Discrimination of coarse and fine aerosols maps from satellite data with a unique algorithm for land and water bodies

We present an unsupervised procedure for near-real-time mapping of advected aerosols, namely desert dust and biomass-burning smoke, at urban scale. A differential, surface-normalized spectral index (DBB-2) is computed from four spectrally overlapping bands of Sentinel-2 MSI (B2, B3, B4, B5; 10 m) and Sentinel-3 OLCI (Oa04, Oa06, Oa08, Oa11; 300 m). The 10m Sentinel-2 map, available every five days in cloud-free conditions, is fused at feature level with the near-daily 300m OLCI map, so that aerosol loading is updated in each 300m cell at least every two days while the sub-cell spatial texture is extrapolated from the latest Sentinel-2 acquisition. The sign of DBB-2 separates scattering-dominated dust (positive) from absorption-dominated smoke (negative), and its modulus tracks the aerosol optical depth (AOD). The clear-sky reference is selected from Sentinel-2 products alone, so no ground-based auxiliary data are needed. Validation against AERONET spectral deconvolution (SDA) AOD at Potenza (Italy, Saharan dust, March–April 2024) and Barcelona (Spain, smoke, October 2020) yields agreement of about 96% and 95%, respectively. A full account of the method is given in Ref. 1.