ABSTRACT
The Standardized Precipitation Index was developed as a probability-based index able to monitor rainfall deficit in a standardized or normalized way. Thus, the performance of this drought index is affected by the use of a distribution that does not provide an appropriate fit for the rainfall data. The goal of this study was to evaluate the adjustment of the gamma distribution for the rainfall amounts summed over several time scales (Pelotas, Rio Grande do Sul, Brazil), to assess the goodness-of-fit of alternative distributions to these rainfall series and to evaluate the normality assumption of the Standardized Precipitation Index series calculated from several distributions. Based on the Lilliefors test and on a normality test, it is verified that the gamma distribution is not suitable for calculating this Index in several timescales. The generalized normal distribution presented the best performance among all analysed distributions. It was also concluded that the drought early warning systems and the academic studies should re-evaluate the use of the gamma distribution in the Standardized Precipitation Index calculation algorithm. A computational code that allows calculating this drought index based on the generalized normal distribution has also been provided.
Key words:
rainfall departures; normality test; goodness-of-fit tests
The dark squares denote that the distribution cannot be used to assess the probability of the rainfall amounts. The open squares indicate the opposite
The dark squares denote non-normal distributions. The open squares denote normal distribution
