Study on the decision making method for probabilistic long range forecasting with global climate model

Min Ah Sun

Department of Environmental Atmospheric Sciences, The Graduate School,
Pukyong National University

Abstract
Recently the importance of long range forecast had attracted in economic policy areas and industries for weather disasters due to climate change. The needs for long-term prediction service in agriculture, fisheries, energy industry (Coastello et al., 1998; Hamlet et al., 2002; Adams et al., 2003), etc., have been increased. However, long-term forecast has the uncertainty (Lorenz, 1963, 1993) from the initial data due to lack of observation and chaotic nature that cannot be predicted for future. Probabilistic forecast that allows for a rational decision making about the uncertainty has been proposed. It has been important to provide more predictability information prior to providing long range forecast with probability. How to determine the probability of long range forecast, prediction results will vary based on how to share by dividing into three equal parts climatology. Therefore, in order to provide information that predictability more predicted long range forecast results, presents a methodology for determining the terciles, analyzed and evaluated for prediction performance.
Probability long range forecast, using the probability density function (Leith, 1973; Madeen, 1976; Zwiers, 1996; Kharin and Zwiers, 2003), which is mainly used in numerical weather prediction generally predicted by below normal, near normal, above normal in three categories. The deterministic method has been presented in three ways to below normal, near normal, and above normal to determine () reference value terciles. The first is intended to use the terciles of climatology to determine and second is and to be applied to the probability density function of the climatology, the probability becomes 33.3% it is to determine. Finally, to obtain the standard deviation of the climatology is to determine between the and . To calculate the Temporal Correlation Coefficient (TCC) value of precipitation and average temperature at 850hPa of summer and winter, we have calculated the Ranked Probability Skill Score (RPSS) and Area of Relative Operating Characteristic (AROC) value and Relative Operating Characteristic (ROC) curve for the prediction performance evaluation.
The probability distribution of summer temperature at 850hPa of each method over the East Asia, the probability of Above normal category is the highest among the three categories. However method 3 is larger probability in Near normal category compared to the other methods. In case of rainfall, method 3 also higher probability in Near normal category compared to the other methods. The RPSS of 850hPa temperature, the method 3 has the best prediction performance over East Asia and performance of Above normal category is best among them. In particular, the significance is higher in Korea peninsula and Japan. In the case of precipitation, the best prediction performance in method 3 and high significance in northern?eastern China. In winter, the largest probability of Below normal category and method 3 has larger probability of Near normal category than other methods. Precipitation also has the highest probability in Near normal category. The RPSS of 850hPa temperature, the method 2 has the best prediction performance over East Asia and performance of Above normal category is best among them. In particular, the significance is higher in northern Japan, eastern?southern China. In case of precipitation, best prediction performance of Below normal category on method 3 and high significance in eastern?northern China, edge of northern?western Pacific ocean. Overall predictable performance score is not high, but method 3 has the highest prediction performance over the East Asia.