Regime Shift in Parameters of Multi-Site Daily Rainfall Stochastically Generation

Many models (e.g. hydrologic, meteorologic, crop yield) need for stochastic daily rainfall generation. Most of the stochastic models are single-station and rather few deal with rainfall correlation structure (occurrence and amount). A plausible shortcoming of the later, however, is due to not considering the possible time-non-stationarity. A total of 36 rain gage stations with 30 years of record were considered in North-, Razavi-, and South-Khorasan Provinces for this study. One stochastic rainfall generation consisting of first order Markov model for rainfall occurrence and Gamma probability density function for rainfall amount for 6 rainy months of November to May was adopted. Results showed that the model parameters (rainfall probability conditioned to previous day rainy or dry state, and two parameters of Gamma distribution) were depended on the month of the year and on the geographical location. Yet significant relations were not found to describe them. It was shown that all parameters were non-stationary in time. Non-stationarity was modeled through regime shift concept. Results showed that nearly all parameters of rainfall amount model (Gamma probability density function) were independent of regime shift. This may be a clue for a plausible regional probability density function. However, the regime shifts for all parameter were not consistent which poses difficulty in modeling the stochastic generation of daily rainfall in the region.