Enhancing PM 2.5 prediction coverage in Northern Thailand using ground station information assisted with aerosol optical depth and weather data from NASA satellite instruments

Air pollution has become a significant issue in Northern Thailand due to forest fires, human activities, and seasonal climate variations. This study primarily addresses the issue of missing data in PM10 measurements obtained from the Pollution Control Department (PCD) stations in the provinces of Chiang Mai, Lampang, and Nan, which limited the capability to predict future pollutant levels. The study proposed three imputation methods: k-nearest neighbors (KNN), last observation carried forward (LOCF), and nearest observation carried backward (NOCB) to solve the existing missing gaps in the dataset. The NOCB method yielded the highest R², the lowest mean absolute error (MAE), and the lowest root mean squared error (RMSE) across varied proportions of mimic missing data. The imputed data were then used for time-series predictions with long short-term memory (LSTM) neural networks, achieving an R² of approximately 0.92 for PM10 predictions. Building on these foundations, the research extends by incorporating the data from the National Astronomical Research Institute of Thailand (NARIT) and the National Aeronautics and Space Administration (NASA) satellite observations to enhance the prediction of PM2.5 levels. The results demonstrated a strong relationship between ground-based PM2.5 and satellite data suggesting that the incorporation of satellite data enables more accurate PM2.5 analysis, even in the absence of ground monitoring stations. The findings indicate that satellite-derived aerosol optical depth (AOD) and other meteorological weather variables from NASA instruments can effectively supplement ground-based PM2.5 prediction, ensuring reliable pollution forecasting and enhanced environmental management in Northern Thailand. This approach not only addresses data gaps but also highlights the potential of utilizing satellite data in regions lacking extensive ground monitoring infrastructure.