Monthly, seasonal and annual 1 km PM2.5 estimation surfaces for the Western Province, Sri Lanka (2025)

This dataset contains the gridded PM2.5 estimation surfaces produced for the Western Province of Sri Lanka using a machine learning model trained on satellite aerosol optical depth and ancillary predictors. Estimates were generated with an XGBoost model fitted to 440 matched station-day observations from five ground-based monitoring stations, using MODIS MAIAC AOD (MCD19A2), ERA5-Land meteorological reanalysis (2 m air temperature, relative humidity, 10 m wind speed), OpenStreetMap-derived road density, ESA WorldCover land cover, and a month index as predictors. The fitted model was applied pixel by pixel to monthly predictor raster stacks for January to December 2025. Contents: Twelve monthly mean PM2.5 surfaces, January to December 2025 Four seasonal mean composites following the Sri Lankan monsoon calendar (North-East Monsoon, First Inter-Monsoon, South-West Monsoon, Second Inter-Monsoon) One annual mean surface for 2025 One pixel-wise uncertainty surface, derived as the standard deviation of five leave-one-station-out annual predictions Four exceedance frequency surfaces, giving the number of months per pixel exceeding each WHO interim target (35, 25, 15 and 10 µg/m³) All rasters are supplied as GeoTIFF at 1 km spatial resolution in WGS 84 / UTM zone 44N (EPSG:32644). Values are PM2.5 concentration in µg/m³, except the exceedance surfaces, which give a count of months from 0 to 12. Ground-based station measurements are not included in this deposit. National Building Research Organisation reference-grade data were obtained under institutional permission, and PurpleAir and IQAir records remain subject to the respective providers' terms of use.