연구보고서
- 저자
- 유진호 박사
- 작성일
- 2016.01.23
- 조회
- 230
- 요약
- 목차
A new approach for improving stability and robustness of statistical post processing of dynamical seasonal prediction is proposed. It is to filter out unpredictable noise variability from predictand variable (observation) by assuming that predictable signal is only related with sea surface temperature (SST) varability in the tropics and sub-tropics. In reality, this assumption may not be true but in the context of statistical post-process aiming extracttion of information from large scale field of Multi Model Ensemble seasonal prediction result, which will consist of climate feedbacks mostly associated with SST variability.
July mean precipitation over the east Asia was examined as a predictand. The filtering is performed by re-constructing precipitation data from leading SVD (Singular Value Decomposition) mode between east Asian preciptation and tropical and subtropical SST at Indo-Pacific basin. Filtered precipitation data is able to synchronize associated circulation patterns in the MME forecast and that of observation to some degree. A statiscal post-process methods based on the stepwise pattern projection method using this filtered observtion shows significant improvement of skill compared to that with original unfiltered observation. It is because of robust and organized selection predictor around tropical ocean in the filtered observation case, whereas predictor selection was not systematic and spreaded over the globe in the case of unfiltered observation in the post-process.
Although there are more issues to be resolved for implementation in the operational forecas, this approach suggest a possibility of improving statistical post-processing by re-constructing observation variable prior to correct forecast result. This is particularly effective in the situation of seasonal forecasts with short hindcast length to obtain robust statistical relationship for the conventional methods.