| description abstract | Study of soil erosion in the reservoir watershed, the main source of reservoir sedimentation that affects the reservoir’s lifespan and capacity, is of vital importance for watershed management. Due mainly to the lack of data, empirical formulas are commonly used to estimate reservoir sedimentation. However, these estimations are far from accurate. Field measurements data of discharge and suspended sediment were collected during three typhoon events in Shihmen Reservoir watershed, Taiwan. Temporal variations of water surface elevation, discharge, and concentration of suspended sediment were measured. A numerical model, Hydrological Simulation Program Fortran (HSPF), developed by the USEPA was adopted to simulate the sediment yield. However, as calibration and verification data are not always available and the parameter-calibration process is complicated and tedious for novice users of the model, an artificial neural network (ANN) model was proposed. Significant amount of the synthetic data from the calibrated HSPF model were first generated to train the ANN model, which in turn was used to estimate the sediment yield. Comparisons of the sediment yield using both the HSPF and ANN model give correlation coefficients of 0.96 for training and 0.93 for validation. Without the complicated parameter calibration process, the ANN model was faster and easier to use than the HSPF model. | |
