Evaluating Performance of Four Statistical Downscaling Models (SDSM) of Precipitation and Temperature Data under the Fifth Assessment Report of the Intergovernmental Panel on Climate Change (IPCC) Scenarios

The investigations show that the increase in greenhouse gasses has led to an increase in the average temperature of the earth’s atmosphere and, consequently the global warming in recent years. Moreover, the rise of greenhouse gasses has also caused changes in other climatic variables such as precipitation. Given the fact that climate change will bring about considerable effects on the water resources, the prediction of the climatic condition and water resources of a region is one of the planning prerequisites for the economic and social development of every country. Therefore, considering the climatic condition and volume of water resources in a country in the future is inevitable. Since the initial effects of climate change are exerted on the meteorological variables of precipitation and temperature, and any change in these variables results in disorder in many hydrological phenomena, it is essential to precisely examine the future changes of these parameters and their simulation methods. Hence, this study was conducted to evaluate the uncertainty of 20 AOGCM models under two emission scenarios of RCP2.6 and RCP8.5 in the two statistical periods (1961-2006) and (2006-2018) in the drainage basin of Karaj River. The changes in the parameters of minimum temperature, maximum temperature, and precipitation in the period of (2020-2040) compared to the baseline period were also investigated.

At first, the downscaling of the parameters of minimum temperature, maximum temperature, and precipitation was carried out using the two statistical periods of (1961-2006) and (2006-2018) and the large-scale NCEP variables as the inputs for the models of Artificial Neural Networks (ANN), LARS-WG, SDSM, and Change Factor method to determine the model’s error and evaluate its performance. For this purpose, the features and functions available in the programming environments of MATLAB, LARS-WG, and SDSM5.5.2 were employed after data normalization. The bootstrap confidence interval method and statistical criteria such as coefficient of determination (R2), Nash–Sutcliffe coefficient (NS), percent bias (PBIAS), and PRS were used to evaluate the performance of the models. Due to the suitable range of each one of statistical evaluation coefficients being within the confidence interval defined by the bootstrap method in the period of (1961-2006), the most appropriate climatic models were chosen for this from the selected AOGCM climatic model of the fifth report. In order to improve the accuracy of the chosen models in the period of (2006-2018) under the scenarios of RCP2.6 and RCP8.5, the most appropriate model for each AOGCM scenario was selected using the statistical indexes and bootstrap confidence interval method. In the next step, the chosen models of each climatic scenario were weighted using the MOTP approach for the final evaluation and selection of the model with the lowest uncertainty. Each one of these weights indicates the capability of each model in simulating the variables of minimum temperature, maximum temperature, and precipitation of the intended month, which was conducted in both scenarios of RCP2.6 and RCP8.5. The weights are presented as percentages, and the highest weight is 100, which is also considered the highest score for each model. The model chosen in this step conducts the downscaling and simulation of statistical parameters with lower uncertainty compared to the other AOGCM models and the downscaling models investigated in this study. In this stage, the confidence interval of the model (MPI-ESM-LR downscaled by the LARS-WG model) was calculated by the bootstrap method to ensure the efficiency of the chosen model. In order to analyze the accuracy of the model in downscaling the climatic elements, the number of station-months within the confidence interval was taken into account. In the last step, the climatic parameters predicted by the chosen optimum model (the model with the lowest uncertainty) in the future period (2020-2040) were compared with those in the baseline period (1999-2018). The changes in the precipitation, minimum temperature, and maximum temperature in the region were also investigated.

In the period of (1961-2006), among 20 AOGCM climatic models of the fifth report, the nine models of MIROC-ESM, CESM1-WACCM, CSIROC-MK3-6-0, EC-EARTH, GISS-E2-H, GISS-E2-R, MIROC-ESM-CHEM, MPI-ESM-LR, and MPI-ESM-MR were selected as the optimum climatic models in the drainage basin of Karaj according to the suitable range of each one of statistical evaluation coefficients for the models being within the confidence interval defined by the bootstrap method. More detailed investigations on the nine selected models in the (2006-2018) period under the RCP2.6 and RCP8.5 emission scenarios showed that the MPI-ESM-LR climatic model simulates and downscales the maximum temperature parameter in the two scenarios of RCP8.5 and RCP2.6 and the precipitation in the RCP8.5 scenario with the lowest uncertainty using the LARS-WG model. The GISS-E2-R2 model has similar performances on the minimum temperature in the RC8.5 scenario. Moreover, the EC-ERTH model simulates and downscales the precipitation with the lowest uncertainty. The three chosen models were weighted using the MOTP approach for the final evaluation and selection of the best model. The results showed that in most of the months, the highest weight percent in the simulation of climatic variables belongs to the MPI-ESM-LR model. This model is more capable of simulating the precipitation, minimum, and maximum temperatures in both scenarios of RCP8.5 and RCP2.6. Eventually, the confidence interval of the MPI-ESM-LR model (downscaled by the LARS-WG model) was calculated by the bootstrap model to ensure the efficiency of the chosen model. According to the results, it was found that among the 72 station-months, the average maximum temperatures in 62 cases were within the confidence interval.Furthermore, the monthly analysis of average and variance of the minimum temperature, maximum temperature, and precipitation showed that in most of the months, these parameters are within the confidence interval. This shows the high accuracy and low uncertainty of the selected model. In the next step, the parameters of minimum temperature, maximum temperature, and precipitation in the period of (2020-2040) were simulated and downscaled by the chosen model. The evaluation of results in this section showed that the minimum temperature would have a growing trend until 2040. The slope of the average minimum temperature curve in the studied statistical period under the RCP2.6 and RCP8.5 scenarios will increase by 0.02% and 0.08%, respectively. The most significant growth of this parameter was estimated to be +0.8ºC and +0.16ºC under the RCP2.6 and RCP8.5 scenarios in September and March, respectively. The precipitation in the RCP2.6 scenario has nor decreasing neither increasing trend. The statistics of Sen Test in the evaluated confidence intervals indicated that the average precipitation had decreased by almost 0.38 mm every year. The annual average in the RCP8.5 scenario has a significant decreasing trend with a slope of -1.31 mm in the low confidence interval (α = 0.1). The most considerable growth and reduction with the values of +0.45 and +0.63 in the slope of precipitation curve in this scenario are seen in July and October, respectively. This parameter has also a decreasing trend in the very high level of (α = 0.001) for the average maximum temperature in both scenarios. Based on the obtained results, this climatic parameter has a significant decreasing trend in the studied confidence intervals in most of the months. The largest reduction of this parameter under the RCP2.6 and RCP8.5 scenarios will happen in March and June by 0.20ºC and 0.14ºC.