Assessment and characterization of the monthly probabilities of rainfall in Midwest Brazil using different goodness-of-fit tests as probability density functions selection criteria.

The probable rainfall is an excellent climatic parameter of information since it consists of the highest/lowest expected rainfall for a particular period of the year considering a certain level of probability. This study aims to evaluate traditional PDF performance using different goodness-of-fit tests to establish a criterion for choosing them on a monthly rainfall scale, in Mato Grosso do Sul, Brazil. The Cramer-von Misses (CVM) and Anderson–Darling (AD) tests were the most rigorous in rejecting the hypothesis of PFD suitable for monthly rainfall data, while the Kolmogorov–Smirnov (KS) was the least rigorous. The application of stricter goodness-of-fit tests as the CVM implies the use of up to 60% fewer series compared to the KS test. However, suitable series by the KS test presented erroneous estimates of probable monthly rainfall. The CVM and AD tests indicate the PDF with the best statistical performance (higher precision and accuracy between the observed and estimated frequency by the PDF) in more than 60% of situations. The most suitable PDFs for total monthly rainfall by the goodness-of-fit tests was gamma (12 months of the year). The exponential and Generalized Extreme Values (GEV) can be used for both the dry and rainy periods, respectively. The parameters of PDF are correlated with geographical variables, describing the total monthly rainfall distribution such as, for example, the influence of the South Atlantic Subtropical Anticyclone, the South Atlantic Convergence Zone in the rainy period, and the orographic effect in the dry period. [ABSTRACT FROM AUTHOR]