Modeling uncertainty of statistical downscaling methods in quantifying the climate change impacts on late spring frost risk over Iran

Abstract

Late spring frosts (LSFs) play a key role in the evaluation of climate suitability for agricultural and horticultural crop production and are considered to be one of the main components of food security. It is expected that future climate change will affect the occurrence of LSFs and its damages. In order to quantify these changes, this study aims to investigate the performance of two downscaling methods (SDSM and ANN), using two GCMs (general circulation model) data of HadCM3 (A2 and B2 scenarios) and CGCM3 (A1B and A2 scenarios). To this end, daily minimum temperature data (Tmin i ) of 50 meteorological stations located across the entire country were gathered and quality controlled for the period 1961 through 2010. The trend analyses of annual minimum temperature series showed a significant increasing trend of 0.3 °C per decade (α = 0.01). Both downscaling models were calibrated for the 40 years of 1961–2000 and the evaluation is conducted for the 10 years of 2001–2010. The Wilcoxon rank-sum and the Levene and bootstrapping tests were used for comparing and finding confidence intervals of averages and variances of downscaled daily minimum temperatures at each month. Results showed that, firstly, downscaling models’ performance in simulating averages is much better than variances. Secondly, SDSM simulates warmer summers and colder winters comparing ANN. Thirdly, the best and the worst results were achieved by ANN_CGCM3_A2 and ANN_HadCM3_B2, respectively. Finally, very low, moderate, high, and very high-risk zones for critical temperature of 0 °C and their areas at the historical period (1961–2010) and three future periods (2011–2040, 2041–2070, 2071–2100) were quantified and compared. Based on the results, SDSM simulated changes in the high and low frost risk zones more than the ANN model. However, since both Very High and No Frost risk zones will drastically change, SDSM results do not always indicate an increasing risk of frost damage. Moreover, the average of the area covered by “low-risk” zone will increase from 1.00% in the current period to 4.43% at the end of the century, and area of “very high-risk” zone will decrease from 8.2 to 4.7% at the same condition.