Regional flood frequency analysis and uncertainties: Maximum streamflow estimates in ungauged basins in the region of Lavras, MG, Brazil

Regionalization techniques are an important alternative to overcome the scarcity of streamflow data and to provide adequate estimates for flood management. This work aims to generate a maximum streamflow regional function and evaluate the uncertainties in estimates for a basin in the South of Minas Gerais, Brazil. Six streamflow gauging stations, that presented stationarity and no trends according to the Mann-Kendal test, were used to develop the regional function by the dimensionless curve method. The annual maximum streamflow series were adjusted to the probability density functions (PDFs) 2-parameter Log-Normal, 3-paramenter Log-Normal, Gumbel, and Gamma. Then, the goodness-of-fit of the PDFs was verified by the Anderson-Darling test. For the regional analysis, heterogeneity and discordancy measures verified the homogeneity of the region, and the jack-knifing technique evaluated the predictive performance of the regional function by the coefficient c of Camargo and Sentelhas. The uncertainty analysis for at-site and regional function estimates was assessed by Monte Carlo simulations and bootstrapping. The performance of the regional function was classified as optimum for most of the watersheds, but as good when considering the prediction in certain return periods. The uncertainties were larger in ungauged basins, especially for greater return periods and drainage areas. The regional function developed can be employed in ungauged sites, but also to improve data in poorly monitored watersheds. Thus, this study demonstrates the adequate applicability of regionalization in ungauged watersheds and provides a simple alternative for maximum streamflow estimates in order to promote sustainable flood risk management.