Analysis of late spring frost dates over Iran under current climate and future scenarios

Abstract

Frost damage, especially in spring, can have profound effects on agricultural and horticultural crops so that in some years, Iranian insurance companies have to pay millions of dollars of cost to the farmers. It seems that the costs tend to change due to changing in frost features under future climate scenarios. Therefore, the purpose of the study is to investigate these changes and examining the ability of SDSM and Artificial Neural Network (ANN) to downscale AOGCM output data HadCM3 (A2 and B2 scenarios) and CGCM3 (A1B and A2 scenarios). To this end, daily minimum temperature data (Tmini ) of 50 synoptic stations all over Iran for the last 50 years (1961–2010) were collected and quality controlled. The trend analyses of Late Spring Frosts (LSF) dates during historical period showed a significant (at the confidence level of 5%) downward trend of 1.4 days per decade, especially after 1980s. Consequently, statistical tests including Wilcoxon rank-sum method, Levene’s test, and bootstrapping were applied, respectively to compare the mean and variance of LSF dates and to determine the uncertainty of LSF dates at the 95% confidence level. Results revealed that, (a) performance of both downscaling techniques in simulating averages is much better than variance, (b) SDSM simulates warmer summers and colder winters in comparison with ANN; (c) the best and worst accordance between observations (during 2001–2010) and different projections obtained from ANN_CGCM3_A2 and ANN_HadCM3_B2, respectively. Finally, different zones of average dates of LSF occurrence (0 °C or colder) and their areas at current condition and three future scenario period (2011–2040, 2041–2070, 2071–2100) were quantified and compared. According to results, SDSM simulated the probable effects of climate change on LSF more intensively compared to the ANN model.