Multi-scale validation of gridded precipitation datasets in Central-West Brazil

Abstract Accurate precipitation estimates are essential for hydrological modeling, water resource management, agricultural planning, and climate impact assessments, particularly in tropical regions with sparse and uneven observational networks. This study evaluates the performance of twelve gridded precipitation datasets at daily, monthly, and annual scales across Central-West Brazil, using observations from 120 INMET automatic weather stations distributed across the Amazon, Cerrado, Pantanal, and Atlantic Forest biomes. Continuous statistical metrics, including RMSE, MAE, PBIAS, Willmott’s d, Spearman’s r, Lin’s concordance coefficient, and a Composite Model Ranking (MR), were combined with rainfall-event detection metrics and station-level spatial analysis. Most datasets reproduced mean precipitation consistently with station observations, especially at monthly and annual scales, although differences in error magnitude, bias, ranking, and event-detection skill were observed across biomes and temporal scales. BR-DWGD showed the best overall performance, with the lowest errors and highest agreement coefficients, while CPC performed strongly at the daily scale and in rainfall-event detection. CHIRPS v3 and the ensemble presented competitive performance, mainly at monthly and annual scales; however, the ensemble did not outperform BR-DWGD. Detection skill decreased as rainfall intensity increased, indicating that moderate and intense daily rainfall events remain difficult to represent for all datasets. The station-level analysis showed that the best-performing dataset varied spatially and by temporal scale. These findings highlight the importance of multi-scale, event-based, and spatially explicit validation for selecting precipitation datasets in heterogeneous tropical regions.