Spatiotemporal imputation and bias correction of Sentinel-3 SYN for intraurban air quality assessment using deep learning.

This work describes preliminary attempts aimed at creating a dataset of daily averages of aerosol optical depth (AOD) on an intraurban scale (300m) using MODIS MAIAC AOD, SEN3 SYN, and AOD from ERA5 reanalysis models. Our preliminary efforts were aimed at understanding the quality of available AOD products by comparing them with daily average measurements provided by the AERONET network for the Italian peninsula during the reference period 2019-2023. MODIS MAIAC AOD proves to be state-of-the-art in satellite AOD reconstruction, while SEN3 SYN correlates less and shows a significant bias when compared with AERONET. Our efforts are thus oriented in two directions: a) evaluating whether SEN3 SYN is mature enough to deliver unbiased AOD products on an intraurban scale on a daily, weekly, and monthly basis, possibly using other sources of information such as DEM, latlon, and LST; b) performing data imputation of missing observations using AOD from reanalysis models such as ERA5. Regarding AOD correction on an urban/intraurban scale, we are evaluating pixel-based approaches such as linear/nonlinear/GAM regression algorithms fueled by the combined use of SEN3 SYN, MODIS MAIAC, ERA5 AOD, and auxiliary data such as MODIS land surface temperature and climatic data. Our findings demonstrate that there is room for further improvement of AOD products by imputing missing AOD values and by further calibrating AOD using regression models fed with available AOD estimates and auxiliary data. This work is part of a collaborative project funded by ASI and called APEMAIA (Assessment of PM Exposure at the intra-urban scale in preparation for the MAIA mission). The project is designed to investigate the potential of MAIA by developing a multi-modular system for extracting PM concentrations at the intra-urban scale using Artificial Intelligence techniques.