Modelling of Aerosol Optical Depth Using Machine Learning Techniques

Abstract

Aerosol Optical Depth (AOD) is a critical parameter in atmospheric sciences, representing the concentration of aerosols in a vertical column of the atmosphere. Accurate prediction of AOD is essential for understanding air quality, climate change, and their impacts on human health. This study explores the potential of machine learning techniques in predicting AOD levels over urban regions in Pakistan, specifically Lahore and Karachi. We employ three machine learning models: Support Vector Regression (SVR), Gradient-Boosting Decision Tree (GBDT), and Random Forest (RF), leveraging various meteorological and environmental datasets. The datasets are pre-processed by removing outliers, handling missing values, and standardizing the data points. Key input features include temperature, relative humidity, wind speed, wind direction, and day of the year. We validate the performance of these models using metrics such as the correlation coefficient (R), Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE). The results indicate that the SVR model, optimized using the Gray Wolf Optimizer (GWO), outperforms the other models with a correlation coefficient (R) of 0.64 for Lahore and 0.54 for Karachi. The optimized SVR model also significantly improves in MAE and RMSE, highlighting its robustness and accuracy in predicting AOD levels. This study demonstrates the efficacy of Machine Learning (ML) techniques in environmental monitoring, providing a reliable tool for predicting AOD. The findings suggest that, by incorporating higher-quality data and a broader range of input variables, further improvements can be achieved. The successful application of these models in Pakistan could pave the way for enhanced air quality management and climate research in other regions.