Autocorrelation And Multiple Testing Procedures In Trend Detection Analysis: The Case Study Of Hydrologic Extremes In Sao Francisco River Basin, Brazil

This paper investigates the presence of monotonic changes in hydrologic extreme indices in the Sao Francisco river basin, which plays a strategic role in Brazil. The study focuses on the following indices: number of consecutive dry (wet) days (CDD/CWD), number of days with precipitation greater than a given value (Rx), number of days with precipitation (Ndays), simple daily intensity index (SDII), annual maximum series of both precipitation (Rxt) and flows (Qmax), and annual 7-day averaged minimum flows (Qmin). The Mann-Kendall test in conjunction with the false discovery rate procedure were applied to series from 178 rain and 102 river gauges, with record lengths ranging from 30 to 88 years. When autocorrelation was significant, Pre-Whitening and Trend-Free Pre-Whitening procedures were applied. Results illustrate the effects of ignoring autocorrelation and multiplicity in the interpretation of the results, suggesting these factors should be carefully analyzed and incorporated in trend detection studies so as to avoid large number of false detection. A relatively large proportion of gauges located at the main two regions of the basin were declared non-stationary (5% significance level) for Ndays (20% -30%), SDII (30%) and Qmin (70%), and 30% of gauges in the Middle region for Qmax.