Air Pollution Modeling from Remotely Sensed Data Using Regression Techniques

Abstract

There is a need for timely information about changes in the air pollution levels in cities for adopting precautionary measures. Keeping this in view, an attempt has been made to develop a model which will be useful to obtain air quality information directly from remotely sensed data easily and quickly. For this study pixel values, vegetation indices and urbanization index from IRS P6 LISS IV and Landsat ETM+ images were used to develop regression based models with Air Pollution Index (API), which were calculated from in-situ air pollutant information. It was found that among the 12 parameters of IRS, highest correlation exists between pixel values in NIR (Near Infra-Red) band (Pearson correlation −0.77) and Normalized Difference Vegetation Index (NDVI) (Pearson correlation −0.68) and both have inverse relationship with API. In case of Landsat, the highest correlation was observed in SWIR (Short Wave Infra-Red) band (Pearson correlation −0.83) and NIR (Pearson correlation −0.78). Both single and multivariate regression models were calibrated from best correlated variables from IRS and Landsat. Among all the models, multivariate regression model from Landsat with four most correlated variables gave the most accurate air pollution image. On comparison between the API modeled and API interpolated images, 90.5 % accuracy was obtained.