains catchments of varying size, this study tries to answer some of these questions. This study is also relevant to a debate which has persisted for the past two decades concerning the influence of landuse practices in the uplands on flooding in the distant lowlands. Eckholm (1975), Myers (1986) and others have promoted the notion that deforestation in the upland agricultural areas of Nepal is causing massive stream sedimentation and consequent devastating floods in the lowlands of Bangladesh and that there will be no forests left in Nepal by the turn of the century. These forecasts of Himalayan environmental catastrophe greatly influenced development activity until the mid-1980s. Many other authors and researchers (e.g., Gilmour 1988; Ives and Messerli 1989; Lauterburg 1993) have tried to show that the linkages are mythical and statements of imminent forest and agricultural demise are unfounded. No one study can answer all of the questions posed and suggested above. In this study, the focus is on headwater catchments. The steep topography and intense rainfall in these areas present a challenging environment for agriculture. These natural factors, combined with the heavily manipulated agricultural lands, render these headwater areas vulnerable to severe erosion. Because of population increase, the headwater areas are under tremendous pressures from landuse intensification. Thus, important baseline information must be established if we hope to improve the diagnosis of this agricultural system to enhance its future viability.  1.2 R e s e a r c h Context  Ives and Messerli (1989) have summarised others' hypotheses of this region's imminent catastrophe calling it the Theory of Himalayan Environmental Degradation. In the theory, it is assumed that population growth is the root cause of all environmental degradation in the Himalayan  3 region. The mountain farmer is seen as an ignorant accomplice in a vicious cycle of resource extraction and environmental demise. The upland degradation is soon followed by widespread flooding and sedimentation causing further disruption and demise for a far greater population. Though based on economic indicators and real data describing landuse change, these statements ignored scientific understanding about highland-lowland linkages. Until the late 1970s, there was an absence of any quantitative measurements of erosion and sediment transport in the Middle Mountains. From the late 1970s through the late 1980s, a number of independent studies (Kandel 1978; Mulder 1978; Laban 1978; Upadhaya et al. 1991; DSCWM 1991; Sherchan et al. 1991) attempted to examine erosion quantitatively in the southern Himalaya and in the Middle Mountains. These initial attempts to assess erosion in the Middle Mountains, reviewed later in Chapter 5, emphasise surface-erosion estimates over fixed spatial and temporal scales. Spatial scales varied from an erosion-plot scale (10 to 100 m) to specific catchment scales (10 to 1000 km) with 2  2  little or no intermediate resolution. Similarly, the results were generally presented as annual rates with little or no temporal resolution. There was a growing awareness in the 1980s that the Theory of Himalayan Environmental Degradation should be challenged on the basis of its inadequate data and fundamental misunderstandings about both the behaviour of highland-lowland systems and the role of the people living within the area. This recognition was formalised at the Mohonk Mountain Conference in 1985 which set the stage for integrated studies of landuse, erosion, and management. Thompson and Warburton (1985) put forth the thesis that uncertainty in the Himalaya is so extreme that uncertainty itself contains the problem and therefore standard scientific approaches to understanding and addressing the Theory of Himalayan Environmental Degradation cannot succeed. In the 1990s, several long-term integrated studies were initiated, incorporating measurements of surface erosion and mass wasting with related precipitation and landuse parameters (Perino 1993; Ries 1994; Overseas Development Agency 1995). These studies are attempting to reach more  4 meaningful conclusions about catchment-scale processes. For instance, Perino (1993) described a paired-catchment study evaluating the consequences of improved farming methods on total basin sediment output. This study integrated agrometeorologic variables with comprehensive sediment and water measurements for basins of about two hectares. Ries (1994) measured rainfall input, surface erosion, and stream suspended-sediment yield in the High Mountains for low, medium, and high landuse intensity. Overseas Development Agency (1995) described results from three years of research looking at the relations amongst soil erosion, water quality, landuse change, management practices, and in-stream fauna. The conclusions of these studies generally point to a strong, indigenous agricultural system helping to sustain - not undermine - food production. The above comments point out how limited the database is on erosion and sedimentation in Nepal. In comparison, we have a reasonable understanding of many catchment-scale processes within small temperate-region basins. However, the characteristics of these Middle Mountain catchments are quite distinct from their temperate counterparts. In particular, aspects such as the steep topography, heavy precipitation concentrated within a few months, soil types, and high level of human manipulation over every part of the landscape are likely to create very different outcomes than have been observed elsewhere. This study tries to fill this gap through an intensive monitoring program over a range of spatial scales from the plot scale (0.01 ha) through basin spatial scales of 100 to 500 to 10 000 ha in size and covering temporal scales from a single flood event (over 300 individual floods are examined) to seasonal and annual timescales during a three-year period. A range of related biophysical measurements are made to describe sediment dynamics within these headwater catchments. The findings of this study are assembled to produce a diagnostic evaluation of the health of these headwater agricultural systems with respect to soil erosion. Ehrenfeld (1992) described ecosystem health as a bridging concept: Health is an idea that transcends scientific definition...it  contains values that are not  5 amenable to scientific methods of exploration... Health is a bridging concept connecting two worlds: it is not operational in science if you try to pin it down, yet it can be helpful in communicating with non-scientists. Equally important, if used with care in ecology, it can enrich scientific thought with the values and judgments that make science a valid human  endeavour. (Ehrenfeld 1992)  The overall diagnosis is non-scientific but draws heavilyfromthe scientific findings. It is put forth because there is an urgent need for a diagnosis. It is this need which was the genetic influence for the research.  1.3 Goals and Objectives Spatial and temporal scales impose a number of constraints on research of this kind. The Jhikhu River basin has been chosen for a case study because it is typical of the Middle Mountains with respect to climate, physiography, landuse, and soil type and is representative of the level of landuse intensity likely to affect most other basins in the region in the near future. Further, there already exists a large inventory of physiographic data in a GIS for this basin (Shah et al. 1994). In applying well-established scientific concepts over manageable spatial and temporal scales in low-order basins of the Middle Mountains of Nepal, this thesis sets out to accomplish four central goals, each with specific research objectives: 1)  Diagnose causes of sediment dynamics  •  Identify the sources of suspended sediment in Middle Mountain streams.  •  Evaluate the importance of topography, rainfall, and landuse (and other management practices) in shaping the sediment regime over relevant spatial and temporal scales. Research questions: • What are the significant sediment sources? • Do seasonal changes in rainfall regime (intensity, duration,frequency,spatial variation), surface condition, and management cause important changes in the sediment regime? • What is the extent of seasonal changes in sediment loss?  6 2)  Assess the efficacy of indigenous management techniques  •  Document all important indigenous soil-management approaches related to erosion and sedimentation.  •  Evaluate the influence of these techniques on sediment dynamics. Research questions: • Are local farmers effective in modifying sediment dynamics? Do their activities reduce or increase basin sediment loss? • To what extent are the farmers able to favourably influence the erosive fate of their soils? At what point does their management regime become ineffective?  3)  Determine sediment and nutrient budgets  •  For the three-year period of study, construct sediment budgets over specific spatial (plot, subbasins, basin) and temporal (event, season, year) scales to determine the relative importance of sediment sources and sediment storage to basin yield.  •  Using phosphorus as a representative limiting nutrient, examine its relative redistribution within sub-basins and net loss from sub-basins; identify the landuse(s) which provide the dominant phosphorus contributions to basin phosphorus loss. Research questions: • • • • •  How does the spatial scale of a basin affect its sediment output? What proportion of the annual basin sediment loss is accounted for in the biggest floods? Can predictive relations be developed for sediment output? What is the significance of surface degradation to basin budgets? To what extent are nutrients redistributed within (but not "lost from") sub-basins?  4)  Prescribe soil/sediment management recommendations  •  If appropriate, suggest management recommendations to improve the effectiveness of present farming techniques.  •  Provide a statement clarifying the contribution of headwater areas to downstream concentrations of suspended sediment.  •  Provide a statement of sustainability regarding the current management regime with respect to erosion. Research questions: • To what extent are the soils redistributed rather than lost in this highland-lowland system? • Are there feasible options available to the farmers to reduce their risk of soil loss?  7 The completion of these four research objectives will make it possible to assess the following hypothesis: soil and sediment dynamics and the indigenous management techniques within headwater Middle Mountains basins do not themselves indicate a deterioration of the health of the agricultural system.  The thesis comprises two parts - one, Biophysical Analyses and two, Management and Implications. The first part focuses on quantitative field measurements and their detailed analyses. Chapter 2 provides a description of the study area and Chapter 3 gives a summary of methods used in the entire study. Chapters 4 through 6 address specific biophysical concerns associated with the causes of sediment dynamics laid out as the first goal. Specifically, Chapter 4 presents the precipitation regime of the study area, developing appropriate temporal and spatial scales for relating the precipitation regime to the sediment regime. Chapter 5 examines the many interacting factors involved in sediment transport in the study area - for example, rainfall characteristics, surface cover, soil characteristics, management, and topography. Chapter 6 uses sediment properties (particle-size distribution, colour, and phosphorus content) to look at some fundamental dynamics associated with suspended-sediment transport. These insights have implications for the causes of observed sediment regimes presented in Chapter 5. The second part is concerned with management and the implications of the findings presented in Part 1. Chapter 7 evaluates the effectiveness of indigenous management techniques. Chapter 8 presents sediment budgets based on the relations developed earlier in Chapters 5 and 6. Finally, Chapter 9 addresses integration and overall conclusions and suggests recommendations for management and further research.  8  2. Study Area 2.1 Jhikhu River basin 2.1.1 Location and physiography The Jhikhu River basin is located 35 km east of Kathmandu in Nepal's Middle Mountains as shown in Figure 2.1 and illustrated in Appendix Al. It has an area of 111 km and elevations ranging 2  from 800 to 2030 m. The topography within the basin, presented quantitatively in Table 2.1, includes large areas of steep land. Landuse is dominated by subsistence agriculture though, with recent political changes, a market economy is rapidly developing. The population in this basin was 32 956 in 1990, growing at a rate of 2.9% per annum (Shrestha and Brown 1995). The Arniko highway, a major east-west corridor in Nepal, passes through the watershed providing good access to Kathmandu and the Tibetan border. The geology of the Jhikhu River basin consists of sedimentary rocks which have been affected by low- and high-grade metamorphism (Dongol 1991). These metasediments include phyllite, quartzite, schists and mica-schists. The geomorphological development of this region has been influenced by tectonic uplift leading to over-steepened slopes that are prone to instability (Saijo 1991). The highly dissected landscape contains ridge slopes formed by soil creep and the weathering of bedrock, erosional slopes formed by landslides, and landforms originating from sediment deposition. The ridge slopes remain active and are the most vulnerable to gully and sheet/rill erosion. Large terraces exist above the Jhikhu River valley bottom, created in the Quaternary Period when this drainage was tectonically dammed (Dongol 1991). Red soils are dominantly Ultisols and some Alfisols whereas the non-red soils are dominantly Entisols and Inceptisols and some Alfisols. The Ultisols are the oldest soils in the study area; they dominate the terraces above the main valley bottom, though they can be found up to elevations as high as 1400 m. These red soils have a generally lower soil fertility than do the non-red soils.  10 Table 2.1  Slope, aspect, and elevation of the Jhikhu River basin (from 1:20 000 mapping).  Slope (•)  Aspect  Elevation (m)  Class Percentage of Basin Area 43.4 0-10 10-20 24.6 20-30 21.1 £>30 10.9 Flat 16.3 NE, N, NW 34.0 E 12.8 29.7 SE, S, SW W 7.2 800-999 39.6 1000-1199 27.7 >1200 32.7  Note: slope does not consider terracing.  The highly dissected landscape signifies a dense natural drainage network. Tributaries of the Jhikhu River are steep, confined, boulder-bed channels. The Jhikhu River, in contrast, is a meandering sand-bed river in some of its lower reaches with a slope of about 0.1" (0.2 %).  2.1.2 Farming system and landuse Table 2.2 shows the landuse breakdown in the Jhikhu River basin for 1990. The landscape is dominated by cultivation on slopes of up to 40'. Annually, up to three crops are grown on the irrigated khet fields and two crops on the rainfed bari land. Rice is the dominant crop on the khet land, though recently cash crops such as tomatoes and peppers have risen in importance. Maize continues to dominate on the upland rainfed fields, with millet and wheat grown during the dry season. Legumes are typically intercropped on many fields. The agricultural system has traditionally relied on the forests for nutrient inputs. Forests provide about 40% of the feed for livestock (Carson 1992) whose dung is incorporated into compost,  11 Table 2.2  Landuse of the Jhikhu River basin in 1990. Landuse  Percentage of .Basin Area  khet  15.4  bari  39.1  forest shrub grassland other  30.2 8.4 4.2 2.7  becoming the prime source of added nutrients to the cultivated fields. Over the past 20 years, synthetic fertilizers have gained in prominence as landuse pressures have increased (Chitrakar 1990). In the past few years, the cost of these amendments has risen dramatically, placing the cash-poor farmers in a difficult situation. Further, much of the forest land is in a degraded condition with crown cover below 10% and little or no surface cover. Landuse mapping over the past 50 years shows a steady conversion of land into cultivation (Schmidt 1992). Schreier et al. (1994) showed that between 1947 and 1981, this conversion was at the expense of forest cover. More recent mapping (1972-1990) documents an increase in forest cover and agricultural land at the expense of both shrub land and grassland. Schreier et al. (1994) explained that these changes in forest cover reflect widespread deforestation earlier this century followed more recently by both afforestation associated with plantations during the past 15 years and increased tree planting on private land (Gilmour 1991).  2.1.3 Regional climate The climate of Nepal is strongly influenced by the southwest monsoon across the Indian Subcontinent and by its own abrupt relief. The seasonal reversal of winds associated with the South Asian Monsoon brings warm humid air to much of northern India and southern Nepal during the summer months. Nepal rises from the Indo-Gangetic Plain to the highest mountains on Earth within  12 only 150 km. These two factors combine to yield annual precipitation of overfivemetres in parts of the country and cold deserts on some leeward mountainsides (Department of Hydrology and Meteorology 1992). Unfortunately, these regional variations are not well described because of the lack of monitoring stations in the country's difficult mountainous regions (Ramanathan 1981). The Middle Mountains physiographic region (see Figure 2.1) does not experience the extremes of climate found in much of the rest of Nepal. At about 25' north latitude, the climate of this physiographic region varies from warm temperate on high mountain ridges to subtropical in the low valleys. Locally, climatic patterns can vary strongly due to elevation and aspect. Air temperatures span O'C to 40"C annually (Kenting Earth Sciences Ltd. 1986).  2.1.4 Local climate In the Jhikhu River basin, the rainy season starts normally in the beginning of June, ending by the end of September. Precipitation in this basin is in the form of rainfall only and varies typically between 900 and 1600 mm annually. Up to 90% of the annual rainfall falls during the monsoon period with a distinct prolonged period of drought throughout most of the rest of the year. The local climate of the Jhikhu River basin can be characterised using data from several stations covering a range of elevations. Sources include a long-term monitoring station run by His Majesty's Government (HMG) and found within the valley lowland at 865 m, other long-term HMG climate stations found in the region of the Jhikhu basin, and stations which have been maintained by the present study for three tofiveyears at several points within the study area. The HMG data are used to describe long-term trends in climate in this region. The data from within the Jhikhu basin provide the best picture of the variation in weather occurring within the study site.  Rainfall  One of the longest records of precipitation measurement available near the study basin is that  13 Figure 2.2  Mean, maximum, and minimum monthly rainfall at (a) Kathmandu airport (HMG Records 1968-1990) and (b) Panchkhal (HMG Records 1978-1994;, means based on 1978-1985 and 1988-1994).  700  J F M A M J  J A S O N D  J FMAMJ  J A S O N D  (b) Panchkhal 865 m 1978-1994  (a) Kathmandu Airport 1336 m 1968-1990 Mean annual rainfall (1968-1990) = 1408.2 mm  Mean annual rainfall (1978-1985 & 1990-1994) = 1251.0 mm Mean Maximum Minimum  14 of the Kathmandu airport. At this station, datafrom1968 to 1990 are currently available in published government records (Department of Hydrology and Meteorology 1992 etc.). These precipitation measurements are summarised in Figure 2.2a. An HMG climate station at Panchkhal is located within the study area itself with data available from 1978-1994 (Department of Hydrology and Meteorology 1992 etc.). These data are summarised in Figure 2.2b. Together these data sets provide a long-term indication of averages and extremes expected in the study site. The Panchkhal station is a valley-bottom station indicative of the subtropical lowland areas while the station at the Kathmandu airport is indicative of the more-temperate regions. Both figures show that average monthly precipitation peaks in July at about 300-350 mm per month. Interestingly, the maximum monthly precipitation peaks in June at both locations. In each case, this peak is from different events, each of unusual magnitude for the month: in 1971 at the Kathmandu airport and in 1978 in Panchkhal. In fact, while the one station experienced very high rainfall, the other experienced a rainfall of average magnitude yet the two stations are only 30 km apart, within the same Middle Mountain physiographic region. Figures 2.3 provides summary statistics of precipitation data gathered by the present study at three points within the Jhikhu River basin. The Baluwa station (1992-1994) is within the Andheri River catchment at 900 m, 4 kmfromthe HMG Panchkhal station (also in the valley bottom). The Bela station (1990-1994) is at 1211 m, in the Andheri catchment and also in the upland Kukhuri catchment. The Dhulikhel station (1990-1994) is at 1500 m. The locations of these stations are indicated on Figure 3.1. As indicated in Figure 2.4a, across these five sites, the average monthly precipitation does not vary greatly. Though the periods of record differ, there appears to be a tendency for the peak monthly rainfall to be higher at the Dhulikhel and Baluwa locations. Comparing the longer-term records (Figure 2.2) with the recordsfromthis study, one can see an obvious difference in the lack of a peak monthly rainfall in June for Baluwa, Bela, and Dhulikhel.  15 Figure 2.3  700 r 600 -  Mean, maximum, and minimum monthly rainfall at (a) Baluwa (1992-1994; means based on 1993-1994), (b) Bela (1990-1994), and (c) Dhulikhel (1990-1994).  (a) Baluwa 900 m 1992-1994 Mean annual (1990-1994) = 1152.6 mm  (b) Bela 1211 m 1990-1994  700  500 -  500 -  400 -  400 -  Ol 300 -  300 -  1  200 -  200  h  100  100 0  r n  i i i i i i r  FMAM J JAS OND  700 600 h  Mean annual (1990-1994) = 1352.9 mm  600 -  0  F-  I  I I I I I I I I  FMAM J JAS OND  (c) Dhulikhel 1500 m 1989-1994 Mean annual (1990-1994) = 1689.0 mm  Mean Maximum  500 1400 300 fi 200  — -—  o 100 0  m  1111111  FMAM J JAS OND  Minimum  16 Figure 2.4  Mean monthly rainfall (a) and maximum 24-hour rainfall (b) at Kathmandu Airport (1968-1990), Panchkhal (1978-1985 and 1988-1994), Baluwa (1993-1994), Bela (1990-1994), and Dhulikhel (1990-1994).  J FMAMJ  J A S O N D  J FMAMJ  (a) Mean monthly rainfall  J A S O N D  (b) Maximum 24-hour rainfall Kathmandu Airport — Panchkhal Dhulikhel Bela Achharya Tol  17 Presumably, this suggests that heavy June rainfall can occur but it is infrequent and has not happened in the Jhikhu River basin during this period of study (1990-1994). Figure 2.4b contrasts the maximum 24-hour rainfall measured at all five sites during the periods of record. Though the mean precipitation in June is lower than in July on average at all sites, this figure reveals a tendency for the highest 24-hour precipitation to occur in June. These comparisons hint at patterns which will be more fully examined in Chapter 4. A striking result in Figure 2.4b is the heavy post-monsoon rainfall measured on October 20, 1987 at the Kathmandu Airport. Finally, all of these descriptive rainfall statistics are summarised in tabular form in Appendix A2.  Temperature  Temperature data for the three valley-bottom climate stations are presented in Figure 2.5. Figure 2.5a shows the mean maximum and mean minimum temperatures and Figure 2.5b shows the extreme maximum and minimum temperatures. Though the data are taken from different periods, they provide a consistent characterisation of the valley-bottom temperature regime: April through September have mean monthly temperatures well above 30°C, temperatures frequently exceed 35*C and, during the monitored periods, the minimum has rarely gone below freezing. Figures 2.5c and 2.5d present the equivalent temperature data for the two high-elevation sites and provide the mean result from the Kathmandu airport for comparison. The mean maximum monthly temperature at the high-elevation sites is several degrees cooler than at the valley-bottom sites. During the indicated periods, the temperature rarely exceeded 35'C and went below freezing especially at the Kathmandu station. Summary statistics taken from these monthly results are given in Table 2.3. Not unexpectedly, elevation influences the mean annual temperature. The high-elevation sites show a slightly lower average mean annual temperature than the valley-bottom sites. However, the lower elevation sites  18 Figure 2.5  Mean-monthly and extreme-monthly maximum/minimum temperatures at (a) three low-elevation sites (Panchkhal, 1978-1994; Baluwa, 1993-1994; Bhimsenthan, 19931994) and (b) three high-elevation sites (Kathmandu airport 1968-1990;, Bela, 19901994; Bhetwaltok, 1993-1994).  (a) Low-elevation sites  (b) Low-elevation sites  J FMAM J JASOND  40  (c) High-elevation sites mean-monthly max/min  J FMAMJ  40  (d) High-elevation sites extreme-monthly max/min  Bela — -  I  — ——  Bhetwaltok  I I I I I I I I I  J FMAM J JASOND  -10  JASOND  ——  I  Bela Bhetwaltok  I I I I I I I I I  J FMAM J JASOND  19 Table 2.3  Descriptive temperature statistics from the six climate stations. Elev. (m)  Mean Annual  Mean Daily Max  Mean Daily Min  Extreme Daily Max  Extreme Daily Min  Period of Data  Panchkhal (flat)  865  21.2  28.1  14.2  39.5  -0.2 (1989)  78-94  Baluwa (north-facing)  865  21.0  28.8  13.1  38.0  1.0  93-94  Bhimsenthan (south-facing)  895  21.2  28.0  14.4  37.0  2.0  93-94  Kathmandu Airport  1336  17.9  n/a  n/a  34.0  -3.5 (1978)  68-86  Bela (north-facing)  1254  20.6  26.2  14.9  35.0  3.0  90-94  Bhetwaltok (south-facing)  1300  20.3  26.1  14.5  39.0 (Jun 9 92)  3.0  93-94  Site  Low Elevation  High Elevation  appear to have lower minima providing evidence for inversions. The annual extremes and mean monthly temperatures are comparable at all sites. There does not appear to be a large effect on temperature due to aspect at these sites.  2.2 Study catchments A total of six smaller catchments and sub-catchments in the lower regions of the Jhikhu River basin are the focus of this study. Other catchments have been examined as part of the larger study not included in this thesis. Some of these catchments are nested as illustrated in Figure 2.6. The topography of these six study catchments is contrasted in Table 2.4. Two sources of topographic data are used in this study's analyses. A topographic map at 1:20 000 scale was produced in 1990 for the entire basin and a detailed digital topographic map at 1:5 000 scale was produced in 1994 for a cross-  20  21 Table 2.4  Topography of the six study catchments based on 1994 1:5 000 mapping. Data for the Dhap catchment are taken from 1990 mapping at 1:20 000 scale.  St. No. Area (ha) Slope f) 0-10 10-20 20-30 >30 Elev. (m) 800-1000 1000-1200 >1200 Aspect Flat N, NE, NW E S, SE, SW W  Andheri  Andheri Mid #2 12 299  Upper Andheri 9 178  Kukhuri  3 558  2 532  Andheri Mid #1 11 401  73.2% 20.2 5.9 0.8  19.2% 25.4 38.2 17.2  16.6% 24.4 39.2 19.8  11.2% 22.7 43.1 23  9.5% 22.0 42.9 25.6  10.3% 21.9 43.6 24.2  98.3% 1.7 0.0  24.8% 34.9 40.3  15.2% 31.8 53.0  0.0% 28.9 71.1  0.0% 14.3 85.7  0.0% 27.7 72.3  35.4% 9.8 7.3 36.4 11.1  6.3% 53.1 9.54 15.8 15.3  5.2% 53.5 9.1 17.1 15.1  3.4% 56.3 9.6 14.7 16.0  3.2% 63.8 12.9 6.6 13.5  3.2% 51.1 0.9 22.3 22.5 .  Dhap  10 72  Note: numbers represent percentage of total area. section of the Jhikhu River valley, containing the Andheri River study catchments. All digital topographic data have been incorporated into a Geographic Information System to produce quantitative information on elevation, aspect, and slope (see section 3.3.4). The results for the Jhikhu basin appeared earlier in Table 2.1. The landuse of these six catchments is presented in Table 2.5. Mapping was carried out in 1990 at 1:20 000 scale for the entire Jhikhu River basin and these results appeared in Table 2.2. Table 2.6 summarises the features of these streams in terms of channel slope and stream order at the hydrometric station (Leopold et al. 1964) and their estimated bankfull discharge (Dunne and Leopold 1978). The streams in these catchments are contrasted in profile in Figure 2.7.  22 Table 2.5  1990 landuse distribution of study sub-catchments within the Andheri basin, mapped at 1:20 000 scale. (Total area based on 1994 1:5 000 photogrammetric map.) Dhap  St. No. Area (ha)  3 558 24.6% 36.8 20.9 3.8 7.6 6.3  khet bari  forest shrub grassland other  Andheri Lower 2 532 6.8% 33.1 32.3 14.6 8.0 5.3  Andheri Mid #1 11 401 8.7% 41.1 24.4 11.3 9.0 5.4  Andheri Mid #2 12 299 8.4% 51.3 22.0 7.2 7.4 3.7  Andheri Upper 9 178 8.9% 53.8 22.5 4.3 6.9 3.7  Kukhuri 10 72 8.2% 55.4 17.9 6.1 8.3 4.1  Note: numbers refer to percentage of total area.  Table 2.6  Slope, order, and bankfull discharge of the study streams at each hydrometric station. Station  Order  Slope  C) 1 2 3 9 10 11 12  Andheri  6 4 5 3 2 3 3  0.3 1.9 1.0 5.1 6.6 2.1 4.8  Bankfull Discharge (m/s) 85 25 20 13 5 unknown unknown 3  catchment  The Andheri catchment is characterised by intensively-managed cultivated uplands and degraded, often-gullied, red soils in the lowland. It is north-facing with a moderate temperature regime. Station 2 is situated at 850 m near the Andheri River's confluence with the Jhikhu River.  23 Figure 2.7  Elevational profiles of the study streams (vertical scale is exaggerated by 10 times).  1750  20  15  10  5  Horizontal Distance (km) from Jhikhu Station (1) Note: numbers denote hydrometric stations  0  24 Kukhuri catchment  The Kukhuri River catchment forms part of the headwater area of the Andheri catchment. This catchment is dominated by well-managed rainfed cultivated steeplands with only a small red-soil component. The Kukhuri River is monitored at 1060 m at station 10, 40 m upstream of its confluence with the Andheri River. Upper Andheri catchment  Along with the Kukhuri River catchment, this catchment forms the headwaters of the Andheri River. Though the characteristics of these catchments are generally similar, the Kukhuri catchment is proportionately steeper and its irrigated land is consequently of lower productivity. A relict landslide (occupying about 1 % of the basin) along with some smaller pockets of mildly-degraded red soils are located in the northwestern portion of the catchment. This upland section of the Andheri River is monitored at Station 9 at 1060 m about 200 m upstream of its confluence with the Kukhuri River (station 10). Mid-Reach Andheri catchments  Stations 11 and 12 are hydrometric stations located between the headwater area (stations 9/10) and the mouth (station 2) of the Andheri River. These monitoring positions better isolate the transition in this basin between the steep, intensively-managed headwaters and the gullied, red-soil, partlyabandoned lowland. Specifically, station 12 (elevation 1020 m) includes all of the Upper Andheri and Kukhuri sub-catchments along with a portion of degraded forest along the Andheri River. Station 11 (elevation 880 m) includes all of the station 12 drainage as well as extensive forest and degraded shrub land and a small amount of the gullied lowland. These are referred to as the Andheri Mid #1 (station 11) and Andheri Mid #2 (station 12) catchments. Dhap catchment  The Dhap River catchment, largely flat and south-facing, provides a strong contrast to the characteristics of the catchments within the Andheri catchment. Dominated by red soils, its land is  25 often in a degraded condition. Its drainage is composed of two distinct units - one in a steep location near the river's confluence with the Jhikhu River and the other flatter and more distant. This affects hydrological behaviour at its mouth accordingly. This river is monitored at station 3 at 825 m, 675 m upstream of its confluence with the Jhikhu River.  26  3. Methods Methods used in the study are described according tofieldand laboratory approaches. A selection of procedures used for data synthesis is presented in the chapter'sfinalsection.  3.1 Field methods 3.1.1 Climate Climate measurements carried out in this study consist of rainfall and temperature using both manual and automated instruments. Rainfall Figure 3.1 shows the locations of all rainfall monitoring stations in the study area. At these locations, a mix of storage and recording gauges has been in place from three to seventeen years with most records being of a three- tofive-yearduration. Appendix A3 provides a summary of the gauges, including gauge number, location, and elevation. Most gauges are located between 850 and 1350 m elevation. Measurement of rainfall intensity has been made atfivesites for three years using recording rain gauges (tipping buckets - Middleton and Spilhaus 1953) complete with pulse data loggers. Rainfall resolution is 1.0 mm at the two monitoring sites within the Andheri Basin (Baluwa and Bela) and 0.25 mm at the other three sites (Bhetwaltok, Bhimsenthan, and Kamidanda). The tipping buckets were installed on a permanent table, one metre from ground level, and have a circular opening of eight inches (324.3 cm). Generally, they were visited monthly, though during the rainy season this 2  frequency was increased. Files were converted into ASCII format and manipulated as described in section 3.3.3. The calibration for each tipping bucket was checked twice annually. Sampling frequency varied from 2 to 10 minutes in the earlier period of monitoring (depending on the resolution of the tipping mechanism) and wasfixedthroughout at 2 minutes for subsequent monitoring.  27 Figure 3.1  Lay-out of monitoring network for rainfall and temperature within the Jhikhu River basin.  Climate Monitoring Network — Major Rivers • 24-hour Rain Gauge A Tipping Bucket (includes a 24-hr rain gauge) O Air (manual, max/min) O Air/Soil (automated) (§) Manual & Automated both present  28 To assure data continuity, a 24-hour rain collector was installed nearby each tipping bucket. Occasionally, the automated gauge record is incomplete as a result of data logger malfunction; in these instances, the supplementary gauge is used to complete the data record (though the short-term rainfall-intensity result clearly is lost). Two types of storage gauges were installed for measuring rainfall over 24 hours and were visited daily at 7:00 am local time. The first is of the type used by the Department of Hydrology and Meteorology of HMG. It has an opening of eight inches (324.3 cm) which stands 1.5 mfromthe 2  surface. The rain flows directly into a cylinder calibrated for the opening. Four of these rain gauges have been in place in the Jhikhu River basin during the study period. The second rain collector was custom designed using off-the-shelf ABS (hardened plastic). This collector was produced inexpensively so that over 50 could be installed in thefield.Its opening is four inches (81.1 cm) and it sits 55 cmfromground level. This compares well with the standard 2  gauge of the Atmospheric Environment Service (of Canada) which has a 64.5-cm opening, sitting 2  30.5 cm off the ground (Storr and Ferguson 1972). The custom gauge has a straight 5-inch-long entranceway, tapered at the bottom, and a storage capacity equal to a total rainfall of 250 mm (double that of the AES standard gauge). The rainfall reading (mm) is made by measuring the volume of water in the storage bottle (ml) and then applying a conversion factor of 0.1233 to account for the size of the collector's opening. The catches from these three gauge types have been compared at seven locations to test for consistency across gauges as shown in Table 3.1. The catch ratio for each measurement at each test site has been calculated for all storms and the results appear in Appendix A3. The well-known negative bias associated with wind exposure of a gauge (Brown and Peck 1962) should be responsible for most of this variability. It is to be expected that the custom gauge best represents rainfall because of its smaller opening and its installation closer to the ground. Keller (1972) pointed out that a rain gauge in a network in uniform terrain with the highest catch should be the gauge least affected by  29 wind. McGuinness and Vaughan (1969) observed a seasonal change in rainfall and snowfall catch and determined that it was in fact a wind effect due to seasonal wind regimes. They also found a large difference in the ability of different instruments to cope with the wind effect. The design of the custom gauge minimises wind bias which serves to increase the catch of the custom gauges relative to the two "standard" gauges. The averages and distributions of all catch ratios at each site are given in Table 3.1 and suggest that the standard gauges  underpredict  rainfall by an average of 20.5%  (0.258/1.258) using the six comparisons which fall in the tight range of 1.23-1.29 having an average of 1.258. (A catch ratio of 1.258 means that 0.258 mm out of each 1.258 mm is not measured by the standard instruments; this represents an underprediction of 20.5% and can be corrected using a multiplier adjustment of 1.258 on the rainfall measured by the standard instruments.) As a result of this finding, all measurements from the standard gauges have been increased by 25.8% from that measured. Table 3.1 Test  Comparison of catch ratios of the rain gauges at seven test sites. Location  Gauges Installed  Catch Ratio  Custom/Tipping Bucket  Custom/HMG  N  X  a  X  a  104  1.29  0.24  N  1  Baluwa  111  1.28  0.18  2  Bela (1)  58  1.00  0.083  3  Bhimsenthan  78  1.23  0.17  4  Kami dan da  95  1.44  0.13  5  Bhetwal tok  82  1.24  0.30  6  Dhulikhel  70  1.27  0.21  7  Bela (2)  93  1.24  0.19  It is expected that the wind bias is variable depending on wind speed. Measurements of wind speed within the study area are unavailable. However, observations during the study period suggest that only infrequently is wind speed significant during heavy rain. As a result, the effect of wind  30 speed is neglected in adjusting rainfall records. Further examination of Table 3.1 indicates that the catch ratios at Sites 2 and 4 are very differentfromthe 1.258 average. Site 4 (catch ratio of 1.44), and the tipping bucket in particular at this site, is exposed to a higher degree of wind than at the other sites and as a result the catch ratio at this site is higher than the average catch ratio at the other sites. At Site 2, the custom rain gauge is installed on a small ridge which effectively raises its point of installation above the ground. As a result, both the custom and standard gauges experience a similar wind bias at Site 2 and the catch ratio is unity. These two anomalies are not used to calculate the average catch ratio. The adjustment is applied to every measurement of the tipping bucket regardless of rainfall intensity. The bias associated with the moderate- and high-intensity rainfall dominates the aggregate bias, here called the catch ratio. Since moderate- and high-intensity rainfall are of the greatest interest in this study, it is reasonable to adjust all rainfall intensities consistently by the overall catch ratio. Temperature  Air temperature was monitored using a combination of manual and recording devices. These sites are shown in Figure 3.1 and are summarised in Table 3.2. Temperature was measured manually at a total of six locations. In each case, the max-min thermometer was installed in a standard Stevenson's screen (Middleton and Spilhaus, 1953). Five of these were maintained by this study and provide a record of the daily maximum and minimum air temperatures during the past four years. The sixth site was maintained by HMG and provides a daily record of maximum and minimum temperature covering 17 years. Automated temperature monitoring, recording every 10 minutes, was begun in 1992 at three sites as shown in Figure 3.1. In these cases, both air and soil temperature (depth of 20 cm) are measured. These sites were chosen to reflect the characteristics of north-facing, south-facing, and valley-bottom microclimates as summarised in Table 3.2.  31 Table 3.2  Installation information of temperature measurements.  Location  Aspect  Baluwa Baluwa Panchkhal Bhimsenthan Bhetwaltok Bhetwaltok Bela Bela Dhulikhel  Flat Flat Flat Flat South South North North North  Elev. (m) 865 865 865 895 1300 1300 1255 1255 1545  Installation Date May 29, 1992 Jun 12, 1992 Nov 1970 May 30, 1992 Jun 6, 1992 May 27, 1993 Jan 1, 1990 Jun 12, 1992 Jun 1, 1989  Type air; manual air/soil; automated air; manual air; manual air; manual air/soil; automated air; manual air/soil; automated air; manual  Note: Automated measurements are sampled at 10-minute intervals. 3.1.2 Stream measurements Flow measurements or suspended-sediment sampling, or both, were carried out at seven hydrometric stations during 1992-1994 as shown earlier in Figure 2.6. These measurements were challenging because the floods are generally of short duration and most occur at night following afternoon and evening convectional showers. A monitoring team was trained for each station to improve measurement success. Bridges were constructed at five of these cross-sections (stations 1, 2, 3, 9, and 10) to take discharge measurements at high flow. Flow measurements were made using Price AA current meters. At low flow, readings were taken by wading in the stream and recording the flow velocity at frequent lateral intervals (Buchanan and Somers 1969). Due to the very brief nature of high-flow conditions and to prevent damage to the meter, flow measurements of the surface were taken and estimates made of the average velocities at three lateral positions in the stream crosssection. Following an approach modified from Hoyt (1912), a conversion factor was applied to the flow velocity of the surface. Rating curves have been constructed for each station using the three years of flow measurements and are discussed later in Section 3.3.1. Event and annual hydrographs have been constructed using both automated and manual  32 records of gauge height (stage). Manual records consist of daily (7:00 am) measurements throughout the years 1992-1994 and flood measurements when possible. Manual measurements form the entire record of stage for stations 3, 11, and 12. Automated readings provide the large majority of the flow record at the other four stations and are augmented when necessary with the manual record. At each of stations 1, 2, 9, and 10, an automated monitoring system was installed in 1992. The equipment consisted of a pressure transducer installed near the gauge, a data logger, and a 12-volt power supply. A two-minute sampling interval was selected during most of the study period. Samples of suspended sediments were taken during flood events at all stations. A DH-48 depth-integrating sediment sampler (Guy and Norman 1970) was used at stations 1, 2, 3, 9, and 10 while dipping-bottle samples were taken at stations 11 and 12. When safe to do so, the rivers were waded in order to sample the thalweg. For safety reasons, sampling at higher flows was carried out at a position at or near the bank. These streams are steep and well-mixed when in flood, permitting this varied approach. Sampling frequency was changed according to station and flood size. In the steeper catchments such as stations 9, 10, and 12, as many samples as possible were taken - generally up to ten samples per flood. At stations 1, 2, 3, and 11, between 5 and 15 samples were taken per flood. At Station 1, as many as 30 samples are available for some floods.  3.1.3 Erosion plots Five erosion plots were built in 1991 and then monitored throughout 1992-1994. Some features of the plots are summarised in Table 3.3 and their locations are indicated in Figure 2.6. Each plot contained two terraces connected by a steep terrace riser. In each case, the cropping practice was the same: maize during the rainy season and either millet or wheat during the dry period. The farmers were encouraged to manage the plots in the same way that they managed their otherfields.Each plot was on red soils because it is widely accepted that red soils are more difficult to manage and erode at  33 Table 3.3 Plot No.  Characteristics of erosion plots in the study.  Elev  Aspect  (m)  Hillslope Angle  Area  Upper Terrace Slope/Length  Terrace Riser Height  Lower Terrace Slope/Length  C)  (m )  (Vm)  (m)  (Vm)  2  USDA Soil Type  1  1240  NNE  30  70  18/4.3  1.6  18/5.0  Haplustalf  2  1240  NNW  27  68  23/10  0.8  23/3  Rhodustalf  3  1240  NNW  27  64  23/9  0.8  23/4.2  Rhodustalf  4  1305  sw  22  108  12/6  1.0  11/6  Rhodustalf (degraded)  5  1260  ssw  25  100  8/10  5.0  15/9  Ustochrept (truncated Rhodustalf)  a greater rate than non-red soils (Turton et al. 1994). Erosion and runoff were measured on an event basis. Three drums were located in series at the outlet of each plot. The soil and water mixture in the drums was sampled after each erosioncausing storm event. The methodology used to achieve this evolved during the study period. For each drum, the total height of soil and water was recorded. The water column was sampled and drained off to leave behind the eroded soil which had settled to the bottom of the drum. The height of this soil was also recorded and this material sampled. Using these measurements and the samples, and knowing the dimensions of the drums, total erosion and total runoff were calculated for each storm event.  3.1.4 Soil properties In the field, a selection of soil properties was measured including compactness and infiltration rate. In addition, a complete profile description was recorded at 20 sites.  34 A penetrometer was used to estimate surface-soil compactness. The instrument was used at ten locations within a field and the results averaged for the entire field. The penetrometer was inserted into the surface using roughly the same force for the surface (0-15 cm) and the subsurface (15-40 cm) soils. The value noted was the highest value reached while the penetrometer was being pushed downward in each vertical section. Soil moisture content and bulk density were measured concurrently in most cases. The penetrometer measurements provide a relative index of the compactness of agricultural fields in the study area. This index can be easily qualitatively calibrated to other quantitative measures of compaction. Infiltration rates of surface soils were measured using double-ring infiltrometers following Booker Agriculture International Limited (1984). For each measurement, three sets of infiltrometers were installed and monitored simultaneously. The test was run for 4-6 hours, though in clay-rich soils this often had to be extended to up to 10-12 hours. Soil profile descriptions were recorded at all sites where infiltration was measured. At each site, the depth and type of each horizon down to the parent material was noted. For each horizon, the following soil parameters were determined (Booker Agriculture International Limited 1984): • colour • texture (of the fraction < 2 mm in size) • structure • consistency • porosity • presence of roots • horizon transition • coarse-fragment content (percentage of sample > 2 mm in size) • moisture content In addition, the landuse, physiographic position and topography of the site were recorded. The indigenous soil classification was also determined for the surface horizon (see Chapter 7). 3.1.5 Soil movement A collection of techniques was employed to provide spatial estimates of soil movement.  35  Erosion and accumulation pins were installed to indicate change in soil level at specific spatial locations. Surveys provided information on the spatial variation of erosion and deposition across an area. Bamboo pegs were used in khet, bari, and in gullies to assess rates of soil loss and accumulation. In the khet where they were used to measure soil accumulation, between 4 and 12 accumulation pins were installed in each individual field. In the first year, pins were installed at entranceways and exits to the fields to estimate deposition and scouring occurring there and to relate these areas to the average field values. It was found that these edge effects can be significant but are generally confined to the perimeter of the field due to the complex routing water takes in the fields. Therefore in subsequent years, pins were placed in only the centre of khet fields. For each pin, an average of all four sides was taken of the soil height relative to the notch showing the original soil level. In the bari, soil pins were used to estimate soil accumulation in permanent and temporary drainage ditches. In each case, three pins were installed - at the beginning, the middle, and at the end of the ditch length. All pins were placed in the lateral centre of the ditch. These pins were also measured on all four sides. Finally, in gullied land, pegs were installed near active gullies to estimate rate of change of gully dimensions. For each monitored gully, about five to ten pins were installed and maintained as permanent control points. They were installed flush with soil level to make them invisible to passersby. Detailed records were kept of the relative locations to facilitate quick relocation of the pins, to have many lines from which to measure the gully sides, and to cope with pins lost to gully growth. Survey techniques were used on several occasions to record the spatial variation of soil erosion and deposition. Surveys of surface erosion and mass wasting were carried out after several major events during 1992-1994. At its most-detailed level, the entire basin was examined on foot for evidence of erosion and deposition. For each example found, its location was noted on the aerial  36 photograph and the following characteristics were recorded: • type of erosion • cause • volume of material moved (estimate) • description of eroded material • connection (if any) to other observed erosion Where deposition was observed, the volume and type of material involved were noted. Later in the study, surveys of stream deposition were carried out to quantify changes in sediment storage within the bed of the Andheri and Kukhuri Rivers. Rather than attempting a complete inventory over time of bed storage, locations were chosen to monitor bed stability and changes in bed storage over time. The following observations and measurements were made at these points of reference: • photographs • rate of sand-bar accretion/loss using bamboo pegs • description of sediment sizes present • grab sample of fine sediment (for colour) This information was used to evaluate the extent of storage within the bed of the Andheri and Kukhuri Rivers. 3.1.6 Mapping Topography, resource attributes, and landuse have been mapped variously at 1:50 000 (1947), 1:20 000 (1972, 1990), and 1:5 000 (1994) within the Jhikhu River basin. For this study, the 1:20 000 topographic base map with 50-metre contours is used for overall assessments within the Jhikhu basin and the 1:5 000 topographic base map with 10-metre contours is used for detailed evaluations within the Andheri basin. The 1990 landuse was mapped using photo interpretation and follow-up field verification (Schmidt 1992). In 1994, the 1:5 000 map was developed from the 1990 map using analytical photogrammetry techniques. The landuse was updated by field checking to yield a 1994 landuse map for use with this 1:5 000 topographic map. Both maps were converted for use in a Terrasoft Geographic Information System.  37 Landuse mapping used in this study follows six categories (Schmidt 1992): bari  cultivated rainfed  khet  cultivated irrigated  forest  trees with a crown cover of more than 10% and a height greater than 2 m  shrub  trees less than 2 m in height or with a crown cover of 1-10% if greater than 2 m  grassland  uncultivated with less than 1 % crown cover  other  outcrop; built-up areas.  Forests in the Middle Mountains are heavily coppiced and harvested but still maintained as "forests" hence the low value of 10% crown cover to distinguish forest and shrub classes. Note that forest, shrub, and grassland definitions are not tied to litter or surface cover. A new legend is developed in this study to evaluate land degradation. Its three classes are: moderately  degraded  - little or no vegetation with minor gullies; or - forests with less than 25% crown cover and no understorey vegetation; or - shrub with crown cover less than 25% severely degraded  - heavily gullied with little or no vegetation productive  - active bari; active  khet  - built-up areas - forests with more than 25% crown cover or with good understorey vegetation - shrub with crown cover greater than 25%. The 25% crown cover value distinguishing forests between the productive and moderately degraded classes recognises that many forests in the study area are in a moderately degraded state crown cover).  (i.e.,  10-25%  38 3.1.7 Interviews and Questionnaires Throughout the period of fieldwork, constant contact was made with the local farmers to improve understanding of their knowledge about their environment. Of particular interest were the techniques they employ to manage their agricultural landscape and the rationale on which these approaches are based. A great deal of this fact-finding was informal and completely unstructured. However, several attempts were made to formalise this part of the study using interviews and questionnaires. A complete list of the interviews which were carried out is provided in Table 3.4. Table 3.4 Year 1991 1992 1992/3 1993 1993 1993  Topics examined for farmer indigenous knowledge. Interview Type unstructured semi-structured structured structured structured structured  Subject Perceptions, attitudes, and approaches Soil classification £7ief-accumulation management Soil classification fiari-erosion management Irrigation-dam management  No. 5 12 32 11 21 32  Interviews were unstructured, semi-structured, and structured. Unstructured interviews were carried out only in the first year of the study to improve overall understanding. Semi-structured interviews were used to pursue a more narrowly-defined subject and to allow the farmer to reveal as much of her or his knowledge as possible. Structured interviews were used to evaluate farmer approaches for specific subjects after unstructured and semi-structured techniques were exhausted. These enquiries consisted of questions which could be answered quantitatively. Questions were written into formal questionnaires so that the results could be compiled easily on a spreadsheet. Questionnaires were brief, taking at most ten minutes to complete. Typically, they were translated to the farmer by a Nepali staff member who also recorded the farmer's response. An example of each questionnaire appears in Appendix A4.  39 3.2 Laboratory methods 3.2.1 Stream Sediment Samples Analysis of stream suspended sediment was carried out in thefieldheadquarters and at the University of British Columbia. Beforefiltering,both the conductivity and the pH were measured using a Hanna HI 9025 meter. The samples werefiltered(generally overnight) through a pre-weighed Whatman 40filter(medium - 0.008-mm mesh). Some clay-rich samples with a high amorphous content werefilteredtwice. Eachfilterpaper was air dried and then sent to Kathmandu from the field where it was oven-dried and weighed. Further analyses of the suspended sediment were carried out at the University of British Columbia. The Munsell colour designation of all samples was notedfirstbefore any destructive analyses took place. Chemical analyses carried out on these sediments were limited to phosphorus, carbon, and pH due to the small sample sizes generally available. Phosphorus was measured as orthophosphate with a Lachat instrument. Total carbon was determined with the Leco Analyser (Lavkulich 1978). pH was measured using a Radiometer Copenhagen PHM62 Standard pH Meter. A sub-sample of no more than 3 g of sediment was taken manually from a thoroughly mixed sample. Sediment particle-size distribution was determined byfirstdry-sieving this sub-sample to 0.250 mm. In each case, the sediment was separated into the following size classes: 1.4-2.0 (mm) 1.0 - 1.4 (mm) 0.710 - 1.000 (mm) 0.500 - 0.710 (mm) 0.355 - 0.500 (mm) 0.250 - 0.355 (mm) The size distribution of the remaining sample (though not less than 1.5 g) was determined using a Sedigraph 5100 Particle Size Analysis System. The two sets of results were reconciled to the original sample using a spreadsheet.  40 3.2.2 Stream water samples  Stream water samples were analysed in Kathmandu in the Water Laboratory operated by the Department of Hydrology and Meteorology (HMG). Conductivity, pH, sediment concentration, and dissolved sodium, potassium, calcium, magnesium and orthophosphate contents were measured. Conductivity and pH were measured in the same way as they were in our field laboratory. Sediment concentration was virtually nil becausefilteringwas normally carried out in thefieldbefore sending the sample for water analysis. Dissolved phosphorus was measured as orthophosphate. Sodium and potassium were analysed using a flame photometer. Calcium and magnesium were measured by titration.  3.2.3 Soil samples  Soil analyses were carried out on soils from upland bari, surface and cumulic soils from irrigated khet, and soil horizons samples taken from soil pits dug nearby where measurements of infiltration capacity were taken. Depending on the type of soil sample, a variety of physical and chemical parameters was determined. Soil chemical analyses included pH, total carbon, available phosphate, exchangeable cations, cation exchange capacity, and base saturation (Lavkulich 1978). The Leco Analyser gave total carbon for a 0.5-g sub-sample through complete combustion. Available phosphorus was measured as orthophosphate using a Lachat Autoanalyser. Exchangeable cations (Ca, Mg, K, and Na) and cation exchange capacity were determined using ammonium acetate extraction and atomic adsorption spectrophotometry. Physical parameters included particle-size analysis and colour. Particle-size distribution was determined using a hydrometer and Stokes' Law (Lavkulich 1978). The coarse fragment content was calculated from dry sieving to 2 mm. Colour was determined according to the Munsell colour system.  41 3.3 Data synthesis  To prepare the data for detailed analysis, some basic assimilation procedures were used. Four important examples are: • • • •  Development of stage-discharge relations Parsing of automated data. Reconstruction of rainfall record from tipping-bucket record. Use of a geographic information system to integrate spatial data  3.3.1 Stage-discharge relations  Stage-discharge relations have been developed atfivehydrometric stations: Jhikhu River below Bhendabaribesi (station 1), Lower Andheri River (station 2), Dhap River at Shree Rampati (station 3), Upper Andheri River above Kukhuri River (station 9), and Kukhuri River at Andheri River (station 10). For each point on the stage-discharge relation, individual velocity measurements are combined to yield an estimate of total flow at the recorded gauge height. Totalflowis determined byfirstmeasuring flow velocity and depth at a selection of lateral points in the stream cross-section. At each lateral point, at least one velocity measurement is taken at a vertical position chosen to represent the average velocity over the entire depth at that lateral position as described by Buchanan and Somers (1969). If the lateral slice is less than one metre in depth, a vertical distance of 0.6 of the depth below the surface is acceptable. If the depth is greater than one metre, two measurements are recommended - one each at 0.2 and 0.8 of the total depth. Where the 0.6 or 0.2/0.8 method has been successfully carried out, the entireflowrateis determined by addition knowing the area of each slice (using stream depth and the amount of the stream width represented by the measurement at that lateral position). As explained in subsection 3.1.2, many flow measurements cannot be carried out in this standard way. Instead, aflowmeasurement is made at the stream surface which is at a higher point in theflowthan the 0.6 and 0.2/0.8 methods suggest. In these instances, a correction factor (CF, modified from Hoyt 1912 based on comparison of surface and 0.6 measurements taken at medium  42 flows) has been applied to the calculated velocity (V  .)  tUT(aa  V = CF x V,  urface  where  as follows:  depth < 1 m CF = 1 depth ^ 1 m CF = 1 - (depth/20) (depth does not approach 20 m)  Once the flows are calculated and summed for each measurement, the stage-discharge pairs are assembled in an x-y plot. Linear regression using a logarithmic transform is used to develop relationships of discharge as a function of gauge height. Thefivestage-discharge relations which are used in this study are provided in Appendix A5.  3.3.2 Automated data Automated data from the twelve data loggers in this study were parsed using in-house algorithms. Typically, long periods pass with little or no change in data reading. This applies to river stage, rainfall, and air/soil temperature. Data parsing follows the general principle of retaining only thefirstand last readings within an allowable, user-defined tolerance. Thus, when a gap in the parsed record exists (other than data gaps due to equipment error etc. which were appropriately flagged), it is known that the "missing" data fall between the values bounding them subject to the tolerance applied by the parsing algorithm. Table 3.5 shows the tolerances used to parse all automated data. The value of zero for the rainfall data indicates that only non-zero readings were retained and therefore the rainfall record can be reconstructed with absolutely no loss of information. In contrast, the stage and temperature record suffer a small loss in resolution according to the magnitude of the tolerance. The tolerances used in both cases were far smaller than any value which would threaten the integrity of the database.  3.3.3 Tipping-bucket rainfall data Any tipping-bucket rain gauge records total rainfall during small time periods continuously. No information is retained over timescales smaller than the basic time period of integration. At high  43 Table 3.5 Tolerances used to parse automated data. Data Type Stage, station 1 Stage, station 2 Stage, stations 9 and 10 Rainfall Air temperature Soil temperature  Tolerance 6.6 cm 3.0-4.4 cm 2.6 cm 0 mm 0.5'C 0.8'C  Note: In each case, if the change in reading is less than or equal to the tolerance, then the subsequent reading is not retained. rainfall rates, this is essentially of no concern but at low rates, especially those falling below the rate of one tip/period, the loss of information can be important. For instance, if one tip is recorded every hour for three hours, did each tip's rainfall occur during one period? two periods? or during some combination of periods up to each entire hour? The tipping bucket does not provide this information so some assumptions must be made to deal with these low-rainfall rates. The assumptions used in this study are as follows: 1)  Single tips are averaged over all preceding sample periods (within the same storm) for which no rainfall was recorded. This value is also imposed upon a preceding single tip if it begins the storm record.  2)  If more than one tip occurs in a tipping period then the tips are not averaged over preceding tip periods regardless of the number of tips recorded in these earlier periods.  3.3.4 Geographic information system The larger study has integrated all spatial biophysical and socioeconomic data into a Terrasoft Geographic Information System (GIS). The GIS has also been used to derive a digital terrain model for the basins under study. The present study uses this GIS and its database for three applications:  44 1) Overlay  techniques  Land and landuse attributes are overlaid to determine spatial relationships. For example, the soil colour map is overlaid with the digital terrain model to reveal the elevation of red soils. Also, rainfall isohyets are overlaid on the soil colour map to determine the soil colour where rain is heaviest. 2) Area  computation  Areas are computed for various physiographic and landuse classes. For example, the area is computed within a basin which is in specific bands of elevation, aspect, and slope. 3) Point  data  A variety of point data are determined from the GIS. For example, stream profiles are developed using point data along stream courses.  45  4.0 Monitoring monsoonal rainfall for studying erosion and sediment transport 4.1 Introduction In Chapter 2, patterns of local and regional climate were characterised descriptively with a focus on rainfall over monthly and annual timescales. However, rainfall delivery is highly variable in time and space and, in the Middle Mountains of Nepal, these variabilities have the potential to affect profoundly erosion and sediment transport. Patterns of rainfall behaviour important to erosion and sediment transport are therefore evaluated here to address the significance of rainfall to the overall diagnosis of basin sediment dynamics over different spatial and temporal scales. A storm classification is developed for use in the erosion analysis. In the study area, erosion and sediment transport can respond quickly to changes in rainfall delivery. The landscape and its surface form a highly heterogenous mountainous environment, resulting from both very steep topography and a high degree of site-specific human manipulation. Surface condition changes both temporally and spatially in response to management. Variability in both rainfall delivery and erosion susceptibility interact to yield highly complex basin sediment dynamics. Rainfall monitoring is a sampling exercise which must pay careful attention to regime variability to be successful in its characterisation. Decades of research examining the mechanics of rainfall measurement have yielded a good understanding of how to sample rainfall effectively at a point. In contrast, understanding how to sample for temporal and spatial variabilities is grossly inadequate, especially in mountainous environments. Logistics, cost, and challenges in measurement all contribute to the paucity of studies. Some specific questions addressed by this chapter include: Temporal  a)  Variability  How do storm-period variables (especially storm intensity, total rainfall, duration, and burst timing) vary between storms?  46 b)  Do these storm-period variables change seasonally?  c)  Does the pattern of spatial variability in storm rainfall change seasonally?  d)  Does storm frequency change seasonally?  e)  How frequently must rainfall be measured to adequately assess storm-period variability?  Spatial  Variability  a)  How much rain is actually falling over the basin?  b)  Where is the rain falling?  c)  How accurately does the gauge network monitor rainfall?  d)  Are there predictable local and elevational changes in rainfall delivery? This study pursues answers to these questions using recording and storage rain-gauge data  gathered over a variety of spatial and temporal scales. Recording rain gauges have been in place for three years at five different sites (see Figure 3.1). Four of these sites are located to provide results for a 2x2 matrix of high and low elevation and south and north aspect. The fifth gauge is located at high elevation and north aspect, far removed from the sites of the other four. About 25 storage gauges provide spatial resolution of 24-hour rainfall on each of two adjacent hillslopes during the rainy season. The monitoring network on the south-facing hillslope was in place for 1992 only while the north-facing network has been in place during each rainy season of 1992-1994. On each hillslope, the network is divided into a high- and low-elevation cluster each of at least 10 gauges. Eight storage gauges have been in place since 1992 providing uninterrupted measurements of 24-hour rainfall at locations throughout the Jhikhu River basin. After discussing findings from other research which are useful to this study, three analyses are presented. Analysis of temporal variability focuses on changes in storm-period variables during a storm and across seasons. Spatial variation regionally with elevation and locally within a storm cell and across a hillslope form the focus of the following section. Findings from these two analyses are assembled in the last section to yield a classification approach to addressing the rainfall parameter in  47 studies of erosion and sedimentation.  4.2 Research background Though there has been a great deal of study about the mechanics of precipitation, there remain large knowledge gaps in topographic patterns of rainfall delivery, especially in the Himalaya. The extent of variability inherent in systems of mountain precipitation, the difficulty of gathering data - including automated data - in the Himalaya, and the common problem of adequate monitoring to address spatial variation are the most common causes for these gaps in understanding. After presenting a review of rainfall behaviour emphasising rainfall delivery at the ground surface, findings from Himalayan studies useful to the present work are highlighted to calibrate the understanding developed in other regions and to provide a context for this study's data and results.  4.2.1 Patterns of rainfall delivery Spatial patterns  Although a wide variety of parameters affects rainfall delivery and its spatial distribution as measured at the ground surface, it is the effect of elevation which has received perhaps the greatest interest. Henry (1919) outlined the early studies (of the 19th century), discussing examples from around the world of changes in precipitation with elevation. He pointed out that there is typically a zone of maximum precipitation - generally below 1000 m in the tropics and between 1400 m and 1500 m in the temperate regions. He recognised that for mountains to cause an increase in precipitation, their axis cannot be parallel to the direction of the winds bringing the storm rainfall and that the steeper the slope, the greater the rate of change of precipitation. Givone and Meignien (1990) recognised this relation in establishing an array of 14 recording gauges perpendicular to the mountain range under study. They examined the detailed meteorological features of the rainfall before studying the influence of topography on it. They found that instability triggered at the footslope of the  48 mountain yielded the heaviest hourly totals at low elevation with the heaviest daily totals occurring at the top of the range. Henderson (1993) examined elevational effects on storm total-rainfall in the Southern Alps of New Zealand and found the maximum rainfall to occur at 10 to 20 km upwind of the elevational maximum. These findings underscore the importance of carefully selecting the stormperiod variables under consideration and the range of elevation for which extrapolation is acceptable. Smith (1989) described the four most likely mechanisms of orographic precipitation: smooth, forced ascent; seeder-feeder mechanism; diurnally-forced convection; and triggered convection by forced ascent or blocking. A common aspect is that precipitation enhancement occurs on the upwind slope and can be so reliable as to be used as a crude indicator of regional wind direction. Unfortunately, despite significant study of precipitation change with elevation, comparatively little is understood about the specific mechanisms involved. He suggested that this lack of understanding must change if we are to better predict spatial variability in mountainous terrain. The rainfall-elevation effect, however, is farfromalone in shaping spatial patterns of rainfall delivery. Spreen (1947) realised this and included slope, exposure, and orientation as independent variables important to rainfall spatial variability. Spreen's ability to predict precipitation locally increased dramatically when these other factors were also considered (R=55% elevation only; 2  R=94% all factors). Importantly, he found that in steep mountainous terrain, these non-elevational 2  factors dominate locally. Sinclair (1993) blamed this lack of "orographic resolution" on the inability of elevationally-driven approaches to predict rainfall spatial variation in an extreme event. These studies have one thing in common: they examine spatial variability over large scales (> 100 km). In the Middle Mountains, and certainly in many other mountainous areas of the world, 2  though the average rainfall over many storms may vary spatially in a predictable way, individual events may not. This temporal variability over afixedspatial scale is a topic rarely discussed in the literature. There are few examinations of the behaviour of individual cells of limited spatial extent (1 to 10 km) in mountainous terrain. For instance, Paturel and Chocat (1993) used measurements from 2  49 a network with average density of one gauge per 20 km by asserting that this figure corresponds to 2  the smallest convective cell likely to have an influence on the drainage network. They noted and did hot explain, however, that a high degree of spatial heterogeneity resulted in their rainfall data. Temporal  patterns  Another common approach to examining dynamics of storm rainfall uses descriptive statistics of myriad storm-period variables. There appear to be as many analyses of storm-period variables as there are authors. For instance as shown in Table 4.1, the definition of a storm (when it is given) is rarely the same from author to author. Minimum total rainfall (R) and time-without-rain (S) requirements are the dominant criteria. How these are applied appears to depend on the nature of available data. For instance, Reid et al. (1981) had 24-hour rain data and established the definition of storm on that basis. Tropeano (1991) discussed 280 events in detail but did not present his storm definition. These studies tend to define a storm to accommodate the data available and do not evaluate the appropriateness of the definition chosen. Depending on the goal of the analysis, there exists the potential of introducing an error at a fundamental level.  Table 4.1  Published storm definitions.  Author(s) Huff (1967)  Storm Definition S 2> 6h R ^ threshold from network R ;> 25 mm anywhere in basin R ^ 0.5 mm in 24 hours with a day of R < 0.5mm on each side None provided S > 1 h and R ^ 2.5 mm S ^ 4h S> 1 h T  Collinge and Jamieson (1968) Reid et al. (1981)  T  T  T  Tropeano (1991) Farmer and Fletcher (1972) Patural and Chocat (1993) Overseas Development Agency (1995)  T  Note: R = total storm rainfall (m) S = length of time without rain before and after a storm (h) T  50 Having established the definition of a storm, there are many approaches to analysing the storm-period variables. Herschfield (1962) presented extensive analyses of storm-period variables and Huff (1967) pointed out the importance of tailoring such analyses to a specific application with specific spatial and temporal scales of concern. Unfortunately, after analysing 261 storms using a quartile-separation approach, Huff (1967) determined that quartile separation does not vary consistently with storm type. The results of these analyses uncover some interesting temporal dynamics in rainfall behaviour at a point. Farmer and Fletcher (1972) examined convectional, high-elevation burst-rainfall behaviour in Utahfindingthat over half of storms yielded more than 50% of the storm rainfall in a 10-minute burst. Also, in most storms the period of heaviest rainfall occurred in thefirstquartile. They used their burst information to define storms for planning purposes in soil erosion and flood protection. Their study illustrates that relevant rainfall statistics ultimately depend upon precipitation mechanisms, in this case semi-desert to desert convectional showers driven by extensive surface heating. Bryant (1991) documented an apparent worldwide maximum envelope of rainfall intensity for any given timescale. Presumably such a relation could help develop an appreciation for the relative severity of different mountain precipitation regimes. Bryant also emphasises the lack of rainfall data over short timescales of less than 5 minutes. Given the many hurdles to adequately measuring storm rainfall distribution in both time and space including the instrumentation concerns reviewed in section 3.1.1, it is not surprising that many researchers have turned to rainfall models for their rain data (e.g.,  Rodriguez-Iturbe and Eagleson 1987). Another approach to estimate rainfall in mountainous terrain is to use streamflow data  distributed back over the catchment and corrected for evapotranspiration (Danard 1971; Anderson 1972; Ishihara and Ikebuchi 1972; Dingman 1981; Griffiths 1981). Unfortunately, it is for flood prediction in mountainous terrain that storm characterisation is most needed (e.g., Collinge and Jamieson 1968; Dingman 1981).  51 4.2.2 The Asian Monsoon The Asian monsoon which brings the rainfall examined in this study occurs because of the seasonal reversal of winds across the Indian subcontinent. Webster (1987) identified the three fundamental driving mechanisms behind the planetary monsoon: 1) the differential heating of the land and ocean creating the pressure gradient which drives the winds from high pressure to low pressure; 2) the rotation of the Earth which introduces a swirl to the winds; and 3) moist processes that store, redistribute, and selectively release the solar energy arriving over most of the tropics and subtropics. The Asian monsoon is notable for its immense extent and the penetration of its influence beyond tropical latitudes (Barry and Chorley 1992). The potential magnitude of rainfall in this region and both spatial and temporal variability of the monsoon are important to the present study and are discussed in this section in relation to synoptic-scale precipitation mechanisms. The Asian monsoon is known for its ability to deliver huge quantities of rainfall. In developing the concept of an envelope of maximum possible rainfall in relation to total measurement time (Bryant 1991), the data for all time periods greater than one week came from Asian monsoon measurements in Cherrapunji India. The highest 24-hour rainfall ever recorded in the Ganges basin was 823 mm (at Nagina in Uttar Pradesh), in the Brahmaputra basin it was 1036 mm (at Cherrapunji) and in Nepal it was 539.5 mm at Tistung (Dhital et al. 1993). The mechanisms responsible for these extreme rates of rainfall delivery include cyclonic storms near the Bay of Bengal, various types of low-pressure (widespread across the Subcontinent), and orographic mechanisms due to complex topography and, at times, extreme relief (Takahashi and Arakawa 1981). The variety of large-scale effects and the contrasts in topography over this vast region result in considerable spatial variation in rainfall delivery across the Subcontinent. Because the coastal cyclonic storms generally do not adequately penetrate the Subcontinent to  52 reach the Middle Mountains near Kathmandu, it is the monsoon depressions (areas of moderate lowpressure covering tens to hundreds of thousands of square kilometres) originating near the Bay of Bengal which bring most of the precipitation to the study area during the period of the southwest monsoon. The area is prone to thunderstorms (Takahashi and Arakawa 1981), likely triggered by the suite of orographic effects resultingfromthe area's complex topography. The Middle Mountains possess adequate relief (in contrast to the Siwalik and Mahabharat Ranges to the southwest) to cause smooth forced ascent and to trigger conditional and convective instabilities. Also, convection is triggered by upslope winds and the considerable diurnal heating of slopes (Barry and Chorley 1992). Undoubtedly, these effects may all be present in addition to topographically-created preferred pathways. Superimposed on the large-scale regional spatial variability in mountain precipitation are significant local variations. For example, Upadhyay and Bahadur (1982) studied rain gauge data in the Dehang catchment in the Brahmaputra basin and found that correlation between adjacent stations became negligible at about 35 km. Dittman (1970, in Domroes 1979) found that there was "no systematic connection" between the daily totals of rainfall for five gauges - located in a mixture of valley bottoms and ridge tops - within a few kilometres of each other in the elevation range of 18602130 m near Jiri, Nepal. LRMP (1984) suggested that rainfall intensity is higher at lower elevations. Despite the extent of these important local differences, they are poorly researched (Bruijnzeel and Bremmer 1989) and, unfortunately, the overall density of rain gauges is also low - in Nepal, 1 in 490 km in 1984 (Department of Hydrology and Meteorology 1984). The World Meteorological 2  Organisation recommends that in mountainous regions of tropical zones rainfall monitoring stations be installed in altitude zones of 500 m per zone and with a minimum density of 1 station for 100-250 km (World Meteorological Organisation 1981). Given the emphasis on valley-bottom gauge sites in 2  Nepal and the other factors mentioned above, there is considerable uncertainty about actual precipitation rates at ungauged mountainous locations.  53 Orography causes total precipitation generally to increase because of the relation between temperature and saturation vapour pressure. But since less moisture reaches still higher (and colder) elevations, precipitation declines (Upadhyay and Bahadur 1982). This behaviour has been observed in, for example, the Canadian Coast Range (Loukas 1994). In the eastern part of the GangesBrahmaputra basin, a decrease in rainfall with elevation has also been observed (Bruijnzeel and Bremmer 1989). In his work in the High Mountains, Ries (1991) found rainfall to be inversely proportional to elevation between 2000 and 3300 m. Rainfall intensities were found to be greater in the valleys than on the hillslopes. Unfortunately, these authors do not report how well gauge catch biases were controlled in their studies. Although the Asian monsoon demonstrates considerable predictability, temporal variability can be significant. Das (1987) illustrated the annual arrival date of monsoon rains across the Indian subcontinent. Near Kathmandu where the study area of the present research is located, the initial monsoon burst typically comes in the second week of June. This start date is defined by an increase in the moisture content of the atmosphere and a sustained increase in rainfall (Das, 1987). Although 80% of annual precipitation falls during the monsoonal months of June, July, August, and September, there is a variable pulse alternating between active and break periods (Barry and Chorley 1992). This wet-season variability in monsoonal rainfall delivery is similar to the widely-recognised variability in tropical rainfall (Jackson 1989). Although breaks in the Asian monsoon are most common in August and September and last on average five days, they may occur at any time during the summer and can last as long as five weeks. Rainfall delivery in the Middle Mountains shows a nocturnal maximum. Winkler, Skeeter, and Yamamoto (1988) found this for hourly precipitation in the United States, particularly at high intensities. This behaviour is likely related to regular diurnal cooling of convective showers (Shaw 1972).  54  4.3 Definitions  How storm-period variables are defined strongly influences how effectively they can be used in this study's diagnosis of basin sediment dynamics. The storm and seven storm-period variables are defined in this section. In each case, the choice of spatial and temporal scales is guided by the application to erosion and sediment transport of interest here (100 m to 10 km in particular). 2  2  4.3.1 Storm  In quantitatively representing storm rainfall-intensity data, the continuous record of rainfall is first divided into distinct events. After storms have been delineated, storm-period variables can be quantitatively extracted. Two steps are applied in defining a storm. In thefirststep, a minimum "time-without-rain" or storm separation time (S^N) is used and is applied to the beginning and end of a period of rain. Because storm events will later be related to flood events, this approach is convenient because it creates a storm characterisation which parallels that of the floods. In the second criterion, "storms" of inconsequence for sediment dynamics arefilteredout through a minimum totalrain constraint (RMJ ). N  At what value should  be set? Two practical considerations narrow the range initially:  •  avoid combining storm rainfall arisingfromdistinctly separate rainfall events and  •  avoid resolution effects of the tipping bucket rain gauge.  Thefirstof these suggests that S  mN  should be less than the time it takes for a synoptic scale event to  come and go - no less than about half a day and generally much more. To assess the quantitative extent of the resolution effect, we divide the poorest resolution (1.258 mm - adjusted) by the lowest intensity of consequence for erosion in a storm event, about 10 mm/h. With these numbers, we expect 7.5 minutes to occur between measurements by the instrument even under continuous rain. If we make SM, less than 7.5 minutes, we will create separate storm events as a result of this resolution n  limit.  55 Tofixthis variable's timescale within this "operating range", consider the variation in the number of storms (N) as a function of S . As illustrated in Figure 4.1a, as S  mN  M1N  becomes small and eventually reaches unity. Conversely, as S  mvi  becomes large, N  becomes small, N will grow large  up to a maximum related to the length of the record and the propensity for and the frequency of rainfall in the specific location. The continuous data for allfiverecording gauges in this study have been subjected to this sensitivity analysis and the normalised results are presented in Figure 4.1b for the range 8^^,= 0.1 to 12 hours. This graph confirms the intuitive trend illustrated in Figure 4.1a. Additionally, between S = 1 and 2 hours, a major change develops in N. In other words, as S mN  mu  is reduced from 12 to 2  hours, little change in total number of storms results, but when the criterion is varied further to 1 hour, a large increase occurs. This is the lumping together of separate flood-causing convective bursts which we set out to avoid. We can get further help in choosing S  mti  events which result from these storms. Ideally, S  within this range by also considering the flood MIN  should be chosen to avoid unnecessary splitting  and lumping so that all the rainfall that contributes to a certain flood event gets accounted for in an individual storm. In these headwater areas, it takes typically only a few hours for a flood to rise and return to a small percentage of its peak flow. If  is significantly less than these few hours, then  undesirable storm splitting will occur. Specifically, in this case, it is best to use the highest value in the range being considered, namely S  MIN  = 2 hours.  Having divided the continuous rainfall record into discrete events, a minimum-size criterion (Rvnn)  is applied to the results. Storms of less than 3 mm are excluded from further analysis because  they are unimportant to erosion and in addition, the measurement of single-tip and double-tip storms can be unfavourably affected by resolution limitations of the instrumentation. The combined application of these criteria yields, typically, 80 storm events per year at each of thefivesites. It is these storms which are the subject of the analyses in the following sections.  56 Figure 4.1  Number of storms in relation to (a) generalised minimum-time-without rain, and (b) specific minimum-time-without-rain at the five study locations.  Nmax  o o N= 1 S=0  S-> large  Minimum time-without-rain (S)  1.0  Storm number normalised Tipping Bucket Resolution as a proportion of the number of storms at 0.1 h. •»—•«—- Coarse  0.8 O CO  Fine  0.6  High Elevation Low Elevation  o 0.4  *"™"  0.2 0  1 8  """"»» «» Ss jn  2 4 6 10 Storm Separation (hr)  ,  12  57  4.3.2 S t o r m - p e r i o d variables  Appropriate storm characterisation underpins any examination of the effect of rainfall on erosion and sediment transport. Storm-period variables must be extracted from the rainfall record so that rainfall aspects most germane to erosion will be quantitatively and comparatively expressed. For example, it is known (Hudson 1981) that rainfall intensity plays an important role in initiating surface erosion, but what is the best expression of this storm-period variable? Which timescale(s) is most appropriate? Short-term rainfall intensity and total rainfall are the two rainfall characteristics which most strongly shape the character of the erosion process (see Chapter 5). High-intensity bursts are effective at initiating surface erosion at the plot scale as a result of the kinetic energy they possess. Antecedent moisture conditions (influenced by prior rain activity) can also greatly affect the ability of these bursts to cause surface erosion. Large rainfall volumes (usually over a long period) shift the emphasis within the sediment budget from surface erosion at the plot scale to slumping, streambank erosion and bed scouring (often farther downstream) due to large, channelised flows. Several storm-period variables are proposed to address surface erosion. A characteristic of tropical and subtropical rainfall is heavy burst rainfall lasting only a few minutes. Several minutes of rainfall at high intensity can cause damaging surface erosion on sloping agricultural fields. In choosing a timescale to characterise this intensity, the resolution of the recording rain gauge is the major constraint in accurately quantifying the short-term rainfall intensity. In this study, most measurements of rainfall intensity are for two minutes, though 5-minute and 10-minute intervals have also been used. If this short-term intensity timescale is set smaller than ten minutes, the intensity parameter cannot be calculated for all the storms. Further, if it is set too close to the resolution of the instrument, then the intensity parameter will be undesirably affected by how the instrument randomly measures the event. For these reasons, the maximum rain that falls in a ten-minute period of a storm  58 (I ) is measured for each storm. 10  The timing of the peak 10-minute burst (T ) and antecedent moisture conditions can both 10  greatly influence the erosivity of the burst. For this reason, two other storm-period variables are considered: T  10  S  The start time of the peak 10-minute rainfall intensity relative to storm start The periodflir)before a storm since the end of the previous storm ("period-withoutrain")  Together, these variables provide a good relative indication of the antecedent moisture conditions as I begins. 10  Several storm-period variables are defined to address the effect of large amounts of storm rainfall. Since total storm rainfall (R T) and duration (T ) are frequently cited in other studies these TO  DUR  are also considered in this analysis. In these steep headwater catchments, the flood-causing rain of a storm often does not occur over the duration of the storm (sometimes it is delayed as long as ten hours). It appears that the bulk of the rainfall of most storms occurs over a period of about one hour. What timescale best represents this characteristic? As the value becomes larger, more storms will not be characterised because of inadequate duration. As the value is made smaller, it will not properly represent the total-flood-causing-rainfall concept. One hour is chosen and 1^, is defined as the maximum amount of rainfall occurring in a 60-minute period of the storm. Consistent with T , T is 10  w  defined as the start time of the peak 60-minute rainfall intensity.  4.4 Storm-period variables  This section describes the temporal variability of rainfall by examining storm-period variables in the study area and assessing the seasonal change in the distribution of their characteristics. To make the storm classification relevant to the observed geomorphic thresholds for erosion over contrasting spatial scales, findings from Chapters 5 and 8 are brought in to the quantitative discussion  59 which follows.  4.4.1 Distributions Seven storm-period variables - total rainfall (R ), event duration (T ), maximum 10T0T  DUR  minute (r ) and 60-minute (1^) rainfall intensities, within-storm timing of 10-minute (T ) and 6010  10  minute (Tgo) peak intensities, and the rain-free period before each storm (S) - have been determined for all the events at the five recording rain gauges. Distributions of these variables are discussed on both annual and seasonal bases. Cumulativefrequencydiagrams are used to examine gross trends and similarities. Occurrence frequencies by class are examined for statistical differences. Class intervals forfrequencydistributions follow Conrad and Pollak (1950). Figure 4.2 shows the cumulative frequency distributions of the seven studied storm-period variables. These cumulative plots exaggerate similarities and highlight only large differences. All these plots indicate a preponderance of storms at the low-value end of the distributions. This is consistent with thefieldobservation that the majority of storms is not flood-producing. Figures 4.2a and 4.2b show the cumulativefrequenciesof I and 1^, in terms of percentage of 10  total occurrences for each of thefiverecording rain gauges within the Jhikhu River basin (19921994). Figure 4.2a reveals a consistent pattern across all gauges: about 75% of storms have a maximum 10-minute rainfall intensity of less than 30 mm/hr. In Chapter 5, this value is found to represent a threshold for initiation of significant surface erosion. Only 10% of all storms exceed 50 mm/hr, a minimum threshold (of I ) determined in Chapter 8 to be required for significant basin soil 10  loss to occur (see also section 4.6). 1^, exhibits the same coherent pattern, the values being correspondingly lower. R  T0T  (Fig 4.2c) shows more variability between sites with 75% of all storms  delivering less than 15 mm and only 10% of all storms bringing more than 25 mm. Although storm duration is often cited in studies describing storm characteristics, the distributions of T  DUR  given in Figure 4.2d reveal the inadequacy of this storm-period variable for this  60 Cumulative frequency distributions of storm-period variables (1992-1994) at the five study locations: (a) maximum 10-minute intensity (I ), (b) maximum 60-minute intensity (!«))> (c) total rainfall (R T), (d) duration (T ), (e) start of maximum 10-minute rainfall intensity (T ), (f) start of maximum 60-minute rainfall intensity (T )> and (g) period without rain before storm start (S)  Figure 4.2  10  TO  DUR  10  100  w  80 cr  1  S u  60  -  /AS  40 20  30 50 I I , I , 0 1, I 0 20 40 60 80 100 Max. 10-Min Intensity (mm/h)  0  1  2  3  4  Time of Max. 10-Min. Intensity (h)  0  20 40 60 80 100  Period Without Rain Before Storm Start (h)  s u Max. 60-Minute Intensity (mm/h)  0  1  2  3  4  Time of Max. 60-Min. Intensity (h)  61 purpose: the result is sensitive to the resolution of the instrumentation. Whereas T (Fig 4.2e) is also 10  sensitive to gauge resolution, T^ is not because the effect of gauge resolution is lost over the longer period. Figures 4.2e and 4.2f show that burst rainfall occurs very soon after the beginning of a storm. For instance, in half of all storms the 10-minute and 60-minute peak intensities begin in thefirst15 minutes of the storm. Figure 4.2g shows how many hours without rain precede storms. About 50% of all storms follow only 12 hours without rain. A further 25% follow 12-24 hours without rain. Thefinal25% can follow long periods without rain, often of many days. This timescale has important implications for surface erosion, especially if the storm rainfall arrives abruptly on dry ground. For the variables not sensitive to the measuring instrument, Table 4.2 summarises the ranges of the variables in the three classes: 0-75%, 75-90%, 90-100%. These distributions provide an overview of the nature of storms in the study area, one which will be useful in section 4.6 when a storm classification system is described and applied to these data. Table 4.2  Frequency of occurrence of four storm-period variables in three classes. Percentage Occurrence First 75% Next 15% Last 10%  R (mm) 0-15 15-25 £25 T0T  1.0  (mm/h) 0-30 30-50 >50  (mm/h) 0-7.5 7.5-15 £15  (h) 0-0.5 0.5-1 £1  S (h) 0-17 17-34 £34  The distributions shown cumulatively in Figure 4.2 were tested for differences using a ChiSquared Goodness of Fit analysis for thefivestorm-period variables not sensitive to the measuring instrument. Occurrence frequencies were classified using afive-classsystem with the class boundaries established so that each class contained about 20% of all occurrences. For each of thefivevariables, three tests were carried out. In thefirst,the distributions at allfivesites were tested in a combined 5 x 5 test with 24 degrees of freedom. In the other two, occurrences from the high- and low-elevation  62 sites were pooled to form two groups which were tested in a combined 5x2 test with 9 degrees of freedom. Only R  T0T  showed a significant difference (0.90 < P < 0.95). Aspect appeared to drive  this difference (0.95 < P < 0.975) though no consistent trend between the two distributions pooled by elevation was observed.  4.4.2 Seasonal distributions  Figures 2.3 and 2.4 demonstrated that averages and extremes of total monthly rainfall vary during the year suggesting that storm characteristics might change seasonally. The data on stormperiod variables are used to determine quantitatively how storm characteristics change seasonally. The majority of the annual rainfall occurs during the months of June through September with a distinct dry season from November to April. During May and June, the rains first start to arrive while the land surface is desiccated and the cultivated fields generally bare. This period is known as the pre-monsoon season. During the months of July, August, and September, vegetative cover is well developed - this season is termed the monsoon season. The date of onset of the monsoon season is variable but in this study, the end of the pre-monsoon season is determined to be June 30, followed by a transition season of 19 days' duration (from July 1-19) at the end of which the "monsoon" season begins. The choice of these dates is made on the basis of stream suspended-sediment regimes (see section 5.4.2) so in a sense, this is to some degree a "land-surface response" monsoon definition rather than a strictly hydrometeorologic one. For consistency, the same seasonal definitions are used to characterise both the rainfall and sediment regimes. Before continuing, it had to be decided whether the 19-day transition season designed to facilitate the analysis of sediment dynamics - will be included with the pre-monsoon season or with the monsoon season. A sensitivity analysis using three options was evaluated for demarcating the end of the premonsoon and the beginning of the monsoon season: (midnight) June 30, July 10, and July 20.  63 Initially, data from the two gauges on the north-facing hillslope were used to assess the sensitivity of all the variables to changes in the date separating the seasons. I i was the most important to surface 0  erosion and therefore was used as the key indicator in examining the data from allfivesites. The seasonal difference in I was the greatest using June 30. As the separation date advanced, the 10  differences tend to decline at allfivesites suggesting a dilution of real seasonal differences as "monsoon" storms of July 1-19 are added to the "pre-monsoon" storm population. Hence, in this seasonal analysis, the pre-monsoon season ends on June 30 (midnight) and the monsoon season begins on July 1. A second set of Chi-Squared Goodness of Fit analyses containing two parts is carried out on these seasonal storm data. In thefirst,the analyses carried out for the combined data are repeated for the separate seasonal groups. These tests examine differences within seasons. In the second, differences  between  seasons are examined by pooling the data within each season in different ways to  yieldfivetests for each variable: all sites (24 degrees of freedom), high elevation, low elevation, north aspect, and south aspect (9 degrees of freedom). To limit the accumulation of Type I Error, only differences limited to 0.5% shouldbe considered significant [(1-0.005) = 0.882] though all 25  differences within 10% are discussed. The within-season analyses revealed no important differences within either season. The test using data pooled by elevation resulted in a difference in T at 90% though no consistent cause or m  result of this difference was evident. Differences between seasons were the most statistically important of all the results. Of the differences observed, the results for S were the strongest. The high-elevation and north-aspect groupings showed differences of 99% and 97.5% respectively while the result for the seasonal contrast for the south aspect pool was 90%. Consistently, S was higher in storms of the pre-monsoon season. This is an important difference because a longer period without rain before the start of a storm suggests drier conditions and can point to more erodible surface soils.  64 The test results suggest that T«, comes earlier in pre-monsoon storms: differences in southfacing pooled data were significant at 97.5% and in low-elevation pooled data at 90%. Early arrival of a storm's heavy rain should increase its erosivity. If combined with a longer pre-storm dry period, the effect can be important. Consistent with the non-seasonal analyses, some differences were found in R . Not xox  surprisingly, R  T0X  for the monsoon storms was shifted toward higher values. All distributions showed  this characteristic. The seasonal contrast using allfivesites was significant at 90% and it was at the south-facing sites where this effect was the greatest (99.5%). In summary, the similarities between seasons far outweigh the differences. Storm characteristics in the study area are conducive to surface erosion as heavy rain can fall over dry ground throughout the year. The seasonal contrasts suggest that the pre-monsoon storm regime is somewhat more liable to generate surface erosion, exacerbating a situation where the surface soil is already vulnerable to erosion due to its bare condition.  4.4.3  Event ranking  Rainfall  Figure 4.3 illustrates the ranking of the largest monitored storms within the Andheri basin by R and I . Though the previous section showed that storm-period variables are not significantly x  10  different between the upland and lowland, these distributions reveal them to be slightly larger in magnitude at the upland recording rain gauge location. These distributions also reveal that there is little correspondence between the upland and lowland areas for the rankings of the largest events only two events appear in the topfivefor both the upland and lowland areas. Only the July 10, 1992 event appears in all categories, suggesting that it is a significant event of the study period. Though the three-year record is inadequate to justify the development of intensity-durationfrequency relations, these rankings are useful to illustrate the relative magnitude of the major  65  Figure 4.3  Ranking of largest monitored storms in upland and lowland of Andheri basin by (a) I and (b) R .  10  T  140  co  »—i  Rank  |- A (a) Ranking of max. storm 10-minute intensity 120 »•* (>30mm/h) 100  1 2 3 4 5  Lowland  Upland  June 15, 1994 Sept 12, 1992 J u l y 24, 1992 June 5 , 1993 J u l y 10, 1992  June 9, 1 9 9 2 J u l y 10, 1992 M a y 27, 1993 June 2 2 , 1 9 9 2 J u l y 18, 1994  80 60 h  i  40 h  X  20  o  0  10  • •  110  a  a •l-H  4 -—>  o  9 0  20  30  40  50  60  70  Rank  Upland Lowland  I (b) Ranking of storm total rainfall I" (>10mm) 4  70 b  Rank 1 2 3 4 5  Lowland J u l y 24, 1992 J u l y 2 1 , 1992 A u g . 10,1993 Sept. 9 , 1 9 9 2 J u l y 10, 1992  Upland July 10,1992 A u g . 10,1993 M a y 27, 1993 Aug. 24,1993 J u l y 30, 1992 June 9 , 1 9 9 2  50 30 10  T  0  20  40  60  Rank  80  100  66 monitored storm events. The upland experienced two or three events noticeably heavier (with respect to both R  xox  and I ) than all the other events at both the upland and lowland stations. Overseas 10  Development Agency (1995) report a maximum measured rainfall intensity of 144 mm/hr (for one minute) from a three-year monitoring study at several nearby locations. High-flow  A ranking of high-flow peaks at Khukuri, Lower Andheri, and Jhikhu stations is presented in Figure 4.4. The continuous nature of the data from Khukuri and Jhikhu suggests that these distributions represent events with low return period of perhaps less than three years. The result for Lower Andheri stands in stark contrast suggesting that the return period of the flood event of July 10, 1992 is significantly higher than that of the other events during the study period. The local farmers indicated that a flood event like that of July 10, 1992 had not been experienced in over ten years. Though it is difficult to provide a confident estimate of this event's return period, it may be reasonable to consider that it was a ten-year flood event. This event was the highest measured at all three scales providing further confidence in this estimate. With analysis of the subsequent three years' data, this estimate can be improved. Though the five biggest events at the Khukuri and Lower Andheri stations exhibit some consistency, the rankings of the high-flow events at these two stations shows no relation to that of Jhikhu (except for the July 10, 1992 event). These changes in ranking with scale hint at spatial variability in the rainfall-runoff process. In addition, a comparison of the flood rankings at Khukuri/Andheri to the rainfall rankings provided in Figure 4.3 suggests that it is the large totalrainfall events of the upland that are most able to shape the high-ranking events of Lower Andheri station. Rainfall is an important control on the flood regime but it is clearly not the only important control. These comparisons will be useful in Chapter 8 where individual events are examined further.  67 Figure 4.4  Ranking of monitored floods at (a) Lower Andheri, (b) Kukhuri, and (c) Jhikhu stations.  (b) Lower Andheri (2), partial  (a) Lower Andheri (2)  140  Rank 1 2 3 4 5  120 100 80 -  Date Jury 10,1992 July 30, 1992 May 27,1993 Sept. 3,1994 Aug. 28, 1994  60 40 20 0  |^r"p»i»M|M"|H"|  i  0  10  20  30  Rank  40  |  50  (c) Kukhuri (10) Rank 1 2 3 4 5  5 A. 4 3  • —  Date July 10, 1992 May 27,1993 June 9, 1992 July 30,1992 Aug. 3, 1992  0  60  20  30  Rank  40  50  60  (d) Jhikhu (1)  120  Rank 1 2 3 4,5  100 h 80 60  •  10  —  Date Jury 10,1992 Aug. 8,1994 Aug. 10,1993 Sept. 3,1994 July 9, 1994  A  •  2  40  1  20 0  i  0  20  r  40  60  Rank  80  100 120  68 4.5 Spatial variation The effect of elevation over three different scales (1, 10, and 100 km) and local variation 2  within individual storm cells (1 km) are examined using 24-hour rainfall measurements. The 100-km 2  2  analysis (the Jhikhu River basin) uses data from long-term climate stations. The 10-km (hillslope) and 2  1-km (storm cell) analyses use data from the detailed wet-season rain-gauge network (see section 2  3.1.1).  4.5.1 Elevation Over the spatial scale of the Jhikhu basin (100 km), how does elevation influence rainfall? 2  Rainy-season (June-September) data from eight long-term monitoring stations within the basin are presented in Figure 4.5a. The graph suggests that total June-September rainfall increases with elevation, perhaps at a rate of about 0.5 mm/m. These data are also plotted monthly in Figures 4.5b through 4.5e: the same trend is evident in each month. The data are too few to justify further quantitative analysis. Data from the detailed north-facing study basin in Figure 4.6a show that on the Andheri hillslope (5 km), June-September rainfall declines with elevation, driven by the behaviour during 2  June and September. The two groups, separated elevationally at 1150 m, were tested using the MannWhitney U-Test. In June and September, the rainfall in these two groups is significantly different (September: 95%, P=0.014; June: 99.9%, P=0.0001). Reporting on rainfall dynamics in the Likhu basin near Kathmandu, Overseas Development Agency (1995) found that the areas which are the first to receive a system are also the areas with the higher rainfall. This is generally the case on this hillslope: systems arrive in the Jhikhu River valley and move up into the Andheri sub-basin. The shoulder months also exhibit a tighter pattern suggesting that wind effects may be minimised during this period. The variability in rainfall measured over the hillslope during July and August may be attributed largely to the negative bias of wind. The high measurements at Gauge 19 (1182 m) situated  69 Effect of elevation on rainy-season rainfall at eight monitoring stations distributed across the Jhikhu River basin (100 km; 1992-1994): (a) June-September, (b) June, (c) July, (d) August, and (e) September.  Figure 4.5  2  1500  (a) June-September  •  o  Study Data HMG data  Note: HMG data are uncorrected.  3  800 1000 1200 1400 1600 Elevation (m) (c) July  (b) June  3  300  c2  200  100  800 1000 1200 1400 1600 Elevation (m)  500  800 1000 1200 1400 1600 Elevation (m)  (d) August  (e) September  500  400 -  *3 300 po  2 200 h 100 800 1000 1200 1400 1600 Elevation (m)  £  200  1-  800 1000 1200 1400 1600 Elevation (m)  70 Effect of elevation on rainy-season rainfall at 20 monitoring stations located on the Andheri basin hillslope (10 km; 1992-1994): (a) June-September, (b) June, (c) July, (d) August, (e) September.  Figure 4.6  2  1500  (a) June-September  1 1000 1200 1400 1600 Elevation (m) 500  (b) June  500 400 -  400 300 -  S  300 r  <2 200 h  (2  200 h  100 800 1000 1200 1400 1600 Elevation (m) 500  (c) July  100 800 1000 1200 1400 1600 Elevation (m)  (d) August  800 1000 1200 1400 1600 Elevation (m)  500  g  400 -  9  300 -  3  200 -  (e) September  100 800 1000 1200 1400 1600 Elevation (m)  71 at the break in slope below the Kukhuri sub-basin (10) suggests that there may be a mid-slope maximum - perhaps due to a modest exhaustion effect - but the data are too few to confirm or refute this hypothesis. Data are availablefrom1992 for the south-facing hillslope and are presented monthly in Figures 4.7a through 4.7c. The July data are too few to infer any trend in rainfall with elevation on the hillslope. In contrast, data for August reveal a strong increase with elevation, one which experiences no exhaustion upslope. This contrast to the north-facing hillslope could be due to the abrupt nature of the south-facing slope. These findings have important implications for rainfall monitoring in these mountainous regions. Spatially, measured rainfall can vary by 50% on a 5-km hillslope reflecting the combination 2  of real spatial variability of rainfall and the variability due to gauge performance as affected by wind patterns. Real rainfall spatial variability is dependent on the elevation of the slope and on topographic peculiarities of the area especially in relation to prevailing patterns of the region whereas the variability due to gauge performance is undoubtedly affected by seasonal changes in wind strength, a well known control on rainfall catch - unstudied here.  4.5.2 Storm cell The rain-gauge clusters located on the Andheri hillslope are used to examine the variability in rainfall delivery within a storm cell (1 km). The monthly changes in total rainfall revealed in the last 2  section suggest that elevation may affect within-storm spatial variability; because the elevation of these clusters is contrasting, the extent of influence of elevation can also be investigated. The 24-hour rain-gauge network, in place for three rainy seasons on the Andheri hillslope, has yielded hundreds of days of rainfall data, each day's data comprising simultaneous measurements at up to 30 locations. The gauge locations form two clusters. One cluster is over steep, intensivelymanaged agricultural land in the headwater Kukhuri basin. The other cluster is over the lower reaches  72 Figure 4.7  Effect of elevation on total rainfall at 18 monitoring stations located on the south-facing hillslope (10 km; 1992): (a) July - only 7 data available, (b) August, and (c) September. 2  (a) July 500  (b) August  400 r  800 1000 1200 1400 1600 Elevation (m) 500  (c) September  400 h  3  300 r 200 h 100 800 1000 1200 1400 1600 Elevation (m)  3  300  c2  200 100 800 1000 1200 1400 1600 Elevation (m)  73 of the Andheri basin where bare and heavily-gullied, red soils prevail. To examine the variation of rainfall within a storm cell, only the daily 24-hour measurements which derive from individual storm events are considered. Initially, following the storm definition developed in section 4.3.1, only storms with an average of at least 3 mm of rainfall (from either the upland or lowland rain-gauge cluster) are used. To remove the data for which more than one storm contributes, the detailed record from the tipping bucket within each rain-gauge cluster is used. Eight discrete 24-hour periods contain two distinct events (none contains three); the rainfall for these events cannot be known. In addition, storm rainfall cannot be known for the five events which straddle the measurement time (7:00 am) and are followed immediately by a 24-hour period with a storm (or storms). After removing other data, 90 events remain for consideration. For each of these 90 storm events, eleven well-distributed gauges are chosen from each cluster. The mean rainfall and its variance are determined from these data for each event in each cluster individually and for the two clusters combined. These descriptive statistics are then tested for significant difference. Both parametric (t-test, 95%) and non-parametric (Mann Whitney U Test) tests are applied due to the uncertainty of the underlying distribution. In most cases, the two tests agree. When the result from one test is at a lower confidence limit, then the result from the other test is used to rule on the difference. In one case only did the two tests outright disagree: for this data pair, it is assumed that no significant difference exists. Using the results from the statistical tests, the data are divided into three groups. Basin events occur where there is no significant difference between the data from the two clusters. Where significant differences are found, the events are termed as either upland or lowland depending on which has a greater mean rainfall. Finally, the coefficient of variation is calculated for each event. The average of the coefficients of variation for all events within each of the three event-area classes (basin, upland, lowland) is presented in Table 4.3. Also, for all rainfall over each area (the upland, the lowland, and the entire basin) the average coefficient of variation is determined regardless of the  74 Table 4.3  Spatial variation within a storm cell and distribution of events according to lowland, upland, and basin area-events for storms of R £ 3mm and R £ 10mm. T  Areal Extent  Events Included  R £ 3 mm ; 90 events low-elevation lowland cluster only high-elevation upland cluster only entire basin basin low-elevation all cluster only high-elevation all cluster only entire basin all R > 10 mm; 63 events low-elevation lowland cluster only high-elevation upland cluster only entire basin basin low-elevation all cluster only high-elevation all cluster only entire basin all  Average Coefficient of Variation  T  Number Percentage of Total of Events Events  T  0.283  22  24%  0.480  26  29  0.459 0.445  88  47 100  0.426  90  100  0.594  90  100  0.244  20  32%  0.307  14  22  0.328 0.338  29 63  46 100  0.332  63  100  0.459  63  100  T  si**  The events with > 3 mm include those with £ 10 mm. Two upland events yielded no rain in the lowland hence CV cannot be calculated.  event designation (as "upland", "lowland", and "basin"). The lowland events (CV=0.28) show less variability than do their upland counterparts (CV=0.48). The upland cluster sits in steep topography and the resulting variability is evidenced in Table 4.3, this despite the fact that the gauges of the lowland cluster are more widely dispersed than those in the upland. The variability of the basin events (CV=0.46) is similar to that of the upland cluster. Not surprisingly, rainfall for all the events over each area type shows generally more variability than do the area events when considered on their own. About half of these events provide basin coverage with the other half split between the upland and the lowland. Though it was not possible to select this set of storm events to be representative of  75 the entire record, there are 90 events and as such it may be reasonable to assume that the doubleevent (and triple-event) days which were removed have the same distribution. The entire set of events considered above includes many events of low R (though greater T  than 3 mm). Events with R < 10 mm over both the upland and lowland areas are rarely floodT  producing and are therefore of limited interest in this study. These events are removed from the data set, the summary statistics are recalculated, and the results appear in the bottom half of Table 4.3. The spatial variability within a cluster is reduced when only the larger events (R ^ 10 mm) T  are considered. Variability in the upland remains higher than in the lowland. For each area type, the variability is always higher if the entire data subset is included rather than exclusively the local events. Note also that there are far fewer large upland events than large lowland events. By subtraction, there are proportionately far more small (3 to 10 mm) upland events (15) than there are equivalent lowland events (2). The previous section described changes in monthly rainfall on the north-facing hillslope in terms of elevational differences and possible exhaustion and wind-bias effects. Though the exhaustion mechanism may be present and the wind bias mechanism is certainly present, it is clear from the findings of this section that 22% of high-rainfall events (R ^ 10 mm) bring a significantly lower T  rainfall to the lowland hence hillslope exhaustion is not an appropriate explanation for all storm events. Convection is strongly "cellular" in this region at times bringing rainfall to one area but not to the other (50% of cases) and at other times, bringing rainfall to the entire hillslope, perhaps with an exhaustion effect. The mechanism for this elevational difference remains open to speculation. Do hillslope exhaustion effects operate during most storms and cause there to be less rainfall over the upland? Does the local, convective nature of showers in this region mean that storms arriving via distinct topographic pathways bring characteristic levels of rainfall? In Table 4.4, the large (R ^ 10 mm) events are stratified according to season. Basin events, T  though common during the monsoon season (49%), occur infrequently during the pre-monsoon  76 Table 4.4  Seasonal distributions of spatial variation within a storm cell and distribution of events according to lowland, upland, and basin area-events for storms of R £ 10 mm total rain. T  Areal Extent low-elevation cluster only high-elevation cluster only entire basin low-elevation cluster only high-elevation cluster only entire basin low-elevation cluster only high-elevation cluster only entire basin low-elevation cluster only high-elevation cluster only entire basin  Number Percentage of Events Coefficient of Variation Included of Events Total Events Pre-monsoon season; 9 events lowland 0.282 45% 5 upland  0.252  4  37  basin all  0.346 0.366  2 11  18 100  all  0.328  11  100  11 all 0.585 Monsoon season; 45 events lowland 0.230 15  100 29%  upland  0.330  10  19  basin all  0.326 0.331  27 52  52 100  all  0.331  52  100  all  0.433  52  100  season. In other words, the pre-monsoon season shows a greater propensity for local rain. However, the spatial variability of the local pre-monsoon events is equivalent to those of the monsoon season. Figures 4.8 through 4.10 illustrate the rainfall delivery over the Andheri basin hillslope for representative basin, upland, and lowland storms. Figure 4.8a occurred on July 10, 1992 (transition season) and Figure 4.8b shows a basin event which is typical of the monsoon season. Although it was an upland event, it brought heavy rainfall to the entire basin to be the most significant event of the study period. Figures 4.9 and 4.10 contrast upland and lowland events of the pre-monsoon and monsoon seasons. Upland storms like the one presented in Figure 4.9a are infrequent but when they  77 Figure 4.8  Rainfall isolines for storm rainfall over Andheri basin for (a) an upland event during the transition season which was also the heaviest event of the study period and (b) a basin event typical of the monsoon season.  (a) July 10, 1992  Lowland  U p l a n d  500  1000 1500 2000  Isoline values in mm.  (b) August 25, 1993  Lowland  Upland  78 Figure 4.9  Rainfall isolines for storm rainfall over Andheri basin for upland events during the (a) pre-monsoon season and (b) monsoon season.  (a) June 9, 1992  \  \  \  /  \ .\ .\  L o w l a n d  •• •  \  \  ^  \  • U p l a n d  5 0 0  1 0 0 0  1 5 0 0  I s o l i n e v a l u e s i n  2 0 0 0 m m .  (b) August 3, 1992  L o w l a n d  U p l a n d  79 Figure 4.10  Rainfall isolines for storm rainfall over Andheri basin for lowland events during the (a) pre-monsoon season and (b) monsoon season.  (a) June 6, 1993  5 0 0  1 0 0 0  1 5 0 0  I s o l i n ev a l u e s i n  2 0 0 0  m m .  (b) July 21, 1992  L o w l a n d  U p l a n d  80 occur, they are a major threat to upland soil erosion. Lowland pre-monsoon storms (Figure 4.10a) appear to be more frequent in occurrence. An important application of these data involves the determination of how accurately point measurements represent the areal average. This is especially important when attempting to construct basin water budgets. To investigate this representativeness, the following equation E = (t) (s) ( n r  5  (Cochran 1977) relating realised error (E) to the standard error (s) and size (n) of the sample is used (t is the appropriate value from student's t-distribution). This equation is solved for the relative or acceptable error (AE=E/x, where x is the sample mean) to give n = (t) (CV) (AE)" 2  2  2  where CV is the coefficient of variation (s/x) and t is a function of n. This equation is solved iteratively to yield the results which appear in Figure 4.11. Figure 4.11 combined with Table 4.4 indicates the number of rainfall monitoring sites required to achieve a specified level of relative error for upland, lowland, and basin events of the premonsoon and monsoon seasons. For example, with a coefficient of variation of 38% (monsoon season) and 56% (pre-monsoon season), in the order of 100 monitoring sites are required to measure basin rainfall (for storms of R £ 10 mm) to a confidence of within 10%. If the gauge requirement T  were limited to, say, 10 sites a level of error of 30% would result. The coefficient of variation of the lowland (0.23 to 0.28) and upland (0.154 to 0.196) events is smaller so fewer gauges are required for the same level of relative error: 10% requires 10-40 monitoring sites, 20% requires 5-20, 30% requires 3-6, and 50% requires 2-4, depending on which area/season combination is under consideration. A 10% level of error appears unworkable for most applications due to the excessive monitoring requirements - a level of error of 30% appears to be the most practical, if acceptable in the management or research application. Certainly, if only one gauge is installed in the basin, the data will essentially tell nothing reliable about storm rainfall delivery within a basin. This result casts  Figure 4.11  Number of rainfall monitoring sites required in relation to the relative error for a in coefficient of variation.  200  0  10  20  30  40  Acceptable Error (%)  50  82 serious concerns on the value of data from thinly-spread (in valleys) HMG gauges, for example.  4.6 Integration The previous sections examined separately spatial and temporal variability in rainfall delivery as measured throughout the study site. This section attempts to integrate these findings to serve two purposes. The first provides guidelines for those attempting to measure rainfall in the Middle Mountains. The second establishes a quantitative reference framework to be used in subsequent chapters in the diagnosis of erosion.  4.6.1 Classification A storm classification system is proposed based on R and Ij because these are the stormT  0  period variables most important to surface erosion (plot scale, on-site) and mass wasting (larger scales, off-site). Findings from later chapters are combined with insights from section 4.4.1 to yield thresholds. In Chapter 5, a 30 mm/h threshold is observed for significant surface erosion at the plot scale (similar to Hudson (1981) who described a 25 mm/h threshold). In Chapter 8, it is shown that serious basin-scale sediment output corresponds roughly to storm events with Ij exceeding 50 mm/h 0  and R ;> 30 mm. Events with R < 10 mm rainfall yield insignificant basin sediment output. Table T  T  4.5 illustrates the class thresholds of each of these parameters and how their combinations are used to derive three storm classes: minor, intermediate, and major. For all five recording rain gauges, storm-period variables have been analysed to determine storm type for every storm and the resulting distributions appear in Table 4.6. Despite the changes in elevation and aspect and the differences in data gaps from gauge to gauge, there is a striking consistency in the five distributions. A large majority (77%) of all events is minor (either R < 10 x  mm or I < 30 mm/h), of little consequence for flood generation. One in five events is intermediate 10  and likely causes floods. The erosional consequences of these events focus on either surface erosion  83 Table 4.5  Definitions of minor, intermediate, and major storm classes in relation to total rainfall and peak 10-minute rainfall intensity. Peak 10-minute Rainfall Intensity (mm/h) 0-30  Table 4.6  Minor  3-10  Total Rainfall (mm)  £50  30-50  10-30  Intermediate  £30  Major  Distribution of storm events in three storm classes (minor, intermediate, major) at five sites. Site  Total  Minor  Intermediate  Major  North facing; high elevation  211  75.4 (159)  20.9 (44)  3.8 (8)  North facing; low elevation  211  76.8 (162)  19.0 (40)  4.3 (9)  South facing; high elevation  200  78.5 (157)  19.0 (38)  2.5 (5)  South facing; low elevation  173  78.0 (135)  18.5 (32)  3.5 (6)  Outside detailed study area  229  75.1 (172)  21.4 (49)  3.5 (8)  Overall average  n/a  76.8%  19.7%  3.5%  (if I £ 50 mm/h) or on slumping and stream bank/bed erosion (if R £ 50 mm) but not both. The 10  T  major events are erosive in terms of both surface erosion and mass wasting as a result of their high intensity and high volume. These events are infrequent, occurring in about 3% of all storms (1 in 33) Table 4.7 presents the seasonal distribution of storm class at all sites. On average, 38% of all storms occur in the pre-monsoon period whereas 62% occur during the monsoon season. The natural variability around these trends is potentially enhanced by the unequal distribution of data gaps at the sites. Though the storm class distribution within each season varies from site to site, it is remarkably  84 Table 4.7  Seasonal distribution of storm events in three storm classes (minor, intermediate, major) at five sites for 1992-1994 data.  Site  Percentage of All Events Occurring Within Each Season  Percentage of Season's Events Occurring Within Each Class  P  M  North facing; high elevation North facing; low elevation South facing; high elevation South facing; low elevation Outside detailed study area(high elev) Overall, High Elev. Overall, Low Elev.  30.3 (64) 43.1 (91) 42.0 (84) 31.2 (54) 41.9 (96) 38.1 37.8  69.7 (147) 56.9 (120) 58.0 (116) 68.8 (119) 58.1 (133) 61.9 62.2  Pre-Monsoon (P) Min Maj Int 2.2 76.9 20.9 (45) (15) (4) 23.4 6.3 70.3 (70) (19) (2) 16.7 0.0 83.3 (70) (14) (0) 72.2 22.2 5.6 (39) (12) (3) 72.9 24.0 3.1 (70) (23) (3) 75.8 21.3 2.9 3.4 75.2 21.4  Overall  37.7  62.3  75.1  21.4  3.5  Monsoon (M) Min Maj Int 74.2 20.8 5.0 (117) (25) (5) 3.4 79.6 17.0 (89) (25) (6) 4.3 75.0 20.7 (87) (24) (5) 80.7 16.8 2.5 (96) (20) (3) 76.7 19.5 3.8 (102) (26) (5) 3.8 77.3 18.9 77.4 18.8 3.8 77.2  19.0  3.8  consistent. In fact, the integrated average of all sites yields essentially the same distribution for each season - 76.8% minor, 19.7% intermediate, 3.5% major (and obviously the same as the non-seasonal distribution). Observed differences in the storm class distributions were tested for significance using a Chi-Squared Goodness of Fit test. Annual and seasonal distributions were tested by site, elevation, and aspect and each test showed no significant difference (90%). In addition, seasonal differences in these categories were found not to be significant. The significant differences in storm-period variables found in section 4.4.1 are insufficient to persist after storm classification. It is the temporal change in storm characteristics at a site that provides the important difference geomorphically. The minor events rarely cause a change in streamflow important to erosion. The intermediate events are significant but tend to be easily managed. It is the major events, because of their combined high intensity and high  85 volume which are seriously damaging. While these differences are statistically nonsignificant, the small differences in measured occurrence of major events may be important if they persist at other times. Due to their importance in initiating erosion, a slight systematic departure infrequencyof occurrence of major events may make some sites more prone to erosion. For instance, at the low-elevation site, 6.3% of pre-monsoon storms, 3.4% of monsoon storms, and 4.3% of all storms are major storms in contrast with the 3.5% average overall. If important, these concerns would need consideration in light of total storm frequency (Table 4.8).  4.6.2 Storm Frequency In the previous section, conclusions were reached about overall and seasonal distributions of events within the study site. This section seeks to extend those findings by examining the annual and seasonal frequency of occurrence of storms within the basin during the three-year study period. Table 4.8 presents the total number of storm events at each site for each year (1992-1994) as recorded by the tipping buckets. Thesefiguresare not exact averages due to the presence of data gaps of variable extent at most sites. Because these data gaps are generally small, the estimates form reasonable approximations, completed using adjacent 24-hour measurements and inference from the nearest recording gauge (with data). Sites had between 68 and 87 events per year with most sites having between 70 and 80. The overall mean is 77.5 events/year. In Table 4.9, the seasonal distributions by storm class (from Table 4.7) are applied to the average annual eventfrequency(77.5) to yield expected annual stormfrequenciesin the study area. On average, of the 77.5 events which occur each year, 48.0 of them occur in the monsoon season while the remaining 29.5 occur in the pre-monsoon season. Only 17.8 of these are non-minor; of the major events, one will occur in the pre-monsoon season of almost every year while one or two should occur each year during the monsoon season. We also expect almost ten intermediate storms during the  86 Table 4.8  Number of storms in each year (1992-1994) at each site as given by recorded data. Site  Total Annual Number of Storms 1992  1993  1994  Mean  North facing; high elevation  71  87  79  79.0  North facing; low elevation  71  68  72  70.3  South facing; high elevation  70  74  78  74.0  South facing; low elevation  72  83  78  77.7  Outside detailed study area; high elevation  85  87  87  86.3  Overall mean  73.8  79.8  78.8  77.5  Table 4.9  Expected average seasonal storm frequency by class (minor,intermediate, major) within the study area. Minor  Intermediate  Major  Total  Pre-monsoon season  21.9  6.2  1.0  29.2  Monsoon season  37.2  9.2  1.8  48.3  Both seasons combined  59.0  15.7  2.8  77.5  Note: Frequencies of occurrence for combined seasons are based on an averaged distribution, hence do not identically equal the sum of the two seasonal distributions presented. monsoon season and about six intermediate events during the pre-monsoon season.  4.7 Summary and conclusions The specific quantitative findings from this chapter are listed in this section followed by the  87 main conclusions derivedfromthese detailed results.  4.7.1 Summary of quantitative findings Thefindingsfromthis chapter can be grouped around four topics: temporal variation, spatial variation, storm classification andfrequency,and guidelines for monitoring and analysis. In making these observations, a storm is defined as delivering at least 3 mm of rainfall and being separated by at least two hours from other rainfall. Temporal variation  1) Storm-period variables • Peak 10-minute and 60-minute intensities begin within thefirst15 minutes in half of all storms. • In 25% of all storms: • the peak 10-minute rainfall intensity exceeds 30 mm/hr; • more than 15 mm total rainfall is delivered; and • more than 24 hours without rain precedes the storms. 2) Seasonal effects • Storm characteristics are similar between the pre-monsoon and monsoon seasons: they deliver both high-volume and high-intensity rainfall capable of causing severe erosion. • In comparison with monsoon storms, pre-monsoon storms: • deliver less total rainfall; • are shorter in duration; • occur after longer periods without rain; • show a delayed occurrence of peak rainfall. Spatial variation  Spatial variation was assessed as a function of elevation and storm cell variability in terms of total rainfall (over specific time periods) and in terms of storm-period variables:  88 3) Storm-period variables • Systematic differences in storm-period variables are not evident between elevation and aspect; differences observed relate to narrow combinations of season and storm characteristic. 4) Elevation • Over the scale of 100 km , wet-season rainfall (June-September) increases with elevation; the trend 2  is insensitive to season. • Over the scale of a hillslope (10 km) rainfall variation shows a trend to increase with elevation 2  though marked exceptions are observed: • total pre-monsoon-season rainfall decreases with elevation; • total monsoon-season rainfall shows a midslope (1050-1150 m) maximum. It is not clear the extent to which these elevation differences are due to hillslope exhaustion and due to the local "cellular" nature of convection. • Over a hillslope (10 km), total variation in rainfall can exceed 50% in a month; a large proportion 2  of this variance may be due to increased wind bias in measurement over complex topography. 5) Storm cell • About half of all hillslope storm events deliver significantly different total rainfall within contrasting 1-km upland and lowland subregions; 2  • Hillslope-wide events are uncommon during the pre-monsoon season; • Low-rainfall events (3 mm < R < 10 mm) are highly variable especially in the mountainous, x  high-elevation terrain where CV=0.48; • Lowland and upland high-rainfall events (R £ 10mm) demonstrate reduced variability. x  • High-rainfall "area" events (isolated over either the lowland or upland) occur more frequently at low elevation than at high elevation and together constitute more than half the total number of hillslope events.  89 Storm Classification and Frequency  6) Storm classification • A matrix classification using I,,, and R provides a convenient basis for classification because its T  basis is compatible with the rainfall-induced variation in character of erosion: • Minor events deliver R < 10 mm or have I < 30 mm/h; T  10  • Major events deliver R ^ 50 mm and have I i ^ 50 mm/h; and x  0  • Intermediate events form the remainder of the storms. 7) Storm distributions • Annually, 76.8% of all storms are minor events; 19.7% are intermediate events; and 3.5% are major events. • Storm-class distributions are insensitive to aspect, elevation, and season. • Of the wet-season storms, 37.7% occur during the pre-monsoon season and 62.3% occur during the monsoon season. 8) Storm classification and frequency • Annually, an average of 77.5 storms were recorded across all five sites: 59.0 minor events, 15.7 intermediate events, and 2.8 major events on average across all sites. • Only 2.8 major storm events are expected annually (1.0 pre-monsoon; 1.8 monsoon). Guidelines for monitoring and analysis  9) Field Instrumentation • When measuring rainfall intensity for erosion studies in the Middle Mountains, use a minimum resolution of 0.25 mm and a sampling frequency not exceeding 2 minutes; maintain these constant during the monitoring period. • To measure mean storm rainfall over a fourth-order tributary basin, data from one gauge is inadequate; in the 5.3-km study basin, 2 to 4 gauges are needed to limit the allowable error 2  to 50% and about eight gauges are required for a maximum allowable error of 25%.  90 • Maintain duplicate instruments at many sites, especially when different types of instrumentation are used and breakdown can cause gaps. 10) Analysis • A storm definition based on a two-hour minimum period without rain respects the inherent characteristics of rainfall in the Middle Mountains and R  mfi  = 3 mm eliminates gauge  resolution effects. • I and I best reflect rainfall characteristics affecting surface erosion and mass wasting respectively 10  w  (provided the sampling frequency is ^ 5 minutes). • S and Teo provide useful relative indices of antecedent moisture conditions.  4.7.2 Conclusions These findings suggest this chapter's three principal conclusions: • Seasonal storm characteristics  Storms of the pre-monsoon and monsoon seasons deliver equivalent high-intensity, highvolume rainfall capable of causing severe erosion. However, in comparison with monsoon storms, pre-monsoon storms deliver less total rainfall, are shorter in duration, occur after longer periods without rain, and show a delayed occurrence of peak maximum 60-minute rainfall intensity. Spatially, pre-monsoon storm cells are smaller than monsoon storms. These conclusions suggest that the erosivity of storm rainfall in these two seasons should be equivalent but the timing of pre-monsoon storms may enhance the relative erodibility of the land surface during this season's significant storms in contrast with the monsoon season. • Spatial variability  Over a hillslope (10 km), variation in total rainfall often exceeds 50% in a month. This 2  variability is driven by a combination of wind bias (due to instrument limitations and complex topography), local storm-cell structure, and hillslope exhaustion. This conclusion raises severe  91 concerns about the usefulness of rainfall data derived from single gauges, notably when these gauges are not within the catchment of concern. The data may provide useful information regarding broad regional trends but they do not provide any reliable information on synoptic-scale flood-producing storm rainfall. • Storm  distribution  Under a three-class storm classification system designed for this study of erosion and sediment transport and using class divisions based on maximum 10-minute rainfall intensity and total rainfall there is no significant difference in the distributions of storm classes between site, aspect, and elevation within either season. Annually, an average of 77.5 storms was recorded across all five sites. Of these, an average of 2.8 major events (more than 50 mm/hr maximum 10-minute intensity  and  greater than 30 mm total rainfall) occurs: 1.0 event during the pre-monsoon season and 1.8 events during the monsoon season.  92  5. Diagnosing headwaters controls on erosion and sediment transport 5.1 Introduction Factors which shape the erosion regimes of the study catchments are examined diagnostically to evaluate their relative effects on sediment dynamics. A wide assortment of geomorphologic, hydrologic, and agricultural techniques is available for measuring rates of erosion over widelycontrasting spatial and temporal scales. Data from erosion plots are used here to evaluate the effect of individual storms at the plot (100 m) scale during all seasons. The sediment-rating-curve technique is 2  used to examine the combined effects of all erosion processes operating within the catchments during individual flood events over various spatial and temporal scales. By contrasting the behaviours of catchments of different character and over different seasons, dominant causes are inferred. Erosionpin measurements and erosion surveys provide corroboration of the findings. The diagnosis provides a foundation for the sediment-budget analysis in Chapter 8. Rainfall is obviously the precursor to erosion in this region. Rainfall intensity patterns of the pre-monsoon season are strikingly similar to those of the monsoon season. However, pre-monsoon storms tend to be of shorter duration and have longer dry periods between them than do monsoon storms. More rainfall occurs at lower elevations in the study area yet peak rainfall intensity tends to be somewhat higher at higher elevation. These distinct patterns evident in the rainfall regime will be examined more closely for their relation to observed patterns of sediment dynamics. Having established how rainfall behaves in the study area, it is now possible to attempt a diagnosis using erosion and sediment-transport datafromthis study. The evaluation begins with an examination of surface erosion from cultivatedfieldsat the plot scale. Findings are tested and conclusions extended by examining stream suspended-sediment regimes over contrasting spatial scales and basin character.  93 5.2 Research background A wide variety of methods has been used to study patterns of fine-sediment transport, including erosion plots (Mutchler et al. 1988), stream hydrometric stations (Gregory and Walling 1973), soil-profile development (Harden et al. 1979), tracers (Bovis 1982), erosion pins, and erosion surveys (ASCE 1975). Each approach focuses on specific processes and spatial and temporal scales. Erosion plots and stream hydrometric stations are the two techniques used in this chapter so a review of these two tools forms the focus of this section. Himalayan studies of direct relevance to this study are also reviewed.  5.2.1 Behaviour of fine-sediment erosion and transport Erosion plots  Erosion-plot soil-erosion research has a long history in agriculture and the Universal Soil Loss Equation (USLE) represents the most-comprehensive application of this work (Wischmeier and Smith 1978). This equation predicts average annual sheet and rill erosion and was developed using data from over 10 000 plot-years' data from standard plots of 72.6 ft (22.1 m) in length and 9% slope throughout the United States Midwest (Wischmeier 1976). It represents the major controls on plotscale erosion using six parameters: rainfall erosivity (R), soil erodibility (K), slope length (L), slope steepness (S), cover and management (C,P). The Universal Soil Loss Equation (USLE) is useful in that it explicitly recognises all the controls on soil loss operating at the plot scale. For conditions within the range of parameters studied, it provides a convenient and accurate assessment of soil erosion for management. Several attempts have been made to improve upon the USLE. The Revised Universal Soil Loss Equation, RUSLE, (Renard et al. 1991) uses modified USLE factors and computerised algorithms. Unfortunately, like the USLE, outside of the conditions for which the equation was developed, it is often not useful. Increasing computer capabilities have led to the development of  94 complex mathematical predictors which model the successive detachment, transport, and deposition processes within small catchments. The Water Erosion Prediction Project (WEPP) model (Laflen et al. 1991; Risse et al. 1995) is the modern replacement to the USLE, building on earlier attempts like the Chemicals, Runoff, and Erosion for Agricultural Management Systems (CREAMS) model (Foster 1988). These mathematical models are limited in their application in developing countries because of their high data requirements. Despite the limitations, researchers have attempted to extend use of the USLE to other regions. The Soil Loss Estimator for Southern Africa (SLEMSA) was developed in Zimbabwe based on the USLE model (Wendelaar 1978 in Elwell 1984). Narayana et al. (1983) used the USLE and river, reservoir, and soil-loss data to estimate an average annual rate of erosion for India as 16.4 tonnes/ha. They used work by Babu et al. (1978) in developing an R-factor map of India based on climate records. Low (1967) proposed an estimator of soil loss for developing countries based on easily-obtained mean terrain attributes. Elwell (1984) stresses that countries which attempt to determine local USLE factors face an onerous task and often lose interest in prediction techniques believing that development costs are well beyond their resources. In situations where a quantitative prediction of soil erosion is not directly necessary, erosion plots can be efficiently used to diagnose patterns of sediment production over small spatial scales (< 1 ha). For example, Lai (1982) studied the effect of terraces on sediment production and delivery in small basins (4 ha) in Nigeria and found that while terraces are effective at reducing the delivery of soilfromthe plot, they do not substantially reduce soil detachment. Young and Onstad (1982) looked at the effect of soil characteristics on soil erosion and found that the degree of aggregation, aggregate stability, and the soil clay content most strongly influenced sheet erosion. Pandey et al. (1983) and Pathak et al. (1984) used plots to assess hydrological aspects of forested and nonforested slopes in the Kumaun Himalaya in India. Pinczes (1982) studied sloping, Hungarian vineyards,findingthat erosion rates correlated well to runoff and rainfall intensity for low-intensity rain events and correlated poorly  95 for high-intensity events. Richter and Kertesz (1987) discovered a strongly seasonal pattern of erosion from plots in Germany and Hungary. Hudson (1981) explained the importance of the kinetic energy of rainfall in determining soil loss from plots. In developing an alternative to the USLE R-factor for use in Africa, he had found that 25 mm/hr represents a threshold of kinetic energy between erosive and non-erosive rain. He pointed out that others went on to modify this concept proposing various combinations of total rainfall, rainfall intensity, and the energy of the storm (Elwell and Stocking 1975; Lal 1976; Morgan 1977). These studies focus attention on specific erosive storms rather than total annual rainfall at a plot. Plot-basin  scale linkage  To address soil loss over larger spatial scales, the concept of a "sediment delivery ratio" has been developed (Glymph, 1954; Maner 1958; Roehl 1962) and applied to erosion data extrapolated spatially over a basin to link the plot to the basin scales. Walling (1983) defined the sediment delivery ratio (SDR) as the ratio of the sediment delivered at the catchment outlet to the gross erosion within the basin on an annualised areal basis. Unfortunately, the SDR is very hard to predict and depends on the sediment sources, the drainage network, relief, slope, soil characteristics, vegetation, and landuse (Walling 1983; Ichim 1990). The SDR concept is a black-box model which lumps all spatial and temporal variabilities into a single number. Spatial heterogeneities have led to distributed application of the delivery-ratio concept often in conjunction with the USLE. Burns (1979 in Walling 1983) focuses on individual sources and on identifying their separate delivery potentials. With adequate spatial resolution and a thorough diagnosis of the delivery potential of important sediment sources, the SDR concept could accommodate spatial heterogeneity. The alluvial erosion and storage component of the basin sediment budget is not generally included in the calculation of "gross erosion" (Piest et al 1975; Walling 1983) and forms the key  96 weakness of the delivery-ratio concept. Church and Slaymaker (1989) examined landscapes in British Columbia (Canada) responding to glacial disturbance and found an increase in specific sediment delivery downstream due to stream channel bed and bank erosion. Though this has been termed a controversy (Bull et al. 1995), it is simply the alluvial component of the basin sediment budget operating over very long temporal scales. This relaxation time of disturbance is also evident over shorter timescales as sediment wedges cascade through a drainage system over decades as a result of anthropogenic disturbance associated with logging activities (Roberts and Church, 1986). Equivalently, it can also occur over short timescales such as the single event. The problem is further complicated by the size distribution of transported sediment (Walling and Moorehead 1989) and the existence of geomorphic thresholds (Schumm 1977). Considering sediment as a homogeneous material is to create another black box akin to the one associated with spatial lumping of the delivery-ratio concept. This will be considered in detail in Chapter 6. Extreme events occur even for a system in equilibrium and can greatly affect the storage term. The relative importance of extreme versus frequent geomorphic events has been well discussed (Wolman and Miller 1960) and will be addressed in Chapter 8. The difficulties associated with linking plot-scale measurements to basin sediment yield are clear. An effective approach uses multiple methods covering a wide variety of spatial and temporal scales. In this study, stream sediment sampling to determine stream sediment yield is the other dominant technique used to assess erosion. Stream sediment yield  Patterns of suspended-sediment transport in streams have long been studied to determine average areal denudation rates for spatial scales that are impractical for the erosion-plot technique (> 1 to 10 ha). The larger spatial scale expands the range of controls under consideration. Campbell and Bauder (1940) popularised the concept of a sediment rating curve to determine the characteristic suspended-sediment concentration for any stream discharge. They identified the straight-line  97 logarithmic relation between suspended load and stream discharge, C=aQ where C is suspendedb  sediment concentration, Q is stream discharge, b is the exponent of the relation (or "slope" on logarithmic coordinates) and a is the coefficient (or y-intercept on logarithmic coordinates). Walling (1974) pointed out that b is normally between 1.0 and 2.0, though smaller values have been reported (e.g., Loughran et al. 1986 found b = 0.68 for a 170-ha basin in New South Wales). This relation has been exploited extensively for determining totalfine-sedimentoutput from basins (e.g., Johnson 1942; Porterfield 1972; Singh and Durgunoglu 1992). Due to the distributed, stochastic nature of sediment production and delivery, the technique of regressing C on Q remains a common approach to estimating basin sediment yield (Shen and Li 1976). The sediment-rating curve represents the net effect of the interaction of sediment availability (supply) and its movement through a basin (transport) (van Sickle and Beschta 1983). Most basins studied in the literature are supply-limited resulting in suspended-sediment concentrations which often range across two orders of magnitude (Walling 1977a) - the combined event, seasonal, and spatial variation prevented Brown and Krygier (1971) from reaching general conclusions. Analysis of variance and multiple regression are frequently used to identify dominant controls (e.g., Walling 1974, McPherson 1975, Griffiths 1981). Unfortunately, these methods rest on the appropriate choice of independent variables and require that each variable be measurable and expressed quantitatively; in addition, the variables can be confounded (e.g., precipitation and discharge). Typical reported high values of annual suspended sediment yield range between 100 and 1 000 fknr 'yr for small basins (1 to 10 km) (e.g., Nordin 1963; Griffiths 1979; Doty et al. 1981; 2  1  2  Tropeano 1991). Tropeano (1991) also reported 5 200 t • km" • yr" for a 0.75 km basin in 2  1  2  northwestern Italy. Church et al. (1989) reported almost 20 000 t • km" • yr" for sub-basins (0.2 to 2  1  400 km) in China's Middle Yellow River basin. 2  Effective use of the rating-curve technique requires that an adequate number of events be sampled, covering a wide range of the controls on sediment yield. In small, especially mountainous  98 basins,flowand suspended-sediment can change rapidly making measurement difficult, increasing error (Walling 1977b), whereas in larger basins more controls can be operating increasing the variability of the net suspended-sediment response. Measurement error and variability in response must be considered when using sediment rating curves to calculate stream event, seasonal, and annual sediment yields (Walling 1977a; Walling and Webb 1981) as discussed in Chapter 8. These concerns are particularly problematic in Nepal (Rakoczi 1985). A variety of mechanisms controls sediment availability and its transport or retention in a basin. To improve the accuracy of the relation, rating curves can be stratified according to these controls. Ultimately, the relation is established byfiveprimary controls - geology, climate, hydrology, topography, and management (adapted from Griffiths 1981). However, to stratify results effectively, measurable parameters important to the controlling mechanisms are the most useful. Effective controls on suspended-sediment rating curves can be grouped into four categories: Hydrology  • • Surface  stream discharge storm-period variables (peak intensity, total rainfall) response  • • •  topography soil characteristics surface cover  •  antecedent soil-moisture conditions  Scale  • • •  spatial/temporal variability in rainfall/runoff travel time antecedentfloodhistory  Management  • •  modification of surface soil/cover structures  •  water diversion  These controls interact to affect the mobilisation, transport, and deposition of sediment, yielding a spatially- and temporally-heterogenous relation. Measurement difficulties and high variability can make it difficult to isolate the effects of some of these controls. Two common approaches are to  99 stratify by hydrograph stage (rising versus falling limb) and by season. Frequently, a much smaller number of controls change with either season or stage and by examining these stratified rating curves, information can be gained about the importance of certain controls. Most of the literature reviewed in the following discussion uses stratification by season to examine, through inference, the operation of the above controls. Hysteretic effects complicate analysis of sediment rating curves by desynchronising changes in sediment and discharge. Though this decoupling introduces difficulties in developing an accurate predictive relation, it also provides new information about upstream sediment dynamics which can assist in evaluating the controls. In this discussion, the general importance of hysteresis is indicated when it arises - a complete discussion of the mechanics and consequences of hysteresis for sediment rating curves is given in Chapter 6. Discharge is generally observed to be the strongest single control on suspended sediment. Higher flows possess a greater ability to carry sediment, especially the larger size classes. Discharge can also induce a change in riparian sediment supply. Sidle and Campbell (1985) suggested that high flows in a gravel-bed stream break the surface armour allowing the fines to be flushed out before rearmouring again at or near the peak flow. Paustian and Beschta (1979) studying forested basins in the Oregon Coast Range found that over 30% of the winter sediment output of gravel-bed streams was stored in the bed. Discharge-induced supply contributes to the well-known clockwise hysteretic behaviour described in Chapter 6. At higher flows, the stream is also able to access sediment supplies not frequentiy available. Sidle and Campbell (1985) observed a steeper sediment-discharge relation at higher flows - that is, a stronger relation of C to Q. These in-stream supply dynamics compelled Van Sickle and Beschta (1983) to propose the partitioning of supply among several compartments accessed at different levels of streamflow. Some authors prefer to use rainfall instead of discharge as the independent variable. Griffiths (1981) derived regional relations for large basins in South Island New Zealand concluding that mean  100 rainfall is the best predictor of suspended sediment output (6 basins of 4-100 km , 27 basins of 1002  1680 km). He reached similar conclusions for North Island basins (Griffiths 1982). Griffiths (1981) 2  noted the dominance of precipitation found in other climatic regions (semiarid, arid, subhumid, tropical). Walling (1974) used multivariate analysis to assess the effect on suspended sediment on an event basis of three rainfall parameters (15-minute intensity, total rainfall, storm kinetic energy) and found them to have a strong relation. Unfortunately, because discharge and precipitation are confounded in small basins, it is difficult with this approach to determine the relative importance of storm-period variables to suspended-sediment output. Burt (1989) stressed that although topography can be a strong control on suspended sediment, it is in the sub-basin scale where it is most important. In the headwaters, topography influences how fast runoff is concentrated into a powerful stream. In very small basins (scale depends on the climate), there is inadequate flow to be competent in transport. In very large basins, tributary inflows are desynchronised reducing the effect of topography. It is in the mid-sized tributary basins where topography can exercise the greatest effect on the flow regime. Tropeano (1991) echoed these ideas pointing out that basin lag time is also a function of basin shape and dimensions - lag time is therefore considerably shorter in headwater basins. Surface condition - expressed through surface cover and soil characteristics - is important in determining the quantity of sediment available for transport. Because most systems are supply-limited (van Sickle and Beschta 1983), this effect can be large. Tropeano (1991) found suspended sediment output from a 6.8-km basin in Northern Italy to be greatest in the summer resulting from recently2  ploughedfields.Walling (1974) also attributed a seasonal change in response in a British basin to a decrease in vegetative protection during winter. With bare ground, storm-period variables - especially short-term versus long-term rainfall intensity - become important to sediment production. Infiltration rate and soil moisture conditions affect runoff and its ability to entrain material. Surfaces with low permeability - e.g., built environments - encourage shorter times of concentration  101 resulting in higher peak flows and greater sediment production. Wood (1977) pointed out that long periods without erosion can cause fines to be in greater supply due to weathering and biological activity. He also suggested that desiccated surfaces may favour drying and crumbling which cause fines to be more readily entrained during the onset of precipitation. Conversely, fines may be harder to entrain when wet, especially if the surface is hydrophobic or clay-rich. Findings of Chapter 4 indicated that antecedent moisture conditions of surface soils during the pre-monsoon season should be drier than those of the monsoon season - the difference may contribute to seasonal erodibility of cultivated soils in the study basins. Spatial- and temporal-scale interactions modify the sediment response measured at a point. Scale influence has already been mentioned in terms of headwater hydrology. Variability in rainfall input and a heterogeneous surface response add further spatial-scale influences to the sedimentdischarge relation Q?orterfield 1972). This effect is lost in very small and very large basins (scale depending on the climate). Studies often find that total output is dominated by sediment production from a limited area of the basin (e.g., Tropeano 1991 found that 30-50% of total tributary input came from 8% of the land area, dominated by badlands). Also, Sidle and Campbell (1985) pointed out that the sediment response of small basins can be dominated by one source such as a landslide (e.g., Brown and Krygier 1971). Antecedent flood history can modify the supply regime. Many authors have noted seasonal, inter-event, and intra-event exhaustion patterns (Colby 1964; Arnborg et al. 1967; Walling and Teed 1971; Wood 1977; Beschta 1978; Walling and Webb 1982). Beschta (1978) observed the seasonal shift after the annual peak flow had occurred. Walling and Webb (1982) found sediment output to increase strongly with "recovery period" (defined as the time between successive events). Wood (1977) found that sediment exhaustion within individual events depends upon the length and severity of the event. Arnborg et al. (1967 in Wood 1977) suggested that the entrainment of sediment deposited on the bed during the recession stage of a previous flood also results in this differential  102 supply regime. In large basins, the different travel times of water and sediment can modify the sedimentdischarge relation. This spatial-scale effect results in a counterclockwise hysteresis loop, in contrast to the clockwise loop commonly found from the effects of supply exhaustion. Watershed management can reduce or enhance the effects of most of the controls described above. Anderson (1981) found that landuse variables accounted for 30% of the variance in sediment rating curves for 61 basins in California. Management comes in many forms; the two of most interest to the present study are structures and modification of surface conditions. Roads are one of the most familiar modifications made to a managed watershed. In British Columbia and the USA Pacific Northwest, the increase in sediment production resulting from construction of forest access roads is well documented (Brown and Krygier 1971; Megahan and Kidd 1972; Beschta 1978; Reid et al. 1981; Reid and Dunne 1984; Anderson and Potts 1987). Other structures such as reservoirs and diversion dams can also strongly modify the downstream sedimentdischarge relation. In agricultural basins, perhaps the greatest effect from management is the surface modification brought about by cultivation and other agricultural activities. Loughran (1986) studied a 1.7-km  2  managed basin in New South Wales, Australia (forestry/agriculture) and found that 93% of sediment output derivedfromvineyards (60% of the area) and only 7% came from the forested land (30% of the area). Using erosion plots, he was able to attribute the difference to the absence of a surface cover on cultivated fields. Though outputfromsloping agriculturalfieldscan be high, it is important to remember that typically only one-third to one-half of the amount erodedfromthe surface actually leaves the basin (Walling 1983). For example, Imeson (1974) examined an 18.9-km basin in England 2  and found that because only a third of the soil erodedfromthe bare agriculturalfieldsactually left the basin, the river channels were an important sediment source accounting for over 35% of the total basin output. Presumably, in steepland agricultural basins like those in Nepal, the effect of surface-  103 cover modification is pronounced. This is discussed further in the following section.  5.2.2 Quantitative Himalayan data Himalayan soil-erosion research has evolved from brief and superficial assessments of the dominant causes of erosion, to measurements of the rate of erosion over limited spatial and temporal scales, to the current focus on combined measurements of precipitation and landuse variables important to erosion and sediment transport. Several good overviews exist (Carson 1985, Ramsay 1986; Bruijnzeel and Bremmer 1989). This review focuses on studies specifically designed to measure erosion rates in the Middle Mountains and other research designed to explain the causes. Early research on erosion in the Himalaya sought to explain the causes of observed erosion using inference from very limited data. In particular, research was focused on determining whether erosion was natural in origin or caused primarily by farming activities. Laban (1977) estimated erosion due to mass movements by counting landslides from an aircraft and noting their association with land use. His results led him to conclude that about 75% were from natural causes and 25% were induced by human activity. Using a geomorphological analysis initiated as a result of catastrophic rainfall during October 1968 (500 mm in 24 hours), Starkel (1972) estimated surface lowering in the Darjeeling Himalaya to be of the order of 60 to 70 t-ha" - yr" over the prior century. 1  1  Mass movements resulting from extreme rainfall events dominated this rate of surface denudation, and there was not a direct relation between the amount of mass wasting on forested versus nonforested slopes. According to Carson (1985), both Sastray and Narayana (1984) and Winiger (1983) found that terracing and related farming practices have a stabilising influence on steep slopes as long as farming is economic; otherwise, slope degradation can occur at a greatly accelerated rate until natural rates resume. Attempts at measuring rates of erosion in the Himalayan region have been based largely on erosion plots and stream sediment yields as summarised in Tables 5.1 and 5.2 respectively. The  104 Table 5.1  Surface erosion rates as determined by field studies using erosion plots within or near the Middle Mountains.  Source  Details of Study  Erosion Rate t • ha" • yr"  Scale (m) 2  Erosion Plots Laban (1978)  Mulder (1978)  DSCWM (1991) Sherchan et al. (1991) Upadhaya et al. (1991) Ries (1994) Overseas Development Agency (1995)  Note:  1  Middle Mountains n/a grassland overused grassland seriously eroded, gullied Kathmandu Valley 10 densely forested well-managed pasture steep, overgrazed Shivapuri 15 terraced; cultivated; mulched/nonmulched; steep Pakhribas 18 cultivated terrace; (various treatments) Kulekhani 90 terraced; cultivated; 5% and 10% slopes; 14 High Mountains traditional cultivations traditional cultivations 76-536 degraded shrub 25-95 degraded forest 71-85 grassland 30-69 - based on extrapolations from partial sampling during  1  10-20 20-50 200-500 0.34 9 35 6-32 18-35 0.8-7 1-9 3-13 6-22 0-19 <1 1992/3.  Bruijnzeel and Bremmer (1989) summarise other early Himalayan measurements.  results from erosion plots indicate that at the spatial scale of less than 100 m , annual erosion rates 2  vary on cultivated land and on non-gullied shrub/forest land from 1 to 35 t/ha. The low annual rates reported by Upadhaya et al. (1991) at the DSCWM Kulekhani site (most fall between 1 and 3 t/ha) may be due to measurement discontinuities and the gentle slopes of the plots. Laban (1978) summarised his measurements and a variety of earlier field measurements concluding that 10 to 20 t/ha was a reasonable estimate of annual surface erosion for well-managed cultivated or grazing land, rising to 20 to 50 t/ha if overused and increasing locally to 200 to 500 t/ha if seriously eroded and  105 Table 5.2  Surface erosion rates as determined by field studies using check dams and hydrometric stations.  Scale (km) Accumulation behind Check Dams (all sites within Middle Mountains) 0.13-0.25 Laban (1978) overgrazed grassland gullied, overgrazed grassland - parallel-dipping phyllitic schists - 30% trap efficiency - 70% SDR 0.18 overgrazed scrubland Laban (1978) severely gullied 0.11-0.19 - weakly consolidated granites and migmatites - 50% trap efficiency -100% SDR 0.11-0.15 Laban (1978) degraded, gullied forest - Mahabarhat Lekh; very steep - metamorphic/sedimentary rocks Basin Sediment Yield Ries (1994) High Mountains: Bamti 0.08 Chhukarpo Low 0.24 Chhukarpo Middle 2.7 Chhukarpo Middle 3.7 Surma 5.7 Middle Mountains 6-585 Kandel (1978) 6 basins; mix of forested, cultivated, degraded land 585 Sharma (1977) Bagmati Trisuli 4110 Karnali 42 890 > 5000 some major Nepalese rivers Ramsay (1986) Williams (1977) Tamur 5770 in Carson (1985) Aran 34 525 Sunkosi 18 985 Saptakosi 59 280 Source  Details of Study  2  Note:  Some values based on conversion of 1 mm =13 t/ha  Erosion Rate t • ha • yr -1  1  22 29  43 125-570  63-420  13 30 7.5 3.7 4 3-46 (excluding premonsoon) 46 19 51 10-70 38 7.6 21 15  106 gullied. Most denudation estimatesfromsuspended-sediment data are for large rivers (5 000 to 50 000 km) and generally rangefrom10 to 70 t-ha" 'yr . Agricultural erosion rates averaged over larger 2  1  1  scales typically result in lower rates due to deposition within the basin. At the large scales assessed in the studies which are summarised in Table 5.2, myriad sediment sources - especially those associated with riparian erosion (Bruijnzeel and Bremmer 1989) - are active making extrapolation to field-scale erosion extremely difficult and uncertain (Ramsay 1986). The sediment sources which contribute to the denudation rates calculated using river measurements are different than those under far narrower consideration in the plot studies. In the High Mountains, Ries (1994) measured annual sediment yield from small basins (0.1 to 6 km) of 4 to 30 t/ha; these rates may be more representative of those 2  expected for agricultural basins however this work was carried out in a physiographic region very different to the Middle Mountains. Many of these studies, unfortunately, are based on significant assumptions in extrapolating their results through space and time. For instance, although many researchers have contended that surface erosion rates are highest in the pre-monsoon period (Overseas Development Agency 1995; Ramsay 1986; Bruijnzeel and Bremmer 1989), data are rarely available to defend this assertion. Impat (1981) found that soil loss was greatest at the beginning of the measurement period in June despite precipitation peaking in August (in Ramsay 1986). Overseas Development Agency (1995) reported that 45-60% of measured soil loss occurred in May (part of the pre-monsoon period) but did not provide details on differences between monsoon and pre-monsoon sampling coverage. The spatial scales of most of these measurements are either very large (> 5 000 km) or very small (< 1 km) 2  2  processes and sediment sources important in the intermediate scales are rarely evaluated. Several studies have recently tried to integrate specific measurements of surface erosion with precipitation and landuse measurables over contrasting spatial and temporal scales to reach more meaningful conclusions about catchment-scale processes. Ries (1994) (Table 5.2) provided perhaps the  107 best example:field-scaleerosion rates are contrasted with basin sediment yields and rainfall delivery in the High Mountains over consistent spatial and temporal scales. Perino (1993) documented a paired-catchment approach for basins of 2 hectares in size within the Phewa Tal and Kulekhani basins in the Middle Mountains. The complete capture of all runoff (including bedload) in a large structure with detailed hydrometeorologic measurements is combined with manipulation for conservation practices. The low basin output of thefirstyear (1992) - less than 1 t/ha - was attributed to exceptionally low rainfall in the study area. Overseas Development Agency (1995) (Table 5.1) provided detailed measurements of surface cover and rainfall storm-period variables and concluded that the lack of surface cover in May of the pre-monsoon season is a strong control onfield-scalesoil loss; unfortunately, they failed to relate this to suspended sediment dynamics at larger basin scales. Despite these recent catchment-scale research efforts, there remain few quantitative data describing erosion in the Middle Mountains over spatial and temporal scales of importance to farming activities and their relation to downstream sediment transport. Specific parameters crucial to understanding the erosion process remain conspicuously absent. These include high-flow sediment samples, rainfall intensity and distribution measurements, and the measurements of sediment storage to address within-catchment variability. Further, measurements remain dominated by monsoon observations, ignoring the pre-monsoon season when it is widely believed transport rates are the highest.  5.3 Surface erosion on cultivated rainfed uplands  Five erosion plots have been monitored throughout 1992-1994 on steep, high-elevation, rainfed, cultivated fields. In this section, the factors which shape surface erosion at this scale are examined. Event, annual and seasonal erosion rates are determined. Event analyses provide insight into the mechanics of the surface-erosion process and strengthen the conclusions from the integrated annual and seasonal analyses.  108 5.3.1 Controlling factors  The factors which determine surface erosionfromthe erosion plots include topography (slope length and steepness), soil characteristics, storm-period variables, and management. In the Jhikhu River basin, management is pervasive, greatly influencing all controlling factors except rainfall characteristics. For instance, management affects the rate of surface erosion through cropping practices and soil characteristics such as the organic-matter content. The topography of the cultivated fields is greatly modified by management through the construction of terraces. The Middle Mountain farmers who cultivate the steeplands have developed a long-standing tradition of terraced agriculture which works to minimise the negative effects for cultivation of the steep topography.  5.3.2 Erosion plots: annual regimes  Table 5.3 presents a summary of the annual rate of soil loss from each of the five erosion plots during 1992-1994. Management practices were held constant across the plots and were characteristic of prevailing management on the rainfed cultivated fields (section 3.1.3). Similarly, the slope of the plots (22'-30°) is typical of much of these cultivated uplands. The results suggest a tremendous range of surface-erosion in these upland fieldsfromalmost none to rates exceeding 40 t'ha" •yr". 1  1  Table 5.3  Annual rate of soil loss (tonnes/ha)fromall plots, 1992-1994. Plot 1 2 3 4 5  1992 18 23 38 0.1 0.1  1993 4.1 34 37 0.2 0.3  1994 42 6.4 6.9 2.9 2.6  109 The significant plot-to-plot variation in soil erosion is largely a reflection of the huge variation in soil properties across the five plots. In particular, infiltration rate as affected by the texture of the surface soil is a controlling influence on the nature of storm runoff and therefore on the rate of erosion. Soil loss from plots 1, 2, and 3 is over an order of magnitude greater than that from plots 4 and 5 (see also Figure 5.3). Table 5.4 contrasts coarse-fragment content, fine-fraction texture, and infiltration rate for the five plots. Plots 1,2, and 3 have a much greater tendency for overland flow as a result of the fine surface texture and the consequent reduction in surface infiltrability. In contrast, the surface (and subsurface) of plots 4 and 5 are highly porous, resulting in infrequent overland flow. Although there are differences in clay content of the surface soils between the plots, it is the much larger difference in coarse-fragment content which is the first order effect and sets apart the erosion rates at the two sets of plots. Plots 1 and 2/3 also differ in their responses largely due to their surface soil characteristics. (Plots 2 and 3 have the same physical characteristics because they are adjacent - see Chapter 3.) Generally, Plot 1 yields less erosion than Plots 2/3. This may be due to the inherent erodibility of  Table 5.4 Plot No.  Surface-soil characteristics of erosion plots. Surface Horizon (#1) (about 0-15 cm) Texture Coarse of Fine Fragment Fraction Content (S/Si/C) (%)  Sub-surface Horizon (#2) (about 15-50 cm) Coarse Texture Fragment of Fine Content Fraction (%) (S/Si/C)  Surface Infiltration Rate Initial (10 min)  Final (>4hr)  cm/hr  1 1.7 44/35/21 1.1 40/32/28 42 32 2 4.2 37/34/29 3.4 47 37/33/30 16 3 4.2 47 37/34/29 3.4 16 37/33/30 4 45.8 52/37/11 68 20 69.3 53/36/11 5 67.3 39/42/19 94 39.8 34/40/26 51 Fine fraction consists of particles < 2 mm in size. Coarse fragments include large gravels. S=sand; Si-silt; C=clay.  110 these two soils: plots 2/3 contain red soils which are highly weathered and of low carbon content (Shah et al. 1994) and lack aggregate stability (though direct measurements are unavailable). In contrast, the soil of Plot 1 is better aggregated to resist surface erosion. Also, infiltrability is somewhat higher in Plot 1 than in Plots 2/3 (see Table 5.4) which might further discourage overland flow in Plot 1. The reversal in erosion rates between 1993 and 1994 at these plots was driven largely by storm patterns (not by soil properties) especially in relation to surface cover. Table 5.4 also shows a large inter-annual variability of erosionfromeach plot of an order of magnitude. This variability can be investigated further by examining seasonal changes in erosion rate, the subject of the next section.  5.3.3 Erosion plots: seasonal regimes Within each plot there is also a large intra-annual variability as shown in Figure 5.1. Table 5.5 shows the percentage of each plot's annual erosion which occurs during the pre-monsoon, transition, and monsoon seasons. In most years and at every plot, more than half of the annual erosion occurs in the pre-monsoon season. In fact, often over 80% occurs in the combined premonsoon and transition seasons, yet only 45% of the annual precipitation occurs during this period. The presence of swelling clays in Plot 1 might explain why a greater proportion of the annual erosion  Table 5.5  Plot No. 1 2 3 4 5  Percentage of each plot's annual erosion occurring in the pre-monsoon, transition, and monsoon season, 1992-1994. pre-monsoon season 92 1 82 60 37 23  93 15 100 100 62 74  94 31 98 96 100 100  transition season 92 64 17 39 50 8  93 10 0 0 6 4  94 57 1 1 0 0  monsoon season 92 35 1 1 13 69  93 75 0 0 30 22  94 12 1 3 0 0  Ill  from this plot occurs during the monsoon season (see the end of section 5.3.4). When the rains first start to arrive in the late spring, in the pre-monsoon season, the land surface is desiccated and the upland cultivated fields bare and vulnerable to erosion. By mid-July, the monsoon season is well under way, normally bringing regular rainfall. The first few weeks of July show erosion-regime behaviour which intergrades between that of the pre-monsoon and monsoon seasons and, in this study, is called the transition season (see section 5.4.2 for more details). The major seasonal change in surface condition which occurs at the plot is a change in surface cover as weed growth and the summer crop develop. In 1994, measurements were made of average maize stalk height and leaf length to document the seasonal development of the crop within the five erosion plots. The results shown in Figure 5.2 indicate a close correspondence between these basins' erosion regimes and the development of surface cover in the plots especially when compared to each plot's chronological erosion history (Figure 5.1). The farmers also frequently intercrop providing a further protection of the soil from intense rainfall. Changes in vegetative cover also explain the large inter-annual variation in soil loss at a given plot. Surface cover does not develop immediately but requires rainfall to get started and then takes several weeks to be complete. If damaging rains occur when the surface cover is only partially complete - which is common - then significant losses are likely at the plot in a single event. If the fields have recently been weeded (typically once in either June or July) then further losses can result. Subsequent rains falling on the same ground with a strong vegetative cover (monsoon season) often cause almost negligible surface erosion. As a result, loss from single events dominates annual soil loss. Table 5.6 shows the percentage of each plot's annual soil loss that occurred in the two mostdamaging events at each plot and their dates of occurrence. At all plots and in almost all years, over 50% of the annual total occurs in only two events. At plots 2 and 3, the rate is over 80%. These two events typically occur during the pre-monsoon and transition seasons.  112 Figure 5.1  Soil loss from the erosion plots on an event basis, 1992-1994.  40 10 1 0.1 0.01 0.001  Note: Scale on ordinate for plots 1 through 3 differs from that of plots 4 and 5. Seasons D = Dry P = Pre-monsoon T = Transition M = Monsoon M +  1992  ^  .2 O  "M  D 3  Plotl +* + 1  1994  D  M  1993 T T  I  i -  M  M  D  Plot 3  +  1994 D  + + +  a  \  1994  1993  1992 D  M  M  D  Plot 4  0.1 0.01  pm  +  4«  o.Ol 0.001 — O  "M"  "TT  Plot 2  1992 40  M  1993  40 IT 10 H 1 1 ^ 0.01 0.001  £  D  D  ~W  +  + +  +  +  +  ++  --  +  +  0.001  "M  D  D  p + "ST  P T  Plot 5  o.i o.oi -  1994  1993  1992  +  + +  o.ooi  •IIIIIII  1992  1993  1994  1995  113 Figure 5.2  Average maximum maize height (a) and maize leaf length (b) at all erosion plots in 1994.  May  Jun  Jul  Aug  Sep  Oct  114 Table 5.6  The percentage of the total annual erosion at each plot which occurred in the two most-damaging events of each year, 1992-1994.  Plot No. 1 2 3 4 5  1992 55 (10/7; 15/8) 96 (10/6; 10/7) 78 (10/6; 10/7) 69 (9/7; 10/6) 60 (21/7; 11/6)  1993 50 (21/7;22/7) 88 (17/5;27/5) 91 (17/5;27/5) 40 (21/4; 10/8) 57 (21/4; 10/8)  1994 44 (9/7;2/7) 87 (9/5;30/5) 82 (9/5;30/5) 60 (19/6; 15/6) 67 (15/6;17/6)  Note: Dates (day/month) represent when the two events occurred. 5.3.4 Erosion plots: event regimes  In Chapter 4, it was shown that rainfall characteristics of storms in the pre-monsoon season differ - but not greatly - from those of the monsoon and transition seasons. In section 5.3.2, it was suggested that soil texture strongly influences the likelihood of runoff to occur (and hence surface erosion) in any season. It was determined that under conditions when storm runoff occurs, it is vegetative cover that is the dominant seasonal control on surface erosion, with other factors such as soil aggregate stability exercising a non-seasonal control. But to what extent do differences in stormperiod variables, and particularly rainfall intensity, contribute to the observed seasonally-variable surface erosion at the plots? Figure 5.3 shows the influence of peak 10-minute storm rainfall intensity on soil loss at the five erosion plots for each event of the three-year period. The seasonal differences observed in the last section are reinforced here and the effect of rainfall intensity is revealed. There appears to be a threshold of about 30 mm/hr below which storms are non-erosive at the plot scale. Above this value, significant surface erosion can occur in the pre-monsoon and transition seasons, increasing with peak rainfall intensity. A threshold of this order was described by Hudson (1981) based on African data.  115 Figure 5.3  Id  The effect of maximum 10-minute rainfall intensity (I ) on soil loss at all erosion plots, seasonally stratified for all events of 1992-1994. 10  25 ~}  (a) Erosion Plot 1  o  Pre-Monsoon  A  Transition  •  O A  §20  O  Monsoon  Dashed line indicates 30 mm/h.  ~l—  0 40 80 120 160 Max 10-Min Intensity (mm/h) 25  25 -j (b) Erosion 0 Plot 2 § 2 0 ~_ CA t*> .  r-  O  -I  I"  § 2 0 -_  (c) Erosion Plot 3  O  CA CA .,  1  "3  o  s5  j80pea120i 160 •  o  0  1.0  1.0  CA  o0.6  §0.8 CA  O  O  H  o0.6  — i J  (e) Erosion Plot 5  H  §0.4 H  §0.4 H lo.2 H rS0.0  T  0 40 80 120 160 Max 10-Min Intensity (mm/h)  0 40 Max 10-Min Intensity (mm/h) (d) Erosion Plot 4 §0.8 -  o  IO  o  0 40 80 120 160 Max 10-Min Intensity (mm/h)  o  §0.2  O  -D  O  o.o JJamflBmaui.  O  • fn jni  A  | • 11 •  0 40 80 120 160 Max 10-Min Intensity (mm/h)  116 High-intensity rainfall clearly occurs at all plots in all seasons yet it is rarely of concern at this scale during the monsoon season. This observation is consistent with the conclusion that vegetative cover is a dominant seasonal control on surface erosion at the plot scale. Above the threshold, a relation between I and soil loss is of limited use because of the lack of data with respect to the tremendous 10  variability in soil loss resultingfromthe timing of heavy rainfall relative to the development of a vegetative cover. Consistent with the limited spatial scale of the erosion plots and the resulting minor extent of channelised flows within the plots, soil loss does not correlate with rainfall intensity over significantly longer periods than that of I (including R )10  T0T  Figure 5.4 relates event runoff coefficient (CR) to the corresponding I . C is defined as the 10  R  percentage of incoming event rainfall which runs off the plot. It tends to be the highest during the pre-monsoon season (declining in value through the transition season) at all plots except Plot 1. With saturated soils often prevailing in the monsoon season, one might expect higher C during this period R  but the opposite is generally observed. This behaviour might be due to a change in dominant runoff mechanism with season for significant erosion events. Figure 5.5 shows seasonal variation in the influence of C on the soil loss at the five erosion R  plots. The greater C , the more likely are overland flow and surface erosion to occur. There appears R  to be a threshold in C below which erosion does not occur, though it is inconsistent across the plots. R  Above some threshold of between 5 and 10% (depending on soil characteristics), erosion can be serious in the pre-monsoon and transition seasons. Permeability also influences the number of events that occur at each plot and is reflected in C . About half as many runoff events occur at Plots 4 and 5 R  as occur at Plot 1, with Plots 2 and 3 intermediate to these. This is a direct result of the coarsefragment texture of the surface soils, influencing the propensity for erosion to be able to occur on a given high-elevation cultivated field. Although within-plot observations during rain events are few in number, the presence or absence of rills combined with an examination of the pattern of rainfall for specific events suggests  117 Figure 5.4  The effect of maximum 10-minute rainfall intensity (I ) on event runoff coefficient (CR) at all erosion plots, seasonally stratified for all events of 1992-1994. 10  (a) Erosion Plotl  o  Pre-Monsoon  A  Transition  •  Monsoon  Runoff is higher than indicated for solid symbols because drums overflowed. 0 40 80 120 160 Max 10-Min Intensity (mm/h)  0 40 80 120 160 Max 10-Min Intensity (mm/h)  0 40 80 120 160 Max 10-Min Intensity (mm/h) 40 (e) Erosion Plot 5  S10  -J£ [ T i l l  0 50 100 150 Max 10-Min Intensity (mm/h)  0 50 100 150 Max 10-Min Intensity (mm/h)  118 Figure 5.5  25  Relation between runoff coefficient (CR) and event soil loss at all erosion plots, seasonally stratified for all events of 1992-1994. -3  (a) Erosion Plotl  §20  I  o  Pre-Monsoon  A  Transition  •  10  > 5  Runoff is higher than indicated for darkened symbols because drums overflowed. Dashed line indicates 10 %.  o  0  1  0  —r  10 20 30 40 Event Runoff (%)  25 §20  (b) Erosion Plot 2  f p  110 •  0  1  *  §20  I  °  o 1  10 ~.  > 5  ° T  0  25  (c) Erosion Plot 3  60  o  S 5  Monsoon  1 — 1  10 20 30 40 Event Runoff (%)  0 10 20 30 40 50 60 70 Event Runoff (%)  —  o  A  A  -  0  o o  D  0  10 20 30 40 Event Runoff (%)  0 10 20 30 40 50 60 70 Event Runoff (%)  119 runoff mechanisms for different storm types. Those storms which bring heavy rainfall on dry ground within the first few minutes (during the pre-monsoon season in particular - see Chapter 4) resulted in Hortonian overland flow. Storms bringing high total rainfall initiated saturation overland flow in those instances when a rate of erosion was measurable; brief periods of Hortonian overland flow when the rainfall intensity was sufficiently high may also have been possible during those storms. In Chapter 4, it was concluded that the most significant seasonal differences in rainfall regimes involved total storm rainfall, the period of time before a storm without rain, and the within-storm timing of storm rainfall: pre-monsoon storms are more likely to bring heavy rain on dry ground than are their monsoon counterparts. This is very important to surface erosion and may be a major factor contributing to the high values of C during the pre-monsoon season. R  There are several possible explanations for the more-frequent runoff at Plot 1 as compared to the other plots. The farmer who monitors this plot tended to sample all the small (minor) events which were not sampled at Plots 2 and 3 because they are insignificant in the soil-loss budget. It is also possible that the soil in Plot 1 possesses a greater aggregate stability and can sustain greater overland flow before being entrained. This is consistent with farmers' comments on this soil using the indigenous classification; it is a clayey soil with high productivity and different behaviour as discussed further in Chapter 7. A final possibility is that the soil of Plot 1 may contain swelling clays which increase cohesion, discouraging entrainment. This is an area of potential research on these cultivated steeplands. If so, this could explain why Plot 1 shows its highest percentage runoff during the monsoon season (when the soil is frequently saturated) instead of during the pre-monsoon season as it occurs at the other four plots.  5.4 Stream sediment regimes  The findings of the last section are expanded by examining sediment regimes in the streams of the detailed study basins especially those nested around erosion plots 2 and 3. As the spatial  dimension under consideration grows larger, so does the variety of sediment sources and mechanisms of erosion. The size of these basins varies between 1 and 100 km - a range in spatial scale that has 2  been examined rarely in the Middle Mountains. Five hydrometric stations in the Jhikhu River basin have been monitored during most of the 1992-1994 study period and two additional stations during 1993-1994. Suspended-sediment sampling and stream-flow measurements were carried out at five of these sites during flood events at all times of the rainy season. The relation between suspended-sediment concentration and discharge forms the basis of the sediment-rating-curve technique. Discharge is determined from a relation with gauge height as explained in section 3.3.1. Observed patterns in suspended-sediment transport can be explained by linking these patterns to storm-period variables, inherent characteristics of the basins, and landuse activities.  5.4.1 Controlling factors A wide range of factors is responsible for shaping a basin's sediment-discharge relation. Rainfall characteristics of importance include peak intensity, spatial variability, and total storm rainfall. Soil-surface condition is important because it directly affects the entrainment process. Management activities can shape both the susceptibility of the surface to erosion and the rainfallrunoff process itself. Soil moisture, antecedent flood history, sediment storage, and sediment exhaustion can also be important factors as discussed in section 5.2.1. The next section characterises and contrasts the prevailing suspended-sediment regimes of all study basins.  5.4.2 Seasonal regimes The entire suspended-sediment data set consists of 2287 high-flow samples and 820 low-flow samples taken during 1992-1994 at seven hydrometric stations. For each sample taken, the measured sediment concentration is coupled with the corresponding discharge and presented in its station's C-Q  121 Figure 5.6  Sediment rating curves for monitored hydrometric stations based on entire data set (19921994): (a) Kukhuri, (b) Upper Andheri, (c) Lower Andheri, (d) Dhap, and (e) Jhikhu.  (d) Dhap; St. 3 lli£b>o ET^ O  •  0 01 ^  io  0.1 13 Discharge (m /sec)  ~•  -  3 —  Discharge (m /sec)  1000  o  • A  .01  0.1 1 10 100 Discharge (m /sec) 3  1992 1993 1994  100  122 graph. Figure 5.6 presents the C-Q distributions for the major hydrometric stations located within the Jhikhu basin stratified only by the year in which the samples were taken. In particular, Figures 5.6a and 5.6b present the results for the steep, headwater catchments (stations 10 and 9 respectively) within the Andheri basin. Figures 5.6c and 5.6d show the data for the Lower Andheri and Dhap catchments (stations 2 and 3 respectively), catchments of similar area but of strongly contrasting topography. Finally, the data for the entire Jhikhu basin (station 1) are found in Figure 5.6e. Stagedischarge relations are not available for the mid-reach Andheri basins (stations 11 and 12) so these distributions cannot be presented. These unstratified C-Q graphs reveal the general nature of suspended sediment transport in these basins. The small size and steepness of Kukhuri basin is reflected in steep C-Q behaviour. As basin scale increases, the distributions shift toward higher discharge values and weaker relations with Q. The graphs in Figure 5.6 show that the data derive from a wide cross-section of the period of study (1992-1994) and hence can be expected to reflect a wide variety of controls present during that period. A comparison of sediment transport in these basins is made possible by developing relations based on functional analysis (Kendall and Stewart 1979) for each station. Section 5.3 demonstrated a strong seasonal discontinuity in the rate of soil loss at the plot scale. This knowledge is used to improve the relations by applying an initial stratification by season before applying functional analysis. Seasons important to erosion and sediment transport need definition. Elevated levels of suspended sediment were readily observed at the beginning of the pre-monsoon season with lower values resulting during the later part of the rainy season. Where is it appropriate to define the extent of each season? To determine the end of the pre-monsoon season, sediment data were added incrementally to those of the beginning of the rainy period (generally early June). High sediment concentrations were maintained until the end of June then started to decline indicating the beginning  123 of a transition season. The same process was carried out from the end of the rainy period adding incrementally sediment data earlier in the monsoon. Levels remained low until those before July 20 were included and this pattern was present at all five hydrometric stations. As a result of these sensitivity analyses, the pre-monsoon season is defined to last until and including the last day of June, the monsoon seasonfromJuly 20 to the end of the rainy season, and the transition season (showing intergrade behaviour)fromJuly 1 to 19 inclusive. The concept of a floating rating curve has also been considered to model the change in suspended sediment concentration through the rainy season. This approach would assume specific (daily) rates of decline in suspended sediment concentration through the entire rainy season. In this situation, the rate(s) would reflect changing supplies due to increased vegetative cover during the premonsoon season and perhaps an additional supply-exhaustion effect during the monsoon season. Though such a model may provide a valid simulation for these basins, synoptic-scale variation in suspended sediment concentration within the data available obscures the observation of a floating rating-curve pattern. For instance, samples taken during May at the Upper Andheri station reveal a heightened response during the early stage of the pre-monsoon season. Unfortunately, storms and floods are lessfrequentat this time than in June and too few samples are available at any station (including Upper Andheri) for adequate development of a floating rating curve model for the premonsoon season. In addition, a systematic decline in suspended sediment concentration within either the transition or monsoon seasons was not evident; only a September drop at Lower Andheri appeared defensible. This remains a useful area for further research C-Q relations derived from these data are used both to compare the sensitivity of the basins with respect to the operating controls and to predict C values for given Q values. The comparison demands functional analysis whereas the prediction requires simple log-linear regression. Mark and Church (1977) have provided equations for the computation of the functional relation in terms of the marginal regression derived for prediction. Because both are required here (prediction - Chapter 8),  124 their equations are used and both relations presented. The results of simple log-linear regression are expressed in terms of C = aQ and are given in b  Table 5.7 for each of stations 10, 9, 2, 3, and 1. The correlation coefficient (R) and the standard 2  error (s) of the regression are provided in terms of the transformed Oog ) values. These values are r  10  used in Chapter 8 for predictive purposes and a standard bias correction is applied in that application to better predict the influential high-flow estimates (Miller 1984). Table 5.7  Sediment-rating-curve relations based on seasonally-stratified data using log-linear regression (1992-1994, assuming Q known without error) excluding data from the transition season.  Station Kukhuri  No. 10  Upper Andheri  9  Lower Andheri  2  Dhap  3  Jhikhu  1  Season P M P M P M P M P M  a, 36.13 16.91 30.33 8.27 12.20 3.15 11.53 8.24 1.44 0.45  b, 1.431 1.520 0.385 0.583 0.586 0.740 0.284 0.429 0.872 0.950  n 45 110 31 75 150 243 62 128 201 489  R 0.397 0.604 0.225 0.390 0.592 0.678 0.402 0.218 0.546 0.688 2  0.413 0.199 0.170 0.218 0.0913 0.0911 0.0312 0.122 0.0937 0.0819  Rating curves based on C = aQ power law relation; n=number of samples; R=correlation coefficient; s=standard error of the regression (log, g/1); P - pre-monsoon season; T - transition season; M - monsoon season b  2  r  0  To determine the functional relations for each station, it is necessary to estimate the ratio of the error variances of C and Q (X = E^/EQ ) as explained by Mark and Church (1977). The 2  calculation of E  2 Q  is straightforward using the standard error of the regression for the stage discharge  relations presented in Appendix A5. A geometrical approach is followed to determine E . It is 2  C  assumed that any additional departures in the C-Q relation between the measured and expected C values (beyond that attributed to Q) can be attributed to real variance in C. Hence by geometry, E  2 C  = E [dC] = E [(C 2  mea8  -C  - b(dQ)]  2  exp  r  125 where C  is the actual measured suspended sediment concentration (g/1), C  mau  exp  is the expected value  based on the marginal regression, b isfromthe marginal regression, and dQ is the error (m/s) 3  r  associated with the measurement of discharge. Using this approach, all synoptic variability in suspended sediment concentration including hysteresis is attributed to the C variate. This is appropriate since adjustments for these synoptic effects are typically unavailable. Following Mark and Church (1977), this relation is used to calculate values of b for each station/season combination: f  b = {(b /R -X)-tV[(b /R -X) +4Xb ]}/2b 2  f  r  2  2  2  2  2  r  r  r  where R isfromthe marginal regression (Table 5.7). Church and Mark (1980, p. 385 and erratum) 2  provide confidence limits for b in terms of the one-tailed Student's t for n-2 degrees of freedom: f  X^tanJtan-^fX-^ + ^sin-^ All calculations are carried out here using log transformed units. 10  Figure 5.7 illustrates the results of these calculations (for the expected relations) and the equations of the resulting lines appear in Table 5.8. This table indicates that the error variance associated with the measurement of C is between 2 and 19 times greater than that of Q. A small set of replicate samples (21 pairs) taken at Upper and Lower Andheri, Dhap, and Jhikhu stations under conditions representative of high flow has a small associated error variance in comparison to the values calculated to yield Table 5.8 (0.0019 versus a range of 0.087-0.45). This contrast suggests that the unstructured variance in C is due largely to the effect of synoptic controls on suspended sediment operating within seasonal scales with a minimal contribution derived from random fluctuations during sampling. The estimated error associated with Q is at times larger than the one suggested by the standard error of the regression for the stage discharge relations. At Dhap station (3), the stage discharge relation is not developed over the highest flows putting a larger error expectation on the high-flow values calculated through extrapolation. In addition, bed control difficulties at Upper Andheri station (9) and some uncertainty in reading the gauge at the highest flows at Jhikhu (1) and  126  Figure 5.7  Seasonally-stratified sediment rating curves based on all entire data set: (a) Kukhuri, (b) Upper Andheri, (c) Lower Andheri, (d) Dhap, and (e) Jhikhu.  Uiooo  (a) Kukhuri; St. 10  "3>  IOOO  e o  (b) Upper Andheri; St. 9  100 h  c  o  10  c  0  Oi.ri  U  +J  c  -  1  •l-H  0.1  <L>  C/3  0.01  0.1 10 100 Discharge (m /sec)  Uiooo  • HD  0.01  1 10 100 0.1 Discharge (m /sec) 3  §1000 (d) Dhap; St. 3 fl 100  I J-l  fl fl-  10  o U  e  0.01  0.01  D  - D M  •  §  H o.i  "8 o.i 00  l  Kin  0.1 Discharge (m /sec)  0.1 Discharge (m /sec)  C/3  0.01  0.1 1 10 100 Discharge (m /sec)  ••  3  Pre-Monsoon Monsoon  127 Table 5.8  Sediment rating-curve relations derived using functional analysis.  Station  Season  Name  No.  Kukhuri  10  Upper Andheri  9  Lower Andheri  2  Dhap  3  Jhikhu  1  X  b,  %  expected  range  expected  range  P  18.47  1.68  1.28-2.11  49.5  29.9-84.5  M  10.44  1.72  1.55-1.90  21.9  17.6-27.5  P  12.18  0.402  0.221-0.585  31.6  20.7-48.2 .  M  9.68  0.616  0.500-0.733  8.70  7.27-10.4  P  11.59  0.598  0.545-0.650  12.4  11.7-13.1  M  12.51  0.755  0.712-0.798  3.154  3.14-3.15  P  2.29  0.299  0.239-0.361  11.5  11.5-11.5  M  8.99  0.462  0.362-0.563  8.13  7.80-8.47  P  15.53  0.907  0.832-0.983  1.34  1.13-1.58  M  14.60  0.976  0.938-1.01  0.42  0.38-0.46  Each a, calculated by applying its respective b to the means of C and Q from each data set (a =C -b Q ). Ranges are based on 90% confidence. X is the ratio of the error variances (E /E ). P = pre-monsoon season; M — monsoon season. f  f  2  C  mean  f  mem  2  Q  Lower Andheri (2) increases the expected error of the Q values for these situations. However, due to their limited nature, these are not incorporated further in the present analysis. The pre-monsoon sediment response is greater by up to an order of magnitude in comparison with its associated monsoon response and shows a tendency to merge with the monsoon-season result at the highest flows (Figure 5.7). This pattern is evident within all basins including that of the largest Jhikhu basin. The similarities suggest that the same controls are operating over all scales (1-100 km), 2  with their relative influence changing with season and scale. The influence of basin character on these sediment regimes can be examined further by contrasting the parameters of these ten equations and b). f  Figure 5.8 compares the expected values and ranges (here at 90% confidence) in b and % f  between all five basins. This comparison reveals the relative vulnerability of these basins to floods, vis-a-vis suspended sediment. The exponent of the relation (b) expresses the extent of coupling f  Figure 5.8  Seasonal variation in % and b with basin area (C= Q ). b  f  a  (a) Relation Exponent (bf) 2.5  :ip Monsoon Season  1  I 0.5 1  10  100 200  J.  2 J.  3  0.5 1  Basin Area (krn^)  10  100 200  Basin Area (km^)  (b) Relation Coefficient (af) 100  Monsoon Season  80 60 O  40  SS  g. -10  <w  o  20  u 0.5 1  100 200  9  3  0 0.5 1  Basin Area (km^)  10  Basin Area (km^)  Error limits based on 90% confidence. 10 - Kukhuri 9 - Upper Andheri  2 - Lower Andheri 3 - Dhap  100 200  1 - Jhikhu  129 between Q and C. Relief has a strong but not exclusive effect on this coupling. The higher b is, the f  greater is the possible transport limitation. The coefficient (aj) provides an indication of the magnitude of the relation between Q and C. Higher values of a, (for equivalent b) indicate a greater sediment f  availability. Considered together, especially in light of basin characteristics (including area), the comparison can shed light on the relative behaviour of these basins. It is important to realise that these two parameters are not independent because they are coupled through the bivariate mean in the linear regression (and functional analysis) and hence they are related to one another through R . For 2  example, a higher % and lower b of one relation in comparison to another may suggest higher f  sediment concentration at low flow and lower concentration at high flow. Figure 5.8a shows the seasonal sensitivity of each basin to changes in discharge. The pattern consistent within each season - suggests a decline in sensitivity with scale within the major tributary basins then an increase to the largest scale (100 km). In particular, Kukhuri demonstrates a highly 2  coupled C-Q behaviour with slopes (range of 1.18 to 2.11) higher than those generally quoted elsewhere (section 5.2.1). In this steep basin where cultivation is found on over 63% of the land area (see Table 2.5), discharge has a dominant influence on sediment regime. At low flows, sediment contribution is negligible (Figure 5.7) while at medium and high flows, sediment concentrations climb rapidly, yielding the highest value recorded in the study, 123.7 g/1. The low values of b for the f  Upper Andheri basin are anomalous: this basin is proportionately as steep as the Kukhuri basin yet yields a range of b similar to the two 5-km study basins. Its larger area (2.5 times the area of 2  f  Kukhuri basin) may contribute to an attenuation of this strong headward coupling. The effect may also be due to the presence of pockets of degraded land in the vicinity of the hydrometric station (easily eroded at low flow), exaggerated by measurement difficulties that were faced only at this station. Additionally, sampling during May when it is suspected that concentrations of suspended sediment are considerably higher (in comparison to those of June) was successful at this station and certainly provided an unusual number of high-C/low-Q pre-monsoon samples to raise the low-Q end of this  130 relation. The relations for the entire Lower Andheri basin (2) show steeper slopes than that of the Upper Andheri basin (9), though having three times the area. Management factors which are not as influential on the sediment regime within the Upper Andheri and Kukhuri basins come into play in the lower reaches of the overall basin. Two of these factors, surface degradation and sediment storage within the irrigation system, are discussed respectively in detail in sections 5.5.2 and 5.5.3. Dhap, though of the same size as Lower Andheri has a significantly lower sensitivity to Q. Dhap is dominated by modest relief containing little steep land like that characterising the upland of Lower Andheri (Table 2.5). The sensitivity of Jhikhu basin increases over that of the major tributaries suggesting a change in dominant processes between the 10- and 100-km scales. The most likely 2  explanation appears to be a loss of sediment to storage within the large valley meanders between the Dhap and Andheri Rivers' confluences with the Jhikhu River. It is only for the major tributary basins (Dhap and Lower Andheri) where the sensitivity to discharge changes significantly between seasons. Both Jhikhu and Upper Andheri show a trend of increasing sensitivity under the monsoon regime in comparison with the pre-monsoon regime, however, the differences are not statistically significant. Kukhuri basin shows no change in sensitivity with season, a fact consistent with its small area and steep topography. Not unexpectedly, aj shows a consistent decline with basin scale as a direct result of increased storage opportunities. The only exception to this is the Dhap basin during the monsoon season - its high coefficient (given the relatively consistent behaviour of b in Figure 5.8a) suggests a high rate of f  sediment output during the monsoon season. This situation is related to this basin's degraded condition and is examined in detail in section 5.5.1. During the pre-monsoon season, the coefficient rises to very high values for small basin areas indicating - especially in light of the strong C-Q coupling observed in Figure 5.8a - a high degree of sediment mobilisation and an important area of concern for soil-erosion management. In Figure 5.9, the regression results for all five stations are overlaid for the pre-monsoon and  131 Figure 5.9  Seasonal sediment rating curves overlain for all study basins showing both the expected functional relations and envelopes representing confidence limits to these relations (at 90%): a) pre-monsoon regime b) monsoon regime.  Soildlines: within Andheri D a s h e d l i n e s : outside A n d h e r i K u k h u r i (10) D h a p (3)  U p p e r A n d h e r i (9) L o w e r A n d h e r i (2)  ( a ) P r e - M o n s o o n  Expected Relations  < 1000  S e a s o n  1000  a •a  100  J h i k h u (1)  Envelope of Expected Relations  100 -  h  10  a O  a o U  1 -  •(^ CS  0.1  B  Discharge (m /sec)  100  0.01 0.01  ( b ) M o n s o o n  Qiooo  Expected Relations  Discharge (m /sec)  1000  100  -  1 0.1  10  Discharge (m /sec)  100  S e a s o n Envelope of Expected Relations  Discharge (m /sec)  100  132 monsoon seasons. These overlays present an "integrated" comparison of af and b between basins. It f  appearsfromthis figure that within the 1 to 10 km range of scale, differences between the relations 2  are due to rotation of the curve as a result of the rapid reduction in specific relief over these small scales. Between 10 and 100 km , the change is dominantly one of translation - basin sensitivity to 2  discharge remains similar but specific suspended sediment transport declines, perhaps due to alluvial storage. This pattern is stronger during the monsoon season because the plentiful sediment supply of the pre-monsoon season raises suspended sediment concentrations at all scales and relatively more at low Q, reducing the dominance of Q during this season. This may provide a conceptual framework within which to examine further the sediment regimes in this agricultural system. The sediment rating curves presented in this section corroborate the findings of section 5.3 in which pre-monsoon surface erosion was seen to be over an order of magnitude higher than the level in the monsoon season. However during the monsoon season and over all studied scales, the rate of sediment productionfromthese upland cultivated fields is not sufficient to equal the amount of sediment which is transported in the streams. It must be concluded that there are other sediment sources present particularly during the monsoon season over scales larger than that of the individual field. These sources are streambed and streambank erosion and terrace damage and are of particular importance during "major" storm events (see section 4.6.1) and at high flow rates. The distribution of floods of the monsoon season tends to have higher peak flows than their pre-monsoon counterparts: this corresponds well to a seasonal increase in erosion due to mass wasting and is backed up by observation (discussed in Chapter 8). Another important source of suspended sediment is revealed by considering the regimes of the transition season. Figures 5.10a through 5.10e show the suspended-sediment data of the transition season for stations 10, 9, 2, 3, and 1 respectively and the seasonal regression lines for reference. In each case, the transition-season data show intergrade behaviour between the other two seasons. The vegetative cover - discussed in section 5.3 - does not develop instantaneously at all locations within  133 Figure 5.10  Suspended sediment data ofthe transition season and the seasonal regression regimes based on the entire data set:(a) Kukhuri, (b) Upper Andheri, (c) Lower Andheri, (d) Dhap, and (e) Jhikhu.  Uiooo  "5  IOOO  0  1 %  o  U *-> fl  B •l-H  T3  .01  0.1 1 310 100 Discharge (m /sec)  <D OO  ^1000  0.01  0.1 1 ,10 100 Discharge (m /sec)  ^1000  .01  0.1 1 10 100 Discharge (m /sec) 3  Discharge (m /sec)  "S iooo  Pre-Monsoon Season  c  o  Monsoon Season  • i-H  3 o fl o  A  U  a •I-H  C/3  .01  0.1 1 310 100 Discharge (m /sec)  Transition  134 the basin and during this period, heavy erosive rains may or may not occur at any particular location within a basin. This mechanism is operating throughout the transition season to a variable degree. At the onset of the monsoon season, the vegetative cover is ostensibly complete and this production opportunity gone. On-site observation during this period supported by the seasonal contrasts provide the evidence for these statements. In addition, the sand component of the suspended sediment load may show a declining availability through the transition season as this fraction (mobilised during the pre-monsoon season) is flushed through the basins (see Chapter 6). The data from the mid-reach Andheri stations (11 and 12) provide further evidence of the trends observed in this section. Seasonally-stratified suspended-sediment data from these stations are given in Figure 5.10 for stations 12 and 11 respectively. In both cases, higher regimes occur during the pre-monsoon season. Also, the transition season displays intergrade behaviour between that of the pre-monsoon and monsoon seasons. Due to the lack of stage-discharge relations at these stations, it is not possible to comment on the relative influence of discharge compared to the other basin scales. In conclusion, the suspended-sediment datafromthe streams reinforce and extend the results of section 5.3. The effect of surface cover in the pre-monsoon is evident clearly and it takes almost three weeks for a complete surface cover to develop in order to virtually eliminate this sediment production mechanism. The presence of significant suspended sediment during the monsoon season points to additional sediment sources beyond surface erosionfromindividual cultivated fields. Controls on suspended-sediment delivery change with scale; sensitivity to discharge declines with basin scale whereas management controls become increasingly complex with basin scale as explored in the following section.  5.5 Landuse signatures Human manipulation through the indigenous agricultural system is pervasive in the study basins, affecting nearly every aspect of erosion and sediment transport. Terracing, cropping practices,  135 Figure 5.11  Seasonally-stratified suspended-sediment data for the mid-reach Andheri River station based on 1993-1994 data: (a) mid-reach #1 (11), and (b) mid-reach #2 (12).  1000 100 -  O  a  a D O  o U +> a  B  10 10.1 10  "3> o 2  Pre-Monsoon  A  Transition  +  •l-H  CO  o  100  200  100  200  Gauge Height (cm)  1000 100 -  h o  o U +^ a  a  T3  <D  10 G a u g e  H e i g h t  ( c m )  Monsoon  136 water diversion, and many other soil and water management techniques control the movement of soil and sediment from its source location to its eventual discharge from a basin. Three analyses are presented here. In thefirst,two basins are compared - Dhap (3) and Lower Andheri (2). These basins are similar in area but have contrasting character and management regimes. This comparison studies the effect of degradation on annual and seasonal sediment regimes. In the second analysis, the effect of degradation is studied within the Lower Andheri basin. This basin contains contrasting landuse between its upper- and lower-elevation areas. The effects of this spatial distribution in surface condition is examined particularly in light of the spatial variability in rainfall delivery which was quantified in Chapter 4. Thefinalanalysis looks at water diversion to estimate the extent to which this management activity recaptures previously eroded soil, redistributing soil from the upland rather than losing it from the larger basin.  5.5.1 Surface degradation: comparison of two basins There arefrequentpockets of degraded land in the Middle Mountains, especially below 1400 m where red soils occur. Though these soils can be highly productive, they can also be difficult to manage and require high inputs of fertilizer and compost. Once these lands degrade, it can be difficult to bring them back into production. To what extent do these degraded areas contribute to the basin sediment budget? How important are they in modifying the patterns which have been discovered in the previous two sections? To help answer these questions, two basins of equal size and of differing character (especially with respect to their level of degradation) are compared in this section. Figure 5.12 contrasts the seasonal sediment rating curves for the Dhap basin with those of the Lower Andheri basin. These two basins are of equal size but the Dhap basin has a more gentlysloping topography and is in a lower elevation range. Despite these characteristics which should reduce this basin's tendency to erode, the sediment rating curve for the Dhap basin is higher than that of Lower Andheri during each season except at high-flow during the pre-monsoon season.  137 Figure 5.12  Seasonal sediment rating curves for Lower Andheri and Dhap basins overlaid for comparison.  1000  100  ao a G  <D  10  C  o U a B  1 -  C/3  0.1 0.01  15 03  ^  Lower Andheri -  PM  :  I  o U  1  Dhap p M -0-  T _  10  3  Discharge (m /sec)  10  c O  0.1  Lower Andheri  100  Dhap  138 The monsoon-season regression curves of these two basins are further separated than are their pre-monsoon counterparts. This enlarged separation develops because the sediment-yield response in Lower Andheri basin drops in the monsoon whereas that of the Dhap basin does not. Recall the anomalous pattern of af shown in Figure 5.8 with b consistendy lower in Dhap basin relative to f  Lower Andheri (due to Dhap's modest relief). These contrasting seasonal changes can be explained by surface cover and soil degradation. The Lower Andheri basin contains large amounts of well-managed cultivated steeplands. These soils experience a seasonal change in surface cover as examined in sections 5.3 and 5.4. The Dhap basin contains a greater proportion of seriously degraded land gullied soil with little or no vegetation. In the Dhap basin, 24% of the land is seriously degraded whereas in the Lower Andheri basin, there is only 15% in this state. This apparently small difference must have a large effect on sediment production within the basin especially given the limited relief of Dhap in comparison to Lower Andheri. Because these degraded lands do not experience a change in surface cover with the onset of the monsoon season, these small areas - if present - play a big role in the basin sediment budget. Thisfindingunderlines the importance of preventing land from becoming degraded. In the Andheri basin, 25% of the area is in a moderately degraded condition (minor gullies or limited crown cover) whereas only 18% of the Dhap basin is moderately degraded. If the moderately-degraded land in the Andheri basin is allowed to slip further out of production, the suspended sediment regime of the Andheri basin could become similar to that of the Dhap basin or even higher due to its steeper topography. Re-examination of Figure 5.7d indicates that the seasonal difference in the regression results for the Dhap basin is a result of low sediment data derived from monsoon-season samples. The highsediment data show no seasonal dependence at any discharge suggesting that the soil surface responds similarly to heavy rainfall in both seasons. The seasonal difference in low-sediment data may be the result of local (<5 km) convective showers over the less degraded parts of Dhap basin. 2  139 It is important to also consider flood frequency to understand the overall significance of the seasonal suspended-sediment behaviour. The sediment loads during the pre-monsoon season, though generally quite high throughout the Middle Mountains, are short-lived because this season is brief and rainfall infrequent. The monsoon season, in contrast, lasts longer and experiences a greater number of flood events. The increased flood frequency in the monsoon season underlines the importance of the lower surface erosion during the monsoon season. In a basin with degraded surface conditions like the Dhap basin, elevated pre-monsoon sediment levels are maintained during almost twice as many floods through the monsoon season. Local management which encourages the rapid return of a complete vegetative cover is therefore the key to reducing net sediment output. To determine the overall and relative importance of these processes, annual budgets need to be calculated and this is done in Chapter 8.  5.5.2 Surface degradation: within Lower Andheri basin The effect of surface degradation on stream suspended sediments can also be examined within the Andheri basin because the basin contains large areas of contrasting landuse. The lowlands are dominated by bare red soils, often gullied, while the majority of the headwaters is under intensivelymanaged, steepland agriculture on non-red soils. In Chapter 4, it was learned that flood-producing rainfall over this basin is local in nature (<5 km) for about half the storms. By linking the rainfall 2  database to the flood database, suspended sediments in floods resulting from storms over each of the two landuse classes noted above can be isolated. Section 5.3 demonstrated that high levels of surface erosion occur on steep cultivated fields (like those in the headwaters of the Andheri basin) during the pre-monsoon season. In the previous section, basin sediment rating curves were used to conclude that the gullied badlands yield high sediment levels though no plot measurements were provided to substantiate those statements. In several gullies of the Andheri lowlands, gully pins were maintained in place to monitor the change in  140 gully dimensions during 1992-1994. Gully growth is highly variable: one gully showed little or no growth in size while the other two changed greatly in dimensions, reflecting the instability of their vertical headwalls. These results confirm that this gullied area is an important sediment source and also suggest that gully management may be focused on the individual gullies which show particularly active behaviour. To carry out the comparison of suspended sediments arising from storm events over these two areas, it must first be clarified what constitutes upland-dominated versus lowland-dominated floods as it concerns suspended sediments. The approach taken here extends the classification described in Chapter 4 in which basin, lowland, and upland events were defined based on the average rainfall for 12 gauges within each of the lowland and upland clusters. If the average of the two clusters was not significantly different, then the event was considered a basin event. If the difference was significant, it was a lowland or an upland event depending on which area received more rainfall. Of the events which are erosive and cause floods, it is necessary to stratify them removing those events which yield suspended sediment from a mixture of upland and lowland source areas. Then the amount of suspended sediment can be related to the source area directly. To accomplish this, the phosphorus content of the sediment is used as a fingerprint to calibrate a threshold which identifies relative source area contributions. The red soils of the degraded lowland are low in phosphorus whereas the non-red soils of the upland are high in phosphorus. This contrast is exploited to identify suspended sediment originating from these two source areas (see section 6.5 for more details). For each flood within Andheri basin, the ratio of the average rainfall in the upland versus that of the lowland was calculated. Beginning with all the suspended sediments for upland and lowland rain events on one C-Q plot, the requirement for a minimum ratio of rainfall in one area to the rainfall in the other area was made incrementally more stringent and the resulting phosphorus content of the sediments at the hydrometric station examined. At a threshold of 1.7 (one area delivering 1.7  141 Figure 5.13  Sediment-phosphorus rating curves stratified by rainfall location and colour for Lower Andheri basin (2) based on (a) rainfall differences only, and (b) both rainfall differences and their relative upland-lowland sediment contributions.  (a) Upland/lowland classes based on rainfall differences only  Discharge (m /sec)  100  • A  Lowland Upland  100 A  AA  A  ArV .  . ^ f M^ 4 ! A A  A  A  • A  •  . AA  A  •  (b) Upland/lowland classes based on rainfall differences & relative sediment contributions  •  0.1 0.01  1  •  0.1  1  • 3  10  Discharge (m /sec)  100  142 times the rainfall of the other area), the phosphorus content of the sediments separated as seen in Figure 5.13. Below this threshold, sediments sampled at Lower Andheri from upland rainfall events contain a significant number of red, low-phosphorus samples and those from lowland rainfall events contain brown, high-phosphorus samples. The threshold value represents the point at which the contribution from one area dominates overwhelmingly that of the other area. The sediments derived from "mixed" events are removed from the C-Q graph altogether. For example, if a lowland event brings 40 mm average rainfall to the lowland, it is retained as a lowland event in this analysis only if it also brings no more than 23.5 mm of average rainfall to the upland cluster. Those area events which do not satisfy this added constraint are termed basin events. In Figure 5.14, the locations of the rain gauges within and near the Andheri basin are overlaid on the red soil map. The areas of red soil in the Andheri basin lowland correspond to the bare and degraded land. This figure shows that if a storm event is a lowland one, it is dominantly over the degraded areas (on red soils) whereas if it is an upland event, it is dominantly over the intensivelymanaged cultivated (non-red) soils. C-Q data from the Lower Andheri (2) station introduced earlier in this chapter (Figure 5.7c) are stratified according to lowland and upland events and the result is presented in Figure 5.15 (data from basin-wide events are not shown). The upland displays a greater sensitivity to Q (as reflected in b) than the lowland and the difference is significant (90%) during the monsoon season. In addition, a, f  for the upland is double that of the lowland during the pre-monsoon season and this difference is significant. The higher response during the pre-monsoon reflects the vulnerability of the steep upland during this season especially given the potential sediment production capability of the degraded lowland given the findings of the previous section. These functional relations are based on only 22 independent flood events. In addition, due to the more frequent occurrence of lowland events, 17 (P - 9; M - 8) of these are lowland events and only 5 (P - 3; M - 2) are from the upland. As a result, the upland data used to develop the relations  143 Figure 5.14  Locations of the rain gauges within and near the Andheri basin overlain on the red soil map.  @ • + —  Red Soil Non-Red Soil Rain Gauge Locations Subwatershed Boundaries Rivers  144 Figure 5.15  Sediment rating curves for Lower Andheri station (2) including sedimentsfromonly lowland and upland events (1992-1994).  1000  •  Lowland (P)  A  Lowland (M)  • 100 -  O  +  fl  o  Upland (P) Lowland (M) Flood #1  • 1—I •*->  c  <D O  10  C  o Uw>  Functional Relations  C B  Lowland  • i-H  TD  ——  <D  M  00  0.1 0.01  —- —  0.1  1  3  Upland - all Upland - w/o #1  Discharge (m /sec) Coefficient (af)  Lowland  40  Upland  p M  P  1.0  M  100  10  Slope (bf)  Lowland P  Upland  M  p  M fl  •g 20  All Data  A l l Data  93  -  o U  -  I  Without T  I  F l o o d 1 J_  g  0.5 -  o  Without Flood 1  oo  3:  0.0  Error limits based on 90% confidence.  145 presented in Figure 5.15a are dominated by a well-sampled  major  event (see section 4.6.1 for  definition of storm classes) of June 9, 1992 (Flood #1) identified on the figure. The effect of this event is investigated by recalculating the functional relation excluding these data. This relation is shown in Figure 5.15a and in the confidence ranges shown in Figure 5.15b and 5.15c. These recalculations show that without this one event, there is no difference during the pre-monsoon season between the upland and lowland flood events. Clearly, an analysis cannot rely on one independent event to reach a conclusion of statistical significance. This finding suggests caution in situations where the number of independent Table 5.9  events  is small.  Annual and seasonal sediment-rating-curve relations for Lower Andheri station based on lowland/upland data using log-linear regression.  Area  Marginal Regression Seas on  a  b  N  R  2  X s  r  b  aj  f  expected  range  expected  range  lowland  P  10.3 0.494 0  44  0.547 0.0589  8.03  0.506  0.4140.600  10.45  9.3511.71  upland  P  21.2 0.704 3  19  0.637  0.124  19.59  0.715  0.5410.891  21.30  20.3222.34  lowland  M  3.69 0.624 4  33  0.551 0.0806 10.15  0.644  0.5090.783  3.77  3.294.33  upland  M  3.39 0.875 8  32  0.892 0.0400  6.87  0.886  0.8130.960  3.36  3.613.13  upland (w/o #1)  P*  11.8 0.561 64  8  0.598 0.0472 4.80  0.585  0.3080.881  12.10  9.5815.52  Upland/pre-monsoon recalculated without the single large pre-monsoon event of June 9, 1992.  A comparison of seasonal sediment yield between the upland and the lowland must also consider the relative frequency of upland versus lowland events. The functional relations suggest that upland events deliver a higher sediment concentration but in section 4.5.2, it was found that almost 43% more lowland events (>10 mm; 1992-1994) were monitored in comparison to upland events. Together, these results suggest that the degraded lowland may leak sediment throughout more of the rainy season whereas the upland's losses are episodic, particularly during the pre-monsoon season.  146 What does this analysis tell us about the effect of landuse on suspended sediment concentration within this 5-km basin? It suggests that event sediment yield is higher for events in the 2  pre-monsoon season originating in the steep, well-managed upland. This increase is due to high-Q samples. This difference is not present during the monsoon season because high-C samples are limited to high-Q conditions. Although the degraded lowland did not generate C values as high as those from the upland, this area yielded higher C values for conditions of low and intermediate flows, especially during the monsoon season. This is consistent with the previous section where it was found that the Dhap basin retained an elevated rate of sediment output during the monsoon season. This analysis is useful because it confirms that we can also see this behaviour within a 5-km basin. 2  According to Burt (1989), the mixing of runoff from widely differing source areas (e.g., in terms of precipitation inputs and surface response) obscures the process-response relations that can generally be identified in smaller basins. Burt pointed out the importance of basin scale in mediating the degree of obscurity: it is in the intermediate-sized basins where variability in surface response (surface cover and condition) and rainfall input is greatest. In the study area, this area appears to correspond to about 5 km . Above and below this scale, process-response relations are identifiable. 2  The analysis presented in this section reveals that surface response and variability in rainfall input contribute formatively to sediment regimes at these limited spatial scales (roughly 5 km). This 2  conclusion has major implications for monitoring in similar mountainous environments.  5.5.3 Sediment storage by water diversion for irrigation  The irrigation system serves as a sediment trap in diverting flow and halting the movement of eroded material out of watersheds. But how effective is this process? To what extent are the farmers able to recapture soil and nutrients lost from their upland agriculturalfieldsand entrained in the fluvial system? To answer this question, two analyses are presented. Thefirstfocuses on the diversion of water while the second looks specifically at the extent of sediment accumulation in the irrigated  147 fields. Water  diversion  As it progresses downstream, runoff is concentrated and thus streams generally grow in size. If rainfall is uniform over the basin and if there are not significant losses to the subsurface nor dramatic changes in the surface characteristics downstream, one would expect the stream to grow in volume in proportion to the ratio of the basin areas which contribute to their outflows. This principle is used here to evaluate the extent of diversion of floodwaters by diversion dams. Figure 5.16 shows the Andheri and Kukhuri Rivers and the locations of the 62 diversion dams which are built along this drainage. This figure also shows the locations of two automated hydrometric stations within this basin. Table 5.10 summarises the pertinent information about each of these hydrometric stations. Though there is irrigated land above the Kukhuri River station (10), it is in the basin headwaters where water is scarce and unreliable. Hence, the flow passing through the Kukhuri station is probably not much different than it would be in the absence of the structures of the irrigation system. The ratio of the contributing areas of the two stations is 7.4, as shown in Table 5.10. Table 5.10  Contributing areas and number of irrigation dams for Kukhuri and Lower Andheri hydrometric stations (stations 10 and 2 respectively).  Total Contributing Area (ha) Irrigated area (ha) Number of dams  Kukhuri Basin (station 10) 72 7 14  Lower Andheri Basin (station 2) 532 37 62  Ratio 7.4 5.3 4.4  Between these two stations, there are 21 large diversion dams. Without these in place, and assuming losses to the subsurface and groundwater are negligible, then we would expect to see about 7.4 times more flow at the Andheri station from rainfall events which provide even coverage over the entire basin. During the period of monitoring, 16 basin rainfall events (see section 4.5.2) were  149 monitored simultaneously at both stations by the automated equipment and are available for comparison. For each of these flood events, the total flow through the two stations has been calculated. The results are presented seasonally and in terms of the ratio of the lowland to upland stations. These results are summarised in Table 5.11. Table 5.11  Flow-ratio comparison of 16 individual floods for Kukhuri (station 10) and Lower Andheri (station 2) hydrometric stations.  Outflow Ratio Class (Andheri/Kukhuri) 0- 1.0 1.0-3.7 3.7 - 7.4 > 7.4 Total  Pre-Monsoon and Transition Seasons 2 3 1 1 7  Monsoon Season 3 5 0 1 9  Total Number 5 8 1 2 16  Percentage of Total 31 50 6 13 100  In all but two cases, the ratios are far less than would be expected under unmanipulated hydrological conditions. In fact, 31% experience  less  flow at the lowland than the upland station.  Further, the seasonal separation suggests that the tendency for a reduction in basin output may be greater during the pre-monsoon and transition seasons than it is during the monsoon season. However, because of the small number of events available, this seasonal consideration will have to be revisited when there are more events monitored in this comprehensive way. The extent of irrigation in the upland compared with the lowland and the surface soil characteristics both serve to make stronger the conclusion of this analysis. The upland has proportionately greater irrigated area than the lowland. If the dams are effective at diverting floodwaters in the steep upland, then this greater area should cause the flow ratio (lowlandflow/upland-flow) to be higher, not lower. Further, the water-holding capacity of the surface soils in the lowland is less than in the upland; this should further cause the ratio to be higher than if this factor was equal. Despite these differences, the low flow ratios persist.  150 The flow-ratio analysis suggests that the diversion dams are very effective at directing floodwaters out of the stream and into the irrigation system. In fact, it appears that a majority of the floodwaters is redirected into the irrigation system. Quantitative water budgets including measurement of runoff diverted into khet on a flood basis is a useful area of further research. In addition, we know from the hydrometric data that large amounts of suspended sediments are carried with the floodwaters. If the hypothesis is correct, then there should also be evidence of soil accumulation within the irrigation system. We know that the farmers annually maintain the canals removing considerable deposition (see Chapter 7) but what of thefieldsthemselves? Sediment accumulation  pins  To examine this, pins were placed in a wide selection of khetfieldsbefore the onset of each flood season during the study period. The pins were collected after harvest and the soil level noted and compared to the level before the pre-monsoon season began. These results, summarised in Table 5.12, suggest that considerable deposition occurs within the khetfieldsthemselves. For instance, of the 25fieldssampled in 1992, 76% showed accumulation and 40% showed more than 0.5 cm. Further, these enriched deposits enhance soil-nutrient condition: lab analyses show that all the base cations are higher in the deposited sediment than in the underlyingfieldsoil (see Figure 7.1). These two analyses suggest that management can recapture large amounts of previouslyeroded fertile soil from upland cultivated fields. Due to its fertility, this redistribution of soil represents a redistribution of wealth from the upland farmers to those in the lowland. To what extent is soil redistributed and not lost from these headwater basins? This question and others will be addressed in Chapter 8 after a more detailed analysis of sediment dynamics is explored in Chapter 6.  Each of the above three sections, considered individually, gives an indication of landuse effects on sediment regimes. Collectively, they provide a stronger indication of the effect on sediment regimes of landuse practices.  151 Table 5.12  Sediment accumulation in irrigated fields measured using pegs.  Year  Accumulation (mm)  1992  1993  n  X  25  10.5  33  1994  47  7.8  Accumulation Distribution (mm) >5,  >15, <25  >25  <£15  6  9  5  3  2  24%  36%  20%  12%  8%  2  11  16  4  0  6%  33%  49%  12%  0%  12  34  1  0  0  26%  72%  2%  0%  0%  <0  21.2  6.1  1.4  >o, <£5  s  1.3  Note: overall average accumulation = 6.6 mm/yr 5.6 Conclusions Soil losses are observed to be strongly seasonal with rates of surface erosionfromupland bari being one to two orders of magnitude higher in the pre-monsoon season than in the monsoon season. High annual rates of erosion (up to 40 t/ha) are due primarily to a lack of surface cover during the pre-monsoon when high-intensity rains may fall on newly-cultivated sloping fields due to unfortunate timing. The high nutrient content of soils during this season means that the seasonal change in nutrient loss is proportionally higher than the change in soil loss. Soil characteristics influence erosion at the plot scale through two mechanisms. For a given storm, they determine the rate of runoff which directly affects the soil's rate of erosion. If the surface is exposed, they determine the soil's erodibility in relation to rainfall intensity. Thresholds of I =30 10  mm/h and C = 5 to 10% were observed below which surface erosion rarely occurs. Above this I R  10  threshold, soil loss increases with I . 10  There is little evidence of seasonal changes in rainfall characteristics sufficient to cause observed seasonal changes in erosion and sediment regimes. In particular, high-intensity rainfall occurs during all seasons within the study period. Higher volume rainfall occurred more often during  152 the monsoon season. The likelihood of heavy rainfall falling on dry ground is measurably higher in the pre-monsoon season and this may contribute to the higher erosion regime of the pre-monsoon season. When coupled with spatial variability in surface condition, rainfall spatial variability can result in significantly different sediment regimes over spatial scales smaller than 500 ha. Degraded land with a year-round lack of surface cover (and often gullied) experiences an elevated rate of erosion due to rain in every season. Because these soil losses have little nutrient value, these losses are of greatest concern for downstream sedimentation rather than the farming system. Indigenous management techniques are their most effective at low and intermediate flows. The indigenous irrigation system captures a significant portion of storm runoff during the pre-monsoon season, thereby capturing a large amount of soil eroded from upland fields. However, they are also vulnerable as the pre-monsoon losses indicate. The limited surface erosion from bari during the monsoon season when rainfall is equally intense illustrates the potential effectiveness of indigenous management in these environments. Spatial and temporal variabilities place constraints on monitoring for erosion and sediment transport studies. Temporally and spatially, surface erodibility and rainfall erosivity operate simultaneously and show interactions. Within basins of 10 000 ha in size, monitoring resolution can be important over spatial scales from 100-1000 ha, to 0.01 ha. Within an annual period, monitoring resolution can be important over time periods of the season and the single event. These conclusions relate to the normal-regime behaviour of these basins. During the threeyear period of study, heavy rainfall with significant erosivity occurred - including one event which may have been a 10-year storm event over 5 km - but absent were events sufficient in magnitude to 2  destabilise these sediment regimes (e.g., deep-seated landsliding). Regardless, the conclusions suggest specific management recommendations germane to the normal-regime behaviour. Recommendations for management are discussed in Chapter 9 following the presentation of sediment budgets.  153  6. Signatures of Erosional Sources and Sediment-Transport Behaviour in the Physical and Chemical Character and the Patterns of Movement of Suspended Sediments 6.1 Introduction  In this chapter, suspended-sediment behaviour is examined in the context of both its behaviour during floods and of the physical and chemical character of the sediment itself. The examination of suspended sediment as a simple, single quantity masks important differences in how its concentration and composition are shaped by the range of controls. Sediment properties serve as sensitive indicators able to reveal a wider range of controls. Controls are generalised in relation to supply and transport limitations. Strengths and weaknesses of the sediment rating curve technique for prediction are identified. Three analyses are presented. An analysis of hysteresis provides further evidence of source and transport controls. Rating curves for individual particle-size classes are developed, illustrating the contrasting behaviours of separate size fractions. Sediment colour and phosphorus content are used as fingerprints to trace the origin of suspended sediment and relate this information to the conclusions reached regarding hysteresis and particle-size behaviour.  6.2 Research Background  In all basins, sediment supply and the competence of transport mechanisms interact through time and space to yield a sediment regime. Chapter 5 showed how bulk behaviour changes as a function of specific controls. The exponent (b) and coefficient (a) of the power-law relation (C=aQ ; b  C - suspended sediment concentration; Q - discharge) were used to discuss the effect of transport and supply limitations respectively. Patterns of hysteresis also reflect supply and transport limitations but their sensitivity is far greater than that of composite rating curves. And these patterns can be examined over various spatial and temporal scales to investigate embedded regime controls. Particle-size controls on sediment yield  154 are important yet poorly studied. Clay, silt, and sand behave in contrasting ways and reveal patterns which can also be linked to supply and transport concerns and related to hysteresis. And, finally, sediment properties can be used as tracers to further identify dominant controls, especially where interactions are occurring. This section reviews existing research findings relevant to these three areas - hysteresis, particle size, and tracers.  6.2.1 Hysteresis Hysteresis in suspended-sediment transport is used to indicate departures from a direct relation between discharge and sediment concentration. Hysteresis can impair predictive ability because the assumption of simple C-Q relations may lead to variously biased predictions. Fortunately, when multiple controls are operating in complex ways on suspended sediment, the presence of hysteresis can actually help to enhance understanding of sediment sources and dynamics. The term hysteresis is normally applied to the event or synoptic scale. Seasonal and annual changes are generally termed rating "shifts". The remainder of this discussion refers to the event and synoptic timescales. Many authors have observed hysteresis (Johnson 1942; Heidel 1956; Walling and Teed 1971; Walling 1974; Wood 1977; Paustian and Beschta 1979; Bogen 1980; van Sickle and Beschta 1983; Sidle and Campbell 1985) and generally attribute the phenomenon to the following primary factors: • supply exhaustion • supply thresholds • kinematic effects (e.g.: lag time, source variability) These controls dominate differentially through time and space and, in some instances, their effects are scale embedded. Since most basins are supply limited (Einstein 1943), supply-related controls are most often cited. Conceptual models of hysteretic behaviour have been presented by several authors in terms of  155 C-Q graphs (Olive and Rieger 1985; Williams 1989; Nistor 1996) and all include both primary and secondary behaviours. Table 6.