Estimating intra-daily PM10 concentrations over the north-western region of Turkey based on MODIS AOD using random forest approach
- Download 49
- File Size 1.71 MB
- Create Date March 21, 2022
Gizem Tuna Tuygun*, Tolga Elbir
Aerosol optical depth (AOD) is a significant predictor that can support PM concentrations in a region without ground-based monitoring. Estimation of intra-daily ground-level PM10 concentrations based on satellite AOD products has been widely used over large areas with the help of machine learning models. This study was the first attempt in the north-western part of Turkey to estimate daily PM10 concentrations based on Collection 6.1 (C6.1) Moderate Resolution Imaging Spectroradiometer (MODIS) DTB AOD products from Terra and Aqua satellites. A space-time random forest (STRF) model outperformed most models with strong predictive power to evaluate the differences between the time windows representing the mean PM10 around the overpass times of the satellites with the synergetic use of space-time information over Turkey was developed in this study. Several spatiotemporal parameters such as MODIS AOD, meteorological, and land-related data were used to improve the overall accuracy of PM10 estimation for 2008-2019. The most significant 13 variables were used to estimate PM10 in the final STRF model. The STRF model performed moderately well, with a moderate correlation coefficient (R) of 0.71 (0.73) for Terra (Aqua) satellite overpass time. The model better estimated PM10 concentrations at Aqua overpass time (R~0.73 and RMSE ~27.3 µg/m3). Moreover, the STRF model developed in this study is the first step to construct a high-quality PM10 dataset over the region with poor air quality due to industrialization and the rapid growth of the regional population.
Keywords: MODIS; AOD; PM10 estimation; Random Forest; Turkey
Proscience vol. 8