The Impact of Model Retraining Frequency and Bias Correction on Predictive Performance in Air Quality Forecasting: Application of NASA’s GEOS-CF and Local Observational Data

Le résumé fourni par la source

To effectively address air pollution's environmental and health impacts, high-definition data and advanced methods capturing both temporal and spatial variations are essential. This paper presents a new approach that uses machine leaning models to improve air quality predictions based on large-scale atmospheric models and in-stu observations. The main focus of this paper is to examine the impact of the model type, refinement and retraining frequency on capturing seasonal variations in regional air quality via comparing different models and retraining intervals, to provide practical recommendations for optimizing machine learning-based air quality forecasting. Our findings reveal that while more frequent retraining significantly improves alignment with observed seasonal trends and improves predictive accuracy, it also increases data demands and computational costs. Importantly, Frequent retraining of models in stable environments tends to produce smaller improvements over time. This highlights the importance of choosing flexible model designs that can adapt well to the constantly changing patterns in air pollution data. These findings provide valuable advice for researchers, policymakers, and environmental organizations working to apply machine learning methods for better air quality management and control strategies.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.