Clustering and assessment of precipitation climates using satellite precipitation products and improved machine learning precipitation models

A deep understanding of the different features of various rainfall regimes is essential for water resources management and meteorological studies. Due to the extensive spatial coverage, satellite precipitation products have afforded researchers an excellent opportunity to investigate various precipitation climates. This study employs cutting-edge satellite precipitation data and improved modelling techniques to innovatively investigate Iran's diverse precipitation climates. By harnessing the power of machine learning (ML), precipitation models were developed that significantly improved the accuracy of satellite-based precipitation estimates. The complexities of Iran's precipitation climates were investigated by combining satellite data and ML models with the K-means++ algorithm. Furthermore, the Mann-Kendall test was employed to scrutinize potential trends in precipitation frequency across varying months and climates to enrich the current body of knowledge further. In addition, a range of probability distribution functions was fitted to the ML precipitation estimates, evaluating the outcomes with the Akaike information criterion (AIC). The results show that both artificial neural network (ANN) and random forest (RF) models improved satellite product correlation results, increasing them from 0.84 to an outstanding 0.93. Moreover, the Mann-Kendall test reveals a positive trend in the frequency of precipitation associated with convective rainfalls during May, as indicated by satellite data, ML models and synoptic precipitation observations. Ultimately, this research holds significant value for water resource management and the ongoing pursuit of refining satellite rainfall products in the context of diverse rainfall regimes.