Stochastic simulation model of daily precipitation for a region

Abstract

A stochastic simulation model of daily precipitation was developed in FORTRAN computer language for the Brazilian Cerrado Region. The model was built using basically two variables: (1) the probability of rain in any one day of each month, and (2) the probability of occurrence of that rainfall within a certain magnitude. The first variable was defined by a transition probability which takes into account the historical series of precipitation, based on the Markov process, where the system jumps from one state to another, in successive assays, considering what happened in the latest state. The second variable, the precipitation magnitude, was also consecutively generated for each "rainy day", according to the probabilities of real occurrence frequencies of such monthly rainfall classes. Three different simulation runs were made in order to evaluate the model and their results compared whith the historical data. Simulated annual rainfall means were above low values of the historical series but deviations ranged from only 0.9 to 3.6%. Monthly means and their standard deviations (SD) were also reasonably dose to the historical means, the December SD of simulations being underestimated. Other variables used to evaluate the fitness of the simulations were: (1) mean number of rainy days per month, (2) monthly frequency of precipitation according to the magnitude, (3) monthly frequency of continuously rainy days, (4) monthly frequency of continuously rainless days. All comparisons showed a good agreement with real data.