Stochastic modelling of daily precipitation for Canada

Abstract

This study assesses the performance of various stochastic models in generating daily precipitation amounts and the durations of wet and dry spells across Canada for the period 1971–2000. The models are tested at 657 stations representing the wide range of climate variability across the country. It is found that the simple firstorder Markov chain (SMC) model is capable of reproducing the statistics of dry and wet spell durations reasonably well. However, the SMC model also yields a substantial over‐dispersion problem, resulting in a considerable reduction of interannual variability in monthly total precipitation. This is mainly attributable to smaller variability in the frequency of wet days. This inadequacy is improved by adding a separate model to simulate the number of wet days in a year. The modification to the SMC model has an advantage over alternative approaches using higher order Markov chains since it requires the estimation of fewer parameters. The generation of daily precipitation amounts is tested using the exponential, gamma, skewed normal, and mixed exponential probability distributions, with one to three parameters. Results indicate that the mixed exponential distribution is superior in general, especially during warmer months, while the gamma distribution is adequate for winter months.