ESTIMATING AIRBORNE PARTICULATE MATTER IN THE NATIONAL CAPITAL REGION, PHILIPPINES USING MULTIPLE LINEAR REGRESSION AND GRADIENT BOOSTING ALGORITHM ON MODIS MAIAC AEROSOL OPTICAL DEPTH

Abstract

The generation of air quality concentration data is imperative for the health and environment of highly urbanized regions. Through remote sensing, air pollutant concentrations can be obtained over large areas for a long time. In this study, particulate matter (PM2.5 and PM10) concentrations were estimated using satellite-derived Moderate Resolution Imaging Spectroradiometer (MODIS) Multi-Angle Implementation of Atmospheric Correction (MAIAC) Aerosol Optical Depth (AOD) values observed in the National Capital Region (NCR), Philippines. Models were generated using multiple linear regression (MLR) and gradient boosting regression to determine the best models for the whole data from 2017 to 2020, dry season, and wet season with a 70–30 split for the train-test sets. Initial models resulted with the best coefficient of determination R2 values of 2.6% and 1.2% using MLR and 2.0% using gradient boosting regression. The results for PM2.5 and PM10 showed the lowest Root Mean Square Error (RMSE) values of 8.79 μg/m3 and 18.99 μg/m3 using MLR and 8.08 μg/m3 and 16.85 μg/m3 using gradient boosting, respectively. The preliminary results indicate the relatively poor performance of models in estimating particulate matter using satellite-derived AOD images. Improvements in the models will include the integration of more in-situ data from air quality monitoring stations and the addition of additional variables and features such as meteorological parameters and geographical layers.