Seasonal autoregressive models for estimating the probability of frost in Rafsanjan

This work develops a statistical model to assess the frost risk in Rafsanjan, one of the largest pistachio production regions in the world. These models can be used to estimate the probability that a frost happens in a given time-period during the year; a frost happens after 10 warm days in the growing season. These probability estimates then can be used for: (1) assessing the agroclimate risk of investing in this industry; (2) pricing of weather derivatives. Autoregressive models with time-varying coefficients and different lags are compared using AIC/BIC/AICc and cross validation criterions. The optimal model is an AR (1) with both intercept and the “autoregressive coefficients” vary with time. The long-term trends are also accounted for and estimated from data. The optimal models are then used to simulate future weather from which the probabilities of appropriate hazard events are estimated.