Bias correction of satellite precipitation products in the Levant

Study Region: This study focuses on the Levant, a climatically diverse and semi-arid Eastern Mediterranean region with strong rainfall gradients, complex topography, and uneven rain-gauge coverage complicating precipitation observation and hydrological assessment. Study Focus: Reliable precipitation estimation remains a fundamental challenge for regional hydrology. This study evaluates ten bias-correction methods, including seven statistical approaches and three machine-learning approaches, applied to four widely used satellite precipitation products: CCS-CDR, CMORPH, PDIR and IMERG. Daily observations from 361 rain gauges for 2004–2016 are used to assess two adjustment-validation strategies: a local station-specific Leave-One-Year-Out (LOYO) framework and a regional Leave-One-Station-Out (LOSO) framework. New Hydrological Insights for the Region: Conventional statistical methods provide limited improvements, whereas machine-learning approaches substantially improve performance, increasing R 2 by up to 0.40 and reducing RMSE by up to 35%. IMERG demonstrates the strongest baseline skill and the most consistent response to correction. Regional models consistently outperform local approaches, with regional LGBM producing the most accurate and spatially coherent correction. Additional rainfall-regime diagnostics show that while ML correction improves daily agreement, it modifies the occurrence of light intensity rainfall and upper-tail behaviour. These findings show that regional machine-learning adjustment can improve the reliability and transferability of satellite precipitation estimates, while distributional diagnostics remain necessary for hydrological applications sensitive to rainfall-regime structure.