Study On Event-based Rainfall-runoff Model For Multiple-scale Watersheds In The Hilly-gully Regions Of The Chinese Loess Plateau

The hydrological processes of small watersheds are closely related to water resourcetheory, land surface ecology, and soil erosion. Severe soil and water loss in the loesshilly–gully region of the Loess Plateau has attracted much concern. Small watersheds arethe basic unit for soil and water conservation. Models of hydrologic processes for smallwatersheds could serve as an important aid for soil and water conservation, water resourceoptimization and flood prediction.In this study, by taking the hilly-gully regions of the Loess Plateau as the study area,the establishment of hydrological process models for different-scale watersheds during asingle rainfall event was studied. In Qiaozi-West watershed （W1-b,1km²）, the modelestablishment, calibration and validation for watershed infiltration, runoff, and flow routingwere studied. In the watersheds of Luoyugou （W100,100km²）, Lvergou （W10,10km²）,Qiaozi-East （W1-a,1km²） and Qiaozi-West （W1-b,1km²）, the Hydraulic Geometrytheory was introduced to study the measuring method of watershed runoff. Preliminaryconclusions of the study are summarized as follows:（1） A method was developed for modifying NRCS-CN model. By introducing steadyinfiltration to the NRCS-CN model, the modified NRCS-CN （MCN） model was developed.The steady infiltration rates for the study watershed were determined as4.8mm h^-1byusing observed initial abstraction, and4.2mm h^-1by using calculated initial abstraction.Both of the MCN and NRCS-CN models were used to simulate watershed runoff processfor the study events. The results showed that the simulation of watershed infiltration andrunoff by the MCN model was better than that of NRCS-CN model by using either calibrated steady infiltration rate, especially in simulating larger infiltration, runoff events;the infiltration simulation using steady infiltration rate of4.8mm h^-1was superior to thatusing4.2mm h^-1for MCN model.（2） Based on Qiaozi-West watershed rainfall-runoff process data, watershed DEM,soil and land use digital maps, the initial abstraction ratio of the NRCS-CN model wasdetermined by Back Calculation （BC） and Event Analysis （EA） methods. The initialabstraction ratios were determined as0.1and0.17by using BC and EA methods,respectively. Using three initial abstraction ratio values of0.1,0.17and0.2, runoffamounts for the study watershed were predicted by NRCS-CN model. Considering both oferror analyses and curve fitting results, the value of0.1was indicated as the appropriatevalue of the initial abstraction ratio for the NRCS-CN model in Qiaozi-West watershed.（3） The observed flow velocity data from the measuring weirs at watershed outletswere fitted with the discharge rate, using both Hydraulic Geometry power function andlogarithmic function models for the Luoyugou （W100,100km²）, Lvergou （W10,10km²）,Qiaozi-East （W1-a,1km²） and Qiaozi-West （W1-b,1km²） watersheds. The coefficient ofdetermination （R²） and model efficiency coefficient （E） were used to evaluate modelcalibration and validation results, respectively. The effect of watershed scale on the modelparameters was examined by using model calibration results from W100, W10and W1-awatersheds. It was found that the parameter k （flow velocity for unit discharge rate） in thepower function model was negatively correlated with watershed size, while parameter m（rate of change of flow velocity） had an opposite correlation with watershed size comparedwith parameter k. In the logarithmic function model, parameter e （rate of change of flowvelocity） had no significant correlation with watershed size, while parameter d （flowvelocity for unit discharge rate） was negatively correlated with watershed size, similar toparameter k The calibration results from the two paired watersheds （W1-a and W1-b） wereused for exploring the effect of watershed land use on the model parameters. Theparameter k in the power function model for W1-a watershed was significantly higher thanthat of W1-b watershed （P<0.001）. The parameter m for the two paired watersheds showedno significant difference. The parameter e in the logarithmic function model for W1-bwatershed was higher than that of W1-a watershed, however the difference was notsignificant （P<0.05）. The parameter d for W1-a watershed was significantly higher than that of W1-b watershed （P<0.05）. Another data set from the study watersheds was used totest the two function models. The results showed that both of the model functions yieldedacceptable results, nevertheless the power function model generally showed superiorperformance to the logarithmic function model for the wide value range of flow velocity.（4） By using GIS tools, the Qiaozi-West （W1-b） watershed was dissected as11slopes,which were connected by a channel. The runoff for each slope in the watershed wascalculated using NRCS-CN model based on watershed rainfall process input for the year1987–2006. A new conceptual method was developed and used to calculate flow routing ofthe runoff from each slope, to derive watershed hydrograph. The predictions for the threeimportant hydraulic variables: runoff, peak discharge rate and time to peak were examined.The absolute error for runoff depth prediction varied from-0.08to7.4mm, the mean was0.35mm; the relative error changed from8%to^-103%, the mean was^-1%. The maximumabsolute and relative errors for peak discharge rate prediction were^-1.85m³s^-1and-63%,and the mean were-0.02m³s^-1and10%, respectively. For the prediction of time to peak,the maximum absolute and relative errors were0.99h and^-109%, and the mean were-0.09h and^-17%. Moreover, the slopes of linear fitting for peak discharge rate and time to peakwere1.09and1.04（both close to1）, with coefficients of determination （R²） both close to1（0.99and0.97）. For runoff prediction, the slope and R²values for the linear fitting were0.83and0.78. The root mean square error （RMSE）, model efficiency coefficient （E）, andcoefficient of residual mass （CRM） were calculated for the simulations of each hydraulicvariable. It was shown that the simulation of peak discharge rate was best, followed by thatof time to peak, and runoff.（5） A new method was suggested to accurately measure the discharge from thewatersheds by measuring the flow velocity and using the relationship between flowvelocity and discharge rate. The power function of flow velocity-discharge rate wasestablished by deriving the inverse function of Hydraulic Geometry power function, takingthe discharge rate as the dependent variable with flow velocity as the independent variable.The inverse power function model was tested by the flow velocity-discharge rate data fromthe measuring weirs of Luoyugou （W100）, Lvergou （W10）, Qiaozi-East （W1-a） andQiaozi-West （W1-b） watershed outlets. According to the calculation of coefficients ofdetermination （R²）, the model calibration of W100was best, followed by those of W1-a, W10and W1-b. Based on model efficiency coefficient （E） calculation results, thesimulation accuracy of W1-a watershed was highest, followed by those of W100, W10andW1-b. The study indicated that the derived power function could be used to determinedischarge rate at study watershed outlets. Therefore, a new method was developed formeasuring discharge rate given the measurement of the flow velocity instead of flow depth.The results of the study contribute to better understanding watershed hydrologicalcycle, and could supply basic tools for the study of water resource optimization, soilerosion, and sediment transport for the small watersheds on the Loess Plateau.