High spatiotemporal resolution PM<inf>2.5</inf> concentration estimation with satellite and ground observations: A case study in New York City

Abstract

High spatiotemporal resolution concentration of fine particulate matter (PM2.5) enables accurate and detailed air quality monitoring, especially for metropolitan cities with high levels of population density. Although ground air quality monitoring stations can provide timely and accurate observations, they are usually very sparsely distributed, and cannot provide PM2.5 concentration data with continuous spatial coverage. Instead, satellite observations, e.g., Landsat 8/Thermal Infrared Sensor (TIRS) and Terra/Moderate Resolution Imaging Spectroradiometer (MODIS), can both obtain data with continuous coverage. However, there is a trade-off between satellite sensors' spatial and temporal resolution. Hence, this study presents an estimation model for PM2.5 concentrations that combines these multi-source data to produce high spatiotemporal resolution concentration maps in urban area. The approach is tested on New York City, NY, USA. Specifically, we first use cloud-free MODIS thermal band images and the corresponding ground-station PM2.5 records to build a local PM2.5 prediction model. Then, we exploit a spatiotemporal image fusion technique to obtain Landsat-like thermal band image series from Landsat 8/TIRS (100 m spatial resolution) and Terra/MODIS (1 km spatial resolution) sensors. Finally, we convert the fused high spatiotemporal resolution thermal band images to PM2.5 concentration maps by the prediction model from step 1. The validation between the estimated and the real PM2.5 values shows that the detailed Landsat-like high spatial resolution PM2.5 estimations are more accurate than the original blurred MODIS one.