%0 Journal Article %T Developing Real-time Multi-model Ensemble and Downscaling of Seasonal Precipitation Forecast Systems: Application of Canonical Correlation Analysis %J Journal of the Earth and Space Physics %I Institute of Geophysics, University of Tehran %Z 2538-371X %A Najafi, Hossein %A Massah Bavani, Ali Reza %A Irannejad, Parviz %A Robertson, Andrew Viliam %D 2018 %\ 04/21/2018 %V 44 %N 1 %P 245-264 %! Developing Real-time Multi-model Ensemble and Downscaling of Seasonal Precipitation Forecast Systems: Application of Canonical Correlation Analysis %K North America Multi-model Ensemble (NMME) %K Downscaling %K Karkheh River Basin %K Multi-Model Ensemble %K Seasonal Precipitation Forecasts %R 10.22059/jesphys.2018.234664.1006904 %X The aim of this research is to evaluate a statistical method for downscaling the precipitation output of a number of Coupled General Circulation Models issuing seasonal forecasts 9 month in advance. Canonical Correlation Analysis (CCA) is applied for post-processing precipitation from the North American Multi-model Ensemble (NMME) project. The analysis is done for a long-term period (1986-2015) in the west of Iran. The area under study includes Karkheh River Basin where a significant reduction in renewable water resources has faced policymakers with challenges in water resources allocation and provision of environmental requirements to Hoor-al-Azim marshland downstream. PERSIANN-CDR biases are computed and corrected against in-situ observations by applying the multiplicative method. Bias corrected Satellite-based rainfall data merged with 23 gauge-based data. The approach for merging station-satellite-based rainfall estimation includes a spatio-temporal LM method which fits linear regression to the deterministic part of universal variation. It exhibits appropriate performance in terms of Correlation, Nash-Sutcliffe Efficiency and mean absolute error and multiplicative bias. After merging, correlation coefficients between the merged data and gauge-based rainfall are between 0.92 and 0.98 for all stations whereas it was between 0.7-0.95 for PERSIANN-CDR. The merged precipitation grided dataset is then used as the reference to evaluate NMME seasonal forecasting systems October-December being the target season. Forecasts initialized on the early October, September and August (lead time-0, lead-time-1 and lead-time-2 months, respectively) are evaluated for individual raw model outputs. Multi-Model Ensemble is also developed by assigning equal weights to individual models. Multi-model Ensemble which consists the 3 best individual models (CCSM4, CMC2 and CFSv2) outperforms all other MME which consist 2 to 8 models (ρ=0.560). It also outperforms CCSM4 which has the highest Spearman correlation of 0.486 among all models. Canonical Correlation Analysis (CCA) is then applied to individual and MME seasonal mean precipitation forecasts to correct biases in the position. Probabilistic forecasts are produced based on the best-guess forecast estimated by regression model (CCA). Predictand is transformed to normal distribution before performing the calculations. Then the forecast is transformed back to the empirical distribution. By assuming that the errors in the best-guess forecast are normally distributed, the variance of the errors is defined by the sampling errors in the regression parameters, and by the variance of the errors in the cross-validated predictions. Then the probabilities of exceeding the various thresholds (below normal, normal and above normal terciles) are calculated for issuing probabilistic forecast from 1986-2015. The goodness index is improved for all models after performing CCA especially for GFDL-aer04 and CMC1 having the most correctable systematic biases. 3 model-based MME is recognized to have highest skill (Spearman correlation=0.623) at 0-month lead time. The models also show high skill for initializations made in the early August and early September. ROC-area for below-normal precipitation is more than 0.5 for almost all models which shows the skill of NMME seasonal forecast systems in meteorological drought prediction. The skill of NMME in forecasting October-December precipitation in the west of Iran can help decision makers in real-time water resources and agricultural planning before water-year starts (In the late September).  %U https://jesphys.ut.ac.ir/article_64863_0b4fda594bbeb9118b963934238b414c.pdf