Groundbreaking new detail in population trends is a gamechanger for ecology

Groundbreaking work from the University of St Andrews in partnership with Cornell University has, for the first time, developed fine scale information on ecological population changes at a continental scale.

In a paper published today in Science, data shows recent bird population trends at 27 km by 27 km scales across north America, the smallest parcels of land ever attempted for an analysis across such a large geographic area.

Previously national and continental monitoring programs could estimate population trends only across entire ranges, regions, or states, but with advances in machine learning and the accumulation of vast amounts of data from citizen scientists, researchers can now look at how well species are doing in areas about the same size as the city of Bath.

The new information has revealed that birds are declining most severely in their traditional strongholds—the very places where they should be thriving. Eighty-three percent of the species they examined are losing a larger percentage of their population where they are most abundant.

The team based at Cornell Lab of Ornithology in the USA was led by Dr Alison Johnston, now a co-director at the Centre for Research into Ecological and Environmental Modelling at the University of St Andrews.

They analysed 36 million observations shared by citizen science birdwatchers alongside multiple environmental variables derived from high-resolution satellite imagery for 495 bird species across North America from 2007 to 2021. They set out to develop reliable information about where birds are increasing or decreasing across North America, but the patterns they uncovered were startling.

Dr Alison Johnson said: “We’re not just seeing small shifts happening, we’re documenting populations declining where they were once abundant. Locations that once provided ideal habitat and climate for these species are no longer suitable. I think this is indicative of more major shifts happening in nature.”

This news follows on the heels of other recent research that documented widespread losses of birds in North America. The 2025 U.S. State of the Birds report showed bird declines in almost every ecological biome in the nation, and a 2019 paper published in Science reported a cumulative population loss of nearly 3 billion birds in Canada and the US since 1970.

“The 2019 paper was telling us that we have an emergency, and now with this work we have the information needed to create an emergency response plan,” added Dr Johnston.

“One interesting finding is that for almost all species we found areas of population increases and decreases. This spatial variation in population trends has been previously invisible when looking at broader regional summaries.”

Areas where populations are increasing provide us with hope, said Johnston: “In areas where species are increasing, that shows us that those species have some potential for recovery, despite declines in their core areas.”

Knowing exactly where on the landscape declines are happening helps scientists start to identify the drivers of those declines and how to respond to them.

Professor Amanda Rodewald from the Cornell Lab of Ornithology who worked on the study said: “It’s this kind of small-scale information across broad geographies that has been lacking and it’s exactly what we need to make smart conservation decisions. These data products give us a new lens to detect and diagnose population declines and to respond to them in a way that’s strategic and precise and flexible. That’s a game changer for conservation.”

The study’s detailed mapping of population changes will help conservation organizations and policymakers better target their efforts to protect declining bird species, which according to the authors is sorely needed to help reverse the declining population trends.

The team used causal machine learning models and novel statistical methodologies that allowed them to estimate changes in populations with high spatial resolution while also accounting for biases that come from changes in how and where people watch birds.

To ensure the reliability of the data the team ran over half a million simulations stacking up more than six million hours of computing, which would take about 85 years to run on a standard laptop computer.