Explainable Machine Learning for Air Quality Prediction, With Satellite and Weather Data

Abstract

Air pollution is a major threat in current times and I consider PM2.5 as a pollutant because PM2.5 has negative impacts on human health. Hence PM2.5 prediction is crucial. The prediction of PM2.5 is a challenging task due to its non-linear relationship with various environmental factors and dust storm events. In this paper, we present a novel machine learning approach for predicting PM2.5 concentration in the Bhubaneswar region of Odisha. Using annotated data sets with corresponding meteorological parameters and satellite Aerosol Optical Depth (AOD) we demonstrate our results. We have collected a total of 147 annotated data sets from 12 October 2025 to 28 June 2026. Along with PM2.5, we have collected ambient air temperature, relative humidity, atmospheric pressure, wind speed, wind direction, precipitation, and AOD 0.55 µm as input parameters for our models. We have implemented four machine learning regressors namely Linear Regression, Random Forest, Gradient Boosting and XGBoost to predict PM2.5 concentrations. We perform experiments with two sets of input features consisting of only meteorological parameters and other consisting of meteorological parameters and AOD.For the purposes of evaluation, we calculate the Model Performance Statistics, including the Mean Absolute Error (MAE), the Root Mean Square Error (RMSE), and finally, the R squared (R2). Additionally, we apply the SHAP (SHapley Additive exPlanations) method to explain our model, that is, to determine which variables influence the outcome. Gradient Boosting algorithm performed the best among all the models. Gradient Boosting when trained on the combined set of input parameters of meteorological data and AOD produced an R2 value of 0.79, an MAE of 11.52 µg/m3 and an RMSE of 15.78 µg/m3. We also observed that inclusion of AOD values improved performance of all the four models. This demonstrates that using combined sets of satellite and local data improves PM2.5 prediction accuracy. Similarly, the use of SHAP provided insights on the contribution of different parameters in the prediction process.