Bird monitoring in Martinique
Are there fewer birds, or just fewer people counting them?
Summary
Twelve years of bird counts from 65 survey routes show no clear long-term trend in abundance, but large changes in how much counting was done. Separating real ecological change from changes in observer effort was the main challenge.
The 65 survey routes, 10 points each. The north and center of the island are barely sampled.
Key results
- 114,495bird observations, 2014–2025
- 13–15observers per year in 2020–2023, down from about 25
- p = 0.43no significant long-term trend in species counts
The problem
A long-term monitoring program counts birds every spring along 65 routes in Martinique, each with 10 listening points. The goal was to measure how biodiversity changed between 2014 and 2025, with honest uncertainty, and to explain species-level patterns. The difficulty: the data mixes real changes in bird populations with changes in how, where and by how many people the counting was done.
What we did
- Audited data quality across three Excel sheets: 123 incomplete visits (not all 10 points counted in one go), species recorded only once, missing weather values, and extreme counts of up to 200 birds in one observation.
- Measured observer effort: number of observers per year, visits per month and habitat, and spatial coverage. Observers dropped from about 25 to 13–15 per year in 2020–2023, most likely because of COVID, and the north and center of the island are barely sampled.
- Built three biodiversity indicators: the median number of birds per complete visit, the median number of sightings per species per year, and the mean number of birds per visit for each habitat.
- Because the counts are heavily right-skewed, used medians and bootstrapped 95% confidence intervals (5,000 resamples) instead of assuming normal distributions.
- Tested for long-term trends with linear regression, overall and per habitat.
- Followed the four most frequently seen migratory species month by month.
Results
No significant long-term trend. The slope for species sightings per year is 0.41 with p = 0.43. Mean birds per visit stays at about 2 to 2.5 in every habitat. Only the peri-urban habitat shows a significant increase (p = 0.0002), but it has few visits, so this should be read with caution.
Effort explains a lot. The dip in observers in 2020–2023 and the rise in incomplete visits in 2025 coincide with changes in the indicators, so part of what looks like ecological change may come from how the counting was done.
Migratory species are stable from year to year, but decline every season from April to June. This could be migration, but observation effort also drops in June, so the two cannot be fully separated.
What I learned
- Check how the data was collected before trusting a trend. A drop in observers or incomplete visits can look like a drop in birds.
- With skewed data, choose the statistic carefully. Medians and bootstrapped intervals gave a much more honest picture than means.
- “No significant trend” is a result, not a failure. Being clear about what the data cannot show is part of the analysis.
Recommendations
- Standardize the counting protocol and record effort (observers, duration, area) for every visit.
- Add automatic data checks to catch incomplete visits and missing values.
- Combine several indicators (abundance, species richness, occupancy) instead of relying on one.