Rainforest bird declines

Seventeen years of standardised bird monitoring showed the upland rainforest birds of the Australian Wet Tropics nearly halving. That evidence supported protection nominations for 14 species under national and international frameworks, and it went live as an interactive app that anyone can query.

The record covers 1,977 surveys at 114 sites, from sea level to 1,500 m. Trend models for 42 species, adjusted for survey effort, quantified the change from 2000 to 2016. Most mid and high elevation species lost more than 40% of their numbers at the lower edge of their range. Upland specialists and species found nowhere else lost almost 50%.

Links

• Paper page: Long-term changes in populations of rainforest birds
• Journal: Plos One
• Live app: Bird population trends explorer

Architecture

The pipeline harmonises counts from several sources into one time series per species and site. Generalised linear models, which relate counts to explanatory variables, estimate each species’ trend with terms for survey effort and habitat change. The effort adjustment matters most: a decline has to be a decline, not a change in who went looking.

A Shiny app publishes the trends interactively, so managers and policymakers can explore each trajectory, its confidence interval and the nomination thresholds. The whole analysis runs in R under version control.

The decision that was hard

A nomination document needs one defensible number per species, and monitoring data offer many candidate analyses. I modelled each species separately, with explicit adjustment for covariates, instead of pooling everything into a community index. A pooled index reads impressively and hides which species is collapsing. Per-species models are noisier, and they name the species a nomination must name.

What was measured

Most mid and high elevation species lost more than 40% of local abundance at their lower edges. Lowland species expanded uphill, increasing by up to 190% in higher areas. Upland specialists and regional endemics declined by almost 50%.

Fourteen species carried enough evidence to support nominations for heightened protection. The nominations ran under national threatened species lists and through the International Union for Conservation of Nature (IUCN).

Role

I designed and ran the analytical workflow, harmonised the time series, and built and deployed the Shiny app. I drafted the manuscript and coordinated the policy communication.

What this taught me about evaluation

An analysis that feeds a decision must survive hostile review, so every adjustment has to be visible and defensible. I hold engineering evaluations to the same bar: publish the method beside the metric. The Shiny app was the other lesson. Results people can interrogate earn more trust than results people must accept.