Predicting Debris Yield From Burned Watersheds: Comparison of Statistical and Artificial Neural Network Models1

Abstract: Alluvial fans in southern California are continuously being developed for residential, industrial, commercial, and agricultural purposes. Development and alteration of alluvial fans often require consideration of mud and debris flows from burned mountain watersheds. Accurate prediction of sediment (hyper‐concentrated sediment or debris) yield is essential for the design, operation, and maintenance of debris basins to safeguard properly the general population. This paper presents results based on a statistical model and Artificial Neural Network (ANN) models. The models predict sediment yield caused by storms following wildfire events in burned mountainous watersheds. Both sediment yield prediction models have been developed for use in relatively small watersheds (50‐800 ha) in the greater Los Angeles area. The statistical model was developed using multiple regression analysis on sediment yield data collected from 1938 to 1983. Following the multiple regression analysis, a method for multi‐sequence sediment yield prediction under burned watershed conditions was developed. The statistical model was then calibrated based on 17 years of sediment yield, fire, and precipitation data collected between 1984 and 2000. The present study also evaluated ANN models created to predict the sediment yields. The training of the ANN models utilized single storm event data generated for the 17‐year period between 1984 and 2000 as the training input data. Training patterns and neural network architectures were varied to further study the ANN performance. Results from these models were compared with the available field data obtained from several debris basins within Los Angeles County. Both predictive models were then applied for hind‐casting the sediment prediction of several post 2000 events. Both the statistical and ANN models yield remarkably consistent results when compared with the measured field data. The results show that these models are very useful tools for predicting sediment yield sequences. The results can be used for scheduling cleanout operation of debris basins. It can be of great help in the planning of emergency response for burned areas to minimize the damage to properties and lives.     

Predicting streamflow response to fire-induced landcover change: implications of parameter uncertainty in the MIKE SHE model

Fire is a primary agent of landcover transformation in California semi-arid shrubland watersheds, however few studies have examined the impacts of fire and post-fire succession on streamflow dynamics in these basins. While it may seem intuitive that larger fires will have a greater impact on streamflow response than smaller fires in these watersheds, the nature of these relationships has not been determined. The effects of fire size on seasonal and annual streamflow responses were investigated for a medium-sized basin in central California using a modified version of the MIKE SHE model which had been previously calibrated and tested for this watershed using the Generalized Likelihood Uncertainty Estimation methodology. Model simulations were made for two contrasting periods, wet and dry, in order to assess whether fire size effects varied with weather regime. Results indicated that seasonal and annual streamflow response increased nearly linearly with fire size in a given year under both regimes. Annual flow response was generally higher in wetter years for both weather regimes, however a clear trend was confounded by the effect of stand age. These results expand our understanding of the effects of fire size on hydrologic response in chaparral watersheds, but it is important to note that the majority of model predictions were largely indistinguishable from the predictive uncertainty associated with the calibrated model - a key finding that highlights the importance of analyzing hydrologic predictions for altered landcover conditions in the context of model uncertainty. Future work is needed to examine how alternative decisions (e.g., different likelihood measures) may influence GLUE-based MIKE SHE streamflow predictions following different size fires, and how the effect of fire size on streamflow varies with other factors such as fire location.     

CRITICAL DATA REQUIREMENTS FOR PREDICTION OF EROSION AND SEDIMENTATION IN MOUNTAIN DRAINAGE BASINS1

ABSTRACT: The potential for understanding and, where necessary, managing sedimentation in humid mountain drainage basins increases with awareness of the conditions that lead to shallow landsliding, debris flows, and catastrophic sedimentation in stream channels. Progress in understanding has involved: improved recognition of source areas and the potential for downstream effects of slope failure; improved understanding of hydrological conditions required for failure; and a general theory of slope stability in shallow colluvium, including the role of plants, fires, timber harvest, and other disturbances. The theory acknowledges spatial variability in topographic and geotechnical terrain characteristics, the stochastic nature of climatic triggering events such as forest fires and rainstorms, and the integrating nature of channel networks in modulating the cumulative effects of transient processes within a basin. Anthropogenic fire regimes, road effects, and timber harvest can readily be included. Continued application and modification of the theory over an expanded geographical range require improvements in field data and their systematic storage in spatial databases. Improvements in digital topographic data for mountain basins, systematic network-wide surveys of channel conditions, and new technology for rapid documentation of soil depths in landslide source areas would enhance the prediction of mass failure, its consequences for channel habitat, and the basin-wide or regional distribution of hillslope and channel conditions. Computations of the probabilities of transient effects throughout basins could then form the basis of ecological risk analyses. Large-scale spatial data sets of a few critical variables are required before this next level of understanding can be developed and applied to sedimentation impacts on ecosystems and other resources.     

Computer simulation of the cumulative effects of brushland fire-management policies

A mathematical model simulates the cumulative volume of debris produced from brushland watersheds. Application of this model to a 176-km² (0.678 = mi²) watershed along the southern flank of the Central San Gabriel Mountains permits assessment of expected debris production associated with alternative fire-management policies. The political implications of simulated debris production are evaluated through a conceptual model that links interest groups to particular successional stages in brushland watersheds by means of the resources claimed by each group. It is concluded that in theory, a rotation burn policy would provide benefits to more interest groups concerned about southern California's brushland watersheds than does the current fire exclusion policy.This research was supported by the College of Agriculture and Life Sciences, University of Wisconsin-Madison, and by the Office of Water Research and Technology, USDI, under the Allotment program of Public Law 88–379, as amended, and by the University of California. Water Resources Center, as a part of Office of Water Research and Technology Project No. A-058-CAL and Water Resources Center Project UCAL-WRC-499. Support was also provided by the California Agricultural Experiment Station, Berkeley, California.     

Evaluating Pre‐ and Post‐Fire Peak Discharge Predictions across Western U.S. Watersheds

This study reviews five models commonly used in post‐fire hydrologic assessments: the Rowe Countryman and Storey (RCS), United States Geological Survey (USGS) Linear Regression Equations, USDA Windows Technical Release 55 (USDA TR‐55), Wildcat5, and U.S. Army Corps of Engineers (USACE) Hydrologic Modeling System (HEC‐HMS). The models are applied to eight diverse basins in the western United States (U.S.) (Arizona, California, Colorado, Montana, and Washington) affected by wildfires and assessed by input parameters, calibration methods, model constraints, and performance. No one model is versatile enough for application to all study sites. Results show inconsistency between model predictions for events across the sites and less confidence with larger return periods (25‐ and 50‐year events) and post‐fire predictions. The RCS method performs well, but application is limited to southern California. The USGS linear regression model has wider regional application, but performance is less reliable at the large recurrence intervals and post‐fire predictions are reliant on a subjective modifier. Of the three curve number‐based models, Wildcat5 performs best overall without calibration, whereas the calibrated TR‐55 and HEC‐HMS models show significant improvement in pre‐fire predictions. Results from our study provide information and guidance to ultimately improve model selection for post‐fire prediction and encourage uniform parameter acquisition and calibration across the western U.S.     

DEBRIS FLOWS THROUGH DIFFERENT FOREST AGE CLASSES IN THE CENTRAL OREGON COAST RANGE1

ABSTRACT: Debris flows in the Pacific Northwest can play a major role in routing sediment and wood stored on hillslopes and in first‐through third‐order channels and delivering it to higher‐order channels. Field surveys following a large regional storm event investigated 53 debris flows in the central Oregon Coast Range to determine relationships among debris flow characteristics and the age class of the surrounding forest. The volume of sediment and wood delivered by debris flows was strongly correlated with runout length. Debris flows that initiated at roads were significantly longer than nonroad related failures, and road related landslides were an order of magnitude larger than nonroad related landslides. Clearcuts and roads tended to have more numerous contributing landslides relative to second growth and mature forests. No statistically significant difference in the average debris flow runout length was detected among the forest age classes, although debris flows initiating in clearcuts and mixed forest and at roads occasionally supported extremely long runout lengths that were outside the range of variability observed in completely forested basins. The size of wood in deposits was not correlated with the size of trees on the adjacent slopes, suggesting that the majority of wood in debris flow deposits was from remobilization of wood previously stored in low order channels.     

A Method for Spatially Explicit Representation of Sub-watershed Sediment Yield,Southern California,USA

We present here a method to integrate geologic, topographic, and land-cover data in a geographic information system to provide a fine-scale, spatially explicit prediction of sediment yield to support management applications. The method is fundamentally qualitative but can be quantified using preexisting sediment-yield data, where available, to verify predictions using other independent data sets. In the 674-km² Sespe Creek watershed of southern California, 30 unique “geomorphic landscape units” (GLUs, defined by relatively homogenous areas of geology, hillslope gradient, and land cover) provide a framework for discriminating relative rates of sediment yield across this landscape. Field observations define three broad groupings of GLUs that are well-associated with types, relative magnitudes, and rates of erosion processes. These relative rates were then quantified using sediment-removal data from nearby debris basins, which allow relatively low-precision but robust calculations of both local and whole-watershed sediment yields, based on the key assumption that minimal sediment storage throughout most of the watershed supports near-equivalency of long-term rates of hillslope sediment production and watershed sediment yield. The accuracy of these calculations can be independently assessed using geologically inferred uplift rates and integrated suspended sediment measurements from mainstem Sespe Creek, which indicate watershed-averaged erosion rates between about 0.6–1.0 mm year^?1 and corresponding sediment yields of about 2 × 10³ t km^?2 year^?1. A spatially explicit representation of sediment production is particularly useful in a region where wildfires, rapid urban development, and the downstream delivery of upstream sediment loads are critical drivers of both geomorphic processes and land-use management.     

“R1-R4” AND “BOISED” SEDIMENT PREDICTION MODEL TESTS USING FOREST ROADS IN GRANITICS1

ABSTRACT: Erosion and sedimentation data from research watersheds in the Silver Creek Study Area in central Idaho were used to test the prediction of logging road erosion using the R1-R4 sediment yield model, and sediment delivery using the “BOISED” sediment yield prediction model. Three small watersheds were instrumented and monitored such that erosion from newly constructed roads and sediment delivery to the mouths of the watersheds could be measured for four years following road construction. The errors for annual surface erosion predictions for the two standard road tests ranged from +31.2 t/ha/yr (+15 percent) to -30.3 t/ha/yr (-63 percent) with an average of zero t/ha/yr and a standard deviation of the differences of 18.7 t/ha/yr. The annual prediction errors for the three watershed scale tests had a greater range from -40.8 t/ha/yr (-70 percent) to +65.3 t/ha/yr (+38 percent) with a mean of -1.9 t/ha/yr and a standard deviation of the differences of 25.2 t/ha/yr. Sediment yields predicted by BOISED (watershed scale tests) were consistently greater (average of 2.5 times) than measured sediment yields. Hillslope sediment delivery coefficients in BOISED appear to be overly conservative to account for average site conditions and road locations, and thus over-predict sediment delivery. Mass erosion predictions from BOISED appear to predict volume well (465 tonnes actual versus 710 tonnes predicted, or a 35 percent difference) over 15 to 20 years, however mass wasting is more episodic than the model predicts.     

SEDIMENT ROUTING FOR AGRICULTURAL WATERSHEDS1

ABSTRACT: A sediment routing technique was developed to route sediment yield from small watersheds through streams and valleys to the outlet of large watersheds. The technique is based on the modified universal sol loss equation and a first order decay function of travel time and particle size. Deposition is dependent upon settling velocities of sediment particles, travel time, and the amount of sediment in suspension. Sediment routing increases sediment yield prediction accuracy and allows determination of subwatershed contributions to the total sediment yield. Also, the locations and amounts of floodplain scour and deposition can be predicted. Another advantage of sediment routing is that measured sediment yield data are not required. The procedure performed satisfactorily in test routings on two Texas blackland watersheds Sediment routing will be useful in flood control evaluation, reservoir and channel design, water quality calculations, environmental impact assessment, and land-use planning.     

Large Woody Debris and Land Management in California’s Hardwood-Dominated Watersheds

Although large woody debris (LWD) has been studied extensively in conifer-dominated watersheds, relatively little is known about LWD in hardwood-dominated watersheds. Field surveys of 32 hardwood-dominated stream reaches in northern coastal California revealed that levels of LWD varied with land ownership and that living trees strongly influenced debris jam formation. Almost half of the channel-spanning debris jams, which stored the most wood and were most likely to form a pool, were formed behind a key piece that was still living. These living key pieces might provide greater longevity and stability than would otherwise be expected from hardwood LWD. Compared to streams on private land, streams on public land had significantly greater LWD loading and debris-jam frequency. Land management practices that remove wood from streams might be contributing to the degradation of salmonid habitat in Californias hardwood-dominated watersheds.     

SPATIAL AND TEMPORAL VARIABILITY IN SUSPENDED SEDIMENT YIELDS OF NORTH ISLAND BASINS,NEW ZEALAND1

ABSTRACT: Specific annual suspended sediment yields and their standard deviations are presented for 47 basins of North Island, New Zealand. Most of the variance in yields is explained by catchment mean rainfall. Rivers with similar flow range have similar suspended sediment concentration ratings, independent of differing watershed lithology and regolith, except for six basins having an abundance of soft fine sediments. Prediction equations for yield and its standard deviation are derived for four essentially arbitrary regions. AU feature rainfall as the independent variable. Differences between regions may owe to variations in intensity, frequency, and duration patterns of storms and, in one area, to bed material size as well. The temporal distribution of annual yields from a basin m be modeled by a two-parameter lognormal function: the prediction equations above may be used to evaluate this function at a site for which suspended sediment data are unavailable.     

Impacts of Climate Variability and Land Use Alterations on Frequency Distributions of Terrestrial Runoff Loading to Coastal Waters in Southern California1

Abstract: The transport of water, sediment, dissolved and particulate chemicals, and bacteria from coastal watersheds affects the nearshore marine and estuarine waters. In southern California, coastal watersheds deliver water and associated constituents to the nearshore system in discrete pulses. To better understand the pulsed nature of these watersheds, frequency distributions of simulated runoff events are presented for: (1) three land use conditions (1929, 1998, 2050); (2) three time periods (all water years 1989‐2002), only El Nino years (1992, 1993, 1995, 1998); and only non‐El Nino years; and (3) three regions (watershed, uplands, and lowlands). At the watershed scale, there was a significant increase (>200%) in mean event runoff from 1929 to 2050 (0.4‐1.3 cm) due to localized urbanization, which shifted the dominant sources of runoff from the mountains in 1929 (78% of watershed runoff) to the coastal plane for 2050 conditions (51% of watershed runoff). Inter‐annual climate variability was strong in the rainfall and runoff frequency distributions, with mean event rainfall and runoff 66 and 60% larger in El Nino relative to non‐El Nino years. Combining urbanization and climate variability, 2050 land conditions resulted in El Nino years being five times more likely to produce large (>3.0 cm) runoff events relative to non‐El Nino years. Combining frequency distributions of event runoff with regional nutrient export relationships, we show that in El Nino years, one in five events produced runoff ≥2.5 cm and temporary nearshore nitrate and phosphate concentrations of 12 and 1.4 μM, respectively, or approximately 5‐10 times above ambient conditions.     

VALIDATION OF AGNPS FOR SMALL WATERSHEDS USING AN INTEGRATED AGNPS/GIS SYSTEM1

ABSTRACT: The AGNPS (AGricultural NonPoint Source) model was evaluated for predicting runoff and sediment delivery from small watersheds of mild topography. Fifty sediment yield events were monitored from two watersheds and five nested subwater-sheds in East Central Illinois throughout the growing season of four years. Half of these events were used to calibrate parameters in the AGNPS model. Average calibrated parameters were used as input for the remaining events to obtain runoff and sediment yield data. These data were used to evaluate the suitability of the AGNPS model for predicting runoff and sediment yield from small, mild-sloped watersheds. An integrated AGNPS/GIS system was used to efficiently create the large number of data input changes necessary to this study. This system is one where the AGNPS model was integrated with the GRASS (Geographic Resources Analysis Support System) GIS (Geographical Information System) to develop a decision support tool to assist with management of runoff and erosion from agricultural watersheds. The integrated system assists with the development of input GIS layers to AGNPS, running the model, and interpretation of the results.     

SOME SUSPENDED SEDIMENT YIELDS FROM SOUTH ISLAND CATCHMENTS,NEW ZEALAND1

: Estimates of specific annual suspended sediment yields, some of which rank among the highest reported in the world, are presented for 33 basins of South Island, New Zealand. Yield from each basin was determined by combining a suspended sediment concentration rating with the complete flow record of each catchment stream gaging station. A multiple regression analysis between sediment yields and climatic, hydrologic and physiographic parameters of each basin demonstrates that most of the variance in yields is explained by catchment mean rainfall. Geology apparently has little influence on sediment yield as suspended sediment concentration ratings, from rivers draining catchments of differing lithology, and regolith, are indistinguishable. Specific suspended sediment yield prediction equations are given for four defined regions covering in area almost all South Island; and except for one area, feature rainfall as the principle independent variable. Differences between regions may be due to variations in intensity, frequency, and duration patterns of storms. It is proposed that a simple power law relationship between yield and rainfall provides useful suspended sediment yield estimates in mountainous regions of temperate maritime climate, provided catchments have not been modified extensively by man.     

Flow,Organic, and Inorganic Sediment Yields from a Channelized Watershed in the South Carolina Lower Coastal Plain

Many small streams in coastal watersheds in the southeastern United States are modified for agricultural, residential, and commercial development. In the South Carolina Lower Coastal Plain, low‐relief topography and a shallow water table make stream channelization ubiquitous. To quantify the impacts of urbanization and stream channelization, we measured flow and sediment from an urbanizing watershed and a small forested watershed. Flow and sediment export rates were used to infer specific yields from forested and nonforested regions of the urbanizing watershed. Study objectives were to: (1) quantify the range of runoff‐to‐rainfall ratios; (2) quantify the range of specific sediment yields; (3) characterize the quantity and quality of particulate matter exported; and (4) estimate sediment yield attributable to agriculture, development, and channelization activities in the urbanizing watershed. Our results showed that the urban watershed exported over five times more sediment per unit area compared with the forested watershed. Sediment concentration was related to flow flashiness in the urban watershed and to flow magnitude in the forested watershed. Sediments from the forested watershed were dominated by organic matter, whereas mineral matter dominated sediment from the urban stream. Our results indicated that a significant shift in sediment quality and quantity are likely to occur as forested watersheds are transformed by urbanization in coastal South Carolina.     

Sediment Production From Forest Roads in Western Montana1

Abstract: Unpaved roads are a primary sediment source in forested watersheds. Validation of erosion models and improvements to road management require information on road erosion rates and the factors controlling erosion. This study measured sediment yields from twenty ～0.05 ha unsurfaced (native) road plots in Belt Supergroup and glacial till parent materials of western Montana, and investigated the factors controlling erosion. Annual sediment yields for individual plots ranged from 0 to 96.9 Mg/ha/yr over 3 years (2002‐2004). Annual mean sediment yield ranged from 2.1 Mg/ha in 2003 to 9.9 Mg/ha in 2004 with an overall mean of 5.4 Mg/ha/yr. The mean of log‐transformed sediment yields for sites in glacial till parent materials was higher than Belt Supergroup parent materials (p = 0.063). A regression model with road slope, time since last grading, roadbed gravel content, and precipitation as predictive variables explained 68% of the variability in sediment yield (F = 28.2; p < 0.0001). Road erosion in western Montana is limited by low erodibility of the dominant parent materials and low rainfall. Management procedures such as reducing the frequency of grading can significantly reduce sediment yields from forest roads.     

Examining Modeling Approaches for the Rainfall‐Runoff Process in Wildfire‐Affected Watersheds: Using San Dimas Experimental Forest

Wildfire can significantly change watershed hydrological processes resulting in increased risks for flooding, erosion, and debris flow. The goal of this study was to evaluate the predictive capability of hydrological models in estimating post‐fire runoff using data from the San Dimas Experimental Forest (SDEF), San Dimas, California. Four methods were chosen representing different types of post‐fire runoff prediction methods, including a Rule of Thumb, Modified Rational Method (MODRAT), HEC‐HMS Curve Number, and KINematic Runoff and EROSion Model 2 (KINEROS2). Results showed that simple, empirical peak flow models performed acceptably if calibrated correctly. However, these models do not reflect hydrological mechanisms and may not be applicable for predictions outside the area where they were calibrated. For pre‐fire conditions, the Curve Number approach implemented in HEC‐HMS provided more accurate results than KINEROS2, whereas for post‐fire conditions, the opposite was observed. Such a trend may imply fundamental changes from pre‐ to post‐fire hydrology. Analysis suggests that the runoff generation mechanism in the watershed may have temporarily changed due to fire effects from saturation‐excess runoff or subsurface storm dominated complex mechanisms to an infiltration‐excess dominated mechanism. Infiltration modeling using the Hydrus‐1D model supports this inference. Results of this study indicate that physically‐based approaches may better reflect this trend and have the potential to provide consistent and satisfactory prediction.     

TARGETING REMEDIAL MEASURES TO CONTROL NONPOINT SOURCE POLLUTION1

ABSTRACT_: The watershed model GAMES is used for the evaluation of a targeting approach to control fluvial sedimentation arising from soil erosion in agricultural areas. The data considered for the analysis consists of output from the application of the model to existing and hypothetical soil and crop management systems in two small watersheds of southern Ontario, one in the rolling uplands and the other in a very flat lowland area. The model output includes estimates of spring sediment yield from field-size cells to the stream outlet for existing agricultural management conditions, and estimates of sediment yield resulting from the successive implementation of two levels of soil erosion controls under four remedial measures strategies. The results reveal that, for the rolling upland watershed exhibiting a wide range of soil erosion and sediment yield rates, targeted control programs can be expected to provide an extremely effective approach to sediment control. For flat lowland watersheds, exhibiting relatively uniform soil erosion and sediment yield rates, the strategy of targeting controls may be somewhat more effective than a random approach to control, but not as efficient as in the case of watersheds in more rolling terrain. It is evident from the study that a screening model such as GAMES provides a very useful tool for the planning and evaluation of erosion and sediment control programs.