Verification of CUACE Model Products in Air Quality Forecasts in Henan Province

Zhu Feng; Tian Li; Wang Xinmin; Kong Haijiang; Li Fei

doi:10.16765/j.cnki.1673-7148.2025.01.002

Abstract：This article verifies and analyzes the air quality forecast performance of CUACE model in Henan Province from April,2017 to March,2021 by using the multi-dimensional grouping verification method.The results show that:(1)There exists overestimation in CUACE's AQI forecasts generally.The model performs the best when the air quality is moderate,but the worst in the case of heavy pollu-tion.The forecast performance of the model is better in June,July,August and September,while it is the worst in March.Its overall forecast performance in Nanyang is the best.(2)CUACE tends to follow"the Doctrine of the Mean"for AQI forecasts,that is,when the observed air quality is good,its forecast result tends to be poor,but when the air quality is poor,it tends to report better.The degree of overesti-mation is the greatest when air quality is good,while the degree of underestimation is the greatest when air is severely polluted.(3)During the intensive pollution stage,there is a systematic deviation in the CUACE forecasts of PM2.5,CO and SO2 concentrations.The predictability of PM10 and O3 is relatively poor,making their correction difficult.(4)The correlation coefficients of PM2.5,O3 and CO predictions decrease with the extension of the model forecast lead time,and each verification indicator of the forecast for pollutant concentration presents peak and trough values in the 24 h cycle.The periodic change feature is the manifestation of diurnal forecast effect variation for each pollutant.(5)The forecast performance of CUACE on AQI is largely affected by the forecast effect of PM2.5.(6)CUACE has the worst forecast per-formance on PM10 because of its weak forecast ability under the sand-dust weather condition.The forecast of PM10 may need to be combined with more specific dust models.Further verification of numerical mod-els for air quality forecasting can provide even better reference for model output correction by adding di-mensions such as weather patterns,using multi-dimensional verification methods simultaneously,and using multi-model product comparisons.