Influence of ENSO and PDO on the exceedance probability of annual maximum flows in the Uruguay Hydrographic Region, Brazil

Código

I-EBHE0180

Autores

Luana Oliveira Sales, Dirceu Silveira Reis Junior

Tema

WG 1.10: Hydrologic Design - Solutions & Communication

Resumo

Sales & Reis (2021) identified gauges in the Hydrographic Region of Uruguay where annual maximum flow series are nonstationary. This paper advances that study by evaluating if large-scale climate indices can explain that behavior. We consider four different statistical models based on the Generalized Extreme Value distribution. The first model (Model 0) assumes stationarity, whereas the second (Model 1), third (Model 2) and fourth (Model 3) models assume the series is nonstationary, allowing the location parameter to vary with covariates (time and ENSO and PDO indices) with a constant coefficient of variation (Cv), with a changing Cv but with a constant scale parameter, and both the location and scale parameters to vary with these covariates, respectively. The generalized maximum likelihood procedure was used to estimate the parameters of all four models. The selection of the best model for each of the 33 gauges used in the analysis was based on the Bayesian Information Criteria (BIC) for the local analysis, complemented with the False Discovery Rate (FDR) for the regional analysis to control the type I Error, but not discussed here. The proportion of gauges in the region where a nonstationary model was more appropriate than the stationary one was 76% (Niño 1+2), 91% (Niño 3), 55% (Niño 3.4), 18% (Niño 4), and 39% for PDO. When ENSO is used as covariate, two nonstationary models were selected for the set of nonstationary gauges: one where the location parameter of the GEV varies linearly with the covariate but maintains the coefficient of variation (? ? ) constant; and the other where ? ? can change freely. Estimates of the 50-year flood, used here to illustrate the importance of evaluating nonstationary modeling, changed substantially over the historical period when Niño3.4 is used in the analysis, with maximum relative change of this flood quantile varying from 24 to 105% depending on the gauge.