Proceedings Article | 29 January 2024
KEYWORDS: Carbon monoxide, Data modeling, Modeling, Atmospheric modeling, Satellites, Vegetation, Satellite imaging, Meteorology, Landsat, Interpolation, Remote sensing
The rising demand for electricity, driven primarily by coal-fired power plants, has escalated concerns over hazardous gas emissions and their impact on air quality and human health. This study focuses on the Pelabuhan Ratu region, where there is a notable gap in understanding the spatial and temporal distribution of particulate matter (PM) and carbon monoxide (CO). To address this, we conducted a ground survey to measure concentrations of CO, PM2.5, and PM10 at various points. Additionally, we utilized Landsat 8 satellite imagery to predict the spatial distribution of these aerosols, while also developing a one-year temporal model. Pelabuhan Ratu's unique geomorphology, encompassing both mountains and coasts, significantly influences pollutant concentrations, which vary with elevation and proximity to the power plant. Employing the Random Forest machine learning algorithm, we predicted concentrations of CO, PM2.5, and PM10 by integrating ground-level gas concentrations with satellite-derived vegetation indices, ambient temperature, altitude, land use, wind direction, and humidity data. Our findings reveal varied predictive accuracies: the CO model exhibited a low correlation value (0.32) and a Root Mean Square Error (RMSE) of 136 ppm, suggesting a less reliable prediction. In contrast, the PM2.5 model showed a moderate correlation (0.474) with an RMSE of 18.4 μg/m3. The PM10 model performed slightly better, achieving a correlation of 0.56 and an RMSE of 55.4 μg/m3. These results underscore the challenges and potential of using integrated ground and satellite data for predicting air pollutant concentrations in complex geographic settings.
