ANALYSIS OF PRECIPITATIONS BY NON-PARAMETRIC TESTS OF MANN-KENDALL AND KRUSKAL-WALLIS

Abstract

The main objective of this research was to verify the long-term rainfall behavior in a location in the Brazilian semiarid region. Century-long monthly precipitation data (1911 to 2017) provided by DNOCS/SUDENE were used in the investigation. These data were used to determine descriptive statistics at monthly, annual, decadal and climatic scales. Next, nonparametric statistical tests were applied: Mann-Kendall (trend) and Kruskal-Wallis (multiple comparisons), both with a statistical confidence level of 95%. It was observed that 67.8% of annual rainfall (661.4 mm) is concentrated in three months (Feb-Mar-Apr) and that there was no significant tendency of monthly and consequently, annual rainfall. The 1930s, 1950s, 1990s, and 2010s were the driest years in the series, showing that the current dry period (2012-2016) is not unexceptional. Although significant trends were observed at 1% rainfall increase between 1930-1980 and 1% rainfall decrease from 1960-2000, when the full series was analyzed there was no significant trend. For climatologies the differences were significant at 1% when comparing Climate_2 (1958-1987), which was the wettest, with Climate_1 (1928-1957) and Climate_3 (1988-2017). It is concluded that the current climate and decade have had significantly less rain, but they are not unique since other decades and/or past climate series have been drier.