Development and Assessment of ADAM3 Ensemble Prediction System

抄録

Forecasts of discrete events, such as precipitation, dust storms, and typhoons, can be deterministic, categorical, or probabilistic. Ensemble-based probabilistic predictions generate more consistent forecasts than individual deterministic models. We introduce and evaluate the Ensemble Prediction System of Asian Dust Aerosol Model 3, which generates probability forecasts of Asian dust. Probability forecasts were produced for 300 μgm⁻³ PM₁₀ mass concentrations, according to the air-quality standards of the Ministry of Environment of South Korea and Korea Meteorological Administration. Crisis-level information was produced to categorize the dust risk level for Asia, using a risk matrix. The model's performance was evaluated using a 2 × 2 contingency table, the Brier skill score, and a reliability diagram. For skill score evaluation via the contingency table, the average hit rate and threat score were 0.46 and 0.34 for the best three sites, Jurihe, Erenhot, and Wulatezhongqi. The Brier skill score was positive for approximately 60% of stations, with the highest (0.410) and lowest (−2.038) values in Erenhot and Yanan, respectively. The reliability diagram revealed overestimated Asian dust frequencies for all stations. Although the stations were located within the same regions, their skill scores differed. Regional characteristics of skill scores should be further investigated in the future.