Multi-temporal burned-area mapping

Multi-temporal burned-area mapping answers the question: where did fire occur, and on which date? The method detects burned surfaces by tracking persistent post-fire changes in surface reflectance across a satellite time-series, distinguishing fire-induced spectral change from transient events such as cloud shadows and phenological shifts.

The MODIS MCD64A1 Collection 6 algorithm (Giglio et al. 2018) uses a hybrid approach.^[1] A burn-sensitive vegetation index (VI) is computed from MODIS shortwave-infrared bands 5 (1240 nm) and 7 (2130 nm), which are sensitive to the spectral signature of post-fire char and ash. A temporal texture metric, measuring variance across the compositing window, distinguishes persistent spectral change from transient artefacts. A dynamic threshold is applied to the combined VI and texture composite. Active fire detections from MODIS thermal anomaly products serve as training samples, anchoring the detection where fire was directly observed from space: 44% of burned pixels in the product coincide with an active fire detection on the same day, and 68% within two days of the mapped burn date.^[1] Collection 6 substantially improved detection of small burns relative to Collection 5.1.^[2]

MCD64A1 v061 provides 500m monthly global burned area with per-pixel burn date assignment and quality flags, covering 2000 to present.^[3] The product is used in IPCC greenhouse gas inventories and the Global Fire Emissions Database (GFED).

The method fails under persistent cloud cover, where burned area may go undetected during the satellite overpass. Rapid savanna re-greening can obscure the burn scar within days of the fire. Small and patchy burns in heterogeneous landscapes are significantly under-detected at 500m pixel resolution.