This repository contains code and data for North American bird declines are greatest where species are most abundant (Johnston et al. 2025).
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. This README.md describes each of the R scripts and data files contained within the archive.
To install all necessary packages to run the R scripts in the code/ directory, run the following line of code in R:
install.packages(readLines("requirements.txt"))
All R scripts in code/ assume that the working directory is the top-level project directory, i.e. the one containing this README file. If you are using RStudio, open the .Rproj file and your working directory will be set correctly. If you're not using RStudio, you can set your working directory manually, e.g. by running the following code in your R session
setwd("/Users/Alison/Documents/REPOS/2024-trends-abundance-paper/")
The R scripts included in the code/ directory of this repository reproduce the figures and results from the paper.
figure-01_species-maps.R: generates trend maps for Great Blue Heron, Wood Duck, and House Wren aggregated to three different spatial scales: range-wide, regional (Bird Conservation Region), and landscape scales (27 km × 27 km grid cells).figure-02_fishbone.R: generates a bar plot showing the minimum, maximum, median, and interquartile range of 27 km trend estimates for each of the 495 species in the study.bam_linear_analysis.R: This runs the linear models for each species relating trend to abundance. It uses thebamfunction in mgcv to include a Gaussian Process to account for spatial autocorrelation. The code in here automatically searches through candidate scales of the spatial correlation and selects an optimal one to use for a final model. See the paper for further details. This script generates a set of CSVs that are required by several of the subsequent scripts, therefore, it should be run prior to running the scripts used to generate Figures 3 and S14-S16.figure-03_bam-abd.R: this script visualizes the results of the BAM analysis frombam_linear_analysis.R, including a histogram of the distribution of slopes of the trend vs. abundance relationship, estimated linear relationships between trend and abundance for each species grouped by breeding biome, and example scatterplots of trend vs. abundance species with negative and positive relationships. Depends onbam_linear_analysis.R.figure-04_community-maps.R: generates community-level trend maps showing the mean trend across species breeding in six different biomes.figure-S01_variation-within-biome.R: generates a histogram showing the distribution of the range of 27 km trend estimates within each BCR-species combination.figure-S02_trend-vs-prop-range-dec.R: calculates range-wide trends and proportion of 27 km grid cells that show a declining trend, then generates a boxplot showing the distribution of the proportion of declining cells binned by range-wide trend.figure-S03-S04_density-prop-range-dec.R: generates density plots showing the distribution of range-wide trends and the distribution of the proportion of cells with declining trends. Density plots are grouped by biome and (in Figure S4) split by breeding and non-breeding season trends.figure-S09-S10_power-error.R: generate plots showing the power and error versus trend magnitude (S9) and relative abundance (S10) for a set of 4 example species Hermit Thrush, Hooded Warbler, Redhead, and Snow Goose.figure-S11_power-error_all-species.R: generates plots showing the power and error versus both the trend magnitude and relative abundance for all species.figure-S12_power-error-histogram.R: generates a histogram showing, for different levels of trend magnitude, what proportion of non-zero 27 km trends have directions that are correctly and incorrectly classified, respectively.figure-S13_rangewide-error.R: generates a histogram showing the distribution of range-wide error across species.figure-S14-S15_bam-edge.R: generate plots showing the relationship between the trend estimates in 27 km cells and the distance of those cells to the species' range edge: a histogram of the slopes of the relationship for each species and the linear relationships by species grouped by biome. Depends onbam_linear_analysis.R.figure-S16_compare-edge-abd.R: based on the results of the BAM analysis, generates distributions of species-specific effect sizes for (a) distance to range edge and (b) relative abundance when both variables were included in the same model. Depends onbam_linear_analysis.R.numbers-for-text.R: produces a variety of numbers that appear in the text of the paper.rangewide-trend-uncertainty.R: estimate rangewide uncertainty in trends.
The datasets required to run the above scripts are archived in Dryad and Zenodo at the following URLS:
All data files from these two archives should be downloaded and moved to a data/ sub-directory within the project directory (where "project directory" is the directory containing this README). They 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.lakes: lake polygons.global_bb: global boundary polygon.