The climate application studies such as agriculture and water resource often requiresthat climate data be available for a very small region, but with a fine time scale. Meanwhile, pure climate researches focus mainly on large-scale atmospheric circulations, so that their results are described in large and coarse scales. Although large-scale circulations have an impact on the local-scale weather and climate, they are not directly utilized for climate application studies due to the scale gap. However, because seasonal predictions for local-scale climates cannot be conducted independently of large-scale circulations, they need to be carried out in the view of pure climate science. Therefore, to produce practical seasonal predictions, we need a method to reduce the scale gap. The related processes are called downscaling.
In this study, we are concerned with the downscaling of the APCC (APEC Climate Center) seasonal prediction for boreal winter and summer season. The target local region is the Nakdong river basin in Korea. It is well known that the EAWM (East Asian Winter Monsoon) and the EASM (East Asian Summer Monsoon) influence the climate of the Korean peninsula during winter and summer, respectively. The downscaling method is based on the response of the climate of the target region to the monsoon circulations. This response is represented by the statistical models embedded into a Weather Generator and multinomial logistic regression models for probabilistic prediction of monsoon strength are established. The downscaling is performed by predicting the monsoon strength by using APCC seasonal prediction as predictor, and producing weather scenarios by running the Weather Generator according to the prediction result.
Many monsoon indices have been proposed for the seasons. We investigate their predictability and relevance to the climate of the target region, and select suitable monsoon indices. Furthermore, based on the selected indices, we produce data for 1000 weather scenarios and verify them by comparing their statistics with those of observational data. Our results confirm that they reproduce various observational statistical characteristics. Therefore, we conclude that the proposed downscaling method is suitable for the purpose of this study.
