The Korea Meteorological Administration has employed the Asian Dust Aerosol Model 2 (ADAM2) to forecast Asian dust events since 2010, where the dust emission flux is proportional to the fourth power of the friction velocity. Currently, the dust emission reduction factor (RF) is determined by the normalized difference vegetation index (NDVI). This study aims to improve the forecasting capability of ADAM2 by developing a daily dust RF using both monthly (January 2007 to December 2016) and real-time moderate resolution imaging spectroradiometer (MODIS) NDVI data. We also developed a look-up table to transform the RF using NDVI and a system to update the RF by producing MODIS NDVI data for the last 30 days. Using these data, new RFs can be produced every day. To examine the impact of RF modification, the current (CTL) and new (EXP) RFs are compared during the period from March to May 2017. The simulations are verified by ground-based PM₁₀ observations from China and Korea. Accordingly, root mean square errors (RMSEs) are reduced by 11.58% when RF is updated using real-time NDVI data. The results suggest that recent daily NDVI data contribute positively to the forecasting ability of ADAM2, in the dust source and downwind regions.

Recently, the Korea Meteorological Administration developed Asian Dust Aerosol Model version 3 (ADAM3) by incorporating additional parameters into ADAM2, including anthropogenic particulate matter (PM) emissions, modification of dust generation by considering real-time surface vegetation, and assimilations of surface PM observations and satellite-measured aerosol optical depth. This study evaluates the performance of ADAM3 in identifying Asian dust days over the dust source regions in Northern China and their variations according to regions and soil types by comparing its performance with ADAM2 (from January to June of 2017). In all regions the performance of ADAM3 was markedly improved, especially for Northwestern China, where the threat score (TS) and the probability of detection (POD) improved from 5.4% and 5.5% to 30.4% and 34.4%, respectively. ADAM3 outperforms ADAM2 for all soil types, especially for the sand-type soil for which TS and POD are improved from 39.2.0% and 50.7% to 48.9% and 68.2%, respectively. Despite these improvements in regions and surface soil types, Asian dust emission formulas in ADAM3 need improvement for the loess-type soils to modulate the overestimation of Asian dust events related to anthropogenic emissions in the Huabei Plain and Manchuria.

Dust events in Northeast Asia have several adverse effects on human health, agricultural land, infrastructure, and transport. Wind speed is the most important factor in determining the total dust emission at the land surface; however, various land-surface conditions must be considered as well. Recently, the Korea Meteorological Administration updated the dust emission reduction factor (RF) in the Asian Dust Aerosol Model 3 (ADAM3) using data from the normalized difference vegetation index (NDVI) of the Moderate Resolution Imaging Spectroradiometer (MODIS). We evaluated the improvements of ADAM3 according to soil types. We incorporated new RF formulations in the evaluation based on real-time MODIS NDVI data obtained over the Asian dust source regions in northern China during spring 2017. This incorporation improved the simulation performance of ADAM3 for the PM₁₀ mass concentration in Inner Mongolia and Manchuria for all soil types, except Gobi. The ADAM3 skill scores for sand, loess, and mixed types in a 24 h forecast increased by 6.6%, 20.4%, and 13.3%, respectively, compared with those in forecasts employing the monthly RF based on the NDVI data. As surface conditions in the dust source regions continually change, incorporating real-time vegetation data is critical to improving performance of dust forecast models such as ADAM3.

To downscale artificial intelligence (AI) medium-range forecasts over the Korean Peninsula, we developed a 1.5 km CorrDiff system trained with ERA5 inputs and UM-LDAPS analyses as targets. For 1.5 m temperature and 10 m wind, the regression stage reduced root-mean-square error (RMSE) and root-mean-square vector error (RMSVE) relative to ERA5 and increased spatial correlation. CorrDiff had smaller absolute biases and greater spatial variability than Regression but higher RMSE and RMSVE and lower spatial correlation. The most consistent improvement relative to Regression occurred for high-temperature events at or above the target 99th percentile, with higher probability of detection (POD) and critical success index (CSI) and lower conditional temperature error. Without additional training or correction, AI Forecasting System (AIFS) forecasts were downscaled through +288 h. In a July 2026 heatwave case at 47 stations, CorrDiff reduced temperature errors and alleviated the underprediction of high-temperature occurrence relative to Regression. CorrDiff had higher POD and CSI than the AIFS for temperatures ≥33 °C but higher RMSE and lower spatial correlation. Generative downscaling should therefore be evaluated for spatial variability, spatial agreement, and event-specific performance beyond deterministic errors; producing high-resolution data does not necessarily imply improved forecast performance against observations.