Reservoir Inflow Forecasting with Clustering vs. Climate Precipitation Prediction; Case Study: Amir Kabir Dam, Iran

Real-time River flow forecasting is essential for optimal operation of water resources systems, especially reservoir dams. However, due to the high complexity of hydro-climatological events, real-time forecasting is very difficult and often error prone. Therefore, improved methods need to be developed for mid-term prediction models for river flows, especially for inflows to dam reservoirs. For this purpose, we used the extended FVNAM (FloodView Nedbør Afstrømnings Model, FVNAM model was developed by Payeshgar Tadbir Afzar Company. For monthly simulation, its linear reservoir routing has been modified), rainfall-runoff model and the SCE (Shuffled Complex Evolution) algorithm for calibration, the inflow to Amir Kabir Dam, which is in the north of Alborz province in Iran, was modeled and the best rainfall-runoff model was determined for predicting inflows to this dam in future years. The RBF (Radial Basis Function) artificial neural network was then used to classify the cumulative precipitation of the rainy season in the historical period at Siah Bisheh Station which is located at the geographical position of 51°19′ longitude and 36°13′ latitude, into five categories, and then the suitable category for precipitation prediction was selected. Finally, the predicted monthly precipitation was used in the rainfall-runoff model to estimate the monthly inflow to the dam. The runoff volume was also estimated using the predicted precipitation by climate models. The results of these two methods were then compared with the observational data. According to the results, the inflow estimated by the precipitation obtained from the classification with the Nash–Sutcliffe efficiency (NSE) of 0.893 proved to be better than the precipitation obtained from climate models. Moreover, the GFDL (Geophysical Fluid Dynamics Laboratory) climate model was considered the best for the 1-month forward time step. In contrast, the CFSv2 (Climate Forecast System Version 2) model was superior for the 2-month and 3-month forward time steps.