Since its establishment in 2005, APCC has made great efforts to develop and improve the long-term seasonal forecast technology using the Multi-Model Ensemble (MME), and as a result, it has established a climate prediction system using various global models of the world's leading businesses and research institutes to provide climate monitoring information and long-term forecast and verification information on its website every month.

For the production of stable climate forecast information, APCC developed the Automatic Forecast System (AFS) in the early days of its establishment. AFS has been improved with the expansion of the forecast period (3 months->6 months), the establishment of the ENSO forecasting system, and the development of the verification system. But it was partially improved by the person in charge, which caused lack of the overall system's efficiency and effectiveness in the system. Current AFS system has many limitations to applying the latest and ever-evolving data processing technology.

Therefore, we have established a more stable and efficient climate forecast system by diagnosing problems of the current forecasting system and readjusting them from a technical standpoint. To this end, we have improved the structure of the climate monitoring and verification system, gained expandability and flexibility, and improved the source code that was heterogeneous. Along with the improvement of structure, it was performed to verify reliability and time efficiency on the new server.

APCC uses four deterministic and one probabilistic MME (PMME) technique to predict the global 6-month MME. The deterministic method consists of SCM, which is simply averages of individual models, and SSE, MRG and SPM, which are statistical postprocessing. The official forecast of APCC provided through APCC's website is SCM (after 2005) and PMME (after 2006). The utilization of the other 3 SCM methods (MRG, SSE, SPM) is quite low, and in particular, it takes up to 45 minutes for SPM. In terms of predictability, the SCM and SPM show similar skill score, but, it shows that the SCM's predictive improvements are mostly on land (Eurasia, Australia, North America), whereas the SPM's forecast improvement is mostly in the oceans (especially in the Arctic and the Southern Hemisphere). Based on a comprehensive analysis of the relative comparative evaluation of the efficiency and predictability, the monthly production of MRG, SSE, and SPM information was judged to be ineffective in terms of operational perspective. Only SCM and PMME is produced to ensure current efficiency. This led to earlier provision of forecast information from 25th to 20th, which will enhance the utilization and competitiveness of APCC seasonal forecast information.

The participating models of APCC MME have changed steadily and predictability has been increasing. Therefore, it is needed to objectively understand the predictability of APCC MME by evaluating the APCC MME data with the other world's leading MME in the same conditions. We collected MME data from 3 agencies (WMOLC, NMME, C3S), which currently provide MME forecasting information, and analyzed the overall predictability of MMEs. Because many studies refer to the sensitivity of observations in the verification of predictability, sensitivity to observations was investigated to ensure objectivity of evaluation. When we analyzed the sensitivity of the observations using various reference data, we found that although there were areas where the skill scores varied significantly depending on the observation data, we could compare predictability under the same conditions because the direction of variation did not vary depending on the MME.

To compare the MMEs of the APCC and other agencies under the same conditions, the MME set was constructed by applying Simple Composite Method (SCM) techniques to individual models over 17 years from 1993 to 2009, the common period of the four agencies’ hindcast. Comparisons were performed for 2m temperatures, precipitation and sea surface temperatures. MMEs from APCC and other agencies showed similar trends in predictability by region and season. The NMME and C3S, which do not have the same individual model, showed similar changes in predictability. This may be because most of the dynamic models have similar strengths or weaknesses. Compared to the predictability of APCC and WMO MME with more models, the predictability of the C3S MME with fewer models is comparable. Since APCC and WMO are a combination of more than 10 different models, the range of individual model’s predictability is wide and the uncertainty is greatly reduced through application of the multi-modal ensemble technique. In the case of C3S, the predictability of the individual models is on average high, so MME is also highly predictable. However, due to the small number of participating models, the significantly low predictability of one or two models can result in lower MME predictability, and the variability of predictability over the season can be a weakness in terms of stable forecasting.