Hossein Emami; Ali Salajegh; Alireza Moghaddamnia; Shahram Khalighi; Abolhassan Fathbabadi
Abstract
Precipitation is of the most important inputs of runoff modeling. The availability of precipitation data with appropriate temporal and spatial accuracy is very important and necessary for watersheds with small and scattered rainfall stations. Nowadays, climatic satellites are practical and widely-used ...
Read More
Precipitation is of the most important inputs of runoff modeling. The availability of precipitation data with appropriate temporal and spatial accuracy is very important and necessary for watersheds with small and scattered rainfall stations. Nowadays, climatic satellites are practical and widely-used tools in precipitation estimations. In this study, first the efficiency of TRMM satellite precipitation data in the monthly time series of Chehelchai Watershed was evaluated using R2, RMSE, NSE and Bias statistical indices by comparing the precipitation data of rain gauge stations (observed) and the values of these statistical indices were 0.54, 22.70, 0.44 and -14.86, respectively. Considering the value of the coefficient of determination (R2), it can be concluded that the TRMM satellite was able to estimate the 0.54 of observed precipitation. In the next step, three base data models including MLP, ANFIS and SVR were used to estimate the monthly runoff. Two different input scenarios were selected :1) observed precipitation data in t and t-1 time steps and runoff in t-1 time step and 2) satellite precipitation data in t and t-1 time steps and runoff in t-1 time step. To compare the accuracy and error of the models, R2 and RMSE of the validation stage were used. The ANFIS model with the values of R2 and RMSE were 0.80 and 0.97 for the first type input combination and 0.78 and 1.02 for the second type input combination, respectively, as the suitable single model for estimating runoff in the study area were selected. Then weighted-mean method was used in the data fusion approach to provide a data driven combination model for each combination of inputs into the model in the studied watershed. This data fusion approach data-driven model improved the values (R2=0.81) and (Bias=-4.85) for the first type input combination and also improved the value (R2=0.79) for the second type input combination.
