Southeast Asia’s haze: fires and forecasts together

Integrating satellite fire detections with atmospheric composition forecasts, and what that combination can and cannot tell us.

Reading this carefully

Shows

How CAMS forecasts expected aerosol conditions to evolve over 14–19 October 2026, set against fires observed the week before.

Doesn’t show

Smoke-plume trajectories. Fire locations and high aerosol together do not establish a transport pathway, and the arrows are modelled 10 m winds.

Connected to

Fire activity and emission strength, winds at several atmospheric levels, vertical mixing, and removal by rainfall. The forecast model represents these; this visualisation does not separate them.

Forest and vegetation fires are important contributors to atmospheric aerosol loading, particularly where seasonal biomass burning affects air quality across national boundaries. Southeast Asia is one such region, and understanding where aerosols are and how they may move requires more than finding the fires.

I put together a small visualisation integrating satellite-derived active-fire detections with atmospheric composition forecasts, to look at how aerosol conditions might evolve across the region.

Data

Fires

NASA FIRMS (VIIRS) active-fire detections, 7–14 October 2026

Forecasts

Copernicus Atmosphere Monitoring Service: five-day forecasts of total AOD, organic matter AOD, surface PM₂.₅ and near-surface winds, initialised 14 October 2026

Terrain

GEBCO 2026

Processing

Python and GIS libraries, animated at six-hour intervals

What the forecasts indicated

Across the study region, modelled mean total AOD decreases by about 43.5% over the five days, and organic matter AOD by about 49%. The reduction is not uniform: areas of elevated aerosol remain in parts of equatorial Southeast Asia even as the regional mean falls. The fire detections give useful geographical context for where organic aerosol loading is enhanced.

What it doesn’t establish

Identifying active fires and observing elevated aerosol does not by itself establish a transport pathway between them. The wind vectors are modelled 10 m winds, while aerosol transport happens across several atmospheric levels. This is an illustration of forecast evolution in relation to observed fires, not a reconstruction of smoke plumes.

The broader aim was to see how several Earth observation and modelling datasets can be combined to communicate a regional process, while staying clear about what each dataset represents.

The animation shows model forecasts, not observed future air quality. Fire detections are historical observations. Data: CAMS/ECMWF, NASA FIRMS, GEBCO Compilation Group (2026).