IMPROVING GEV SHAPE PARAMETER ESTIMATES FOR SUB-DAILY IDF CURVES UNDER SHORT RAINFALL RECORDS THROUGH HIERARCHICAL BAYESIAN INFERENCE

Código

V-END_II-EBHE0165

Autores

Larissa Zaira Rafael Rolim, Tomás Antonio Quezado Duval, Carlos Hennrique Ribeiro Lima, Dirceu Silveira Reis Junior

Tema

14 - Eventos e sistemas meteorológicos extremos;

Resumo

This study evaluates Bayesian hierarchical GEV models for estimating sub-daily IDF relationships under short rainfall records in Rio Grande do Sul, Brazil. The work compares two Bayesian structures for the GEV shape parameter: one varying by station and another varying by station and duration. Conventional independent GEV fits using maximum likelihood and L-moments are used as benchmarks. Results show that the Bayesian model with station-specific, duration-independent shape parameter achieved the best predictive performance according to PSIS-LOO and WAIC. It also substantially reduced uncertainty in 20-year return levels, with a median 95% interval width of 16.4%, compared with 19.3% for the station-duration Bayesian model, and 35.7% for L-moments. The conventional MLE approach showed severe instability under short-record conditions, especially for extrapolated return levels. The analysis indicates that stabilizing the GEV shape parameter is essential for reliable sub-daily IDF estimation. Bayesian hierarchical regularization reduces the influence of isolated extremes, produces more physically plausible tail estimates, and improves return-level stability.