Flood-Rich and Drought-Rich Periods in Brazil: Detection and Attribution

Código

I-EBHE0095

Autores

Vinícius Bogo Portal Chagas, Pedro Luiz Borges Chaffe, Günter Blöschl

Tema

WG 1.08: Deep Explanation & Evaluation for Practices in Hydrological Changes (DEEPHY)

Resumo

Streamflow extremes often exhibit decadal variabilities when analyzed in time series spanning about a century or more. These nonlinear variabilities can be analyzed in terms of flood-rich and drought-rich periods, when the most extreme floods and droughts occur within short periods of time rather than sparsely distributed over the century. The frequency of occurrence of these nonlinearities and their causes remain unknown in many regions, particularly in the tropics, as is the case in Brazil. Here, we examine flood-rich and drought-rich periods in 319 streamflow gauges and 312 rainfall gauges in Brazil from 1940 to 2020. These clustering periods are detected by computing annual maxima and minima time series and using scan statistics to verify whether events exceeding thresholds of 5, 10, and 20-year return periods follow the Bernoulli process and binomial distribution. We compensate for the multiple testing problem and reduce the impact of spatial autocorrelation by controlling for the false discovery rate with a significance level of 5%. The causes of streamflow time clustering are investigated by contrasting it with rainfall-rich and -poor periods, water abstractions, and sea surface temperature oscillations in the Atlantic and Pacific through binomial regressions. Results show that the number of basins with significant flood-rich and drought-rich periods is higher for lower return-period thresholds. For a 5-year threshold, flood-rich periods were observed in 17 % of the basins (3.5 times the false positive rate), while drought-rich periods were observed in 81 % of the basins (16.7 times the false positive rate). For a 10-year threshold, those figures are 8 % and 69 % respectively; for a 20-year threshold, they are 1 % and 54 %. Brazil experienced a long dry period from the 1940s to the 1960s, flooding in the 1980s, and record droughts from the 2000s onward. The higher frequency of basins with drought-rich periods is linked with a sharp increase in water abstractions since the late 1990s, an interannual persistence of water deficits, and a higher frequency of rainfall-poor periods than rainfall-rich periods. Flood-rich periods are aligned with rainfall-rich periods mostly in terms of extreme rainfall (7-day maximum annual rainfall) rather than the average annual rainfall. In contrast, drought-rich periods are aligned with rainfall-poor periods mainly in terms of average annual rainfall as opposed to seasonal rainfall, with the addition of a significant impact of water abstractions. Flood-rich periods are satisfactorily estimated with the binomial regressions using ocean temperature data (R2 from 0.7 to 0.9), with considerable impacts of the Pacific Decadal Oscillation and the Tropical Southern Atlantic Index. Drought-rich periods are also satisfactorily estimated with the regressions, although with most noticeable impacts of the Pacific Decadal Oscillation and the Atlantic Multidecadal Oscillation. The findings noted here highlight the nonlinearity of flood and drought change when investigating time series longer than a few decades.