연구보고서
- 저자
- 이우섭 박사
- 작성일
- 2018.04.24
- 조회
- 611
- 요약
- 목차
This project was designed to improve the stability, predictability, and utilization of the APEC Climate Center (APCC) seasonal forecasting system. For reasonable and efficient operation of the APCC seasonal prediction system, the preprocessing process, Hindcast, and forecast MME were improved and developed. The pre-processing is combined with the Automatic Forecast System (AFS) for operational efficiency, and the main program is developed to integrate pre-processing for the individual model. It is possible to operate the individual models simultaneously. Also pre-processing source code was corrected according to procedure and function naming conventions. To maintain and operate AFS more easily and conveniently, the scripts of run process and Multi-Model Ensemble (MME) models were improved, and the existing AFS 32-bit system was upgraded to new
64-bit system. The scripts of run process were migrated from Ruby script language to Python script language to eliminate maintenance problems such as compatibility issues due to Ruby version upgrades. Also, in order to improve the readability of
NCAR Command Language (NCL) scripts in AFS, a Java-based NCL code beautifier was developed, and the source codes of MME models were thus improved. To analyze and improve the problems arising from differences between 32-bit and 64-bit architectures, the results of MME models (SCM: Simple Composite Method; GAUS: GAUSsian fitting method; MRG: Multiple Linear Regression; SSE: Synthetic Super Ensemble method; SPM: Stepwise Pattern projection Method) were compared. Results indicated that the SCM model was the same for both 32-bit and 64-bit architectures, but the MME models (except SCM) exhibited differing results. In particular, the results of MRG and SSE models were significantly different. This is due to the underflow occurring in the calculation process of the Singular Value Decomposition (SVD) algorithm using the float data type.
This study also assessed the real-time one-month lead forecasts of three-month (seasonal) mean sea surface temperature (SST), temperature, and precipitation on a monthly basis issued by the APCC for 2016. The current level of the APCC operational multi-model prediction system performance is shown. On average, the level of real-time forecast skill during the period of 2016JFM-2016/17DJF is generally higher than that of Hindcasts (1982-2003) and the recent 8-year (2008-15)
period, mainly due to the strong El Nino in 2015/16 boreal winter and global warming. Moreover, this study evaluated the prediction skill of monthly mean temperature over East Asia during the winter season, and the characteristics of extreme events forecasted by APCC MME. By analyzing the cause of low prediction skill, we tried to develop an APCC MME-based dynamical-statistical hybrid model. Two hybrid models were constructed. One is based on the relationship between the observed
T2m and the large-scale key predictors forecasted by APCC MME, and the other is based on the dominant temperature modes that affect interannual variability of temperature over Korea. The MME hybrid models are more skillful than the APCC MME forecast models.
The final objective of this project is to provide tangible guides for easier manipulation and better interpretation of climate information from APCC MME so that APCC forecasters can produce more reliable climate predictions of temperature
and rainfall in Korea. One product developed is a computing tool for the consolidated climate forecast of Korea (named CLIMT-K), which automatically updates information from the APCC MME verification and the relationship with ENSO/Cryosphere,
comprehensively suggests tercile probability forecasts tailored for the Korean Peninsula, and freely provides easy access to final graphics. The tercile probability forecast system built in CLIMT-K is based on past large-scale circulation resemblance, and it is superior to APCC PMME in category prediction for some target seasons and variables. The other product is guidance suggesting a table of maps of global temperature and precipitation from reanalysis and MME/individual model Hindcast data regressed onto climate indices, which have information about 1) the responses of atmospheric circulation to climate indices and 2) correlations between temperature and precipitation over Korea and climate indices.