Stream Flow Prediction Using Support Vector Machine Based on Discharge and Precipitation Time series on Upstream Stations (Case Study: Taleh Zang Hydrometric Station)

In this research, the support vector machine model was used to predict the Taleh Zang stream flow base on the data from 10 hydrometric and 8 rainfall stations upstream of the watershed in 20 years (1992-2011). To this aim and in the first step, influence of applying flow rate time series, rainfall, and a combination of these two parameters as an input; and in the next step, influence of number of upstream hydrometric and rainfall stations on forecast results were examined. The results were compared by using three statistical indicators: correlation coefficient (R2), Root-Mean-Square Error (RMSE), and standard Error (SE). The results showed that, using rainfall statistics along with the stream flow as an input to the model with R2of 0.884, RMSE of 38.41, and SE of 0.28 in comparison to the stream flow statistics as an input to the model with R2 of 0.871, RMSE of 40.20, and SE of 0.29 will improve accuracy of the forecast. Whereas, only application of rainfall time series with correlation coefficient of 0.225, RMSE of 157.73 and SE of 0.62 greatly increases error values in the results. Moreover, through increase of number of upstream hydrometric and rainfall stations, the model would be capable of forecasting stream flow with more accuracy.