Performance of Series Model CMIP5 in Simulation and Projection of Climatic Variables of Rainfall, Temperature and Wind Speed (Case Study: Yazd)

Introduction
According to the IPCC (2014) definition, climate change is a change in the state of the climate in which the average or its modifiable properties varies for a long period. To date various versions of climate change models have been presented for different purposes. These models are including (1) the assessment models of the Intergovernmental Panel on Climate Change (FAR), (2) the models entitled SAR, (3) the report model of TAR, and (4) the model report titled AR4 (CMIP3) and, finally, the 5th Assessment Report models, AR5 (CMIP5). These models use new emission scenarios called "RCP". These scenarios have four key lines called RCP2.6, RCP4.5, RCP6 and RCP8.5. General circulation models are considered as the most reliable tools for simulating climatic variables. These models can simulate the present climate and illustrate the conditions of the future climate under specific scenarios. Although these models are very helpful in the investigation and predictions based on future changes in climate, the outputs of these models are based on a large grid scale (250 to 600 km). Therefore, the application of these models is not suitable on a regional or local scale. The most important tool to create a bridge between a regional/local scale and GCM scales is downscaling. Among statistical downscaling methods, SDSM has been widely used for the downscaling of climate variables in the world. Most of the studies on climate change have been used in AR4 models. The majority of the studies in the word used AR5 models to investigate changes in climatic variables. As mentioned, most of the studies about climate change modeling have been performed using AR4 models. Therefore, studying and updating of that with CMIP5 data is necessary to reduce the uncertainty of modeling climate change in recent decades. Thus, in this study, three variables of rainfall, averaged temperature and maximum wind speed are modeled using AR5 models. These parameters are modeled according to the basic period of 1961-2005 and the future climate change will be simulated in a 95-year period from 2006 to 2100. The investigation on changes of the maximum wind speed in this region, as windy areas, is affected by dust storms every year. This can be of great importance in the studies of dust and wind erosion in the future.
Material and In this study, we use three types of data including daily rainfall, average temperature and maximum wind speed from synoptic station. Second sources of data are NCEP variables and the third are CanESM2 outputs.
For analysis of General Circulation Model, we used in this study the second generation of Canadian Earth System Model (CanESM2) developed by Canadian Centre for Climate Modeling and Analysis (CCCma) of Environment Canada. This is the only model that made daily predictor variables available to be directly fed into SDSM. Also this model has three scenarios such as RCP2.6, RCP4.5 and RCP8.5.
The model SDSM has four main parts including identification of predictors, model calibration, weather generator and generation of future series of climate variables. After reviewing data quality control, predictive variables were determined by NCEP. Calibration of the model was carried out using 70% of the observational data during the statistical period of 1961-1992 to determine the coefficients of the equation for modeling rainfall data, temperature and wind speed. The coefficients obtained in the calibration phase were used for 30% of the remaining data (1993-2005) for model verification. The performance of the SDSM was evaluated based on comparison of the results of verification and observational data in 1993-2005. The performance of the model on downscaling of temperature is higher than rainfall and wind speed. The reason for this is that the temperature parameter is continuous variable and there are no zero values.
The results of evaluation of performance and model uncertainty showed that NSE, RMSE, R2, PBIAS, RSR and Pearson correlation coefficient for mean temperature were 0.96, 1.64, 0.96, 1.63, 0.18, and 0.97 based on downscaled data by the NCEP and 0.88, 1.3, 0.91, 6.8, 0.33, 0.49 based on the downscaled data of CanESM2, which is of a fairly good value. The assessment criteria of rainfall and wind speed are less than the average temperature, which can be explained by the fact that the rainfall data and wind speed are not normal due to the presence of zero values. Based on the results of the Man-Kendall test, observation and modeled rainfall under two scenarios of RCP2.6 and RCP8.5 have no significant trend. While the data generated by the RCP4.5 scenario shows a significant decreasing trend. The results of the Mean-Kendall test on average showed that the series modeled by RCP scenarios, as well as observational data, have a significant incremental trend. Incremental trend of temperature indicates the existence of climate change in the study area. The maximum wind speed showed that the observation and modeled data by RCP2.6 and RCP4.5 scenarios had a significant decreasing trend, while the generated data based on the RCP8.5 scenario had an increasing trend. Based on the results of CanESM2 model and SDSM, study area is not excluded from the climate change phenomenon. The trend model predicted an increase in average temperature in future under three RCP scenarios. The average temperature will increase by 1.54 degrees. According to the RCP2.6 scenario, the average temperature is 4.5 degrees, the scenario RCP4.5 is 7.7 degrees and the RCP8.5 scenario is 18.12 degrees higher than the base period. Unlike temperature variable, there is no significant change pattern for rainfall. According to RCP2.6 and RCP8.5 scenarios, rainfall will be increased in the future, while under RCP4.5 scenario, it will be decreased. The maximum wind speed will be increased by an average of 0.53 m/s compared with the base period. The RCP2.6 scenario is 4.9%, the scenario RCP4.5 44.4%, and the scenario RCP8.5 53.5% of increase compared with observational data