Which eBird data were used to generate the Status and Trends Products?
What environmental data were used for the Status and Trend products?
How are seasons defined for each species? Why are there gaps between seasons?
Why are pre-breeding and post-breeding migrations sometimes separated?
What is the difference between “modeled area” and “no prediction”?
Why are models available only for the Western Hemisphere?
Why are some islands areas of “No Prediction”
What are Bird Conservation Regions and how have they been defined?
Some of the maps have errors? Why does this happen?
Why do some products show as “Unavailable” or cannot be clicked on?
Can I download the data?
How can I find out more about the data analysis used to generate these products?
Why does it look like some species disappear during some parts of the year (e.g., shorebirds)?
How do we ensure that the eBird data are accurate?
What happened to the eBird Status and Trends data products for the 2016 calendar year?
How do the maps match what we know about each species’ biology?
The yearly update to the Status products has been completed, representing over 800 species, including many, new global examples. Estimates are for the year 2019. Trend maps are in development.
The eBird Status and Trend products provide basic ecological information for over 800 species globally, describing their ranges, abundances, environmental associations, and population trends. To generate these products, we use statistical and machine learning analyses designed to combine eBird data with a range of environmental data. The result is a prediction of the occurrence and abundance of species across the Western hemisphere at weekly intervals. These predictions are the building block of the Status and Trends products and are summarized in several ways to produce the different products.
There are three, currently available, major products for eBird Status and Trends:
Different products require different volumes of data and the most data-intensive products such as trend maps are only available for some of the species and seasons. We only present products for species and seasons that have passed analytical and expert quality review tests.
Yes, any eBirder can contribute! These analyses were only possible thanks to the eBird submissions of hundreds of thousands of eBird users. Our ability to update and improve the Status and Trends products in the future continues to depend on the contributions of eBirders like you!
If you submitted checklists that meet all the requirements answered under “Which eBird data were used to generate the Status and Trends Products?“, then you have already contributed data to the Status and Trends! Any future checklists you submit that meet these requirements will automatically be included in analyses for the updated Status and Trends Products.
Remember, any observation is useful, whether it is from today or your field notebooks from 15 years ago. Whether it is from a hotspot with amazing birds, or a place with few species – all checklists are valuable. To ensure your eBirding checklists are most useful to scientific efforts like this, you can make your checklists:
Relative abundance is the count of individuals of a given species detected by an expert eBirder at the optimal time of day, while expending the effort necessary to maximize detection of the species. Relative abundance predictions have been optimized for search effort and user skill, specific for the given region, season, and species, in order to maximize detection rates.
For each species, relative abundance was estimated for all 52 weeks of the year across a regular spatial grid with a density of one location per 2.96 km × 2.96 km grid cell across the Western Hemisphere. Estimates at each location and date were made based on the local habitat, elevation, and topography at that location.
Because detecting birds in the environment can be difficult, we know that there are always some individual birds that are missed by eBirders. For this reason, we refer to the quantity estimated as a relative measure of abundance. Although the relative abundance estimates will underestimate the true abundance, they do provide a standardized index that can be used to compare abundance in different regions. For example, if relative abundance is 10 in one area and 5 in another area, then we would estimate abundance is twice as high in the first area, even if we’re not sure of the actual number of individuals in the area.
For each species and season, we summed the relative abundance estimates across the selected region and then divided it by the sum of the relative abundance estimates across the entire seasonal range. The result is presented as a percentage.
This will be a reasonable estimate if the whole population is within the “modeled area.” For this reason, the reporting of this value was contingent on the majority of a species’ known seasonal range being within the modeled area. If the percentage of total population value is missing, but the season was mapped in other products, it is because an expert reviewer believed that a significant portion of the species’ seasonal range was not included, but that the distribution within the modeled area was correct.
Note that since the state/province and Bird Conservation Regions (BCR) boundaries do not have offshore coverage, species with significant abundance offshore (e.g., Brown Pelican) do not include the offshore portion of their population in these calculations.
Species’ ranges were defined as the areas where the species is expected to occur on at least ten percent of the predicted hypothetical checklists in a given week. This is equivalent to a single skilled eBirder starting at the optimal time of day, expending the effort necessary to maximize detection of the species for each day of the week, and detecting the species on at least ten percent of the checklists within a week. Each species’ range was estimated for all 52 weeks of the year at 2.96 km × 2.96 km grid cell locations across the Western Hemisphere. To create easy-to-read range boundaries, the 2.96 km grid data were aggregated to an 8.89 km grid and spatially smoothed. Both the smoothed, aggregated seasonal boundaries and the raw 2.96 km grid cell boundaries are available for download.
Percentage of region occupied is calculated as the percent of the selected region that is covered by the range of the species. Note that since the state/province and Bird Conservation Regions (BCR) boundaries do not have offshore coverage, species with significant water-based abundance values (e.g., Brown Pelican) do not include the offshore portion of their population in these calculations.
Percentage of range in a region is calculated as the fraction of a species’ total North American range that falls within the selected region.
This will be a reasonable estimate if the whole population is within the “modeled area”. For this reason, the reporting of this value was contingent on the majority of a species’ known seasonal range being within the modeled area. If the percentage of total population value is missing, but the season was mapped in other products, it is because an expert reviewer believed that a significant portion of the species’ seasonal range was not included, but that the distribution within the modeled area was correct.
Note that since the state/province and BCR boundaries do not have offshore coverage, species with significant abundance offshore (e.g., Brown Pelican) do not include the offshore portion of their population in these calculations.
Days of occupation in a region is the number of days that a species is present in the selected region. A species is defined to be present in a region when at least 5% of the region was within the species range during the given season.
To assess and communicate the quality of the model estimates, we evaluated several metrics that describe the model’s ability to predict the observed patterns of species occupancy and abundance. These metrics are made available primarily for interested researchers.
To quantify the performance of the range estimates we used the Area Under the Curve (AUC) and Kappa statistics to describe the models’ ability to correctly classify occupied and unoccupied sites. AUC measures a model’s ability to discriminate between locations where species are and are not likely to occur. Technically, it is the probability that the model will rank a randomly chosen positive observation (species detected) higher than a randomly chosen negative one (species not detected). Cohen’s Kappa statistic was designed to measure the same metric, but taking into account the background prevalence. To quantify the quality of the relative occupancy predictions we also evaluated AUC and Kappa (shown in the second row). To quantify the quality of the abundance estimates we computed Spearman’s Rank Correlation (SRC) and the percent Poisson Deviance Explained (P-DE) (Shown in the third row). SRC measures how well the abundance estimates rank the observed abundances and the P-DE measures the correspondence between the magnitude of the estimated counts and observed counts. More details about predictive performance metrics and how they were calculated and interpreted can be found in Fink et al. (2019).
Checklists used in Status and Trends products must meet the following conditions to be included in analyses.
See the eBird Help Center for more information about eBird checklists.
The analyses used to produce the Status and Trend products rely on matching bird observations with characteristics of the local environment. We used data on elevation, topography, and habitat to describe the local landscapes where eBirders searched for birds. Each checklist location is matched to the environmental data within approximately a 1.5 km radius around the location. For elevation and bathymetry, we calculated the mean and standard deviation of each within the checklist radius using the Global Bathymetry and Elevation Digital Elevation Model. For topography the aspect and slope within the checklist radius were calculated as the mean and the standard deviation (Amatulli et al. 2018). Habitat was described using the MCD12Q1 dataset from NASA and the FAO-Land Cover Classification System, which includes classes for land cover, land use, and hydrology by year. Water cover was described by the Aster Global Water Bodies Dataset, covering 2000-2013 at 30 meter spatial resolution, and with ocean, river, and freshwater categories (NASA/METI/AIST/Japan Spacesystems 2019). Islands were identified using the global shoreline dataset (Sayre et al. 2019). To describe tidal mudflats, we used a high resolution intertidal change dataset (Murray et al. 2019). The land cover, water cover, and intertidal datasets were summarized as the percentage of land cover and edge density within the 1.5 km radius of each checklist. We used the NOAA VIIRS Nighttime Lights composite for 2016 with the cloud mask, outliers removed, and non-lights set to zero (vcm-orm-ntl). The nighttime lights values were calculated as the mean and standard deviation within the checklist radius. Finally, we summarised road density for five types using the GLOBIO Global Roads Inventory Project (GRIP) (Meijer et al. 2018).
Breeding and non-breeding season dates are defined for each species as the weeks when the species’ population does not move. For this reason, these seasons are also described as stationary periods. The dates were defined by experts in the status and distribution of Western Hemisphere birds based on the weekly abundance maps. The selected dates were then checked to make sure that they generally matched expected patterns of phenology for the species.
Migration periods are defined as the periods of movement between the stationary non-breeding and breeding seasons. Note that for many species these migratory periods include not only movement from breeding grounds to non-breeding grounds, but also post-breeding dispersal, molt migration, and other movements. For some species, the transition between stationary and migratory seasons is not clear. Both breeding and nonbreeding ranges are often represented within the migratory seasons since some individuals will have arrived in those areas while other individuals of the species are still migrating. In these cases transitional weeks were excluded to provide the clearest picture of individual seasons. For some species, this resulted in seasons that appear shorter than expected, especially when considered within specific regions.
Season dates are defined specifically to be used with eBird Status and Trends products. These dates should not in general be used to delineate the migration and breeding phenology of species, although in many cases Status and Trends dates may approximate these phenological dates. In addition, the dates used for Status and Trends are distinct from the corresponding seasonal dates defined in Birds of the World.
Some species have pre-breeding and post-breeding migration seasons combined into a single migratory season. These species (e.g., Magnolia Warbler, Black-throated Gray Warbler) use fairly similar areas for both their migrations. However, some species such as Rufous Hummingbird use different paths for their two migrations. For these species we split the map to show pre-breeding migration (green) and post-breeding migration (yellow) separately. If at least 40% of the area used for one migration season is not covered by the other migration season, then we show them as distinct colors.
On the relative abundance and range maps the light gray shows the “modeled area,” where there was sufficient data to run a model, but the species was predicted to be absent. Sufficient data required there to be, on average, at least 0.1% spatial coverage of 3 kilometer grid cells within the region for a given week. The dark gray refers to areas of “no predictions” where there was insufficient data to assess whether the species was present or absent. That is, there was less than 0.1% spatial coverage of 3 kilometer grid cells within the region for a given week.
The Status and Trend products were mostly limited to the Western hemisphere because this region currently has the highest density of eBird data. In 2021, we plan to model all species globally.
Some islands have insufficient data to predict whether a species is present or absent (see above in What is the difference between “modeled area” and “no prediction”?). In eBird Status and Trends, we use an island dataset (Sayre et al. 2019) that enables us to distinguish between islands with and without a particular species. The benefit of this method is that the models can, in effect, distinguish the geographic barriers relevant to islands, constraining species to or excluding species from specific islands. The distributions in the Caribbean for both White-winged Dove and Mourning Dove are good examples of this method in action. However, a consequence is that for species that show variation between islands, each island now needs more information before we can make predictions to it. As a result, many islands that are not frequently eBirded, such as a number of the islands in British Columbia, are often represented as “No Prediction.”
Bird Conservation Regions (BCRs) are regions with similar bird communities, environmental conditions and resource management issues (Sauer et al., 2003). We used the established Bird Conservations Regions for the United States and Canada in accord with the NABCI Bird Conservation Regions, several of which span the border into Mexico. Because many of the Mexican Bird Conservation regions are too small to support the spatial scale of the Status and Trend analysis, we aggregated the Mexican Bird Conservation regions into larger regions in consultation with partners at CONABIO. Since Bird Conservation regions have not been defined in Central America, South America, and the Caribbean we defined regions based on countries, aggregating the smallest countries or groups of countries into units that were large enough to support the spatial scale of our analysis. See the map below for the BCRs that we use, including the customized versions for Mexico, Central America, South America, and the Caribbean.
Note that summaries of the eBird Status and Trends data products can be computed across user-defined regions by downloading the data and using the ebirdst R package.
Like any predictive models, the Status and Trend models make errors when predicting species ranges and abundance. When predicting ranges there are two types of errors: predicting species absence in areas that are actually occupied and predicting species presence in areas that are actually unoccupied. There can also be errors in the estimates of relative abundance, with estimates that are higher or lower than the actual counts.
The predictive models used to generate the Status and Trend products account for gaps in eBird data by sharing information from nearby areas. This works well when (1) there is sufficient eBird data to capture patterns of species’ occurrence and abundance and (2) when the environmental data together with the other predictors used in the models do a reasonably good job describing the ecological characteristics that are important to the species.
Error rates generally increase in regions when one or both of the above conditions is not met. First, in regions where the density of checklists is low (e.g. central Canada and the Amazon basin) there is little information to learn patterns of species’ occurrence and abundance. In these areas, incorrect extrapolation can be a risk. For example, species that occur on the coast of Greenland, such as Iceland Gull, are often extrapolated inland, where they likely do not occur, but where there is no data to inform their absence.
Second, error rates also tend to be higher when species’ detection rates are low. Even if there are many checklists in a region, having very few detections of a species limits the amount of information available to characterize the environment that the species is associated with. Third, error rates increase in regions where the environmental data fail to describe important ecological features for each species.
See the link What are the “Predictive performance metrics”? for more information about how we assess the quality of model estimates.
Species products may be missing for various reasons:
Data products for the most recent results (estimates for 2019 released in December 2020) are not available, but will be made available in the future. Currently, the previous release (estimates for 2018) is available for download, as are the most recent range boundaries:
ebirdst_download()function, which will tell you where the data has been downloaded. From there, you can use the GeoTIFF files in any software you like. Alternatively, you can download directly from the Amazon S3 bucket (arn:aws:s3:::ebirdst-data). There is a CSV file in the root of the bucket that describes all the species and run names. Accessing in this way requires either the AWS CLI (e.g.,
aws s3 cp s3://ebirdst-data/abetow-ERD2018-EBIRD_SCIENCE-20191109-a5cf4cb2/results/tifs/abetow-ERD2018-EBIRD_SCIENCE-20191109-a5cf4cb2_hr_2018_abundance_median.tif .) or direct access to the URL endpoint (e.g., https://ebirdst-data.s3-us-west-2.amazonaws.com/abetow-ERD2018-EBIRD_SCIENCE-20191109-a5cf4cb2/results/tifs/abetow-ERD2018-EBIRD_SCIENCE-20191109-a5cf4cb2_hr_2018_abundance_median.tif).
All downloads are available through or on the Download page. All downloadable products can be used for research, presentations, or display on webpages provided they are for non-commercial purposes and properly attributed; see our Recommend Citation below or here.
Recommended citation: Fink, D., T. Auer, A. Johnston, M. Strimas-Mackey, O. Robinson, S. Ligocki, W. Hochachka, C. Wood, I. Davies, M. Iliff, L. Seitz. 2020. eBird Status and Trends, Data Version: 2019; Released: 2020. Cornell Lab of Ornithology, Ithaca, New York. https://doi.org/10.2173/ebirdst.2019
To generate the eBird Status and Trend products, we use statistical and machine learning models that combine the bird observation data from eBird with environmental data from NASA. The analytical process is designed to correct for biases and gaps within the eBird data. You can read more about the methods in a number of scientific publications. Fink et al. 2019 provides an overview of analytical methodology used for the Status and Trend products. More details about the modeling strategy used to estimate species’ relative abundance is described in Johnston et al. 2015. Details about the methods used to scale the relative abundance model across large spatial and temporal extents can be found in Fink et al. 2010 and Fink et al. 2013. For a more comprehensive description of the eBird database from a scientific perspective see Sullivan et al. 2014.
For some species, it may appear as though they are not where you expect them to be, when you expect them to be there. For example, we know Whimbrel occurs along the coast of Chile during the winter, however this may be very difficult to see on the animated maps. For a species that sticks mainly to coasts, its range for much of the year will only be about one pixel wide. The resolution of each pixel is 2.96 x 2.96 km. So, when the map extent spans much of the Western Hemisphere, these very small pixels can be difficult to see.
Using the ebirdst R package, the data you see on these maps may be downloaded and plotted within your area of interest. This makes the pixels appear much clearer when zoomed in as can be seen in the maps shown below. Note that for the range maps, the 2.96 km grid data were aggregated to an 8.89 km grid and spatially smoothed, making the range for these species easier to see.
There are more than 5,000 automated filters that are active during the submission process for every checklist submitted in eBird. The common filters that may be triggered are for species that are rare for a region and/or season and abnormally high counts of a species for a region and/or season. eBird will ask that additional information (description of the bird(s), counting process, pictures, sound recordings, etc.) be provided for these observations. There are also more than 2,000 eBird reviewers worldwide that examine these checklists as they are submitted. These reviewers work to make sure that each rare sighting is validated before the data from each checklist is available to be used in any analyses.
The 2018 data products are still available via the ebirdst R package as they have been for the past year and will continue to be available into the future.
The 2016 data products, and the ebirdst R package developed to work with them, are still available. The Installation section of the package README describes how to install a previous version of the package, which maintains functionality for working with older versions, including access for downloading them. The release notes describe which calendar year of predictions correspond to which package version.
Read this if you are thinking of comparing estimates across years: there have been a number of important changes between yearly estimates. These changes were made to improve the quality and scope of the data products, taking advantage of increased data volume and quality as well as methodological improvements. These changes were not designed to facilitate cross-year comparisons. For this reason, we do not suggest using individual yearly estimates for direct comparisons.
In addition to the quantitative assessment of the data products based on the predictive performance metrics, we want to know, qualitatively, how well the abundance and range information match up with each species’ known biology. For this, each species map is reviewed, week-by-week, by species distribution experts. Individual seasons exhibiting serious inaccuracies are rejected. Species distribution experts evaluate the information over the entire range and if most of it is accurate, the species and season combo is retained. Small regions of false positives or false negatives are acceptable (and highlight areas where more data is needed). Any season (for a single season or all seasons for a single species) that is rejected by an expert reviewer is excluded from all Status and Trends data products, including the visualizations online and the downloadable data. This way, we know that the maps online and in the downloadable data are not only accurate according to the predictive performance metrics described above, but also accurate in terms of species biology according to expert reviewers.
Amatulli, G., Domisch, S., Tuanmu, M.N., Parmentier, B., Ranipeta, A., Malczyk, J. and Jetz, W. 2018. A suite of global, cross-scale topographic variables for environmental and biodiversity modeling. Scientific data, 5, p.180040.
Fink, D., Auer, T., Johnston, A., Ruiz-Gutierrez, V., Hochachka, W.M., Kelling, S. (2019) Modeling avian full annual cycle distribution and population trends with citizen science data. Ecological Applications. 00( 00):e02056. 10.1002/eap.2056
Fink, D., Damoulas, T., & Dave, J. (2013, July). Adaptive Spatio-Temporal Exploratory Models: Hemisphere-wide species distributions from massively crowdsourced eBird data. In AAAI.
Fink, D., Hochachka, W. M., Zuckerberg, B., Winkler, D. W., Shaby, B., Munson, M. A., … & Kelling, S. (2010). Spatiotemporal exploratory models for broad‐scale survey data. Ecological Applications, 20(8), 2131-2147.
Friedl, M.A., D. Sulla-Menashe, B. Tan, A. Schneider, N. Ramankutty, A. Sibley and X. Huang. 2010. MODIS Collection 5 global land cover: Algorithm refinements and characterization of new datasets, 2001-2012, Collection 5.1 IGBP Land Cover, Boston University, Boston, MA, USA.
Hansen, M. C., DeFries, R. S., Townshend, J. R. G., & Sohlberg, R. 2000. Global land cover classification at the 1km spatial resolution using a classification tree approach. International Journal of Remote Sensing 21:1331-1364.
IUCN. 2001. IUCN Red List categories and criteria: Version 3.1. Prepared by IUCN Species Survival Commission. World Conservation Union, Gland, Switzerland and Cambridge, United Kingdom. Ii + 30 pp.
Johnston, A., Fink, D., Reynolds, M. D., Hochachka, W. M., Sullivan, B. L., Bruns, N. E., … & Kelling, S. (2015). Abundance models improve spatial and temporal prioritization of conservation resources. Ecological Applications, 25(7), 1749-1756.
Meijer, J.R., Huijbegts, M.A.J., Schotten, C.G.J. and Schipper, A.M. (2018): Global patterns of current and future road infrastructure. Environmental Research Letters, 13-064006. Data is available at www.globio.info
Murray N. J., Phinn S. R., DeWitt M., Ferrari R., Johnston R., Lyons M. B., Clinton N., Thau D. & Fuller R. A. (2019) The global distribution and trajectory of tidal flats. Nature. 565:222-225. http://dx.doi.org/10.1038/s41586-018-0805-8
NASA/METI/AIST/Japan Spacesystems, and U.S./Japan ASTER Science Team. ASTER Global Water Bodies Database V001. 2019, distributed by NASA EOSDIS Land Processes DAAC, https://doi.org/10.5067/ASTER/ASTWBD.001. Accessed 2020-12-14.
Sauer, J.R., Fallon, J.E. & Johnson, R. 2003. Use of North American Breeding Bird Survey data to estimate population change for Bird Conservation Regions. The Journal of Wildlife Management 67:372–389.
Sauer, J. R., D. K. Niven, J. E. Hines, D. J. Ziolkowski, Jr, K. L. Pardieck, J. E. Fallon, and W. A. Link. 2017. The North American Breeding Bird Survey, Results and Analysis 1966 – 2015. Version 2.07.2017 USGS Patuxent Wildlife Research Center, Laurel, MD
Sayre R. et al. (2019) A new 30 meter resolution global shoreline vector and associated global islands database for the development of standardized ecological coastal units, Journal of Operational Oceanography, S47-S56, DOI: 10.1080/1755876X.2018.1529714
Sullivan, B. L., Aycrigg, J. L., Barry, J. H., Bonney, R. E., Bruns, N., Cooper, C. B., … & Fink, D. (2014). The eBird enterprise: an integrated approach to development and application of citizen science. Biological Conservation, 169, 31-40.