Analysis of wet and dry behavior of weather through Markov models

Abstract

This study attempts to model the daily rainfall climatology of Sri Lanka using Markov models at three selected weather stations where long term data records were available. Both the first and the second order Markov models were able to forecast the occurrence of daily rainfall to an accuracy of ∼72±4%. The accuracy of predicting wet days during the wet season was very much higher than in the dry season, while predicting dry days during the dry season was very much higher than in the wet season. The mean number of wet spells per month and mean length of a wet spell per month has been forecasted using standard probability distributions combined with transition probabilities based on the first order Markov model. The deviation of the simulated values from the observed values was lower than the statistical variations which indicate that the model is suitable for simulating wet and dry spells in Sri Lanka.