Statistical post processing MRI-CGCM3 output for seasonal precipitation forecast over Khorasan-Razavi province, Iran

In recent years, due to frequent climatic extreme events, demands for reliable seasonal precipitation forecasts have been increased. The seasonal to interannual climate forecasts have become essential for policy makers and risk managers in planning several activities, including those related to agriculture, water resource management and many others which directly or indirectly affect the society, especially in water resources and agricultural, environmental and health sectors. Although numerical models are being increasingly used to generate operational seasonal forecasts, the reliability of these products remains relatively low. In this regard, for improving the accuracy of seasonal precipitation forecasts, the outputs of the MRI-CGCM3 seasonal forecast model have been used for the statistical post-processing of the model precipitation over the synoptic weather stations of Khorasan-Razavi Province in Iran over the period of 1981-2007. We had the model output data from the Tokyo Climate Centre. The outputs of MRI-CGCM3 are available to registered National Meteorological and Hydrological Services (NMHSs) on the website of the TokyoClimateCenter of the Japan Meteorological Agency (JMA/TCC).
Regression-based post processing methods have proven useful in increasing forecast skills. The current study tests this hypothesis applying both linear regressions to the correction of climate hind casts produced by MRI-CGCM3 general circulation models. Statistically significant predictions are produced from the model output with no forecast skill prior to post-processing. MRI-CGCM3 has produced 30 years of reforecast covering a period of 1981-2008. The reforecast data was used to produce post-processing multivariate relations between reforecast parameters and the observed precipitation in the training period of 1981-2001. Model variables and indices which were used in the post-processing were WIO rain, Z2030, Z5060, WIO SST, T850, T2m, SST, NINOWEST SST, WNP RAIN, NINO3 SST, Z3040, H500, SLP, SAMOI RAIN, MC RAIN, DL RAIN, THMD, THTR and total precipitation. The skill of multivariate post-processing was evaluated using Mean Square Skill Score, Mean Bias Error, relative error and categorical skill score over the training and evaluation periods. Categorical skill score is determined by computing the skill of the post processed and the raw model data in forecasting five precipitation categories i.e. above normal, above normal to normal, normal, normal to below normal and below normal.  The area of study covered Khorasan-Razavi province stations including Mashad, Golmakan, Ghuchan, Sarakhs, Torbate-Heydarieh, Kashmar and Sabzvar.
Post processed precipitations were compared to the observed precipitations to investigate the capability of the statistical post processing method. After post processing, the bias and relative error decreased from 107.43 to 2.99 and 66.15 to 0.78 at Mashad station, respectively. Station average bias error decreased from 94.3 to 3.5mm and categorical skill was improved from 25.3% in raw data to 62.2% in the post processed data. The bias and relative error were significantly decreased in the other stations. The skill of the post-processing of precipitation was compared to the observed precipitation for all months. The result showed that the multiple regression method can be significantly used to increase the accuracy of the model predictions over Khorasan-Razavi province.