Abstract: | This study introduces an innovative methodology for enhancing satellite precipitation estimates in the mountainous areas of the United Arab Emirates (UAE) through the application of Long Short-Term Memory (LSTM) networks. The primary objective is to reduce and correct biases in the CMORPH and IMERG satellite precipitation datasets by integrating covariates such as topographic data, minimum temperature, and distance from the coast. Recognizing the notable discrepancies often observed between satellite-derived estimates and in-situ gauge data, especially in complex terrains where orographic effects significantly influence rainfall patterns, this research leverages LSTM’s capability to handle time series data and capture long-term dependencies to model precipitation dynamics. Daily rainfall data spanning 2004 to 2021, alongside topographical and climatic variables, were utilized to train the LSTM model to identify and rectify the biases present in CMORPH and IMERG compared to ground observations. Results demonstrate that the LSTM-based approach markedly improves the accuracy of these satellite precipitation products. The incorporation of terrain and climatic covariates enhances the ability to capture orographic precipitation effects, facilitating more effective bias correction and leading to higher-quality precipitation data. The study’s effectiveness was evaluated using statistical measures including the Nash-Sutcliffe Efficiency (NSE), Root Mean Square Error (RMSE), Normalized Mean Absolute Error (NMAE), Probability of Detection (POD), and Critical Success Index (CSI). The findings indicate that NSE scores improved by nearly 45% for most stations, and RMSE values decreased from 3.4 mm to 0.5 mm. Additionally, the adjusted products demonstrated a 30% increase in accuracy for capturing actual rainfall events, as reflected by the POD and CSI metrics. This research contributes significantly to hydrology and climate science by refining satellite-based precipitation estimates in mountainous regions and exemplifies the potential of advanced machine learning approaches in environmental studies. The outcomes have substantial implications for water resource management, agricultural planning, and disaster mitigation strategies in the UAE and other regions with similar topographical challenges. |
---|