The first deliverable is a decision-focused sampling plan with an uncertainty target and a cost boundary.
A detection count is not population abundance
Animals under a canopy, underground, underwater or outside the sampled area may be absent from the images. Decide whether the decision concerns occurrence, abundance, density, seasonal change or behaviour. Each requires a different sampling unit, revisit schedule and validation plan. A larger image collection does not automatically answer a better ecological question.
Measure the three benefits separately
Evaluate missed detections, personnel exposure and end-to-end cost as separate outcomes. Flying may reduce difficult ground access while leaving launch-site travel, recovery and emergency exposure. Processing can consume the field savings. Treat simultaneous gains in safety, accuracy and cost as a hypothesis to test against the current method, using the same ecological endpoint.
Specify the sample before collecting it
Stratify by meaningful habitat and select plots or transects using a documented random or systematic design. Opportunistic flights around detected animals can help exploration but should not silently replace the inference sample. Record inaccessible plots, weather, visibility and actual sampled area. Join the flight log to the sampling ledger so missing coverage remains visible.
Start with a paired pilot
Compare methods on matched plots and times, using independent observations or known tagged individuals where appropriate. Hodgson and colleagues used replica seabird colonies with known counts in their 2018 experiment. That is a strong counting benchmark, but not a guarantee for moving animals or obscured habitats. A pilot succeeds when it meets the required uncertainty and delivery cost, not merely when the aircraft completes its route.
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.