Abstract:
Particulate Matter 2.5 micrometres and smaller (PM2.5) could be measured by an instrument
at ground-based observatory. But the ground-based observatory cannot provide the data
covering wide area. Currently, remote sensing is used to be an application to estimate PM2.5.
To address capability of PM2.5 estimation method, the study carried out comparing two
methods for estimating PM2.5 concentrations between Multiple Linear Regression (MLR)
and Principal Component Analysis - General Regression Neural Network: PCA-GRNN.
By using Aerosol Optical Depth (AOD) from Himawari-8 satellite and physical data such
as the Digital Elevation Model (DEM), Normalized Diff erence Vegetation Index (NDVI) and
meteorological data during January to December 2018. The estimation results from those
two methods were evaluated by PM2.5 concentration from ground-bases measuring.
The evaluated results show that the PCA-GRNN obtained the root mean square error
(RMSE) of 17.76 and R2 of 0.566, while the MLR obtained RMSE of 33.90 and R2 of
0.012. Therefore, it is indicated that PCA-GRNN is an appropriate method to estimate
PM2.5 concentration over Northern Thailand more than the MLR.