Evaluating the hydrological utility of satellite-based rainfall products using neural network models over the Ghare Ghieh River basin, Iran

In recent years, gridded precipitation data derived from satellite rainfall products have become critical data sources for hydrological applications, especially in ungauged basins where rain gauges are sparse or nonexistent. Also, in streamflow simulations, since the existing rainfall–runoff modelling methods require exogenous input with some assumptions, neural networks can be an efficient solution. In this paper, to simulate daily streamflow on the Ghare Ghieh River basin in northwestern Iran, the Levenberg–Marquardt Neural Network (LMNN) and the Particle Swarm Optimization Neural Network (PSONN) models are proposed. These models are trained and tested with different input patterns from ground-based data for water years of 1988–2008. Then, three satellite-based precipitation datasets, including TRMM-3B42V7, TRMM-3B42RT, and PERSIANN with 0.25° × 0.25° resolutions from 2003 to 2008, are used as inputs for the best-trained models which were selected in the testing step. These products are evaluated before and after calibration in streamflow simulation, and the Geographical Difference Analysis method is used to calibrate them. The results showed that the PSONN model performed better than the LMNN model. Also, in both models, before calibration of satellite precipitation products, TRMM-3B42 showed better performance in streamflow simulation, and after calibration, TRMM-3B42RT performed much better.