Validating the IMERG remote sensing precipitation data for extremes analysis using the new hybrid depth duration frequency model

Abstract

This study introduces a novel hybrid model for estimating depth-duration-frequency (DDF) estimates by integrating four probability distribution functions (PDFs) – Gumbel, GEV, GPD, and EP–through a maximum likelihood-based weighting scheme. Addressing a critical gap in extreme precipitation analysis, where model selection often fails to capture the variability across diverse climate zones, this hybrid model dynamically allocates weights to each PDF component according to the prevailing climate conditions at each location, optimizing DDF estimates for both wet and dry climates. Using Integrated Multi-Satellite Retrievals for GPM (IMERG) data, DDF estimates were calculated across multiple durations and return periods and validated against NOAA Atlas 14 precipitation frequency estimates (PFE) from 2360 stations across the continental United States (CONUS). Results indicate high correspondence between IMERG-based DDF estimates and Atlas 14 PFE, with an average correlation coefficient of 0.71, an average relative bias of 45.6 %, and an NRMSE of 12.1 mm across return periods. The model demonstrated increased agreement over longer durations and in regions with higher rainfall, with correlation coefficients rising from 0.569 for 0.5-h durations to 0.768 for 72-h durations. Spatial analysis shows the hybrid model's robustness, particularly in capturing trends across both wet and dry regions, suggesting its utility for extreme rainfall estimation in ungaged and climatologically diverse areas. This hybrid approach provides a versatile and regionally adaptive tool for engineers, hydrologists, and policymakers, offering improved precision for flood risk management and climate resilience planning. The hybrid model ensures that the prevailing PDF based on regional susceptibility, gets the highest weight, thus dominating influence of the model. Future work aims to extend the Hybrid model application beyond CONUS, enabling broader global applications in diverse, data-scarce regions.