Remote Sensing Letters, Год журнала: 2024, Номер 16(2), С. 136 - 145

Опубликована: Дек. 16, 2024

Recent years have seen increasing academic attention to surface ozone pollution due its significant impacts on air quality and human health. To overcome the spatial coverage limitation of ground monitoring, we proposed a novel approach that integrated Generative Adversarial Network (GAN) with Light Gradient Boosting Machine (LGBM) for full-coverage estimation in Yangtze River Delta Urban Agglomeration (YRDUA), using monitoring data Sentinel-5P satellite data. We assessed performance GAN-LGBM model against other decision-tree-based models (XGBoost, LGBM) three cross-validation (CV) methods: sample-based, space-based, time-based. The results demonstrated consistently outperformed across all evaluation metrics validation scenarios, achieving highest correlation coefficient (R2) 0.94 sample-based CV. Spatiotemporal evaluations further showed robustness ability capture complex concentration patterns. This study introduces promising method explores potential incorporating unsupervised learning methods into regression address correlations environmental datasets.

Язык: Английский