Abstract Climate change has intensified extreme precipitation patterns, raising questions about the accuracy of traditional methods for constructing Intensity-Duration-Frequency (IDF) curves. However, the performance of Generalized Extreme Value parameter estimation methods across diverse climate zones remains underexplored, particularly for the southeastern United States. The study aims to assess the accuracy of L-moments and Generalized Maximum Likelihood Estimation (GMLE) for estimating the 24-hour precipitation component in IDF curves. This study provides a multi-station, multi-GCM comparison of L-moments and GMLE specifically for Alabama, addressing a gap in region-specific evaluations. The analysis was conducted using the historical simulations from ten Global Climate Models (GCMs), which were downscaled using the Local Constructed Analogs (LOCA) method. The daily precipitation data were extracted from the simulations and used to construct the Partial Duration Series (PDS) of extreme daily rainfall events. The analysis was carried out across fifteen stations across Alabama, and the results were compared with the NOAA Atlas 14 24-hour estimates. Results reveal significant spatial variability: TaiESM1 and GFDL-ESM4 models demonstrated the highest accuracy (mean MAE = 4.24 and 7.67, respectively). The L-moments method outperformed GMLE at 10 of 15 stations, especially for skewed data, with the lowest error at Paint Rock (MAE = 2.09). Conversely, GMLE performed better at stations with normally distributed data but produced greater inter-model dispersion. At Tuscaloosa, some GCM estimates for the 500-year return period reached 58 inches, over four times the NOAA reference (14 inches). The findings have direct practical implications. Employing L-moments with TaiESM1 and GFDL-ESM4 models for inland Alabama provides more reliable estimates for long-return periods critical for major infrastructure design. While coastal stations warrant more cautious approaches due to higher uncertainties (e.g., Dauphin Island, MAE = 26.20). They also support region-specific model selection for non-stationary IDF curves and inform infrastructure design under changing climate conditions. Graphical Abstract The visual abstract illustrates a framework for 24-hour IDF curve estimation applied to 15 representative stations across Alabama using LOCA-downscaled GCM data (1950–2014). This localized framework serves as a foundation for climate-resilient precipitation frequency analysis within the studied stations. The framework is based on a sequential five-step process that begins with daily precipitation data extraction from ten GCMs, followed by Partial Duration Series (PDS) development, GEV distribution parameters estimation by employing L-Moments and GMLE methods, precipitation depth calculation for 2–500-year return periods, and finally NOAA Atlas 14 evaluation by employing RMSE and MAE metrics. The main findings indicate that TaiESM1 and GFDL-ESM4 GCMs are recommended as the best-performing models for use in Alabama. The L-Moments method is recommended as the best approach due to its reliability and superior performance over GMLE at 10 out of 15 stations, as well as its robustness in dealing with data skewness and sample sizes. The optimized framework is recommended as a basis for developing IDF curves by employing L-Moments and TaiESM1 and GFDL-ESM4 models, which can form a basis for addressing the urgent need for non-stationary hydrologic design standards.
Opare et al. (Mon,) studied this question.