Satellite-based air quality assessment over HSIA, Dhaka: Using statistical and deep learning models

Abstract

This study evaluates the air quality at Hazrat Shahjalal International Airport (HSIA), Dhaka, from 2019 to 2024 using Sentinel-5P satellite data and advanced forecasting models. Concentrations of NO2, SO2, CO, and O3 were analyzed for spatiotemporal patterns using Google Earth Engine (GEE). Results show that NO2 and SO2 levels peak during winter, with significant reductions observed during the 2020 COVID-19 lockdown and rebounds in subsequent years. Time series decomposition reveals strong seasonal cycles for all pollutants. Forecasting was performed using Seasonal ARIMA and Stacked LSTM models for 2025-2029. SARIMA yielded better performance for NO2 and CO due to their linear-seasonal nature, while LSTM outperformed for O3, capturing nonlinear patterns. While NO2 and SO2 are projected to decline post-2024, CO and O3 are expected to remain stable with minor fluctuations. These findings highlight HSIA as a persistent pollution hotspot and support the use of satellite-based remote sensing and deep learning approaches for scalable, cost-effective air quality monitoring and management in urban airports and densely populated urban areas.