Monthly Drought Modeling Using Post-Processed Output of CFS.v2-RegCM4 System during 1982-2010 (Case Study: Iran)

Document Type : Research Article


Academic member


Iran and its neighboring countries are located in an arid and semi-arid region, so they have been always affected by various climate disasters. Drought as one of the most important climatic hazards in the region, has attracted lots of researches. Drought could have negative impacts on different areas such as agriculture, water resources, industry, transportation and urbanization. Drought is basically a qualitative phenomenon that could have various definitions in different sectors. In a basic categorizing, drought could be climatic, hydrological, agricultural and social droughts. Researches have always keen on defining appropriate indices for each kind of drought. Standardized Precipitation Index (SPI) and Standardized Precipitation and potential Evapotranspiration Index (SPEI) are good examples of climatic drought indices which have been widely used and popular between researchers.

While monitoring the changes of these indices are a good way of drought analysis, effective drought management also needs accurate forecast of drought. A widely applied approach to forecast drought is through using the outputs of Coupled atmosphere- land- ocean General Circulation Models (CGCMs). CGCMs simulate the atmosphere-land-ocean system in a simplified way. Although, these models have developed a lot in recent years, the forecasts are not well enough to be used in small regions. Thus, downscaling and post-processing methods have widely applied to increase the resolution and accuracy of CGCM’s outputs, respectively.

Downscaling methods could be in a way grouped into two categories: 1. Statistical, 2. Dynamical. Dynamical methods are based on physical theories and using parametrization and initial hypothesis. Statistical methods investigate the relationship between modeled variables and their corresponding observation in a period of time. This relationship is then applied to forecast that variable. Both the mentioned methods have their own advantages and disadvantages. In recent years, combined dynamical and statistical approaches have been widely used. In this study, SPI and SPEI were forecasted using the outputs of CFS.v2. Firstly, the outputs of CFS.v2 in the reforecast period (1982-2010) were downscaled to 30km horizontal resolution using RegCM4. Due to limitation in computational power, the dynamical downscaling was limited to 1-month lead time. Then needed outputs for the rest of analysis were post-processed using Decision Tree (DT) and Support Vector Machine (SVM). The outputs of CFS.v2-RegCM4 system for precipitation and ERA5 were used as inputs and ideal outputs, respectively, of DT and SVM for calibration and validation. These post-processed outputs were then applied to calculate SPI. In order to calculate modeled SPEI, firstly, ET was computed using Hargreaves-Samani approach and the needed climatic modeled variables by CFS.v2-RegCM4 system. The reanalysis ERA5 data was utilized to calculate reanalysis ET. Then, in order to calibrate and validate DT and SVM models, modeled ET and reanalysis ET were applied as input and ideal output, respectively. Finally, the postprocessed precipitation and ET were employed to compute modeled SPEI. These modeled SPI and SPEI were spatially compared with SPI and SPEI computed using ERA5 data. The results showed that both DT and SVM improved the accuracy of precipitation and ET forecasted by CFS.v2-RegCM4 system. However, DT model performed much better than SVM. The ability of DT in post-processing precipitation was higher than it for ET. Comparing spatial pattern of modeled SPI and SPEI with reanalysis SPI and SPEI showed that DT model could replicate these values at an acceptable level of accuracy.


Main Subjects