Research on modeling and correction methods for XCH4 remote sensing bias uncertainty through the integration of ground-based observations
-
Abstract
To improve the accuracy and reliability of TROPOMI methane column concentration products for regional-scale applications,this study developed a bias-correction framework based on collocated TROPOMI satellite observations and Total Carbon Column Observing Network(TCCON) measurements from 2019 to 2021, integrated with multi-source environmental variables,including ERA5 reanalysis products,aerosol optical depth,and terrain elevation. First,a seasonal-trend decomposition method was applied to separate the observation bias into trend, seasonal, and residual components, thereby characterizing its systematic variations and random fluctuations. Subsequently,a Bayesian-optimized extreme gradient boosting(XGBoost) model was employed for bias correction, and the trained model was used to reconstruct corrected TROPOMI methane column concentrations. Finally,an interpretable machine learning approach was applied to quantify the contributions of environmental factors to bias variability. The results indicate that TROPOMI methane column concentration biases are jointly influenced by meteorological conditions,underlying surface characteristics,and regional environmental factors,among which soil temperature,2 m air temperature, 2 m dew point temperature, boundary layer height, and near-surface wind fields exhibit the strongest contributions. After bias correction, the mean bias across the nine study sites was substantially reduced, while the average coefficient of determination(R2) increased from 0. 58 to 0. 91,and the average root-mean-square error(RMSE) decreased from 19. 48 nmol/mol to 5. 99 nmol/mol. The proposed bias-correction framework effectively mitigates the systematic errors and random fluctuations in TROPOMI methane column concentration products,significantly improving the accuracy and consistency of satellite observations.
-
-