A stochastic multi-station model for daily rainfall generation in North-East of Iran: Effect of time non-stationarity

For many models (e.g. hydrological, meteorological, crop yield) stochastic daily rainfall generation is required. Most of the stochastic models are single-site, while there are rather few ones that deal with the rainfall correlation structure (occurrence and amount) as a multi-site approach. A plausible shortcoming of these models, however, is due to not considering the possible time-non-stationarity. A total of 36 raingauges stations in North, Razavi and South Khorasan provinces, northeast of Iran with 30 years of record were considered in this study. A stochastic rainfall simulation model for 6 rainy months of November to May was adopted, in which, first order Markov approach for rainfall occurrence and Gamma probability density function for rainfall amount were involved. Model parameters (rainfall probability conditioned to rainy and dry for previous day for rainfall occurrence and two parameters of Gamma distribution) were found to be dependent on the month of the year and geographical location; yet, no significant relations were found to describe them. It was showed that all parameters were non-stationary in time, such that considering this behavior, increased the accuracy of simulations.