Accuracy Evaluation of Regional Climate Models Output in Iran

  All studies in the field of climate change impact assessment needs climate data with different spatial and temporal scales. The lack of temperature and precipitation data with high spatial resolution is a major limitation to analyzing future climate change. In addition, the output of the models has error that needs to be corrected; otherwise they will make a significant bias for assessing effect of climate change. Therefore, identifying the best regional climate model for downscale the global climate models is essential to better understanding of climate conditions in the local and regional scale. In the last few years, using various regional climate models in order to produce a multi-member set of the downscaled data in the CMIP5 project by World Climate Research Program (WCRP) in action with Coordinated Regional climate Downscaling Experiment (CORDEX) with the aim of producing regional climate change forecasts, was established as an input to researches on the impacts of climate change and ways for adaptation to it. The main objective of this research is accuracy evaluation of different model outputs of the CORDEX project with different domain and resolution in Iran. In the CORDEX project there are two domain that covering Iran. These two domains are North Africa-Middle East (CORDEX-MNA) including latitude of 7° S to 45° N and longitude 27° W to 76° E and South Asia (CORDEX-WAS) that includes latitude 13° S to 44° N and longitude is 27° E to 107° E (Figure 1). To do this research, first daily output of precipitation, maximum and minimum temperatures in the period of 1990-2005 for three regional climate models with a special resolution of 0.22 ° and 0.44 ° that performed by three international meteorology institutes, available at ESGF web site (Table 1). Daily observation data that recorded in 304 synoptic stations in Iran for the three variables were collected from Iran Meteorology Organization and transferred to a matrix with 3044×5844 dimensions. Then, several scripts were written in the MATLAB software for extract the model data in the domain of Iran and compare output model and observational data with two conditions. The first condition is in the output models resolution of 0.44° (spatiotemporal matrix with dimensions of 5844×740), the observation station should have a distance of less than 25 km, and the next condition is in the resolution of 0.22 ° (spatiotemporal matrix with dimensions of 5844×3218) should have a distance of less than 12 km. The difference between observation data and its corresponding estimated cell were investigated with statistical method such as Mean Error (ME), Pearson Correlation Coefficient, Root Mean Square Error (RMSE) and Standard Deviation (SD). Also, we were used Box-Whisker plots and Taylor Diagram to find the best regional climate model.
Result and The precipitation accuracy of regional climate models output presented by different meteorological institutes (Table 1) was evaluated with observational data in two domain, CORDEX-MENA and CORDEX-WAS, in Iran (Fig. 4). The calculation of the outputs mean error of different models showed that none of the models have a suitable estimation of precipitation values in research domain. The HadRM3P model shows the lowest RMSE calculated with comparing to observational data for the maximum temperature across Iran except the central parts. However, for the minimum temperature RegCM4.1 model shows the lowest difference with compare to observation data in most parts of the research domain. For annual precipitation using the Box-Whisker plot, which compares the correlation coefficients between the observed data and the corresponding cells in the northern and southern half of Iran, in general, none of the models have an accurate estimate of precipitation in Iran (Fig. 8a). This plot for different models showed that the outputs of the HadRM3P and RegCM4.1 models, respectively, for maximum and minimum temperatures in most cells, have more than 0.8, correlation coefficient (Fig. 8b and c). The correlation of rainfall data shows that most models in the central and mountainous regions of Iran do not have high correlation coefficient with observational data. Spatial distribution of correlation between maximum temperature model outputs and observational data in Iran shows that the two HadRM3P and RCA4-WAS0.44 models have a strong correlation coefficient, respectively and changes in the correlation coefficient in the HadRM3P model are low in both the northern half and the southern half of Iran. The RegCM4.1 model had the stronger correlation in the northern half in compare to the southern parts of Iran. Also, the mean difference of estimated model output with observation data of this variable in the whole of Iran is less than 1°C and this model is the most appropriate model among the available models for minimum temperature in Iran