Satellite remote sensing is vital for global CO2 monitoring, yet active remote sensing suffers from discontinuous coverage despite overcoming passive technique limitations in cloudy or low-light conditions. Although geostatistical and machine-learning methods have been applied to XCO2 reconstruction, balancing spatiotemporal resolution and accuracy remains challenging. To address spatiotemporal correlation and heterogeneity, this study proposes an eigenvector spatial filtering-based geographically and temporally weighted regression model (ESF-GTWR). Leveraging DQ-1 satellite data from June to December 2022 over China, the model employs eigenvector spatial filtering to capture spatial correlation and integrates GTWR for spatiotemporal heterogeneity, addressing nonstationary characteristics while enhancing accuracy. Validation shows an adjusted coefficient of determination (Adj. R²) of 0.86 ± 0.02 (16.3–47.7% improvement over the geographically weighted regression (GWR) model) and stable root mean square error (RMSE) of 0.86 ± 0.11 ppm (32.9–50.0% reduction), demonstrating superior performance over traditional methods.
Zhu et al. (Fri,) studied this question.