Optimization of Probability Density Functions Applicable for Hourly Rainfall

In order to improve the calculation accuracy of rainfall probability distribution in related applications, this study aimed select a theoretical function from applicable functions for three classes class-conditional density (CCPD) hourly series. The are generalized gamma (GΓD), normal (GND), and Weibull distribution. For reason that it is hard distinguish advantages disadvantages by plot error analysis fitted values, optimization criteria proposed, which Bayesian information criterion (BIC) estimated both annual average (AAR) continuous (AACR). results show using 15 regions, relative fitting deviations CCPD1 were less than 2.3% 3.3% ln(CCPD1). goodness-of-fit values all above 0.98 greater 0.94 effect was relatively poor perspective while could be used fit CCPD. GΓD had highest CCPDs, but there concern about overfitting due its broad spectrum. GND, with fewer parameters, comparable performance GΓD, when mean deviation 0.6%, coefficient determination 0.999, ln(CCPD1), they 1.45% 0.989. At same time, GND performed well estimating AARs, an 8.6% 0.92 correlation fifteen indicating can reflect spatial variation characteristics AAR. Moreover, form simple. follows parsimonious principle, suitable whole domain. Therefore, recommended as based on criteria. genetic algorithm adopted obtain approximate solution parameters through optimization, simplify derivation steps. multiplicative additive errors objective functions, gave comprehensive consideration ends curve.