Evaluating Satellite and Reanalysis Rainfall Products for Extreme Events in Rio de Janeiro State, Brazil

This study evaluates the accuracy of satellite-derived and reanalysis precipitation datasets (CHIRPS v2.0, PERSIANN-CDR, ERA5, AgERA5, and ERA5-Land) and examines long-term rainfall trends across the State of Rio de Janeiro. The methodological framework combines performance assessment against ground-based observations using statistical metrics, including Percent Bias (Pbias) , Root Mean Square Error (RMSE) , and Willmott’s Index of Agreement (d) , with trend detection based on the Mann–Kendall test. The results indicate a clear performance hierarchy, with CHIRPS v2.0 exhibiting the highest overall agreement with observations, characterized by lower bias and reduced error magnitudes, while ERA5-based products show intermediate performance and PERSIANN-CDR displays the lowest consistency, with pronounced bias and spatial variability. For extreme precipitation indices, including RX1Day and R95p, all datasets systematically underestimate rainfall intensity, particularly in regions influenced by complex terrain and coastal processes. Trend analysis reveals a consistent divergence between gridded datasets and observational records, as satellite and reanalysis products indicate declining trends in total precipitation (PRCPTOT) and attenuation of extremes, whereas station data show a more heterogeneous pattern, including significant increases in annual precipitation along the Green Coast Region and intensification of daily and seasonal extremes in coastal and metropolitan areas. These findings highlight the limitations of current gridded datasets in representing localized precipitation processes and emphasize the need for region-specific validation and bias correction when applying satellite-derived products in hydrological and climate-related analyses.