Enhancing the TRMM precipitation product in diverse regions of Iran through an intelligent-based post-processing approach

The significant role of remote-sensed precipitation data lies in alleviating the absence of readily accessible daily precipitation data, particularly in developing nations. The TMPA 3B42 V7 precipitation data, with (0.25^0×0.25^0) spatial resolution, widely used in satellite precipitation products, requires evaluation and correction at local scales. The aim of this study was to examine the reliability of the TMPA 3B42 V7 precipitation data in various climate regions in Iran. To achieve this, the data were compared with that of 103 synoptic stations from 2012 to 2017 using agreement indices, including the correlation coefficient. The findings reveal a notable association between the two datasets in terms of monthly and annual timeframes while displaying limited coherence on a daily level. In consideration of the disparity between satellite- and ground-based precipitation data, linear regression model (LRM) and artificial neural network (ANN) (specifically, a three-layer cascade-forward neural network (CFN)) were utilized to adjust the satellite precipitation data. The results demonstrate favorable performance for both in LRM and ANN models. The RMSE decreases to 20.8 and 20.7, while the NSE increases to 0.639 and 0.643, respectively. Specifically, the performance of the LRM model in correcting annual precipitation and the ANN model in correcting monthly precipitation surpasses that of other models. Additionally, the artificial neural network (ANN) displays inadequate performance in arid and semi-arid highlands located in the central region of Iran, as well as in the rectification of monthly and annual precipitation in the western vicinity of the Caspian Sea.