Abstract
The study aimed to assess the accuracy of automated methane emission monitoring systems at oil and gas fields in Azerbaijan using satellite data and a radiation transfer model. The methodology included analysing Sentinel-5P and GHGSat satellite data for 2024, applying MODTRAN and SCIATRAN models to incorporate atmospheric factors, and validating the results with ground measurements using Los Gatos Research spectrometers and Picarro G2401 gas analysers. The results demonstrated that GHGSat determined localised emissions (R2=0.89, RMSE= .7 ppb) with greater accuracy, while Sentinel-5P demonstrated underestimation of concentrations in high humidity (R2=0.72, RMSE=12.4 ppb). Data correction using the MODTRAN and SCIATRAN models improved the accuracy of the measurements: RMSE decreased to 7.8 ppb for Sentinel-5P and 4.2 ppb for GHGSat. The highest methane emissions were detected on the Apsheron Peninsula (2.8 ppb), which is associated with leaks and gas processing processes. The seasonal analysis demonstrated an increase in concentrations in winter (3 ppm) and summer (2.7 ppm) due to a decrease in the dispersion rate and intensification of mining. Machine learning methods (XGBoost, Random Forest) improved the forecasting accuracy (R2=0.91 for XGBoost) by identifying key factors: wind speed, temperature, mining intensity and humidity. The findings highlight the need to integrate satellite data with ground-based measurements and radiative transfer models to improve monitoring accuracy and develop emission reduction strategies