A hotspot is a lead. Identification, missed detections and duplication must be addressed before it becomes an ecological estimate.
Thermal imaging does not see through everything
Thermal contrast depends on the animal and its background; sun-warmed surfaces can reduce separation or create confusing hotspots. Vegetation and terrain still occlude animals, and underwater observation has different optical constraints. RGB offers richer identification cues but suffers from shadows and glare. Validate the acquisition window against local activity, canopy and weather instead of assuming that one time of day is universally best.
Align time, geometry and resolution
Matching a thermal hotspot to an RGB animal requires time synchronization, camera calibration, field-of-view alignment and an estimate of surface geometry. Movement can invalidate a match even across a short delay. Preserve raw frames, calibration parameters, transformation versions and alignment uncertainty. Side-by-side images are useful for review, but are not proof of pixel-level correspondence.
Separate candidate detection from confirmation
Rocks, structures, warm ground and other species can all generate candidates. Keep machine detections, reviewed observations and deduplicated individuals as separate records. Corcoran and colleagues validated automated koala detection against radio-collared animals, illustrating why detection probability and precision answer different questions. A high detection rate alone can conceal substantial false-positive review work.
Check usability before leaving the site
Capture a short representative segment and inspect target size in pixels, blur, thermal contrast, positioning and storage. A more capable model cannot recover animals that were never visible. Define reacquisition criteria beforehand. Report unobservable areas as missing observation rather than zero animals; this protects both inference and the field budget.
Evidence & further reading
Check the original study conditions, species and location before using a result in a plan. Do not transfer findings universally across species and settings.