Executive Summary

The implementation of a high-resolution TRIP river-routing model within the GloSea6 seasonal prediction system establishes an improved basis for evaluating how refined freshwater pathways alter upper-ocean structure, atmospheric mean states, and prediction skill across seasons. Building on prior short-hindcast experiment in the last year, this assessment leverages the APCC testbed to examine numerical stability, reproducibility, and quasi-operational performance. The high-resolution TRIP configuration produces clear summer improvements in the eastern Indian Ocean―reducing unrealistic freshwater spreading, enhancing vertical salinity stratification, and reducing SST biases―while coastal responses include improvements near the Amur River and mixed outcomes near the Yangtze due to complex local dynamics. Large-scale circulation regions such as the Pacific and ENSO-dominant region show limited sensitivity to freshwater restructuring. Winter impacts are weaker overall, with minor high-latitude cold-bias reductions that do not lead to improved ACC-based prediction skill. APCC testbed experiments confirm operationally stable run-time behavior with only manageable blow-up events. Collectively, the implementation maintains large-scale prediction skill while improving upper-ocean and coastal features in key regions, providing a technically validated foundation for future operational adoption and further optimization of river-discharge processes.

This study evaluated the predictability and applicability of high-resolution forecast data from the perspective of extreme climate prediction. Two cold wave and two heavy rainfall cases were analyzed to assess whether forecast skill improved compared to the operational system. For the cold wave cases, both events showed improved representation of low temperature over complex terrain, which can be attributed to the enhanced topographical detail captured by the higher resolution. In addition, the large-scale circulation patterns associated with cold surges were more realistically simulated at longer lead times. However, direct prediction of cold anomalies was limited because the model hindcast used to compute anomalies exhibited a strong cold bias, which suppressed the simulated temperature deviations. When the observed climatology was used instead of the model hindcast, both the circulation fields and the Extreme Forecast Index (EFI)-based cold wave predictions showed notable improvement, demonstrating the added value of the high-resolution forecast data.

For the heavy rainfall prediction, contrasting results were obtained between the two cases. In heavy rainfall event of July 2023, the high-resolution forecast successfully reproduced favorable conditions for extreme rainfall at 3-4 week lead times, outperforming the operational model. Although the predicted rainfall anomalies were slightly displaced northward compared to observation, the high-resolution model reproduced organized heavy precipitation, which was also reflected in improved EFI prediction. In contrast, for the 2022 heavy rainfall event, neither the operational nor the high-resolution forecast captured the occurrence of the extreme rainfall, indicating that increasing spatial resolution alone does not guarantee improved forecast skill for all extreme precipitation events.

A seasonal verification framework was established for the objective assessment of climate prediction model performance, specifically targeting the Arctic climate and the East Asian monsoon system. This verification framework incorporates diagnostic elements tailored to each climate mode, moving beyond simple statistical verification. Quantitative improvement rates for diagnostic elements are displayed via a scorecard, facilitating an intuitive comparison of model performance shifts. The framework not only serves as a universal basis for model evaluation but also provides a scientific insight to guide model development and enhancement. To demonstrate its utility, a preliminary comparative assessment was conducted on the forthcoming prediction system, GloSea-GC5.0, against the current operational system, GloSea6-GC3.2.

Diagnositics of the Arctic climate reveal that the interannual variability of the winter Arctic Oscillation (AO) index has improved by more than twofold. Furthermore, the process of AO development driven by late autumn Eurasian snow cover shows improvements in simulating patterns resembling the negative phase of the AO over the Eurasian continent. While these results appear to be driven by slight improvements in upper-level zonal wind anomalies and the intensity of low-level downward Eliassen-Palm (E-P) flux, the model still underestimates the stratospheric polar vortex structure and wave propagation, indicating weak troposphere-stratosphere interaction. Regarding sea ice, although the model underestimates the magnitude and extent of Arctic warming associated with Barents-Kara Sea ice loss, improvements in the regional vertical temperature profile and barotropic structure have led to a better simulation of the winter Warm Arctic-Cold Eurasian (WACE) pattern. However, compared to ERA5, the model exhibits a shallower and weaker vertical structure and fails to simulate significant upper-level wave propagation patterns. This limitation hinders the transmission of Arctic surface forcing through the upper atmosphere to the mid-latitudes. Therefore, to enhance the simulation of Arctic-mid latitude teleconnection, it is necessary to improve the representation of physical processes, specifically the vertical structure of the Arctic region and upper-level wave propagation.

For the East Asian summer monsoon, GC5.0 showed a reduction in the mean monsoon rainband bias compared to that in GC3.2, attributable to an improved representation of the North Pacific subtropical high. However, a deterioration in the prediction skill for interannual variability of East Asian precipitation was observed, likely due to an increased bias in tropical-extratropical teleconnection, such as the Pacific-Japan teleconnection pattern. Meanwhile, the simulated winter mean temperature over East Asia suffers from a cold bias, accompanied by a deeper East Asian trough and a more accelerated and southward-shifted jet stream. The improvement in prediction skill for the interannual variability of surface air temperature stems partly from enhanced climate mode predictability and internal responses. In both models, the simulation of the Siberian high and its associated monsoon dynamics remained problematic, which is a critical source of error that limits the overall predictability of the East Asian winter monsoon.

This study addresses the limitation of traditional climate model evaluation system relying heavily on hindcast data, which can overestimate actual forecast performance. A python-based verification package (FcstVerif), featuring a comprehensive web dashboard, was developed to directly evaluate and visualize the real-time forecast performance of seasonal prediction models. The package automates the process of data preprocessing, score evaluation, and visualization across various atmospheric variables and climate indices (ENSO, IOD). Applying this system to the KMA’s GloSea6GC3.2 model (Jan 2022 – Dec 2024) revealed a significant in global sea surface temperature (SST) prediction accuracy (ACC) starting from the October 2022 initialization. This improvement is considered to be attributed to the doubling of the forecast ensemble size (from 42 to 82) with a notable reduction in cold-bias in the Southern Hemisphere, demonstrating the package’s utility in continuously monitoring and confirming the efficacy of model improvements in operational forecast.

As the demand for sub-seasonal climate forecast information continues to grow, the importance of accurate climate prediction is also increasing. The accuracy of climate forecast system is primarily assessed using hindcasts representing the potential maximum forecast skill, which is known to differ from the that of real-time forecasts. In this study, the predictability and uncertainty characteristics of key climate modes affecting the Korean Peninsula on a sub-seasonal time scale were diagnosed. It was found that the predictability of summer climate modes is difficult to utilize directly due to the poor skill. Based on the diagnose systematic bias characteristics, an indirect utilization method through statistical correction was proposed. For winter climate modes, predictability was evaluated based on spatial predictability, and the optimal lead times for individual climate modes were identified and presented. The proposed methods for utilizing climate mode predictions, derived from a series of research results, are expected to improve the prediction skill and utility of operational forecast data. Additionally, the results can provide clue for resolving issues in the climate forecast system.