Open Collections

UBC Undergraduate Research

The effect of forest harvesting on streamflow recession curves at Carnation Creek, British Columbia Nelson, Brett 2012

You don't seem to have a PDF reader installed, try download the pdf

Item Metadata

Download

Media
[if-you-see-this-DO-NOT-CLICK]
[if-you-see-this-DO-NOT-CLICK]
Nelson_Brett_FRST_498_Graduating_Thesis_2012.pdf [ 1.54MB ]
Metadata
JSON: 1.0075628.json
JSON-LD: 1.0075628+ld.json
RDF/XML (Pretty): 1.0075628.xml
RDF/JSON: 1.0075628+rdf.json
Turtle: 1.0075628+rdf-turtle.txt
N-Triples: 1.0075628+rdf-ntriples.txt
Original Record: 1.0075628 +original-record.json
Full Text
1.0075628.txt
Citation
1.0075628.ris

Full Text

THE EFFECT OF FOREST HARVESTING ON STREAMFLOW RECESSION CURVES AT CARNATION CREEK, BRITISH COLUMBIA.  by  BRETT NELSON  Dr. R. D. Moore, Primary Advisor Dr. S. J. Grayston, Secondary Advisor  A thesis submitted in partial fulfillment of the requirements for the Degree of Bachelor of Science in Forest Sciences Hydrology Specialization  UNIVERSITY OF BRITISH COLUMBIA Vancouver, British Columbia  APRIL 17th 2012  Abstract At the Carnation Creek Experimental Watershed on southwestern Vancouver Island, British Columbia, the effect of harvesting, regeneration and road building were analyzed through the use of stream discharge data collected at a weir on the catchment outlet. The study was separated into a Pre-Logging period from 1971-1975, Logging from 19761981, and two Post-Logging periods from 1982-1985 and 1985-90 respectively. The current study focussed on the effects of harvesting on streamflow recession curves, which are an indicator of the ability of coastal watersheds to maintain low flows (base-flow) during the water-limited dry season. Approximately 30 years of discharge data along with rainfall and temperature were segmented into corresponding forestry operation periods. Using the linear relationship between log[|dQ/dt|] and log[Qm] according to storagedischarge theory, multiple linear models were created and a regression was used to investigate the significance of each of the forestry operations. It was found that the effect of the roads increased lateral slope interception of sub-surface flow and directed water along the ditch systems to the channel at a greater rate, steepening the recession curves at all discharge levels in the short term, but only persisting at low discharge levels. Harvesting increased the water table height, because of the reduction in transpiration via loss of interception, and low flows and total flows increased over both post-logging periods, which partially offset the effect the roads had on the recession. Regeneration began to occur over the harvested sections of the catchment and it was found their effect between post-logging periods was only significant with the inclusions of extremely low discharge levels. However, a second logging pass in 1987, removing 21% of the forested catchment in the headwaters is believed to have confounded the regeneration effects. i KEYWORDS: discharge, low flows, hydrograph, logging, roads, ditches, culvert, watershed, catchment, riparian harvest, cut-block, riparian buffer.  Table of Contents 1. Introduction ............................................................................................................................................ 1 2. Methods.................................................................................................................................................. 4 2.1 Study Site ......................................................................................................................................... 4 2.2 Data Collection ................................................................................................................................ 6 2.3 Forest Operations ............................................................................................................................. 7 2.4 Analysis ........................................................................................................................................... 8 3. Results ...................................................................................................................................................10 3.1 Regression Assumptions .................................................................................................................12 3.2 Variance Inflation Factor ................................................................................................................12 3.3 Discharge Levels ............................................................................................................................14 3.4 Road Effects ...................................................................................................................................16 3.5 Forest Harvest and Regeneration ....................................................................................................17 4. Discussion .............................................................................................................................................20 4.1 Road Effects ...................................................................................................................................20 4.2 Forest Harvest and Regeneration ....................................................................................................22 4.3 Study Limitations and Errors ..........................................................................................................24 4.4 Improvements to Research ..............................................................................................................25 5. Conclusion ............................................................................................................................................27 6. Works Cited ..........................................................................................................................................29 7. Appendices ............................................................................................................................................32 7.1 Appendix A – Residual Plots ..........................................................................................................32 7.2 Appendix B – Shapiro-Wilk Normality Tests.................................................................................34 7.3 Appendix C – FULL Model VIF Results .......................................................................................35 7.4 Appendix D – Coefficient Estimates (Simple Model) ....................................................................37 7.5 Appendix E – ANOVA Tests ........................................................................................................38  Index of Figures Figure 1. A Map of the Carnation Creek Watershed ...................................................................................... 5  Figure 2. The logging roads and bridge network of ....................................................................................... 7  Figure 3. The Mean Annual Precipitation (MAP) Graphs .............................................................................11  Figure 4. The Mean Annual Temperature (MAT) Graphs ............................................................................12  Figure 5. The Summer (A) and Winter (B) Discharge Graphs ......................................................................14  Figure 6. Pre-Logging vs. Post-Logging 1 - Low (a) and High (b) Non-growing season. ...........................17  Figure 7. (a-c) The Pre-Logging and Post -Logging 1 growing season comparison and (d-f) The PostLogging 1 and Post-Logging 2 growing season comparison .........................................................................18  Figure 8. Pre-Logging vs. Post-Logging 1 and Post-Logging 1 vs. Post-Logging 2 - Low discharge level (with extreme low-flow data) ........................................................................................................................19  Index of Tables Table 1. ANOVA Test p-values ....................................................................................................................15  Table 2. Winter Linear Model Coefficients. ..................................................................................................16  Table 3. Shapiro-Wilk Normality Tests.........................................................................................................34  Table 4. Linear Model Coefficient Estimates ................................................................................................37  1. Introduction There have been numerous studies showing that forest harvesting can increase: annual water yield, water table levels, and overland flow (Smerdon et al., 2009; Hetherington et al., 1998; Hicks et al., 1991; Keppeler and Ziemer, 1990; Harr et al., 1979). Harr (1979) reported that summer low flows increased almost four-fold following clear-cut harvest in a Western Oregon watershed. Additionally, road construction has been shown to reduce basin lag time and increase catchment connectivity (Hetherington et al., 1998). The vegetation removal effects have been seen to persist for one to two decades, and streamflow return to pre-logging levels is not guaranteed. This is due to the interaction between site geologic, vegetative, and climatic characteristics (Keppeler and Ziemer, 1990). The potential effect of forest harvesting on low flows is of particular interest for natural resource managers because low flow periods are associated with limited availability of water resources for human use, reduction of habitat availability for fish and other aquatic organisms, and an increased risk of elevated stream temperatures which can threaten the survival of many aquatic species (Price, 2011; Smerdon et al., 2009; Brandes et al., 2005). A greater knowledge of forest harvesting effects on low flows is fundamental to developing and continuing the sustainability and practicality of coastal watershed management. An important tool in understanding the low flow hydrology of a catchment is the streamflow recession curve, which is the portion of a streamflow hydrograph where discharge decreases steadily during periods of little or no precipitation (Smakhtin, 2001). Recession curves are useful indicators of a catchment’s ability to release water stored as  1  soil moisture and groundwater over extended periods of time, thus maintaining flow and aquatic habitat during periods of dry weather. In particular, the slope of the recession curve is a measure of how quickly water is released from storage within the catchment (Waterloo et al., 2007; Moore, 1997). It is influenced by soil characteristics, such as depth and permeability, hillslope gradients, the nature of the underlying bedrock, and the effects of evapotranspiration (Tallaksen, 1995). Forest harvesting results in a reduction in vegetative transpiration and interception and loss of precipitation, at least over the short term, allowing more water to infiltrate into the soil and flow by subsurface flow paths to the stream channel (Smerdon et al., 2009; Waterloo et al., 2007). Forest harvesting can therefore result in increased soil moisture storage (Smakhtin, 2001) and higher water tables (Smerdon et al., 2009; Hetherington et al., 1998; Hicks et al. 1991; Keppeler and Ziemer, 1990). Consequently, forest harvesting should generate increases in both annual water yield and low flows, as was found by Keppeler and Ziemer (1990) and Hicks et al. (1991), among others. However, as a forest regenerates, increases in transpiration and interception loss can become greater than a mature forest, especially for early seral stage deciduous species, resulting in more extreme low flows over the medium term (e.g., Hicks et al., 1991). Waterloo et al. (2007) found during their study that even with higher rainfalls the lowest flows coincided with forested catchments, due to their large uptake of water in the hydrologic cycle. Forest cover increases basin storage by augmenting the infiltration rate into the soil, despite increased interception. However, by extracting water from the soil that would otherwise continue to maintain flows during periods of dry weather, forest transpiration increases the rate of streamflow recession (Federer, 1973). Therefore, the removal of 2  forest cover should result in less steep recessions, at least for the first few years or so following logging. Over the medium to longer term, recession curves generally become steeper as the forest regenerates. Roads across hillslopes can intercept subsurface flow and redirect it to ditches and culverts and eventually to the stream at a much faster rate than it would in the absence of roads (Winkler et al., 2011). Furthermore, due to the low permeability of road surfaces, much of the precipitation falling onto roads can also be directed as overland flow to ditches. Roads can thus result in decreased travel times and greater connectivity of the land to the channel network, producing faster streamflow response to rainstorms, as well as greater peak discharge. This redirection of flow also means that less water follows the slow subsurface pathways that supply the low flows during recession periods. The increased responsiveness of the catchment is opposite to the effect of harvesting. Therefore, the construction of logging roads is expected to result in a steepening of recession curves that persist through time (assuming the roads are not rehabilitated). Analyses of the stream discharge data aims to separate the effects of the harvesting by growing season and non-growing season (summer and winter). During winter, the transpiration is severely reduced, so the effect of roads on the landscape can be isolated. If there is a significant recession curve steepening in winter, then it can be assumed that the roads are responsible (1). Comparing early post-logging discharge data with prelogging data in the growing season should show a partial offset of recession curve steepening due to reduced transpiration. However, regeneration will increase transpiration and steepen the curve. If there is significant differences between pre-logging and post-logging and post-logging 1 to 2 growing season periods, then the effect of 3  harvesting and forest regeneration is likely culpable (2). Moreover, the logging roads and culverts can be thought of as permanent additions to the landscape and hydrologic regime and their effects should persist through time. If there are no changes to the recession data in the non-growing season over the post-logging period, then it is probable that the effect of roads does not vary significantly over that time (3).  2. Methods 2.1 Study Site This study draws upon data collected at the Carnation Creek Experimental Watershed, a single-watershed experiment site located near the southwestern coast of Vancouver Island, BC. The Carnation Creek experiment was initiated by the Department of Fisheries and Oceans (DFO), and is continued today as a part of the Forest-Fish Interaction Program (FFIP) conducted by the Research Branch of the B. C. Ministry of Forests. It was originally conducted to explore the effects of clear-cutting on coastal watersheds stream channel morphology and fish populations. It has grown into a long-term multidisciplinary study, which is continuing to investigate the effects of forestry-related operations on coastal catchments (Province of British Columbia, 2009). Over 200 papers have been published on the data collected from this site and the results and implications from the studies influenced the Forest Practices Code (FPC), later superseded by the Forest and Range Practices Act (ibid). Carnation Creek has a drainage area of 11 km2 and has a main channel length of approximately 7.8 km. The site falls within the Coastal Western Hemlock zone (CWH), 4  receiving around 2100 to 5000 mm of precipitation every year with 95% of that falling as rain primarily in the fall and winter months. Scrivener (1975) reported that after intense storms, the annual precipitation leaving this catchment as runoff could be up to 90%.  Figure 1. A Map showing the Carnation Creek watershed with an inset showing its location on southwestern Vancouver Island. The climate stations (A-L), the years of harvest, harvest boundaries, hydrologic weir locations and their sub-catchment basins are shown. From Hetherington et al. (1998).  The terrain is fairly rough, maintaining steep gradients up to 80% and an elevation ranging from 0 to 800 masl. The soil profile is a shallow ~ 0.7 m veneer with a coarse textured colluvium, and some variable dense till deposits found overlying bedrock. The last 3 km of stream before the outlet is a floodplain, which is composed of a gravelly alluvium (Hartman and Scrivener, 1990). The soil drains rapidly and extremely well, with preferential flow pathways and large macro-pores allowing quick movement of water  5  through the sub-surface system to the channel network (Fannin et al., 2000). The original mature tree species found in the catchment were Western red cedar (Thuja plicata), Sitka spruce (Picea sitchensis), amabilis fir (Abies amabalis), Douglas-fir (Pseudotsuga menziesii) and Western hemlock (Tsuga heterophylla). Red alder (Alnus rubra) and Bigleaf maple (Acer macrophyllum) were the dominant riparian tree species (Province of British Columbia, 2009; Hartman and Scrivener, 1990). Stream assessments found several species of anadromous salmonids, Oncorhynchus keta, O. kisutch, O. mykiss, O. clarki, Cottus aleuticus and C. asper, inhabiting the lower reaches of the stream, as well as a land locked population of cutthroat trout (O. clarki) upstream in the catchment (Province of British Columbia, 2009).  2.2 Data Collection Permanent weirs and climate stations have been in continuous operation since April 1971, collecting data on water temperature, depth, and stream discharge. Temperature and 24 hour precipitation measurements were obtained from 10 sites in and around the catchment study area. They were serviced weekly and recalibrated monthly if necessary. Weighing precipitation gauges were favoured over tipping buckets. The data was transferred to digital records for analysis, and the precipitation had a measurement error of ± 0.2 mm (Hartman and Scrivener, 1990). Stream discharge data from the original study was obtained from 5 different weirs around the catchment, but for our purposes we only used the discharge data from the V-notch weir at Station B collected with Stevens digital recorders. The stage-discharge curves 6  developed for each weir were calibrated and updated frequently to ensure instrument precision. 2.3 Forest Operations Construction of roads began in January 1975, and were usually built far from the stream (Hartman and Scrivener, 1990). They used D-8 Caterpillars, shovels, and rock drills to construct the majority of the roads. The roads were fully benched and consisted of a very coarse gravel that overlaid the bedrock.  Figure 2. The logging roads and bridge network of the Carnation Creek catchment site. From Hartman and Scrivener (1990).  Riparian harvest treatments included a 1300 m long leave-strip where between 1-70 m widths of riparian forest was left intact from the catchment outlet upstream. Directly  7  upstream from the leave strip was a 900 m long clear-cut treatment, removing all riparian vegetation. The timber was felled and yarded onto and across the stream and all valuable logs were removed from the channel. This was designated as an “intensive clear-cut” treatment. Extending 900 m beyond the first clear-cut was a “careful clear-cutting” treatment, where no activity was permitted in the creek, and small streamside vegetation was retained (Province of British Columbia, 2009). A cable and metal grapple system was used to yard the merchantable material to the roads. However, at several of the bottom valley sites, skidders with rubber low-pressure tires were used for transportation. Slash was mostly piled and burned on site, with some broadcast burning on slopes. Some of the soils were scarified and some untouched, depending on their aspect and location within the catchment (Hartman and Scrivener, 1990).  Through 1976-81, 41% of  catchment was harvested, focussing on the valley bottoms. In 1987, a further 21% of the catchment was harvested, focussing on the headwaters and slopes (Province of British Columbia, 2009). Reforestation efforts took place a year after harvest periods, with the primary seedling species being Douglas-fir, western red cedar, amabilis fir, western hemlock and Sitka spruce. In contrast, some sites received only natural regeneration (Hartman and Scrivener, 1990). 2.4 Analysis All of the data processing and analysis was conducted using the R programming language (R Development Core Team, 2011). The first step in the analysis was to identify recession periods in the streamflow records. In order to qualify as part of the recession, two criteria were to be stipulated. Firstly, there should be no precipitation on the current 8  or preceding days. Secondly, the stream discharge on the current day had to be less than the discharge of the previous day. Theoretical considerations and empirical studies indicate that the relation between the absolute value of the rate of change of daily discharge (|dQ/dt|) and the mean daily discharge (Qm) should follow a power law (Brutsaert and Nieber, 1977). Thus, a plot of |dQ/dt| against Qm on a log-log axes will follow a linear relationship according to storagedischarge theory. Therefore, log(|dQ/dt|) is used here as the response variable and log(Qm) used as a covariate. Initial analysis indicated some nonlinearity in the relation between log(|dQ/dt|) and log(Qm) over the full range of discharge. Therefore, data were stratified by discharge ranges, within which the relation was visually close to linear. To account for the effect of transpiration on recession rates, daily maximum air temperature (Tmax, °C) was included as a predictor variable, as transpiration rates should be correlated with air temperature. To account for changes associated with forest harvesting, the time series was split into four periods: (1) pre-harvest, “Pre” (1970 to 1975); (2) during-harvest, “Logging” (1976 to 1981); (3) initial post-harvest, “Post 1” (1982 to 1984); and (4) later post-harvest, “Post 2” (1985 to 1990). The analysis periods differed in the number and distribution of extremely low streamflow values. To minimize the effect of these differences on the analysis, the data were subsetted again to remove days on which Qm was equal to and below 0.02 m3s-1. This removed 89 observations from the data set. The analysis involved fitting a series of linear models relating log(|dQ/dt|) to log(Qm), Tmax and P, where P is a factor representing period, with four levels (Pre, Logging, Post 1  9  and Post 2). To account for seasonal variations in transpiration rates, data were classified into two seasons (S), summer and winter, with analyses being conducted separately for each season. Summer included the period from May to September, while winter extended from December to February. This resulted in the exclusion of quite a few data points, but the reasoning behind this was to separate the growing season, in which transpiration could be significant, from a mid-winter period when transpiration is likely to be negligible (Shimokura and Shibano, 2003). In addition to a residual analysis to determine if the assumptions of multiple linear regression (MLR) have been met, we also calculated a variance inflation factor (VIF) for each of our models’ parameter coefficients in order to account for the adverse effects of multicollinearity (Graham, 2003; Farrar and Glauber, 1967). Type III partial sum of squares F-tests were conducted on constructed models through the exclusion of P variables for the purpose of observing their significance (Kozak et al., 2008). All statistical significances of p-values were compared using an alpha (α) level of 0.05, but additional alpha p-values are listed for convenience.  3. Results Looking at the mean annual precipitation (MAP) in Fig. 2 shows that there is some slight variation in rainfall between periods. There appears to be a trend towards a decreasing MAP over the study period. However, since the linear models we constructed included Qm as a covariate, we assumed that any changes in precipitation would be correlated and described using the mean discharge. The maximum and minimum mean annual temperatures (MAT) graphs (Fig. 3) show that there are no clear deviations to 10  temperature in summer; however there are some increasing temperatures in winter. This possibility will be accounted for in our linear model with the inclusion of Tmax as an additional covariate, because we assume that it will be the maximum temperature that will have more weight on the evapotranspiration effect on the recession curves. A short study period with multiple catchments would have eliminated possible errors from climate variables and trends. However, a long-term study allowed the correlation between climate variations to be recorded with a larger dataset, and greater statistical power was achieved (Hartman and Scrivener, 1990).  Figure 3. The Mean Annual Precipitation (MAP) over the entire duration of the study, split into the harvest periods. Lowess regression lines were fitted to the data to show the variation over time and between the periods.  11  Figure 4. The maximum and minimum mean annual temperatures over the study period, separated into non-growing and growing season. The harvest periods were defined.  3.1 Regression Assumptions In Appendix A, the residuals plots for log(Qm) and Tmax are given for the linear models, which are separated by growing and non-growing season, and discharge level. They show that that there is independence of observations and equal error residual variance around the lines. The winter data deviated slightly from equal variance around the line and had several large outliers. This could be due to the reduced number of sample data because of the strict season requirements. The results of Shapiro-Wilk normality tests are located in Appendix B. A few of the linear models had p-values below the set alpha level of 0.05 and this meant that they had failed to meet the normality assumption. When normality of the error residuals is not satisfied, this can greatly bias the significance of the variable in the model as well as the parameter coefficient estimates. 3.2 Variance Inflation Factor When we initially performed our ANOVA tests on the models we included Qm, Tmax and P and all the interactions between them. However, we noticed that some of the model 12  parameters (Qm, Tmax and P) were non-significant predictors of the response variables. An example of this is below from a comparison between Pre and Logging Low discharge level. Model 1: log(dQ/dt) ~ log(Qm) * Tmax Model 2: log(dQ/dt) ~ log(Qm) * Tmax * P Res.Df RSS Df Sum of Sq F Pr(>F) 1 46 14.973 2 42 11.121 4 3.852 3.6369 0.01241 * --Call: lm(formula = log(dQ/dt) ~ log(Qm) * Tmax* P) Residuals: Min 1Q -1.14873 -0.28499  Median 0.00001  3Q 0.36418  Coefficients: Estimate Std. Error (Intercept) 21.1888 10.7142 log(Qm) -4.4771 3.0709 Tmax -1.1444 0.5863 P2 -29.3751 24.3688 log(Qm):Tm 0.3367 0.1685 log(Qm):P2 9.4920 7.6003 Tmax:P2 1.4588 1.1944 log(Qm):Tmax:P2 -0.4581 0.3711 ---  Max 0.89659  t value Pr(>|t|) 1.978 0.0546 . -1.458 0.1523 -1.952 0.0576 . -1.205 0.2348 1.998 0.0522 . 1.249 0.2186 1.221 0.2288 -1.235 0.2239  RSE 0.5146, 42DF, R2=0.3905,R2a: 0.2889,F=3.843 7 and 42DF, p-value: 0.002583  There is significant difference between the models (p-value = 0.012) attributed to period, yet in the model, all the parameters involving period are non-significant. We deduced that this was due to a strong interaction between Tmax and log(Qm). We performed variance inflation factors (VIF) - which detect for multicollinearity - on the models and found that each of them had significant correlation. Where a VIF value of 5 or 10 indicated high levels of interaction between the terms, we found levels of 1000 and higher (O’Brien, 2007). The VIF values for all the parameters of the Full model can be found in Appendix C. Graham (2003) and Farrar and Glauber (1967) report that multicollinearity can lead to improper inferences with respect to coefficient estimates, but the predictive power of the  13  model is not hindered. Due to the redundancy effect of including Tmax in the model, it was removed to increase the validity of the specific predictor variable (log(Qm)). 3.3 Discharge Levels  Figure 5. The Summer (A) and Winter (B) mean discharge against the rate of the change in daily discharge for all periods.  Initial graphs of the stream discharge data were very complicated and convoluted. In order to retrieve some semblance of order and results from the data, the period comparison was limited to two, and discharge levels were separated. The ultimate goal was to be able to fit linear models to the data, and the discharge levels were founded on trends garnered from Fig. 5 above. In summer, a Qm ≥ 0.25 m3s-1 appeared to show similar results, so this was accorded High. Additionally, 0.020 m3s-1 < Qm ≤ 0.05 m3s-1 and 0.050 m3s-1 < Qm < 0.25 m3s-1 were labeled Low and Medium respectively. The winter data did not show as much variation over the mean discharge (Qm), so they were segmented into only two discharge levels, Low, where 0.020 m3s-1 < Qm ≤ 0.25 m3s-1 and High, where Qm > 0.25 m3s-1. Using this method allowed us to develop specific linear models to each comparison and discern significant differences between them. 14  Table 1. The p-value results from the type III ANOVA F-tests comparing all the periods and discharge levels. a Discharge levels are defined in section 3.2. b Reduced model is the Full model without third order interactions e.g. Qm:Tmax:P. c Simple model has removed Tmax. d Simple model with data points where Qm ≤ 0.02 m3s-1 are included in the regression.  Season Summer --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------Winter ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------  Period Comparison  Discharge a Level  Full Model  Reduced b Model  Simple c Model  Pre vs. Post 1  All  6.213e-14 ***  2.277e-14 ***  2.804e-15 ***  Pre vs. Post 1  Low  0.4893  0.5412  0.3706  Pre vs. Post 1  Medium  1.144e-14 ***  8.907e-15 ***  1.613e-15 ***  Pre vs. Post 1  High  Pre vs. Post 2  All  Pre vs. Post 2  Low  Pre vs. Post 2  Medium  Pre vs. Post 2  High  Pre vs. Logging  All  Pre vs. Logging  Low  Pre vs. Logging  Medium  Pre vs. Logging  High  Logging vs. Post 1  All  Logging vs. Post 1  Low  Logging vs. Post 1  Medium  Logging vs. Post 1  High  0.5246  0.5942  0.7263  2.2e-16 ***  2.2e-16 ***  < 2.2e-16 ***  0.76  0.8833  0.9986  2.2e-16 ***  2.2e-16 ***  < 2.2e-16 ***  0.4701  0.3859  0.2287  3.226e-08 ***  7.862e-09 ***  3.45e-06 ***  0.01241 *  0.009826 **  0.01225 *  2.865e-07 ***  1.314e-07 ***  3.277e-08 ***  Simple d Model  ----------0.01078 *  ------------------------------0.02809 *  ------------------------------0.008687 **  0.1214  0.08754  0.05453 .  2.2e-16 ***  2.2e-16 ***  < 2.2e-16 ***  -------------------------------  0.001788 **  0.000641 ***  0.000264 ***  0.0001311 ***  2.362e-09 ***  8.9e-10 ***  4.391e-10 ***  0.03023 *  0.02101 *  0.07944 .  -------------------------------  Logging vs. Post 2  All  2.2e-16 ***  2.2e-16 ***  < 2.2e-16 ***  Logging vs. Post 2  Low  0.001006 **  0.000393 ***  0.0001548 ***  Logging vs. Post 2  Medium  2.2e-16 ***  2.2e-16 ***  2.2e-16 ***  Logging vs. Post 2  High  0.917  0.8109  0.9203  Post 1 vs. Post 2  All  0.1278  0.1087  0.1346  Post 1 vs. Post 2  Low  0.3634  0.2517  0.09923 .  Post 1 vs. Post 2  Medium  0.1228  0.8767  0.9125  Post 1 vs. Post 2  High  0.3098  0.2569  0.2535  Pre vs. Post 1  All  0.0005184 ***  0.0005605 ***  0.0003253 ***  Pre vs. Post 1  Low  0.0009649 ***  0.0003459 ***  6.272e-05 ***  Pre vs. Post 1  High  0.13  0.08707 .  0.03284 *  Pre vs. Post 2  All  0.001338 **  0.0008018 ***  0.01114 *  Pre vs. Post 2  Low  0.0008471 ***  0.0004193 ***  0.002013 **  Pre vs. Post 2  High  0.4404  0.522  0.3531  Pre vs. Logging  All  0.01109 *  0.04163 *  0.03998 *  Pre vs. Logging  Low  0.007551 **  0.003261 **  0.01691 *  Pre vs. Logging  High  0.1218  0.07402 .  0.2124  Logging vs. Post 1  All  0.4265  0.5397  0.3563  Logging vs. Post 1  Low  0.644  0.4716  0.01549 *  Logging vs. Post 1  High  0.6769  0.505  0.9203  Logging vs. Post 2  All  0.02201 *  0.06787 .  0.5619  Logging vs. Post 2  Low  0.838  0.7457  0.5692  Logging vs. Post 2  High  0.03151 *  0.01463 *  0.6468  Post 1 vs. Post 2  All  0.1559  0.1242  0.1005  Post 1 vs. Post 2  Low  0.6855  0.5428  0.09106  0.2436  0.1608  0.1943  Post 1 vs. Post 2 High Significance codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1  0.009235 **  ------------------------------0.02724 *  ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------  15  3.4 Road Effects The results of the significance tests for the winter period comparisons shown in Table 1 allow us to secularize the road effects on the recession streamflows. Pre and Post 1 comparisons were found to be significant at a Low discharge level for all model types, but only significant in the High discharge level with the simplified model. Pre vs. Post 2 was only significant in all models at Low and there was a significant difference between the Logging vs Post 1 period at Low for the simple model. Pre and Logging was found to have a p-value ≈ 0.017 at Low discharge and there was no difference between the postlogging periods at any discharge level or model. However, it should be noted that Pre vs. Post 1 Low and Pre vs. Logging Low linear models did not satisfy the Shapiro-Wilk tests for normality (Appendix B). Table 2. The estimated coefficients for the winter linear models, showing only periods where there was found to be significance when tested.  Period Pre-Logging Low Pre-Logging High Logging Low Post-Logging 1 Low Post-Logging 1 High Post-Logging 2 Low  b0 Intercept -1.62 -0.09 -0.25 1.44 0.29 0.15  b1 |log(Qm)| 3.22 2.20 2.35 1.34 2.11 2.10  16  Figure 6. The pre-logging vs. the post-logging 1 period at both low (a) and high (b) discharge levels during the Winter.  Figure 6a show that as smaller mean discharges are reached, there is a greater difference in the |dQ/dt| between the Pre and Post 1 periods. During Post 1, |dQ/dt| decreases more slowly at a lower mean discharge. 3.5 Forest Harvest and Regeneration The harvesting effects were studied by using a comparison between the Pre and Post 1 summer periods. F-tests were performed on these models (see Table 1) and it was found that there was a highly significant difference between the two periods at the Medium discharge level for all model types (Simple Model p-value = 1.6 x10-15). The mean discharge parameter coefficients were found to be 1.76 for Post 1, 1.84 for Pre and Post 2 was equal to 1.70. The intercept parameters for Pre, Post 1 and Post 2 were 1.1, 0.31, and 0.42. A graphical visualization of the harvest effects is seen in Figure 7.  17  Figure 7. (a-c) The Pre and Post 1 growing season comparison showing non-significant differences for Low and High, but significance at Medium. (d-f) The comparison of Post-Logging 1 vs. Post-Logging 2 highlighting the effect of forest regeneration.  The effect of the restored transpiration to the catchment can be observed by comparing the summer Post 1 and Post 2 periods. Figure 7 graphs d-f show the period comparison at  18  all the discharge levels. However, the periods are not significantly different, with Low, Medium and High p-values ~ 0.10, 0.91, and 0.25 respectively (see Table 2).  Figure 8. The low discharge level for Pre vs. Post 1 and Post 1 vs. Post 2, with the inclusion of Qm data below 0.02 m3s-1.  The omittance of the data that fell below the mean discharge threshold, Qm > 0.02 m3s-1, was due to the uneven distribution of those extremely low flows across the study periods. However, we did an analysis on the Low discharge levels, the only levels that would be affected, with the omitted values inserted back into the dataset and found that there was some very strong significance gained. Figure 8 above shows the Pre vs. Post 1 and Post 1 vs. Post 2 new graphs, which differ from the previous graphs in Figure 7. There is a greater separation between the Pre and Post 1 slopes, with the greatest change occurring in the post-logging periods The parameter estimates in Appendix D show that the new pre-logging slope is shallower compared to the original, 1.38 to 1.56. As well, the intercept doubles from 0.64 to 1.3. The new slopes for Post 1 and 2 are 0.84 and 0.61,  19  compared to 1.2 and 1.6 with the data omitted. This led to Pre vs. Post 1, Pre vs. Post 2 and Post 1 vs. Post 2 comparisons becoming significantly different (Table 2).  4. Discussion 4.1 Road Effects Figure 6a singles out the effect of the roads on the catchment and shows that from the pre period to the post 1 period there was a significant steepening of the recession curve. The road network created more hill flow mechanisms, Hortonian overland flow along the road surfaces and pathways (i.e. ditches) for the water to reach the stream. As well, the cut banks allowed the recapture of through-flow back into the ditch drainage system, which would otherwise slowly move downslope. The data indicates that during both High and Low discharge levels there are differences between the Pre and Post 1 periods, even though Figure 8b does not show this very clearly. Looking at the estimated coefficients for the comparison (Appendix D) shows that they are indeed similar, but there are enough data points to increase the degrees of freedom coupled with a strong normal distribution, both of which could increase the statistical power to detect even subtle changes to the streamflow (Table 2). Roads and drainage systems are unique in that they persist on the landscape with a fixed effect. Unless the roads are dismantled and semi-natural drainage patterns are reestablished, then the effect of the roads should not change over time. There were no significant differences between the post-logging periods during winter. This agrees with our hypothesis that the effect of the roads does not change over time. However, looking at Logging vs. Post 2 winter shows some significance at the High discharge level for the Full and Reduced models, unusual because there is no significance during the same level 20  in the summer and none for the previous period of Post 1. Removal of the two highest outliers in the winter Logging vs. Post-Logging 2 yielded a non-significant p-value of 0.0517. There could be several explanations for this occurrence. It could possibly be attributed to additional road building and forest operation effects that would have taken place to allow for additional harvesting. A second pass of logging began in 1987, taking place in winter, so it is reasonable to assume that this could have had some effect on the stream discharge (i.e. diverting more water to ditches and culverts and faster flows to the stream). An additional explanation could have something to do with the decreased MAP occurring during the Post 2 period, compared to the Logging period. If there were overall less water flowing through the system then the data would show a lowered mean discharge. Then large events, like the ones we removed, would have a large effect, increasing the variance and mean, but would not be indicative of the actual site processes. However, given the high VIF values we found for those models parameters (Appendix C), the lack of normality (Appendix B), and the fact that once Tmax was removed there was a definitive non-significance between the periods, we have reason to believe our hypothesis was correct in predicting a persistence of the effects of the roads over the course of the study period. Looking at the coefficient estimates output in Appendix D shows that the winter Low Pre and Post 2 log(Qm) coefficients are 3.2 and 2.1 respectively. This means that because the slope was shallower throughout the Post 2 period, the rate of change in daily discharge continued at a higher level during the same mean discharge. With a p-value of 0.002, there appears to be a long-lasting effect of the roads in steepening the recession curve from pre-harvest levels. 21  It is interesting to note that there is no difference between the Pre vs. Post 2 High discharge level. This would seem to indicate that for a short period after road building, the effect on streamflow recession curves extended to all discharge levels, but after a time it is only noticeable at the lowest flows. Because the effect of the roads was not changing appreciably, and there was negligible evapotranspiration occurring, then the forest regeneration must have been extending beyond its transpirational effect. 4.2 Forest Harvest and Regeneration The effect of the roads during the Post 1 period is similar to the transpirational effects of the forest cover observed in the pre-growing season, in that both steepened the recession curve during low mean discharges. As well, large woody debris (LWD) was removed systematically from the channel at most locations, so less blockage and sediment build up at those locations probably helped increase discharge levels during all periods (Province of British Columbia, 2009; Hetherington, 1988). Watersheds containing highly valuable fish populations and spawning grounds, endemic riparian species, or sensitive ecosystems in or downstream could be negatively affected by changes to the recession curves at low flows. At the Carnation Creek site there were actual increases in salmonid productivity as a result of slightly increased growing temperatures (Province of British Columbia, 2009). Because of forest harvesting, the percentage of precipitation falling on the catchment, which is subsequently being drawn out of the soil by hydraulic lifting associated with plant respiration, has been decreased. More water is able to infiltrate the groundwater system and feed the stream system through recession periods. However, the ability of the harvest to offset the effect of the roads is dependent on the mean discharge, and the  22  reduction in transpiration is temporary as regeneration begins to establish around the catchment. The difference between the Pre and Post 1 and 2 periods means that the effect of the roads is strong all year round, and because there is a decreasing slope in almost all comparisons from Pre, one can conclude that the reduced transpiration partly offsets the effect of the roads, but not completely. The one exception is Post 2 Low discharge level, where the slope had increased past the Pre levels - 1.56 to 1.59. I am skeptical of those results for a number of reasons. The first is due to the large number of points we omitted from the analysis. The distribution of extreme low flow data points among the periods was as follows: 7 points in late summer of Pre period, 21 in Post 1 period, and 60 in Post 2 period. They fell between June and October, with the majority of points coming during August. Because of the high concentration of low-flows occurring in the Post-Logging 2 period, the exclusion of these led to ignorance of information on low-flows, which may or may not have been the result of the roads and the forest regeneration. The second is that there was an additional 21% of the catchment harvested during the Post 2 period, which focussed on the headwaters. Without a control sub-catchment, or a group of subcatchments that were able to account for the second logging pass, the results of the full interaction between harvesting and roads remains ambiguous. In the comparison between the summer Post 1 and Post 2 periods there was found to be no significance. There are a number of possibilities for the lack of differences between the two periods in the Simple model. There could not be enough time difference between them, so that any increased steepening would not be noticeable. In addition to this, there could be a logarithmic increase in the effect of regeneration, so that as the time since the 23  first establishment of the trees increases, the change in their regeneration effect becomes smaller. Eventually this would reach some asymptotic plateau where no change would take place. Such a phenomenon could very likely be the case as the roots of grasses, shrubs and the regenerating trees are able to remove most of the soil water for their respiration uses at only 5 years following disturbance (Winkler et al., 2011). It is interesting to note that Hetherington et al. (1998) found that at the end of summer, the clear-cut areas soils were wetter than the forested soils. Even though forests may increase infiltration rates, provide cooler soil temperatures, and reduce solar radiation, the effect of their water requirements - especially during the water limited growing season removes a large portion of the water in the upper soil profile. The Post 1 vs. Post 2 comparison with the omitted values yielded a significant relationship at the Low discharge level, p-value ≈ 0.027. The Post 2 slope estimate is shallower, indicating a steepened recession curve, which is what would be expected from a regenerating forest.  4.3 Study Limitations and Errors Some improvements and errors to the study and analysis have been identified for the purposes of increasing applicability of the data, as well as for the transferability to other watersheds being managed in similar coastal forests around the world. It would have been ideal to segment the harvest periods into more even lengths of time, the Post 1 and 2 periods more specifically. Additionally, we would have liked to exclude the data collected after 1987, when a second logging harvest was conducted on the catchment. Alternatively, there could be a short Logging 2 period followed by an additional Post 3 period added to the time series.  24  Looking back at Figure 4 shows the time of the clear-cuts was highly variable and patchy across the logging period. It was never a consistently applied effect, but instead had periods of regeneration, harvesting, and burning mixed throughout. It would be informative and useful to have a study on the specific effects of harvesting on recession curves by undergoing a full clear-cut during the first year and measuring the initial effect and the recovering hydrologic system. There is a possibility that we could have used Mean Daily Flow (MDF) and obtained from Mean Annual Runoff (MAR) to observe changes to this low-flow indicator. The separation of summer and winter periods could have been precisely calculated yearly by data when plant respiration started in the region, specifically defined per annum. Nevertheless, this would have accounted for only transpirational effects, as evaporation is possible year round. Because the parameter coefficients had such high variance due to multicollinearity, the use of Tmax was limited. I would have possibly liked to use Tavg or Tmin to see if the VIFs were reduced. There is a technique called "regularization" via ridge regression, which penalizes parameter coefficient estimates that are too high. This reduces the variance in the parameter estimates and provides more meaning to the fitted linear models. Using that technique, we might have been able to continue using Tmax. 4.4 Improvements to Research Hetherington (1988) observed that errors in low measurements could have been confounded by a small leak in the weir and the resulting inaccuracy of low flow measurements. Using the Bamfield climate data for precipitation, daily temperature and discharge data to increase the data pool, or using it to remove outliers would have been 25  helpful when analyzing streamflow data. There was a portion of the catchment that was left alone for the entirety of the study, so no forest management occurred in its reaches. It could have been used as a control, similar to a paired watershed, and comparisons made against on a temporal scale (Province of British Columbia, 2009). This could have improved the error due to climatic variation found in long-scale studies. As well, I would have liked to use the other weir data from around the experiment site, corresponding to separately managed sub-catchments. Noteable would have been to see if a correlation exists between changes to tree volume, species type, and the resulting increase in depth and expanse of a soil water retaining layer following harvesting. Using the correlation to compare the different effects of forest management prescriptions: thinning, pruning, and fertilizer treatments would also have been interesting, but there might not be enough change to detect using streamflow data. However, if the treatments were large enough, the effect on |dQ/dt| might be observable. In the case of future watershed experiments, if possible, road building should commence for a cutblock. In this case harvesting would be delayed for a few years so that a prelogging period could be observed isolating the effect of road building and culverts, separate from the effect of logging. This would be interesting, but might be a costly delay for the forest managers. It would have been interesting to have obtained extended discharge data up to the present for the purpose of better understanding the effects of forest regeneration on recession curve steepening following forest harvest. Because of the small time frame (in regeneration terms) on the analyzed post-logging periods, the results are ambiguous.  26  An experimental design involving the road system of a cutblock being deactivated according to the Watershed Restoration Act (Atkins et al., 2001 in Jordan et al., 2010) would provide the opportunity to study the effect of road dismantling and the return time, in such a case, of natural flow paths. Discharge data could continue to be collected and observed to determine if the effect of filling in cut-banks and removing bridges and culverts would be able to bring the recession curve back to pre-logging levels. This could fuel research towards determining the time lengths of the restoration process and underscore the ecological responsibility that Forest Companies have for the land after harvesting, which includes “Free growing” status.  5. Conclusion At the Carnation Creek Experimental Watershed site, stream discharge data was collected over a period of 30 years and separated by Pre-Logging, Logging, Post-Logging 1, and Post-Logging 2 times. The results showed clearly that the effect of roads, isolated in the non-growing season, had a large steepening effect on the recession curve and that it persisted year round. This steepening was affected at the low and high mean discharges during winter in the short term, but persisted only during the lowest flows in the long term. Forest harvesting, which focussed on the valley bottoms during the first pass, partially offset the steepening by the roads, but only at a medium discharge level. However, when extreme low-flow data was re-submitted into the linear models, both low and medium discharge levels were found to be significantly differ from pre-logging to the post-logging periods. This offset was only temporary as regeneration of the catchment began to steepen the curve once more. A significant effect of regeneration from post-  27  logging period 1 to 2 was only acquired at the low discharge level with the inclusion of the omitted data. This is likely due to the catchment already reaching maximum transpiration effects in a short period. The research conducted at Carnation Creek sheds a stronger light on the lasting effect of forest harvest operations on coastal watersheds. The results from past multi-disciplinary experiments completed there, and ones continuing today, have influenced forestry policies in the province of British Columbia and around the world. The ability for multiple groups, organizations and a variety of specialized scientists to cooperate for a single mission is a strong testament to the desire to maintain our natural landscapes with more sustainable management operations.  28  6. Works Cited Province of British Columbia – Fish-Forestry Interaction Research (FFIP), 2009. Carnation Creek Project. [online] Available at: http://www.for.gov.bc.ca/hre/ffip/CarnationCrk.htm [Accessed Nov. 17, 2011] Beschta, R. L., Bilby, R. E., Brown, G. W., Holtby, L. B. and T. D. Hofstra, 1987. Stream temperature and aquatic habitat: Fisheries and forestry interactions. In: Streamside management: Forestry and fishery interactions, 1986. Salo, E. O. and T. W. Cundy (editors). Contrib. 57. Seattle, WA. Inst. Of For. Res., Univ. of Washington, Seattle, WA. Hartman, G. F., Scrivener, J. C. and M.J. Miles, 1996. Impacts of logging in Carnation Creek, a high-energy coastal stream in British Columbia, and their implication for restoring fish habitat. Can. J. Fish. Aquat. Sci. 53(Suppl. 1): 237–251. Brandes, D., Hoffmann, J. G. and J. T. Mangarillo, 2005. Base flow recession rates, low flows, and hydrologic features of small watersheds in Pennsylvania, USA. Journal of the American Water Resources Association 41(5):1177-1186. Brutsaert, W. and J. Nieber, 1977. Regionalized drought flow hydrographs from a mature glaciated plateau. Water Resources Research 13: 637-643. Farrar D. E. and R. R. Glauber, 1967. Multicollinearity in regression analysis: the problem revisited, The Review of Economics and Statistics 49(1):92-107. Fannin, R. J., Jaakkola, J., Wilkinson, J. M. T. and E. D. Hetherington, 2000. Hydrologic response of soils to precipitation at Carnation Creek, British Columbia, Canada: Reproduced/modified by permission of American Geophysical Union. Water Resour. Res. 36:1481-1494. Federer, C., 1973. Forest transpiration greatly speeds streamflow recession. Water Resources Research 9: 1599-1604. Graham, M. H., 2003. Confronting multicollinearity in ecological multiple regression. Ecology 84:2809–2815. Harr, R. D., 1979. Effects of streamflow in the rain-dominated portion of the Pacific Northwest. In: Proceedings of Workshop on scheduling timber harvest for hydrologic concerns, Pacific Northwest Forest and Range Experiment Station, Portland, U.S., Department of Agriculture, Oreg., 1-45. Hartman, G. F. and J. C. Scrivener (editors), 1990. Impacts of forestry practices on a coastal stream ecosystem, Carnation Creek, British Columbia. Department of Fisheries and Oceans, Ottawa, Ont. Can. Bull. Fish. Aquat. Sci. 223. Hetherington, E. D., 1998. Watershed Hydrology. In: Hogan, D. L., Tschaplinski, P. J. and S. Chatwin (editors). Carnation Creek and Queen Charlotte Islands fish/forestry workshop: applying 20 years of coastal research to management solutions. B.C. Min. For. Res. Br., Victoria, B.C. Land Manag. Handb. 41, pp. 33-40. 29  Hetherington, E. D., 1988. Hydrology and logging in the Carnation Creek watershedwhat have we learned? In: Proc. workshop: applying 15 years of Carnation Creek results. Chamberlin, T. W. (editor). Pacific Biological Station, Nanaimo, B.C., pp. 11-15. Hicks, B. J., Beschta, R. L. and R. D. Harr, 1991. Long-term changes in streamflow following logging in western Oregon and associated fisheries implications. 1991. Water Resources Bulletin 27: 217-226. Jones, J. A., 2000. Hydrologic processes and peak discharge response to forest removal, regrowth, and roads in 10 small experimental basins, Western Cascades, Oregon. Water Resources Research 36: 2621-2642. Jordan, P., Millard, T. H., Campbell, D., Schwab, J. W., Wilford, D. J., Nicol, D. and D. Collins, 2010. Forest management effects on hillslope processes – Chapter 9 In: Pike, R. G., T.E. Redding, R. D. Moore, R. D. Winker and K. D. Bladon (editors). Compendium of forest hydrology and geomorphology in British Columbia. B.C. Min. For. Range, For. Sci. Prog., Victoria, B.C. and FORREX Forum for Research and Extension in Natural Resources, Kamloops, B.C. Land Manag. Handb. 66. www.for.gov.bc.ca/hfd/pubs/Docs/Lmh/Lmh66.htm (Accessed Nov 2011). Keppeler, E. and R. Ziemer, 1990. Logging effects on streamflow: water yield and summer low flows at Caspar Creek in northwestern California, Water Resources Research 26: 1669-1679. Kozak, A., Kozak, R. A., Staudhammer, C. L. and S. B. Watts, 2008. Introductory Probability and Statistics: Applications for Forestry and Natural Sciences. Cambridge: Cambridge University Press. 1-408. Moore, R. D., 1997. Storage-outflow modelling of streamflow recessions, with application to a shallow-soil forested catchment. Journal of Hydrology 198: 260-270. O'Brien, R. M., 2007. A caution regarding rules of thumb for variance inflation factors. Quality and Quantity 41(5):673-690. Price, K., 2011. Effects of watershed topography, soils, land use, and climate on baseflow hydrology in humid regions: a review. Progress in Physical Geography 35: 465-492. R Development Core Team, 2011. R: A language and environment for statistical computing, R foundation for statistical computing, ISBN 3-900051-07-0, available at: http://www.R-project.org. Scrivener, J. C., 1975. Water, water chemistry and hydrochemical balance of dissolved ions in Carnation Creek watershed, Vancouver Island, July 1970-May 1974. Fish. Mar. Serv. Tech. Rep. 564. Shimokura, J. and H. Shibano, 2003. Effects of forest restoration in mountainous basins on the long-term change in baseflow recession constants. Water Resources Systems— 30  Hydrological Risk, Management and Development (Proceedings of Symposium IS02b held during IUGG 2003 at Sapporo, July 2003). IAHS Publ. no. 281. 2003. Smakhtin, V. U., 2001. Low flow hydrology: a review. Journal of Hydrology 240: 147186. Smerdon, B. D., Redding, T. E., and J. Beckers, 2009. An overview of the effects of forest management on groundwater hydrology. BC Journal of Ecosystems and Management 10(1): 22–44. Tallaksen, L. M., 1995. A review of baseflow recession analysis. Journal of Hydrology 165: 349-370. Waterloo, M. J., Schellekens, S., Bruijnzeel, L.A., and T. T. Rawaqa, 2007. Changes in catchment runoff after harvesting and burning of a Pinus caribaea plantation in Viti Levu, Fiji. Forest Ecology and Management 251: 31–44. Winkler, R. D., Moore, R. D., Redding, T. E., Spittlehouse, D. L., Smerdon, B. D. and D. E. Carlyle-Moses, 2011. The effects of forest disturbance on hydrologic processes and watershed response. Chapter 7. In: Pike, R. G., Redding, T. E., Moore, R. D., Winkler, R. D. and K. D. Bladon (editors). Compendium of forest hydrology and geomorphology in British Columbia. B.C. Min. For. Range, For. Sci. Prog., Victoria, B.C. and FORREX Forum for Research and Extension in Natural Resources, Kamloops, B.C. Land Manag. Handb. 66, 179-202. www.for.gov.bc.ca/hfd/pubs/Docs/Lmh/Lmh66.htm [Accessed Nov 2011].  31  7. Appendices 7.1 Appendix A – Residual Plots  32  33  7.2 Appendix B – Shapiro-Wilk Normality Tests Table 3. The W test statistics and p-values of linear model residuals using a Shapiro-Wilk normality test.  Period Comparison  Discharge Level  ---------------------------------------------------------------------------------------------------------------  Pre vs. Post 1 Pre vs. Post 1 Pre vs. Post 1 Pre vs. Post 2 Pre vs. Post 2 Pre vs. Post 2 Pre vs. Logging Pre vs. Logging Pre vs. Logging Logging vs. Post 1 Logging vs. Post 1 Logging vs. Post 1  Low Medium High Low Medium High Low Medium High Low Medium High  0.9804 0.9952 0.9553 0.9902 0.9924 0.9212 0.9654 0.9918 0.9461 0.9763 0.9869 0.9744  0.3064 0.7768 0.1245 0.5945 0.1924 0.002332 0.1497 0.03312 0.004182 0.1237 0.00455 0.2997  -----------------------------------------  Logging vs. Post 2 Logging vs. Post 2 Logging vs. Post 2 Post 1 vs. Post 2  Low Medium High Low  0.9936 0.9878 0.9864 0.994  0.8482 0.002233 0.6864 0.8055  Season Summer  W (test stat)  p-value  34  ---------------------  Post 1 vs. Post 2 Post 1 vs. Post 2  Medium High  0.9931 0.9679  0.4263 0.4071  Winter  Pre vs. Post 1 Pre vs. Post 1 Pre vs. Post 2 Pre vs. Post 2 Pre vs. Logging Pre vs. Logging Logging vs. Post 1 Logging vs. Post 1 Logging vs. Post 2 Logging vs. Post 2 Post 1 vs. Post 2 Post 1 vs. Post 2  Low High Low High Low High Low High Low High Low High  0.8262 0.9815 0.9849 0.814 0.9683 0.7636 0.9799 0.6835 0.9876 0.7129 0.9869 0.7983  2.06E-07 0.9849 0.4402 5.69E-09 0.03129 8.48E-10 0.2273 5.32E-11 0.4498 7.94E-12 0.6113 5.84E-09  ---------------------------------------------------------------------------------------------------------------  7.3 Appendix C – FULL Model VIF Results FULL MODEL |log(dQ/dt)| = |log(Qm)| * Tmax * P Summer  Pre-Logging vs Post-Logging 1 – MEDIUM Qm Tm P Qm:Tm Qm:P Tm:P  Qme:Tm:P  70.899  1284.704  70.818  1067.314  189.880  1110.029  1190.646  Pre-Logging vs Logging LOW Qm Tm P Qm:Tm  Qm:P  Tm:P  Qm:Tm:P  111.0589  25913.2011  29890.4624  28532.7548  Pre-Logging vs Logging MEDIUM Qm Tm P Qm:Tm Qm:P  Tm:P  Qm:Tm:P  122.4904  1340.8997  1636.8703  633.9565  122.2065  26912.7832  1252.1269  473.3785  266.0358  1542.7001  Pre-Logging vs Post-Logging 2 MEDIUM Qm Tm P Qm:Tm Qm:P Tm:P  Qm:Tm:P  98.50343  1140.05448  1329.93133  Logging vs Post-Logging 1 LOW Qm Tm P Qm:Tm Qm:P  Tm:P  Qm:Tm:P  942.4994  37620.4959  43955.8644  Logging vs Post-Logging 1 MEDIUM Qm Tm P Qm:Tm Qm:P  Tm:P  Qm:Tm:P  48.50840  1088.13421  1039.11345  83.90610  2763.5658  46.14674  1119.53724  38260.4530  979.38251  185.93440  3310.6260  111.80573  1330.54212  45263.5092  911.21611  35  Logging vs Post-Logging 1 HIGH Qm Tm P Qm:Tm Qm:P  Tm:P  Qm:Tm:P  106.63796  455.69421  465.39758  Logging vs Post-Logging 2 LOW Qm Tm P Qm:Tm Qm:P  Tm:P  Qm:Tm:P  1564.485  51236.734  63717.344  Logging vs Post-Logging 2 MED Qm Tm P Qm:Tm Qm:P  Tm:P  Qm:Tm:P  59.04932  1005.53322  1078.48261  15.79913  6014.272  52.17458  409.67210  42925.198  1001.00505  125.52866  10294.648  111.73833  414.70749  52039.472  1082.43413  Winter Pre-Logging vs Post-Logging 1 LOW Qm Tm P Qm:Tm Qm:P 5.924  230.0319  128.0639  200.174  120.432  Pre-Logging vs Post-Logging 2 LOW Qm Tm P Qm:Tm Qm:P  Tm:P  Qm:Tm:P  359.355  308.610  Tm:P  Qm:Tm:P  164.13838  375.07257  457.28022  Pre-Logging vs Logging LOW Qm Tm P Qm:Tm  Qm:P  Tm:P  Qm:Tm:P  9.482334  288.03581  556.241031  602.372525  Logging vs Post-Logging 2 HIGH Qm Tm P Qm:Tm Qm:P  Tm:P  Qm:Tm:P  13.00545  39.06798  35.81650  10.52616  227.26329  246.567591  21.46418  144.84047  260.038753  31.09494  277.70883  262.315168  26.50309  28.79800  36  7.4 Appendix D – Coefficient Estimates (Simple Model) Table 4. The simple model intercept and log(Qm) coefficient estimates. The gray column indicates coefficients where extreme low-flow data is included in the regression.  Season  Period  Discharge Level  Dependent Variable  b0  b1  b0  b1  Intercept  |log(Qm)|  Intercept  |log(Qm)|  Low  |log(dQ/dt)|  0.6356804  1.5641538  1.28213  1.378929  Medium  -----------  1.100859  1.839675  -----------  -----------  High Low Medium High  -----------------------------------------  -0.28643 -1.9003 -0.25825 -0.09023  2.3234 2.56671 2.2694 1.96925  -----------1.90033 ---------------------  ----------2.5667 ---------------------  Low  -----------  1.595655  1.2130  2.84022  0.83740  Medium  -----------  0.3081784  1.7615  -----------  -----------  High  -----------  -0.01256  2.095797  -----------  -----------  Low  -----------  0.53390  1.59421  3.91146  Medium  -----------  0.418238  1.700284  -----------  -----------  High  -----------  -0.04554  1.94557  -----------  -----------  Low  |log(dQ/dt)|  -1.62  3.216438  -----------  -----------  High Low High  -------------------------------  -0.0863 -0.248 0.299  2.198623 2.35121 2.043610  -------------------------------  -------------------------------  Low  -----------  1.4431  1.340731  -----------  -----------  High  -----------  2.93E-01  2.1087  -----------  -----------  Low  -----------  0.154991  2.10427  -----------  -----------  High  -----------  9.06E-02  2.17378  -----------  -----------  Pre-  Summer Logging ---------------------------------------------------  -----------  -----------  -----------  -----------  -----------  -----------  Winter -------------------------------  -----------  -----------  -----------  -----------  PreLogging PreLogging Logging Logging Logging PostLogging 1 PostLogging 1 PostLogging 1 PostLogging 2 PostLogging 2 PostLogging 2 PreLogging PreLogging Logging Logging PostLogging 1 PostLogging 1 PostLogging 2 PostLogging 2  0.60812  37  7.5 Appendix E – ANOVA Tests Summer Pre-Logging vs Post-Logging 1 Medium Analysis of Variance Table Model 1: y ~ Qmean * P Model 2: y ~ Qmean Res.Df RSS Df Sum of Sq F Pr(>F) 1 196 100.55 2 198 142.34 -2 -41.791 40.73 1.613e-15 *** --Call: lm(formula = y ~ Qmean) Residuals: Min 1Q -2.34860 -0.55893  Median 0.02459  3Q 0.62177  Max 1.82919  Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 1.3381 0.3409 3.926 0.000119 *** Qmean 1.5668 0.1535 10.209 < 2e-16 *** --Residual standard error: 0.8479 on 198 degrees of freedom Multiple R-squared: 0.3449, Adjusted R-squared: 0.3415 F-statistic: 104.2 on 1 and 198 DF, p-value: < 2.2e-16  Pre-Logging vs Post-Logging 2 Medium Analysis of Variance Table Model 1: y ~ Qmean * P Model 2: y ~ Qmean Res.Df RSS Df Sum of Sq F Pr(>F) 1 260 103.24 2 262 164.98 -2 -61.74 77.746 < 2.2e-16 *** --Call: lm(formula = y ~ Qmean) Residuals: Min 1Q Median -1.78151 -0.56702 -0.05287  3Q 0.62415  Max 1.96863  Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 1.3276 0.2740 4.845 2.17e-06 *** Qmean 1.5005 0.1219 12.314 < 2e-16 *** --Residual standard error: 0.7935 on 262 degrees of freedom Multiple R-squared: 0.3666, Adjusted R-squared: 0.3642 F-statistic: 151.6 on 1 and 262 DF, p-value: < 2.2e-16  Pre-Logging vs Logging  38  Low Analysis of Variance Table Model 1: y ~ Qmean * P Model 2: y ~ Qmean Res.Df RSS Df Sum of Sq F Pr(>F) 1 46 13.521 2 48 16.373 -2 -2.8521 4.8514 0.01225 * --Call: lm(formula = y ~ Qmean) Residuals: Min 1Q Median -1.3439 -0.3372 -0.1700  3Q 0.4555  Max 0.9287  Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 3.6022 1.0914 3.301 0.00183 ** Qmean 0.7747 0.3307 2.342 0.02337 * --Residual standard error: 0.584 on 48 degrees of freedom Multiple R-squared: 0.1026, Adjusted R-squared: 0.08387 F-statistic: 5.486 on 1 and 48 DF, p-value: 0.02337  Pre-Logging vs Logging Medium Analysis of Variance Table Model 1: y ~ Qmean * P Model 2: y ~ Qmean Res.Df RSS Df Sum of Sq F Pr(>F) 1 376 173.43 2 378 190.08 -2 -16.65 18.048 3.277e-08 *** --Call: lm(formula = y ~ Qmean) Residuals: Min 1Q Median -1.6765 -0.5088 -0.0458  3Q 0.5034  Max 2.3761  Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 0.38255 0.22068 1.734 0.0838 . Qmean 2.04807 0.09765 20.973 <2e-16 *** --Residual standard error: 0.7091 on 378 degrees of freedom Multiple R-squared: 0.5378, Adjusted R-squared: 0.5366 F-statistic: 439.8 on 1 and 378 DF, p-value: < 2.2e-16  Logging vs Post-Logging 1 Low Analysis of Variance Table Model 1: Model 2: Res.Df 1 80 2 82  y ~ Qmean * P y ~ Qmean RSS Df Sum of Sq F Pr(>F) 40.175 49.364 -2 -9.1897 9.1498 0.000264 ***  39  --Call: lm(formula = y ~ Qmean) Residuals: Min 1Q -1.80364 -0.59476  Median 0.01979  3Q 0.74740  Max 1.15409  Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 4.1118 1.1341 3.626 0.000499 *** Qmean 0.5351 0.3415 1.567 0.120951 --Residual standard error: 0.7759 on 82 degrees of freedom Multiple R-squared: 0.02908, Adjusted R-squared: 0.01724 F-statistic: 2.456 on 1 and 82 DF, p-value: 0.121  Logging vs Post-Logging 1 Medium Analysis of Variance Table Model 1: Model 2: Res.Df 1 324 2 326 ---  y ~ Qmean * P y ~ Qmean RSS Df Sum of Sq F Pr(>F) 161.42 184.38 -2 -22.962 23.045 4.391e-10 ***  Call: lm(formula = y ~ Qmean) Residuals: Min 1Q Median -2.40908 -0.45502 -0.05679  3Q 0.46849  Max 2.69964  Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -0.04158 0.24114 -0.172 0.863 Qmean 2.11563 0.10417 20.310 <2e-16 *** --Residual standard error: 0.7521 on 326 degrees of freedom Multiple R-squared: 0.5586, Adjusted R-squared: 0.5572 F-statistic: 412.5 on 1 and 326 DF, p-value: < 2.2e-16  Logging vs Post-Logging 2 Low Analysis of Variance Table Model 1: y ~ Qmean * P Model 2: y ~ Qmean Res.Df RSS Df Sum of Sq F Pr(>F) 1 119 45.708 2 121 52.969 -2 -7.2617 9.4529 0.0001548 *** --Call: lm(formula = y ~ Qmean) Residuals: Min 1Q -1.94270 -0.40241  Median 0.02167  3Q 0.41775  Max 1.22065  Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 2.2107 0.7627 2.898 0.00445 **  40  Qmean 1.1409 0.2260 5.048 1.59e-06 *** --Residual standard error: 0.6616 on 121 degrees of freedom Multiple R-squared: 0.174, Adjusted R-squared: 0.1671 F-statistic: 25.48 on 1 and 121 DF, p-value: 1.593e-06  Logging vs Post-Logging 2 Medium Analysis of Variance Table Model 1: Model 2: Res.Df 1 388 2 390 ---  y ~ Qmean * P y ~ Qmean RSS Df Sum of Sq F Pr(>F) 164.10 203.94 -2 -39.832 47.089 < 2.2e-16 ***  Call: lm(formula = y ~ Qmean) Residuals: Min 1Q Median -1.92259 -0.45322 -0.07069  3Q 0.41757  Max 2.71056  Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 0.08884 0.21042 0.422 0.673 Qmean 2.02073 0.09078 22.259 <2e-16 *** --Residual standard error: 0.7231 on 390 degrees of freedom Multiple R-squared: 0.5596, Adjusted R-squared: 0.5584 F-statistic: 495.5 on 1 and 390 DF, p-value: < 2.2e-16  Winter Pre-Logging vs Post-Logging 1 Low Analysis of Variance Table Model 1: y ~ Qmean * P Model 2: y ~ Qmean Res.Df RSS Df Sum of Sq F Pr(>F) 1 57 11.253 2 59 15.803 -2 -4.5497 11.523 6.272e-05 *** --Call: lm(formula = y ~ Qmean) Residuals: Min 1Q Median -1.03842 -0.38120 -0.06555  3Q 0.29976  Max 1.06296  Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -0.1504 0.4652 -0.323 0.748 Qmean 2.3412 0.2588 9.048 9.52e-13 *** --Residual standard error: 0.5175 on 59 degrees of freedom Multiple R-squared: 0.5811, Adjusted R-squared: 0.574 F-statistic: 81.86 on 1 and 59 DF, p-value: 9.517e-13  41  Pre-Logging vs Post-Logging 1 High Analysis of Variance Table Model 1: y ~ Qmean * P Model 2: y ~ Qmean Res.Df RSS Df Sum of Sq F Pr(>F) 1 63 13.969 2 65 15.569 -2 -1.6001 3.6082 0.03284 * --Call: lm(formula = y ~ Qmean) Residuals: Min 1Q -0.91928 -0.25544  Median 0.03581  3Q 0.19693  Max 2.30808  Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 0.04059 0.11937 0.34 0.735 Qmean 2.21243 0.14043 15.76 <2e-16 *** --Residual standard error: 0.4894 on 65 degrees of freedom Multiple R-squared: 0.7925, Adjusted R-squared: 0.7893 F-statistic: 248.2 on 1 and 65 DF, p-value: < 2.2e-16  Pre-Logging vs Post-Logging 2 Low Analysis of Variance Table Model 1: y ~ Qmean * P Model 2: y ~ Qmean Res.Df RSS Df Sum of Sq F Pr(>F) 1 79 18.768 2 81 21.962 -2 -3.194 6.7223 0.002013 ** --Call: lm(formula = y ~ Qmean) Residuals: Min 1Q Median -1.38063 -0.31796 -0.04818  3Q 0.34044  Max 1.10244  Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -0.1884 0.3698 -0.51 0.612 Qmean 2.3405 0.1953 11.98 <2e-16 *** --Residual standard error: 0.5207 on 81 degrees of freedom Multiple R-squared: 0.6394, Adjusted R-squared: 0.6349 F-statistic: 143.6 on 1 and 81 DF, p-value: < 2.2e-16  Pre-Logging vs Logging Low Analysis of Variance Table Model 1: y ~ Qmean * P Model 2: y ~ Qmean  42  Res.Df RSS Df Sum of Sq F Pr(>F) 1 83 19.138 2 85 21.115 -2 -1.9771 4.2874 0.01691 * --Call: lm(formula = y ~ Qmean) Residuals: Min 1Q Median -0.9038 -0.3722 -0.1053  3Q 0.3884  Max 1.0115  Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) -0.5768 0.3691 -1.563 0.122 Qmean 2.5696 0.1970 13.046 <2e-16 *** --Residual standard error: 0.4984 on 85 degrees of freedom Multiple R-squared: 0.6669, Adjusted R-squared: 0.663 F-statistic: 170.2 on 1 and 85 DF, p-value: < 2.2e-16  Logging vs Post-Logging 1 Low Analysis of Variance Table Model 1: Model 2: Res.Df 1 78 2 80 ---  y ~ Qmean * P y ~ Qmean RSS Df Sum of Sq F Pr(>F) 14.217 15.820 -2 -1.6034 4.3984 0.01549 *  Call: lm(formula = y ~ Qmean) Residuals: Min 1Q Median -0.86968 -0.28090 -0.08049  3Q 0.28776  Max 1.05839  Coefficients: Estimate Std. Error t value Pr(>|t|) (Intercept) 0.2296 0.3169 0.725 0.471 Qmean 2.0778 0.1711 12.146 <2e-16 *** --Residual standard error: 0.4447 on 80 degrees of freedom Multiple R-squared: 0.6484, Adjusted R-squared: 0.644 F-statistic: 147.5 on 1 and 80 DF, p-value: < 2.2e-16  43  

Cite

Citation Scheme:

    

Usage Statistics

Country Views Downloads
France 8 0
China 4 4
United States 3 0
Canada 1 0
City Views Downloads
Unknown 4 1
Shenzhen 4 4
Roubaix 4 0
Ashburn 2 0
Wilmington 1 0
Ottawa 1 0

{[{ mDataHeader[type] }]} {[{ month[type] }]} {[{ tData[type] }]}
Download Stats

Share

Embed

Customize your widget with the following options, then copy and paste the code below into the HTML of your page to embed this item in your website.
                        
                            <div id="ubcOpenCollectionsWidgetDisplay">
                            <script id="ubcOpenCollectionsWidget"
                            src="{[{embed.src}]}"
                            data-item="{[{embed.item}]}"
                            data-collection="{[{embed.collection}]}"
                            data-metadata="{[{embed.showMetadata}]}"
                            data-width="{[{embed.width}]}"
                            async >
                            </script>
                            </div>
                        
                    
IIIF logo Our image viewer uses the IIIF 2.0 standard. To load this item in other compatible viewers, use this url:
http://iiif.library.ubc.ca/presentation/dsp.52966.1-0075628/manifest

Comment

Related Items