North American bird declines are greatest where species are most abundant
Data files
Mar 29, 2025 version files 5.41 MB
-
basemap.gpkg
5.05 MB
-
bcr-srd-lookup.parquet
321.24 KB
-
README.md
8.42 KB
-
trends-season-dates.csv
33.85 KB
Abstract
Efforts to address declines of North American birds have been constrained by limited availability of fine-scale information about population change. Using participatory science data from eBird, we estimated continental population change and relative abundance at 27 km resolution for 495 bird species from 2007-2021. Results reveal high and previously undetected spatial heterogeneity in trends; although 75% of species were declining, 97% of species showed separate areas of significantly increasing and decreasing populations. Populations tended to decline most steeply in strongholds where species were most abundant, yet they fared better where species were least abundant. These high-resolution trends improve our ability to understand population dynamics, prioritize recovery efforts, and guide conservation at a time when action is urgently needed.
This archive contains data for North American bird declines are greatest where species are most abundant (Johnston et al. 2025).
Abstract
Efforts to address declines of North American birds have been constrained by limited availability of fine-scale information about population change. This archive accompanies a paper that describes the use of data from the participatory science project eBird to estimate continental population change and relative abundance at 27 km resolution for 495 bird species from 2007-2021. The archived materials, divided across Dryad and Zenodo, are composed of all data and R code that are necessary to conduct the analyses and produce the figures from this paper. The README.md file for this archive, which is part of the Dryad component of the archive, describes each of data files contained within the archive.
Content
The full archive is split between Dryad and Zenodo; the Zenodo components of the data cannot be archived in Dryad because they inherit open-access non-CC-0 licenses that are not allowed in Dryad archives. Compter code to work with these datasets is available in a seperate Zenodo archive (https://doi.org/10.5281/zenodo.14883008). In order to use the combined data and code, one must download the Dryad and Zenodo components and place them in a data/ sub-directory of the folder containing the code. This README file first provides additional details about each of the included data files.
Datasets
These datasets are primarily tabular data in either CSV or Parquet format (Parquet files can be read into R using the arrow R package), and there is a single GeoPackage containing spatial data. The datasets and associated columns are as follows
trends-season-dates.csv: a list of all 495 species included in the study. This file is located in the Dryad component of the archived materials. The columns of the data table are:
species_code: 6-letter eBird species code.common_name: English common name.breeding_biome: breeding biome of the species.season: season for which the trend was estimated.start_date/end_date: dates defining the season boundaries inMM-DDformat.
ebird-trends_2007-2021.parquet: trends estimates for all species 27 km resolution for the 2007-2021 time period. In addition to the trends, a set of species-specific simulation-based metrics capturing the ability of the model to correctly classify trends is included. These metrics are estimated across a suite of simulation scenarios for each grid cell grouped into bins according to trend magnitude. This file is located in the Zenodo component of the archived materials. The columns of the data table are:
species_code: 6-letter eBird species code.season: season for which the trend was estimated.breeding_biome/breeding_biome_label: breeding biome of species.srd_id: unique integer ID for each 27 km grid cell.longitude/latitude: coordinates of 27 km grid cell centers.abd: relative abundance estimates for 2014, the midpoint of the 2007-2021 time period.abd_ppy_median: median percent per year trend in relative abundance. Note these are percentages expressed out of 100, e.g. a value of 5 expresses a 5% change in relative abundance per year.abd_ppy_{lower/lower}: the 10th and 90th percentile of percent per year trend in relative abundance.abd_ppy_{sd/var}: the variance and standard deviation of trend estimates based on a ensemble of 100 folds.distance_to_edge_km: distance between the grid cell and the range edge.directional_power: simulation based estimate of the proportion of trends estimates that are non-zero and correctly classify the trend direction.directional_error_nz: simulation based estimate of the proportion of non-zero trends estimates that incorrectly classify the trend direction.bin_{min/max}: trend magnitude bin boundaries over which the error and power metrics are estimated.
ebird-trends_abd-binned-performance_2021.parquet: simulation-based power and error estimates for each species, grouped by relative abundances binned using ten equally spaced quantiles. This file is located in the Zenodo component of the archived materials. The columns of the data table are:
species_code: 6-letter eBird species code.season: season for which the trend was estimated.abd_bin_midpoint: mid-point of the relative abundance bin.n_nonzero: number of non-zero trend estimates.prop_nonzero: proportion of trend estimates that are non-zero.directional_power: proportion of trends that are both non-zero and correctly classify the trend direction.directional_error: proportion of trends that are both non-zero and incorrectly classify the trend direction.directional_power_nz: proportion of non-zero trends that correctly classify the trend direction.directional_error_nz: proportion of non-zero trends that incorrectly classify the trend direction.
ebird-trends_ppy-binned-performance_2021.parquet: simulation-based power and error estimates for each species, grouped into half percent trend magnitude bins. This file is located in the Zenodo component of the archived materials. The columns of the data table are:
species_code: 6-letter eBird species code.season: season for which the trend was estimated.ppy_bin_midpoint: mid-point of the trend magnitude bins.n_nonzero: number of non-zero trend estimates.prop_nonzero: proportion of trend estimates that are non-zero.directional_power: proportion of trends that are both non-zero and correctly classify the trend direction.directional_error: proportion of trends that are both non-zero and incorrectly classify the trend direction.directional_power_nz: proportion of non-zero trends that correctly classify the trend direction.directional_error_nz: proportion of non-zero trends that incorrectly classify the trend direction.
ebird-trends_range-wide_folds_2021.parquet: range-wide trend estimates for each of the 100 folds that make up the ensemble. This file is located in the Zenodo component of the archived materials. The columns of the data table are:
species_code: 6-letter eBird species code.season: season for which the trend was estimated.breeding_biome: breeding biome of species.fold: integer indicating the fold number.n_cells: number of 27 km cells within the range.abd_ppy: abundance-weighted mean rangewide percent per year trend.prop_decline_range: proportion of 27 km cells with declining trends within range.
ebird-trends_simulations_focal-species_2021.parquet: simulation results for a subset of four species: Hermit Thrush, Hooded Warbler, Redhead, and Snow Goose. This file is located in the Zenodo component of the archived materials. The columns of the data table are:
species_code: 6-letter eBird species code.season: season for which the trend was estimated.scenario_id: unique integer ID for each simulation scenario (1-10).realization: each simulation scenario is repeated for 10 realization, this integer identifies the realization.srd_id: unique integer ID for each 27 km grid cell.simulated: simulated trend.estimated: median model estimated trend.lower: 10th percentile of model estimated trends.upper: 90th percentile of model estimated trends.nonzero: whether the confidence intervals overlap zero.
bcr-srd-lookup.parquet: lookup table identifying which Bird Conservation Region (BCR) each 27 km grid cell falls into. This file is located in the Dryad component of the archived materials. The columns of the data table are:
region_code: unique ID for each BCR.srd_id: unique integer ID for each 27 km grid cell.coverage_fraction: proprtion of the 27 km grid cell covered by the BCR.
basemap.gpkg: spatial data for creating basemaps in the mapping scripts. All data come from the Natural Earth project. This file is located in the Dryad component of the archived materials. This file is located in the Zenodo component of the archived materials. Layers included are:
land: land boundary polygon.country_lines: country boundary lines.state_lines: state boundary lines.slakes: lake polygons.global_bb: global boundary polygon.
The data files in this archive provide estimates of population change and relative abundance at 27 km resolution for 495 bird species from 2007-2021 using participatory science data from eBird. Additional data files are included in the archive that provide simulation-based model performance metrics. For further details, consult the associated paper North American bird declines are greatest where species are most abundant (Johnston et al. 2025).