S-MESH: Surface PM2.5 daily 1km dataset over Central Europe using CAMS, Low-Cost Sensors and AOD

These datasets consist of estimated daily surface PM_2.5 concentrations over Europe at ~1km spatial resolution for the years 2021-2022 in tiff file format generated using our Satellite and ML-based Estimation of Surface air quality at High resolution (S-MESH) model. The S-MESH model for PM_2.5 generates surface PM_2.5 over Europe by downscaling CAMS regional forecast using LCS, satellite AOD and other meteorological parameters through a stacked XGBoost model. The downscaled S-MESH products have accuracies similar to CAMS regional interim reanalysis. More details on this methodology and study can be found in the related research article  https://doi.org/10.1016/j.envres.2025.123558

The S-MESH framework was built in 3 versions for comparison as mentioned below:

1. Baseline Model - contains no LCS data and air quality reference station observations are used as training target features
2. LCST Model - LCS data is used as training target feature
3. LCSI Model - LCS dataset elements are used as additional input features while air quality reference station observations are used as training target features

The attached dataset files are zipped into 6 folders each corresponding to a model type and year. Individual folders can be unzipped from command line using "tar -xvzf filename.tar.gz". Each tiff file represents daily PM_2.5 and is a single band image with an extent of 2.5°E-23°E and 46°N–55°N in EPSG:4326 - WGS 84 projection. 

This work was primarily supported by a PhD fellowship from the Norwegian Research Council under grant agreement 321851.