Table './tdc/tbl_dc_meta_control_57' is marked as crashed and last (automatic?) repair failed
Phakphum Paluang. The Estimation of air pollutants from biomass burning emissions and prediction of air quality using multilayer perceptron artificial neural network in Northern Thailand. Master's Degree(Geography and Geoinformatics). Chiang Mai University. Library. : Chiang Mai University, 2024.
The Estimation of air pollutants from biomass burning emissions and prediction of air quality using multilayer perceptron artificial neural network in Northern Thailand
Abstract:
Air quality issues exceeding the standard Air Quality Index (AQI) threshold are prevalent in many countries, particularly in Northern Thailand during the dry season (January-April), largely due to open biomass burning. This study had three objectives: 1. To estimate the air emissions from open biomass burning and agro-industries. Open biomass burning covers forest areas, and agricultural waste burning in rice, maize, and sugarcane plantations, while agro-industries are assessed based on bagasse used in sugar factories. Data on burned areas and the amount of bagasse from each sugar factory were used as primary sources for air emissions estimation. Burned areas were identified using the Random Forest (RF) algorithm, with training datasets of 100 data points per area, generated by the Geo-Informatics and Space Technology Development Agency (GISTDA). The amount of bagasse was estimated according to guidelines from the Department of Alternative Energy Development and Efficiency. 2. To understand the influence of meteorological data and air emission inventory on ground-level PM2.5 concentrations. This objective focuses on using the Pearson correlation coefficient to identify factors influencing the level of PM 2.5 mass concentration changes. 3. To develop a model for estimating ground-level of PM 2.5 mass concentrations using a Multilayer Perceptron Artificial Neural Network (MLP-ANN), evaluating the models performance through Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). The results for the first objective showed that the RF algorithm effectively identified burned areas, though it detected more areas than reported by GISTDA, primarily due to differences in satellite image spatial resolution. This discrepancy led to higher estimated emissions than in previous studies. In 2019, air emissions were the highest, with PM 2.5 emissions reaching 90,424.8 tons/year, followed by 87,822.5 tons/year in 2020 and 59,367.3 tons/year in 2021. Similarly, emissions from sugar factories peaked in 2019 at 3,604.8 tons/year, then decreased to 2,182.6 tons/year in 2020 and 2,045.0 tons/year in 2021. For the second objective, it was found that all factors were correlated with the level of PM 2.5 mass concentration variations under different conditions, with location-specific and temporal differences being key determinants in defining these relationships. Finally, for the third objective, the optimal network architecture of 8-16-1 achieved the highest model efficiency, with MAE and RMSE values of 0.0187 and 0.0282, respectively. However, the model underestimated the level of PM 2.5 mass concentrations compared to ground station observations, primarily due to spatial differences, pollution sources, meteorological factors, and transboundary pollution influences, which pose significant c