Evaluating persistence and identifying trends and abrupt changes in monthly and annual rainfalls of a semi-arid region in western India

Abstract

In this study, 43-year (1965-2007) monthly and annual rainfall of ten rainfall stations in a semi-arid region of western India are analyzed by adopting three tests for testing normality, and by applying autoregressive technique for examining persistence. Gradual trends are detected by three tests, and their magnitudes are estimated by Sen’s slope estimator. Further, abrupt changes are identified by using four tests and are further confirmed by two tests. The box-whisker plots revealed that the rainfall of June and September are right-skewed for all the stations. The annual rainfall of Bhinder, Dhariawad and Gogunda stations is found considerably right-skewed. The normality tests indicated that the rainfall in July does not deviate from the normal distribution at all the stations. The annual rainfall is found non-normal at five stations. The monthly rainfall of June, July and August has persistence at 3 (Mavli, Salumber, and Sarada), 2 (Kherwara and Sarada) and 1 (Mavli) stations, respectively. The annual rainfall is persistent at Girwa and Mavli stations. The significantly increasing trend is detected at Mavli in the monthly rainfall of July and in the annual rainfall (p-value>0.05), whereas the negative trend at Dhariawad in August month is found significant (p-value>0.10). This study revealed that the presence of serial correlation does not affect performance of the Mann-Kendall test. The mean values of the trend magnitudes for rainfalls of June, July, August and September months are 0.3, 0.8, -0.4, and 0.4 mm year-1, respectively, and the overall mean value at the annual scale is 0.9 mm year-1. It is found that the standard normal homogeneity test and Pettitt test are biased towards the end of the series to locate the change point. However, the Bayesian test has a tendency to look for a change point at the beginning of the series. The confirmed abrupt changes in rainfall occurred in the year 2003 (Bhinder) in June; years 1974 (Mavli) and 1989 (Dhariawad and Salumber) in July; years 1972 (Sarada), 1990 (Dhariawad), and 2003 (Mavli) in August; years 1977 (Dhariawad), 1991 (Sarada), and 2004 (Kotra) in September; and the year 1972 (Mavli and Sarada stations) at the annual scale. It is suggested that the significantly increasing trend of rainfall may have linkages with the climate change and variability. Moreover, this study recommends use of multiple statistical tests for making reliable decisions while analyzing a hydrologic time series.