연구보고서
- 저자
- 문자연 박사
- 작성일
- 2019.06.12
- 조회
- 204
- 요약
- 목차
This project was designed to evaluate the current status of subseasonal prediction and predictability and thereby provide results that can contribute to develop suitable prediction systems for a subseasonal time scale. In this study, three models (ECMWF, NCEP, and ECCC) that have common forecast specifications (e.g., initial date and interval, lead time, reforecast periods) are selected to provide a consistent evaluation.
The first aim of this project is to evaluate the general performance of reforecast, real-time forecast, and multi-mode ensembles (MMEs) of 3 models for 11 variables and 12 verification regions. In terms of climatology, the models display a disorganized long-term mean (LTM) pattern for some regions and variables and tend to underestimate interannual variation. Similarities in LTM spatial patterns decrease, although absolute biases increase as lead times increase, in all three models; however, this was expected. Predictability measured by signal to noise ratio (SNR), which is different from actual prediction skill, was the highest for NCEP but lowest in ECCC for the same ensemble size and references. Prediction skill analysis suggests that forecasts are reliable up to 2 or 3 weeks in most regions; the ECMWF has the longest predictability among three models. There is clear dependency between prediction skill and the months, regions, and variables; the features are very similar for reforecasts and real-time forecasts. Interestingly, global subseasonal prediction skill during boreal winter season was shown to be dependent largely on the variability of the ENSO, and the association becomes clearer with longer lead weeks. From the comparison of the four varying MME methods, which have different combinations of reference period and weightings, we found that the grand ensemble merging MME (MME_g) has a skill that is at least comparable to, and perhaps even better than, the ECMWF model.
The Madden-Julian Oscillation (MJO), one of major sources of subseasonal predictability, and the Ural blocking affecting cold surges in East Asia were investigated to verify the skill and climatic features simulated by the models. According to the verification based on the prediction skill of the MJO, the ECMWF, NCEP, and MME closely resembled observations while the ECCC model underestimated the intensity, eastward evolution speed, and life cycle of the MJO. The prediction skills for the lead time, season, phase, amplitude, and the location of the MJO convection showed contrasting results based on the model and variables used. However, the ECMWF model outperformed in detecting the highest MJO prediction skill.
It is suggested that a modified index for Ural blocking be applied for subseasonal forecasts. Individual models succeeded in capturing Ural blocking with one week of lead time; however, as this increases the skill decreases. Although predicting a blocking event 3–4 weeks ahead is not reliable, the performance can be improved by up to two weeks by applying a simple MME.