Banca de DEFESA: FLAVIA FERREIRA BATISTA

Performance, Limitations, and Improvement of IMERG Precipitation Estimates over Northeastern Brazil: Applications in the Parnaíba River Basin

Satellite precipitation, bias correction, Quantile Regression Forest, Quantile Mapping, extreme indices.

The intensification of climate variability and extremes in recent decades has increased the challenges of hydrometeorological monitoring in semi-arid regions such as the Parnaíba River Basin, where the scarcity and poor distribution of rain gauge stations hinder adequate spatial representation of precipitation. In this context, improving remote sensing-based estimates has become essential, with the Integrated Multi-satellitE Retrievals for GPM (IMERG) product standing out as one of the most prominent recent initiatives. This study assessed the ability of IMERG to reproduce key features of the precipitation regime in the basin and sought to apply robust bias correction techniques to the estimates. Initially, IMERG V06 was analyzed by comparison with rain gauge data, using various statistical metrics and focusing on eight extreme precipitation indices. The results indicated satisfactory performance of IMERG V06 for more frequent events but revealed limitations in capturing extremes, especially during dry periods and in areas of elevated topography. The IMERG Final product showed greater daily agreement compared to the Early and Late runs, with improved results for annual total precipitation and reasonable performance regarding extreme indices. With the release of version V07, a comparative analysis between IMERG V06 and V07 was conducted using the Brazilian Daily Weather Gridded Data (BR-DWGD), previously validated against in situ observations. Good agreement was observed, supporting the use of BR-DWGD as a reference for regional comparisons. The results showed notable improvements in V07, with reductions in systematic errors and better spatial representation of precipitation, particularly in the central and southern regions of the basin, where RMSE decreased by up to 20% and mean bias remained below 5%. To capture the spatial variability of the rainfall regime, cluster analysis was applied to identify homogeneous regions with respect to precipitation behavior. This segmentation allowed differentiated evaluations of IMERG versions, considering the environmental specificities of each region. Based on these results, a bias correction strategy was implemented for IMERG V07 using statistical and machine learning methods, especially the hybrid approach combining Quantile Regression Forest (QRF) and Quantile Mapping (QM), which led to reductions of up to 30% in RMSE and improvements greater than 58% in the KGE index, in addition to advances in rainfall detection and reduction of false alarms. These findings demonstrate that the integration of regionalized validation, advanced correction methods, and multiple data sources is essential for improving climate monitoring, with direct implications for water resource management, agricultural planning, and early warning systems in the Parnaíba River Basin.