Atmospheric Pollution Retrieval Using Path Radiance Derived from Remote Sensing Data

With increasing pollution and its effects on human life, its monitoring and mitigation have become a vital problem. While an air quality ground monitoring station has limited coverage, remote-sensing-based monitoring of air pollution offers the advantage of large area coverage and a synoptic view of pollution dispersion. This study demonstrates the retrieval of particulate matter 10 (PM10) and sulfur dioxide (SO2) using the multitemporal Landsat ETM+ remote sensing imagery of the study area of Vadodara, India. Atmospheric correction is generally applied on satellite images as a part of the preprocessing task in remote sensing applications. In this study, the atmospheric path radiance derived from the atmospheric correction task was experimented with for estimating the amount of PM10 and SO2 using prediction modeling. The traditional multiple linear regression (MLR) and the artificial neural network (ANN) modeling were applied on the processed remote sensing data, and the results thus obtained were assessed using independent test data. The correlation coefficient (R) for the prediction modeling of PM10 was 0.78 for the MLR and 0.72 for ANN algorithms, respectively. For the SO2 estimation, the correlation coefficient (R) was 0.77 and 0.69 for MLR and ANN models, respectively. The test data were evaluated using the root mean square error (RMSE), which was 36.90 and 24.54 using MLR and ANN modeling, respectively for PM10 and SO2 modeling. During the estimation modeling for SO2, the RMSE was 2.70 and 4.19 for MLR and ANN, respectively. The results confirmed the feasibility of the atmospheric-correction-based prediction modeling of PM10 and SO2. Comparison of the prediction outcomes of MLR and ANN modeling showed that ANN provides better results than the classical regression-based prediction model. Using MLP, the test data prediction error (RMSE) was reduced by 33.50% for PM10 and 34.93% for SO2 over the RMSE reported by MLR estimation.