A RAINFALL GENERATOR FOR AGRICULTURAL APPLICATIONS IN THE TROPICS

A new rainfall generator, based on a third-order Markov chain, has been developed to simulate the year-to-year variation in rainfall amount that is observed in the tropics. To do this, some of the parameters of the model are themselves sampled randomly. The model has been fitted to over 40 sites in the tropics. The model was tested in detail for three sites in Central and South America and in the African Sahel. Few significant differences were found between the historical and simulated means and variances of yearly and monthly rainfall amount and raindays per month. There were some problems with simulating wet and dry sequences of longer than 9 days. When historical and simulated rainfall records were used to drive a crop simulation model, there was no significant difference in the mean and variance of maize yield response. The results reported suggest that the rainfall model performs adequately for many applications.