APPLICATION OF GREY MODEL AND ARTIFICIAL NEURAL NETWORKS TO FLOOD FORECASTING1

The main focus of this study was to compare the Grey model and several artificial neural network (ANN) models for real time flood forecasting, including a comparison of the models for various lead times (ranging from one to six hours). For hydrological applications, the Grey model has the advantage that it can easily be used in forecasting without assuming that forecast storm events exhibit the same stochastic characteristics as the storm events themselves. The major advantage of an ANN in rainfall‐runoff modeling is that there is no requirement for any prior assumptions regarding the processes involved. The Grey model and three ANN models were applied to a 2,509 km² watershed in the Republic of Korea to compare the results for real time flood forecasting with from one to six hours of lead time. The fifth‐order Grey model and the ANN models with the optimal network architectures, represented by ANN1004 (34 input nodes, 21 hidden nodes, and 1 output node), ANN1010 (40 input nodes, 25 hidden nodes, and 1 output node), and ANN1004T (14 input nodes, 21 hidden nodes, and 1 output node), were adopted to evaluate the effects of time lags and differences between area mean and point rainfall. The Grey model and the ANN models, which provided reliable forecasts with one to six hours of lead time, were calibrated and their datasets validated. The results showed that the Grey model and the ANN1010 model achieved the highest level of performance in forecasting runoff for one to six lead hours. The ANN model architectures (ANN1004 and ANN1010) that used point rainfall data performed better than the model that used mean rainfall data (ANN1004T) in the real time forecasting. The selected models thus appear to be a useful tool for flood forecasting in Korea.     

A TRIANGULAR MODEL OF DIMENSIONLESS RUNOFF PRODUCING RAINFALL HYETOGRAPHS IN TEXAS1

ABSTRACT: A synthetic triangular hyetograph for a large data base of Texas rainfall and runoff is needed. A hyetograph represents the temporal distribution of rainfall intensity at a point or over a watershed during a storm. Synthetic hyetographs are estimates of the expected time distribution for a design storm and principally are used in small watershed hydraulic structure design. A data base of more than 1,600 observed cumulative hyetographs that produced runoff from 91 small watersheds (generally less than about 50 km²) was used to provide statistical parameters for a simple triangular shaped hyetograph model. The model provides an estimate of the average hyetograph in dimensionless form for storm durations of 0 to 24 hours and 24 to 72 hours. As a result of this study, the authors concluded that the expected dimensionless cumulative hyetographs of 0 to 12 hour and 12 to 24 hour durations were sufficiently similar to be combined with minimal information loss. The analysis also suggests that dimensionless cumulative hyetographs are independent of the frequency level or return period of total storm depth and thus are readily used for many design applications. The two triangular hyetographs presented are intended to enhance small watershed design practice in applicable parts of Texas.     

A NOTE ON AREAL RAINFALL DEFINITION

A discussion is presented of the likely sources of error in defining areal rainfall on a storm basis. These include the instrumental error, sampling fluctuations over the area, and network density. The analysis of dense raingage data provides some perspective of the magnitude of the errors that might be encountered from the natural variability of rainfall. Except for one watershed in Arizona, the coefficient of variation, based on a sample of storm totals from the individual gages in various size areas, remains relatively constant with increasing area for a particular storm. The error due to rainfall variability over the area is probably the most important and must be considered in experiments which attempt to resolve small-area hydrologic problems.     

MODELING RAINFALL-RUNOFF PROCESS INCLUDING LOSSES1

ABSTRACT: Existing discrete, linear rainfall-runoff models generally require the effective rainfall of a given storm as the input for computing the runoff hydrograph. This paper proposes a method for estimating, simultaneously, the optimal values of model parameters and the rainfall losses frem the measured rainfall hyetograph and the runoff hydrograph. The method involves an ARMA model for the rainfall-runoff process and a nonlinear iterative technique. The number of model parameters to be estimated for the ARMA model is much less than the unit hydrograph model. Applications of the model to three different watersheds show that the computed runoff hydrographs agree well with the measurements.     

A DISTRIBUTED DYNAMIC WATERSHED MODEL1

ABSTRACT: A distributed watershed model was developed to mathematically simulate overland and channel flow for a single-event storm. The modeled watersheds in the study were subdivided into rectangular grid elements. All hydrologically significant parameters, such as land slope, rainfall and precipitation excess, were assumed to be uniform within each element. The Green-Ampt method was adopted to generate precipitation excess for each element during the simulation period. A two-dimensional diffusion wave model was used for overland flow routing and an iterative Alternative Direction Implicit scheme was used to solve the simultaneous overland flow equations. Once the overland flow became inflow to the channel, a one-dimensional dynamic wave flood routing technique, based on a four-point, implicit, non-linear finite difference solution of the St. Venant equation of unsteady flow, was applied. A limited number of comparisons were made between simulated and observed hydrographs for areas of about one square mile. Given the appropriate parameters, the model was able to accurately simulate runoff for single-event storms. This paper describes a distributed watershed model developed to simulate overland and channel flow. Comparisons were made between simulated and observed hydrographs for three watersheds. The model was able to accurately simulate the runoff for single-event storms using 61-m by 61-m (200 ft by 200 ft) watershed grid elements.     

A PARAMETRIC MODEL CALIBRATED WITH A PHYSICALLY BASED MODEL FOR RUNOFF PREDICTION FROM UNGAGED STREAMS1

ABSTRACT: Recent developments in the numerical solution of the governing partial differential equations for overland and channel flow should make possible physically based models which predict runoff from ungaged streams. However, these models, which represent the watershed by sets of intersecting planes, are complex and require much computer time. Parametric models exist that have the advantage of being relatively simple, and once calibrated are inexpensive to use and require limited data input. In this study, a procedure was developed for calibrating a parametric model against a physically based model, utilizing base areas of one acre and one square mile, with the expectation that base areas can be combined to model real watersheds. Simulation experiments with the physically based model showed that, for the one-acre base area, the dominant parameter (cell storage ratio, K) related to the slope and friction of the planes, whereas for one square-mile areas, the dominant parameters (K plus a lag factor, L) relate to channel properties. These parameters decreased exponentially as rainfall intensity increased.     

Effects of Forest Composition and Spatial Patterns on Storm Flows of a Small Watershed1

Abstract: The PRMS_Storm model was built as a storm event, distributed hydrological model for studying the hydrological effects of forest composition and spatial distribution on storm‐flow volume and peakflow rates in the Xiangshuixi Watershed in the Three Gorges Reservoir Area, in the Yangtze River Basin in southwestern China. We developed three simulation scenarios based on forest composition and their spatial arrangements across the watershed, including all mixed conifer‐evergreen broadleaf forests (Scenario 1), all mixed evergreen broadleaf forests (Scenario 2), and mixed conifer + evergreen broadleaf + shrub forests (Scenario 3). We examined 11 storm events observed during 2002‐2005. Compared with the existing forest covers, modeling results suggested that the amount of overland flow was reduced by 21, 23, and 22%, and the interflow increased by 16, 88, and 30%, for Scenarios 1, 2, and 3, respectively. During the same time, peakflow rates were reduced by 20.8, 9.6, and 18.9%, respectively. The reduction of peakflow rates was most significant when rainfall intensity exceeded 0.8 mm/min and events with a short duration and effect was minor when rainfall intensity was below 0.5 mm/min. In general, we found that Scenarios 1 and 3 were preferred for reducing storm‐flow volume and peakflow rates due to their higher interception rates, large soil water holding capacity, and higher soil infiltration capacity. The modeled results suggested soil properties are important in affecting the flow processes and thus forest composition and forest spatial distributions will affect storm‐flow volume and peakflow rates at the watershed scale. To maximize flood reduction functions of a watershed, high priority should be given to those forest types (Scenarios 1 and 3) in reforestation practices in the study region. This study suggests both forest composition and spatial pattern are important reforestation designs for flood reduction in the Three Gorges Reservoir Area.     

Estimating Lag to Peak between Rainfall and Peak Streamflow with a Mixed‐Effects Model

We test the use of a mixed‐effects model for estimating lag to peak for small basins in Maine (drainage areas from 0.8 to 78 km²). Lag to peak is defined as the time between the center of volume of the excess rainfall during a storm event and the resulting peak streamflow. A mixed‐effects model allows for multiple observations at sites without violating model assumptions inherent in traditional ordinary least squares models, which assume each observation is independent. The mixed model includes basin drainage area and maximum 15‐min rainfall depth for individual storms as explanatory features. Based on a remove‐one‐site cross‐validation analysis, the prediction errors of this model ranged from ?42% to +73%. The mixed model substantially outperformed three published models for lag to peak and one published model for centroid lag for estimating lag to peak for small basins in Maine. Lag to peak estimates are a key input to rainfall–runoff models used to design hydraulic infrastructure. The improved accuracy and consistency with model assumptions indicates that mixed models may provide increased data utilization that could enhance models and estimates of lag to peak in other regions.     

ESTIMATION OF THE HIEC1 LOSS PARAMETERS FOR ROUTING THE PROBABLE MAXIMUM FLOOD1

ABSTRACT: Twenty storm events were used to select design values of the HEC1 loss parameters STRTL and CNSTL in order to route the probable maximum flood, PMF, through the Englewood watershed, Ohio. The parameter STRTL represents the initial volume of water lost due to interception and incomplete saturation of the soil prior to the storm. The parameter CNSTL represents a continuous loss rate and depends only on the watershed. When optimized from each storm event, STRTL varied between 0.0 and 3.4 inches with an average of 1.0 inch; CNSTL varied between 0.02 and 0.26 inch/hour, and it followed a normal probability distribution with a mean of about 0.1 inch/hour. The absence of correlation between optimum CNSTL values and each of total rainfall, total loss, and runoff duration supported the selection of the mean CNSTL as a design value. PMF routing through the Englewood watershed revealed that the PMF at the outlet is not sensitive to STRTL, but highly affected by CNSTL variations. The insensitivity to STRTL was due to the presence of a dam at the outlet of the watershed that caused the buildup of water in the watershed, thereby masking the storage effect of STRTL. The peak PMF increased by about 27 percent when the design CNSTL was decreased to 0.05 inch! hour, and decreased by about 18 percent when the design CNSTL was increased to 0.15 inch/hour.     

STORM RUNOFF PREDICTION BASED ON A SPATIALLY DISTRIBUTED TRAVEL TIME METHOD UTILIZING REMOTE SENSING AND GIS1

ABSTRACT: In this study, remotely sensed data and geographic information system (GIS) tools were used to estimate storm runoff response for Simms Creek watershed in the Etonia basin in northeast Florida. Land cover information from digital orthophoto quarter quadrangles (DOQQ), and enhanced thematic mapper plus (ETM+) were analyzed for the years 1990, 1995, and 2000. The corresponding infiltration excess runoff response of the study area was estimated using the U.S. Department of Agriculture (USDA), Natural Resources Conservation Service Curve Number (NRCS‐CN) method. A digital elevation model (DEM)/GIS technique was developed to predict stream response to runoff events based on the travel time from each grid cell to the watershed outlet. A comparison of predicted to observed stream response shows that the model predicts the total runoff volume with an efficiency of 0.98, the peak flow rate at an efficiency of 0.85, and the full direct runoff hydrograph with an average efficiency of 0.65. The DEM/GIS travel time model can be used to predict the runoff response of ungaged watersheds and is useful for predicting runoff hydrographs resulting from proposed large scale changes in the land use.     

Fecal Coliform Export From Four Coastal North Carolina Areas1

Abstract: Fecal coliform (FC) bacteria in coastal waters impair the use of these waters for shellfish harvesting and recreation. This study was designed to quantify and compare FC levels and export in two coastal watersheds with different land uses. Continuous monitoring of rainfall and discharge at three sites in the Jumping Run Creek watershed and one site in the Pettiford Creek watershed were conducted during a 4.5‐year period. Primary land use in the drainage area of one of the three Jumping Run Creek sites is low density industrial, while the other two are residential. Land use in the Pettiford Creek watershed is managed national forest. Nonstorm or base‐flow grab and flow‐proportional storm‐event samples were collected and analyzed for turbidity, conductivity, suspended sediment, nitrogen, phosphorus, and FC. Geometric mean FC levels for the Jumping Run Creek monitoring sites ranged from 593 to 2,096 mpn/100 ml, while the mean level at the Pettiford Creek site was 191 mpn/100 ml. Levels of most other parameters were greater in storm discharge from the Jumping Run Creek sites as compared to Pettiford Creek indicating that pollutant export from a watershed increases with development. Statistical analysis of the monitoring data suggested that FC levels in stormwater samples consistently increased with storm rainfall, but were not consistently correlated with any other parameter, including total suspended solids. Multivariate analysis indicated that the weekly FC export for each of the four sites was lowest during the December‐February quarter. Export was highest during the spring and summer at the Jumping Run Creek sites, while for the Pettiford Creek site, FC export was highest during September‐November. The cause of the seasonal variability was unknown but was thought to be associated with human activity in the watersheds.     

Coupling PCSWMM and WASP to Evaluate Green Stormwater Infrastructure Impacts to Storm Sediment Loads in an Urban Watershed

We coupled rainfall–runoff and instream water quality models to evaluate total suspended solids (TSS) in Wissahickon Creek, a mid‐sized urban stream near Philadelphia, Pennsylvania. Using stormwater runoff and instream field data, we calibrated the model at a subdaily scale and focused on storm responses. We demonstrate that treating event mean concentrations as a calibration parameter rather than a fixed input can substantially improve model performance. Urban stormwater TSS concentrations vary widely in time and space and are difficult to represent simply. Suspended and deposited sediment pose independent stressors to stream biota and model results suggest that both currently impair stream health in Wissahickon Creek. Retrofitting existing detention basins to prioritize infiltration reduced instream TSS loads by 20%, suggesting that infiltration mitigates sediment more effectively than detention. Infiltrating stormwater from 30% of the watershed reduced instream TSS loads by 47% and cut the frequency of TSS exceeding 100 mg/L by half. Settled loads and the frequency of high TSS values were reduced by a smaller fraction than suspended loads and duration at high TSS values. A widely distributed network of infiltration‐focused projects is an effective stormwater management strategy to mitigate sediment stress. Coupling rainfall–runoff and water quality models is an important way to integrate watershed‐wide impacts and evaluate how management directly affects urban stream health.     

PROBLEMS IN WEIGHTING OF HYETOGRAPHS1

ABSTRACT: It was found that the conventional weighting factor application to hyetograph ordinates results in artificially attenuated storm patterns. A modified weighting procedure is suggested which allows adjustments in the storm timing, peak intensity, and volume but conserves the storm pattern observed at the raingage nearest to the watershed point of interest. The systematic underestimation of peak flood flows, which result from conventional hyetograph weighting, can be avoided by conserving the hyetograph shape from the raingage nearest to any subarea of a modeled watershed and merely applying weighting factors to the rainfall volumes and temporal center of gravity of several hyetographs.     

RESPONSE OF GREEN‐AMPT MEIN‐LARSEN SIMULATED RUNOFF VOLUMES TO TEMPORALLY AGGREGATED PRECIPITATION1

ABSTRACT: Data collection frequency in automated systems is user determined and can range from seconds to hours or days. Currently, there is no standard or recommended frequency interval for collecting precipitation data from automated systems for input to event‐based models such as Green‐Ampt Mein‐Larsen (GAML). Data from 47 storm events at seven locations were used to simulate the response of GAML excess rainfall to temporally aggregated precipitation data. No difference in model efficiency was recognized when comparing one‐minute interval data (R²= 1.00) to five‐minute data (R²= 1.00). Very little model efficiency was lost at a 10‐minute (R²= 0.96) interval. After 10‐minutes, decline in efficiency became more rapid with R²= 0.16 at one hour. The combined effect of time interval with respect to drainage area, hydraulic conductivity, maximum 30‐minute intensity, and total precipitation also revealed similar results.     

Evaluation of the MIKE SHE Model for Application in the Loess Plateau,China1

Abstract: Quantifying the hydrologic responses to land use/land cover change and climate variability is essential for integrated sustainable watershed management in water limited regions such as the Loess Plateau in Northwestern China where an adaptive watershed management approach is being implemented. Traditional empirical modeling approach to quantifying the accumulated hydrologic effects of watershed management is limited due to its complex nature of soil and water conservation practices (e.g., biological, structural, and agricultural measures) in the region. Therefore, the objective of this study was to evaluate the ability of the distributed hydrologic model, MIKE SHE to simulate basin runoff. Streamflow data measured from an overland flow‐dominant watershed (12 km²) in northwestern China were used for model evaluation. Model calibration and validation suggested that the model could capture the dominant runoff process of the small watershed. We found that the physically based model required calibration at appropriate scales and estimated model parameters were influenced by both temporal and spatial scales of input data. We concluded that the model was useful for understanding the rainfall‐runoff mechanisms. However, more measured data with higher temporal resolution are needed to further test the model for regional applications.     

EVALUATING HSPF IN AN ARID,URBANIZED WATERSHED1

ABSTRACT: The Hydrologic Simulation Program‐FORTRAN (HSPF) is a powerful time variable hydrologic model that has rarely been applied in arid environments. Here, the performance of HSPF in southern California was assessed, testing its ability to predict annual volume, daily average flow, and hourly flow. The model was parameterized with eight land use categories and physical watershed characteristics. It was calibrated using rainfall and measured flow over a five‐year period in a predominantly undeveloped watershed and it was validated using a subsequent 4‐year period. The process was repeated in a separate, predominantly urbanized watershed over the same time span. Annual volume predictions correlated well with measured flow in both the undeveloped and developed watersheds. Daily flow predictions correlated well with measured flow following rain events, but predictions were poor during extended dry weather periods in the developed watershed. This modeling difficulty during dry‐weather periods reflects the large influence of, and the poor accounting in the model for, artificially introduced water from human activities, such as landscape overwatering, that can be important sources of water in urbanized arid environments. Hourly flow predictions mistimed peak flows, reflecting spatial and temporal heterogeneity of rainfall within the watershed. Model correlation increased considerably when predictions were averaged over longer time periods, reaching an asymptote after an 11‐hour averaging window.     

GENERATION OF HOURLY RAINFALL1

ABSTRACT: A rainfall model was developed to divide daily rainfall into storms and distribute storm depths over storm duration for input into the Stanford Watershed Model.     

STORM CHARACTERISTICS OF CONVECTIVE-SCALE PRECIPITATION1

A classification scheme for convective precipitation, having applications in both analysis and modeling of meteorological and hydrological events, is presented. The method is based upon observations of rainfall at the ground, radar scans of storm events, and visible and infrared satellite imagery of larger storm systems. Empirical and theoretical frequency distributions are derived for total storm rainfall, rainfall duration and time between storms for each of the convective categories. This stratification is directly applicable to the experimental design and evaluation of weather modification projects and may be useful for the development and interpretation of meteorological and hydrological models. When atmospheric conditions limit storm development to cells, rainfall was seldom observed. Small clusters also produce small amounts of rainfall but have a longer lifetime than cells and are likely candidates for cloud seeding attempts to encourage their growth to large clusters. Large and nested clusters usually produce large amounts of natural precipitation. A few large storms account for most of a season's rainfall.     

Land-use forecasting and hydrologic model integration for improved land-use decision support

This paper develops a methodology for integrating a land-use forecasting model with an event scale, rainfall-runoff model in support of improving land-use policy formulation at the watershed scale. The models selected for integration are loosely coupled, structured upon a common GIS platform that facilitates data exchange. The hydrologic model HEC-HMS is calibrated for a specific storm event that occurred within central Washington State. The land-use forecasting model, What If? is implemented to forecast future spatial distributions of low-density residential land-uses under low and high population growth estimates. Forecasted land-use distribution patterns for the years 2015, 2025, and 2050 are then used as land-use data input for the calibrated hydrologic model, keeping all other parameters constant. Impacts to the stream discharge hydrograph are predicted as the study area becomes increasingly developed as forecasted by What If?. The initial results of this integration process demonstrate the synergy that can be generated through the linkage of the selected models. The ability to quantifiably forecast the potential hydrologic implications of proposed land-use policies before their implementation offers land-use decision-makers a valuable tool for discerning which proposed land-use alternatives will be effective at minimizing storm water runoff.     

Curve Number Derivation for Watersheds Draining Two Headwater Streams in Lower Coastal Plain South Carolina,USA

The objective of this study was to assess curve number (CN) values derived for two forested headwater catchments in the Lower Coastal Plain (LCP) of South Carolina using a three‐year period of storm event rainfall and runoff data in comparison with results obtained from CN method calculations. Derived CNs from rainfall/runoff pairs ranged from 46 to 90 for the Upper Debidue Creek (UDC) watershed and from 42 to 89 for the Watershed 80 (WS80). However, runoff generation from storm events was strongly related to water table elevation, where seasonally variable evapotranspirative wet and dry moisture conditions persist. Seasonal water table fluctuation is independent of, but can be compounded by, wet conditions that occur as a result of prior storm events, further complicating flow prediction. Runoff predictions for LCP first‐order watersheds do not compare closely to measured flow under the average moisture condition normally associated with the CN method. In this study, however, results show improvement in flow predictions using CNs adjusted for antecedent runoff conditions and based on water table position. These results indicate that adaptations of CN model parameters are required for reliable flow predictions for these LCP catchments with shallow water tables. Low gradient topography and shallow water table characteristics of LCP watersheds allow for unique hydrologic conditions that must be assessed and managed differently than higher gradient watersheds.