Prediction of fugitive landfill gas hotspots using a random forest algorithm and Sentinel-2 data

Abstract Methane from landfill gas is an important contributor of anthropogenic greenhouse gas emissions. Regular monitoring of fugitive landfill methane hotspots is of great practical and environmental importance. The Toronto Green lane landfill is selected in this study, and an analytical approach was developed to identify fugitive gas emission hotspots using the higher resolution Sentinel-2 imagery and a random forest (RF) algorithm. Landsat-8 imagery is traditionally used to identify landfill emission hotspots; however, the resulting spatial resolution is often inadequate for increasingly stringent environmental regulations. Various statistical, climatological, and spectral variables on final model performance were examined. The proposed RF approach reduced data spread, with standard deviations between 84.8% to 99.0%. The presence of water vapor might reduce the accuracy of spectral reflectance and lower model performance. The most significant bands in constructing the RF model are Sentinel-2 spectral band 1 and band 9, with a relative importance of 1.0 and 0.8, respectively. Results showed that the predicted land surface temperatures (LSTs) agree well with LSTs estimated from a conventional approach using Landsat-8 imagery, with R > 0.96. The proposed method maps the fugitive gas emission hotspots at the Green lane landfill with a higher spatial resolution.