연구보고서
- 저자
- 민영미 박사
- 작성일
- 2018.04.24
- 조회
- 537
- 요약
- 목차
This work was motived by the need for the development of an integrated system from finding new contents for providing seasonal forecast information to the development of a prediction method, as well as to the operation and construction
of a prediction system. Considering the current monthly availability of APCC data, its internal capability and difference from other operational centers, four contents were potentially selected from analysis of the current status of leading centers that provide multi-model ensemble (MME) seasonal forecasts and comprehensive reviews of previous researches. The selected contents are ENSO type (or phase) and strength (or intensity), global drought, East Asia Winter Monsoon (EAWM), and climate extremes. The purpose of the first-year research, which was proposed as a multi-year project, is to evaluate the predictability and the prediction skill of the selected contents and the possibility of developing it into a climate service at the APCC.
We first developed a probabilistic MME prediction system (PMME) for ENSO type and strength forecasts, based on an uncalibrated multi-model ensemble with equal model weights, and a parametric Gaussian fitting method, which is the most
appropriate for use in an operational prediction system. Then, we examined the skill of probabilistic forecasts for ENSO type and strength from individual models and from the MME with a large set of retrospective (1982-2010) and real-time forecasts (2013-2016). The results indicate that the developed PMME prediction system is capable of providing more detailed probabilistic forecasts of ENSO type and strength than are typically provided. Furthermore, the possible causes of the failure and success of the model calibration and combination methods in a real-time basis and many operational issues when developing/improving the prediction system are also discussed.
A MME-based global drought prediction system has been developed as a prototype for the period of 1983-2005 and its prediction skill has been evaluated. The model-simulated precipitation can be calibrated based on the mean bias that
performs better than the mean/variance correction. Six-month Standardized Prediction Index (SPI6) prediction shows high skills due to its auto-correlation and robust skills compared to that of persistent run. Categorical forecast skills for the occurrence, intensity, and spatial extent of drought events show that the SPI prediction of bias calibrated MME precipitation has the best performance. The well-designed and well?validated drought prediction system can be further developed for real operational practice.
By analyzing the characteristics of the existing EAWM indices, we present the EAWM index, which is representative of winter temperature fluctuations in East Asia. Using this index, we developed a dynamic - statistical hybrid prediction model and tried to provide reliable winter season prediction information. This hybrid model shows a reasonable skill of 0.6 or higher, which significantly improves the inter-model difference and represents the same EAWM phase for most years. Moreover, it shows consistently significant skill in all individual models, and is maintained with a lead time of four months. Therefore, it is expected that reliable seasonal forecast of EAWM can be produced.
Finally, we examine the skill of probabilistic forecasts for climate extremes during the period 1982-2005, defined by the upper and lower 15th percentile of a climatological distribution, from individual models and the MME in both deterministic and probabilistic sense. Our results reveal several significant issues. Firstly, the superiority of the multi-model concept is more clearly evidenced in a probabilistic framework. Secondly, a parametric Gaussian fitting method along with a pooling approach is the most appropriate way to estimate forecast probability and to combine predictions of individual models for global climate extremes. Thirdly, the current systems can predict extreme temperatures, but the predictability of extreme precipitations on seasonal scales is still limited. Additionally, sensitivity experiments on different percentiles (10/15/20%) are discussed.