Method
How each layer is made, what it shows and what it cannot show.
Air
Retrievals. Sentinel-5P TROPOMI offline Level-3 products (NO₂, SO₂, CO, HCHO, O₃, UV aerosol index) as ingested by Google Earth Engine, with its quality screening; MODIS MAIAC aerosol optical depth at 550 nm (MCD19A2 v6.1, 1 km); FIRMS MODIS active fires (confidence ≥ 50). Region 60–98° E, 5–37° N.
Aggregation. Each day is averaged first and the month is the mean of daily means, so days with two overpasses do not count twice. The number of valid days per cell is kept in the source grids. Cells are 0.1° (about 11 km), close to TROPOMI's 5.5 × 3.5 km pixels. The latest month is partial until it ends; it is redrawn at each update.
Columns are not breathing air. A column counts all molecules between ground and space. Winter NO₂ columns rise partly because the shallow winter boundary layer traps emissions, and haze can hide the surface from the sensor. For surface PM2.5 the atlas uses ground monitors (OpenAQ) and a calibrated model; a daily 1 km surface PM2.5 and population exposure layer is in preparation.
Colour. Fixed ranges per layer so months can be compared; perceptually uniform, colour-blind safe maps (cividis, magma, ColorBrewer); low values fade so the land stays readable; cells with no valid retrieval are transparent, not zero.
Snow
MODIS MOD10A1F cloud-gap-filled daily snow cover, hydrological years 2001–2025; Sen slopes of snow-cover days with Mann–Kendall tests and false discovery rate control, by range and elevation band. Change classes per pixel. Full method in the uib-snow-cover repository.
Updates
The source repositories run incremental downloads to the present day (only new and the last two months are fetched again), then
build/build_data.py redraws the newest maps, rewrites the NetCDF and STAC files, and the site is redeployed. Every step is logged.
Code
All code is open: third-pole (this atlas), south-asia-smog, uib-snow-cover.