U.S. State-Level eBird Data Summaries

Welcome everybody and thank you all for your interest in our project to help support information needs of state and federal agencies.

A little bit of background in who we are, I am Viviana, I’m an assistant director of the center for Avian population studies and the lead of the Conservation Science Program.

You will also see on this website a video presentation from Orin who is a research associate or in conservation science program but he’s also a researcher on the Status and Trends team that you will learn about.

You will also see work in videos that are the result of the hard work that has been done by Andrew a postdoc in our program and also by our student interns Tristan and Archie.

Believe it or not eBird turns 20 this year and along with all of these data we have really focused our efforts and our program for the past five years and working with partners to make sure that all of this scientific information is accessible for informing conservation and management actions included in these data are large volumes of data for each state in the US for example in New York alone we have about 2.9 million checklists from over 54 000 eBirders for the past 15 years the eBird status and Trends project team has worked on statistical advancements and innovations and data visualizations to generate information for all species in the U.S. and beyond this has resulted in unprecedented information on bird populations such as this animation relative abundance of the Yellow Warbler throughout its full annual cycle.

We now have information thanks to all these efforts we now have information on the relative abundance at high spatial and temporary resolutions for almost 2,000 species worldwide including all 450 species in North America that you can find on our website.

Most recently we have generated information on species trends population trends at a 27 kilometer resolution which can also be summarized by state by BCR or across the entire range sir conservation science program is focused on making sure that partners and stakeholders understand and have access to this information to best serve conservation and management actions just one example of a recent collaboration how we use eBird status and Trends projects is a recent collaboration with the fish and wildlife service on using eBird relative abundance on Bald Eagle populations we were able to use kind of Max information a Max annual relative abundance which means that the layer that you see on the top left is the relative abundance of bald eagles across the entire full annual cycle summarize in one layer we use this information to define low risk Collision areas in collaboration with the fish and wildlife service for bald eagles to help facilitate the permitting process for wind energy development for this specific project we have used information eBird relative abundance and trends to create an online platform to make state-level kneeper data summaries available such as maximum year-round abundance that you just saw or the percentage of populations in the breeding season we’ve also developed a workflow to enable annual debate updates of all information we hope that this information is used by state and Wildlife agencies to support their multiple information needs such as potential updates to swap plans in preparation for initiatives such as the hopefully potential passing of recovering America’s Wildlife Act we’d like to thank you again for your interest and we hope you find this information useful.

End of Transcript

I’m Orin Robinson a research associate in the Conservation Science program within the Center for Avian Population Studies or “CAPS” at the Cornell Lab of Ornithology.

I’m going to talk about the eBird data that goes into the Status and Trends models and how we use various filters within eBird and then statistical methods to account for the biases in citizen science data while we are creating these status and trends products.

So the first line of defense, when it comes to data quality in eBird and the eBird products are several thousand filters that are working in the background as you are putting data into eBird.

If we look at the right-hand side of the screen there’s an example of one of these filters where I’ve put in 15,000 Green-winged Teal on Cayuga Lake here in Ithaca, New York.

Now we often see Green-winged Teal, it’s not rare to see a Green-winged Teal on Cayuga Lake, but it would be an incredibly rare event to see 15,000 of them and you can see that this blue box has popped up that was triggered by one of these thousands of filters that are running that suggests that this count is higher than expected for a given location and a given date

It asks for you to provide some comments to show how you got to a count of 15,000 Green-winged Teal..

In the middle panel here I have shown that there was a Snowy Owl here in Ithaca, New York and again another one of these these boxes has been triggered. It shows that you have found a bird that is not on the local eBird checklist and again it asks for you to provide some comments to support your making that identification, whether it be describing field marks, describing the behavior of the bird, what it was doing when you saw it, or the specific habitat in which you saw the bird. For both of these pictures, you would need to add your comments to the checklist to further validate your observation.

There are 5,000 or more of these internal filters that are running. And then after that there are more than 2,000 human reviewers across the world and more than 20% of these checklists are reviewed by a human person. If a checklist gets flagged it will almost certainly be reviewed by one of these humans and they will work with the eBird observer to validate that observation.

So now that we have some data that has been through those filters and has been validated we can download that data and work with it. But we also have to handle all the biases that are inherent in big data, especially citizen science data. There are some issues that we need to be cognizant of when using citizen science data that may not be issues in well-structured surveys.

First, there are people who have an affinity for certain species or groups of species. This leads to a preferential recording of those species. We also know there’s a large variation in the observation process, particularly in effort. Effort is how long you spend birding, how far you travel while you’re doing it, the time of day, the number of observers that are there. We know that all of these things affect the detectability of certain species or of all species.

We also know that most folks are going to sample close to home or close to roads, areas of high biodiversity, highly accessible places. This creates a spatial bias in the data.

There are also temporal biases in this data, and there are there are a couple of different temporal biases that we see. One is that most folks are going to sample when they’re able to, and for most people that is on the weekends or around holidays. So we see more data in those time periods on weekends and around holidays.

We also tend to see more data, especially in North America, we see more data during the month of May. We get more eBird checklists in May than in any other month, and that is for two reasons. One, it’s starting to warm up the weather’s very nice so people just want to be outside, but also that’s when the neo-tropical migrants are passing through a large portion of North America. So that is a second kind of temporal bias.

We also know that there are wide-ranging skill levels within citizen science data and a wide range of behaviors of those observers. We have to account for all of these things when we model this data, even after all the filtering we discussed.

So the first step in quieting some of that noise is to put restrictions on the data that we analyze so essentially here we’re imposing a protocol on the eBird data. There’s a really good R package called “auk” that makes it easy to do this. Here you can see a screenshot from that package “auk” where we have filtered the time that the birders spent on a checklist, the distance they traveled, the years over which we want to conduct our analyzes, the number of observers. There are many other things you can filter for here and you know you would really just need to filter for what made sense for your analyzes. This five hour five kilometers is not set in stone — this is just an example.

We also discussed the spatial bias in the data. Remember that this comes from sampling close to home, easily accessible areas, places where you know a lot of birds are going to be. We address this by spatially sub-sampling the data.

So the example here: each dot represents an eBird checklist, and what we’ve done is put a grid over all of these eBird checklists. And from within each grid cell we select one checklist and run the analyzes on those checklists. In our analyzes we will do this hundreds and hundreds of times selecting different checklists at each iteration.

You can also begin to handle temporal bias here by doing this spatial sub sampling over the period of a week, so you would select one checklist per grid cell per week to run your analysis, or a month or three-day chunk. Whatever makes sense for your analysis.

We know there’s a huge difference in the skill levels among eBirders. We have people who are brand new to birding all the way to people who have been birding for a long time and are just incredibly skilled. We have to be able to account for this difference in skill level among individual eBirders.

We do this with checklist calibration index. We quantify the expertise of citizen scientists. We know that this has a large effect on the detectability, we know this varies seasonally, and we know that including it will improve our estimates of distribution and abundance.

So now that we have some data that has been filtered as it was collected and these various large sources of bias have been accounted for. What do we do with all of this data?

So with our Status models we take the eBird checklists and translate those into these dynamic distribution models for over 2000 species. These models show the abundance distribution by week at a very fine scale spatial resolution.

There are some challenges here when you do this. First, there’s the ecological process that is what links birds to particular locations at particular times of the year. To account for this ecological process we have many environmental variables within this model. We have variables that describe habitat types on the landscape, we have variables that are related to urbanizationm we have variables that are related to tidal mud flats, islands, roads, elevation. All of these things help account for the variation in the environmental processes. Then we have the observation process which is what we’ve spent the first part of this talking about. This is a process by which the eBird data is collected. We collect many effort variables like we talked about and we also use spatial and temporal filtering to help quiet the noise in the observation process within the eBird data.

Another issue is stationarity and scaling. This comes up when we think about a species that may have different habitat associations in different parts of its range or different habitat associations at different times of the year. If you were to model the entire range and the entire year all together those habitat types would essentially be fighting for importance within that model.

Here’s an example with Wood Thrush that prefers different habitat yypes on the breeding grounds than it does on it’s overwintering grounds. To account for this we use this spatiotemporal exploratory model framework. We divide the data into the boxes that you see on the right. These are not the exact boxes that are used this is just for demonstration purposes. We have hundreds of overlapping boxes. In each of those boxes a local model will be run, so that way we are capturing the local environmental processes.

In allowing them to emerge within this model we then make our predictions through these overlapping models at various locations so that the estimate at a given location can borrow information from multiple locations throughout the modeled surface. The end result then is this dynamic map of a species relative abundance and distribution at 3×3 kilometers across its entire range for every week of the year.

We also provide static maps with the seasons averaged. All of this data is made freely available for over 2,000 species on the Status and Trends website. We’ve also recently calculated trends for over 400 species of birds that breed in North America and hope to add more species and regions very soon. The trends are for the breeding season for each species from 2007 to 2021 and at 27 by 27 kilometers.

A lot of the time what you see with large-scale trend analyzes are range-wide estimates like we see here with Warbling Vireo. Looking at this from a range-wide scale you would you would think that the Warbling Vireo is doing fairly well: we have a slight increase in its percent change per year which suggests a growing population. But we know it’s not growing uniformly and there may be some pockets of decline. There may be some parts where it’s growing faster than others.

So we may wish to try to model this trend in a more spatially explicit manner. We also see this done at the regional scale, so a state or bird conservation region as we see here. In that type of analysis would look something like this. Here we can see that it isn’t just uniform growth across the range. There are regions of decline, some regions of large increase, some regions where there’s less increase than the average. What we have done with eBird is calculate these trends at much finer scales but also across the entire range of the species.

When we zoom in on this Warbling Vireo trend and look at the regional or BCR scale versus the landscape scale that we are using with eBird trends we can see the variation the magnitude even within these regions. The story gets richer here because the landscape scale more accurately captures population trends compared to the regions. That’s critical if we want to identify the drivers of the change here. The landscape scale is also more actionable. By that I mean the information is specific and useful and can directly inform decisions because we can scale down to the level where most conservation decisions are made, and that is much finer than the BCR scale. This example demonstrates that very well. We see that the highlighted state of Wyoming is part of a bird conservation region on the left that is showing a strong decline of this species. However when we look at it from the landscape scale we see that the part of that BCR that’s driving all of this change is in the far northwest portion of that bird conservation region and the species is actually growing rather well in the state of Wyoming.

So that concludes this presentation. I will be happy to answer any questions or comments that you may have. My email is listed here, and I thank you for your time.

End of Transcript

Welcome everyone, my name is Andrew Stillman and I work in the Center for Avian Population Studies at the Cornell Lab. In this video I’ll review the state-level data summaries that are available in table format from this website. To download the data for your state, navigate to the “Data by State” section at the bottom of this website, and click the link labeled “summary tables” to download a zipped folder.

So, let’s take a look inside one of these downloads. I’m showing an example here for the state of Colorado, and you’ll see that there are three different files inside.

The first one gives information based on relative abundance using results from eBird Status, the second one gives information on population trends using eBird Trends, and the last file is the eBird product terms of use. The CSV files can be opened in Microsoft Excel or any spreadsheet software you might use.

So, let’s open up this first spreadsheet labeled with the words “regional_status”. The file name will end with the data production year, and it will start with a state abbreviation. In this case, “CO” for Colorado.

Here’s the dataset. First, it’s important to note that the birds included on this spreadsheet are the species with non-zero percent of population estimates within the state during at least one season of the year. If a species appears to be missing, it’s possible that its percent of population was low enough to round to zero, or perhaps the species was omitted from models for some other reason.

Let’s move on to the columns included in this dataset.

The first four columns include the name for the state, followed by the species code, common name, and scientific name for each species. These species codes are the same ones used to access data products using the eBirdst package in R.

The next set of four columns gives the percent of the species global population which occurs in your state, broken up into four seasons. If you picture the global population of a bird species as a big pie, these columns give you the slice of that pie, as a percentage, within your state.

For example, the first column here tells us that 1.39% of the Olive-sided Flycatcher population occurs in Colorado during the breeding season, compared to 3% during post-breeding migration, and 0% during the nonbreeding season.

For resident species, such as Clark’s Nutcracker, all four values will be the same.

It’s important to remember that these values are rounded. Sometimes for a species with just a few local breeding occurrences in the state, you might see a value rounded down to 0.

The next set of three columns gives a twist on this same concept.

First, they show the season in which the percent of a bird’s population reaches its maximum within your state, and the specific week in which it occurs. The dates here are week midpoints. This is calculated using weekly percent of population estimates for every species, and then finding the maximum layer through all weeks of the year.

The next column gives that maximum percent estimate which corresponds to the week. So, in other words, these three columns show the week in which your slice of the pie is biggest, and how big of a slice you have.

For example, we can see that during the week with August 30^th as the midpoint, nearly 7.5% of the global population of Olive-sided Flycatchers is present in Colorado.

If a species is a resident throughout the full year, you’ll see the words “year round resident” in these columns.

In this next column, we’ve ranked each state in terms of the percent breeding population for each species. The first number is your state’s rank, and the second number is the total number of states with a non-zero breeding presence of the species. So, this tells us, for example, that 19 states have a non-zero breeding percent of population for Olive-sided Flycatcher, and out of these, Colorado has the 9^th-highest percent of population during the breeding season.

Last, we’ve included two columns giving the primary and secondary breeding habitat categories for each species. These categories come from the Avian Conservation Assessment Database created and maintained by Partners in Flight. We’ve included them here to help filter the data by major habitat associations if that’s applicable to your use case.

Now let’s switch over to the trends database, which will have the words “regional trends” along with the production year and state abbreviation.

This database begins with the same 4 columns giving the state name and species names. It also ends with the same two columns giving habitat categories from Partners in Flight.

You might notice that the number of species in the trends database is often smaller than the number of species in the status database – that’s because the status and trends data products come from different models, and there’s a different list of species that we’re able to model for each one.

Let’s focus in on this middle section.

The “trend period” column gives the years over which the trend is modeled. In this case for this video, trends are modeled from the year 2012 to 2022. Then the “season code” column gives the season during which the trend is modeled. You’ll generally see three options here: breeding season trends, nonbreeding season trends, or year-round trends for resident species.

The next set of three columns gives the trend estimate for your state. The units for this is the percent change in the species population over the trend period. So, a value close to 0 means very little change, a negative value means a declining population, and a positive value indicates an increasing population. These three columns give the median trend estimate followed by the lower and upper 80% confidence intervals.

As a comparison against state-level trends, we’ve also included the range-wide trend estimates for each species. So, this will show the median trend across the species entire range along with 80% confidence intervals.

Thank you, and I hope you find these data summaries helpful.

End of Transcript

Welcome everyone,

In this video I’ll review the state-level maps produced using results from eBird Status and Trends.

The spatial data we’re providing here are also available on the eBird Status and Trends website for the entire range of the species. But for this state-level website we’ve clipped them down to state boundaries to make the files a bit more manageable, and we’ve also provided some helpful map images.

To download the data for your state, navigate to the “Data by State” section at the bottom of this website. This video covers the zipped folders you can download by clicking the links labeled “breeding abundance,” “full-year maximum abundance,” or “full-year maximum abundance maps.” These first two links will download folders containing .tif raster files with spatial data for each species. The third link will download a folder with a mapped image of full-year maximum relative abundance in PNG format.

First, let’s take a look at the “breeding abundance” spatial data. When you open this folder, you’ll see a .tif raster file giving the breeding season relative abundance of each species. You’ll need access to some sort of spatial data software in order to open these files, such as ArcGIS, QGIS, or R.

The spatial resolution of each raster is approximately 3×3 km, which is the highest resolution currently available. Each raster has been reprojected to the NAD 83 Alber’s Equal Area coordinate reference system, which plots well within the U.S, and each .tif file is cropped to your state. 

It’s important to remember that relative abundance is a modeled data product giving a species-specific index to the number of birds in each gridcell. It’s not the same as population size. In addition, relative abundance is standardized within a species, but not among different species.

If you’re looking for access to range-wide maps of breeding season relative abundance, you can download these from the eBird Status and Trends website by navigating to the species of interest and clicking the downloads tab at the top right of the page. Relative abundance maps are also available for other seasons and at weekly timescales.

Next, let’s talk about files labeled “full-year maximum abundance.” As before, this folder will have a .tif raster file for each species. They are at approximately 3×3 km resolution and cropped to the area of your state.

But this time, instead of showing relative abundance, these maps show the maximum relative abundance value that each 3×3-km gridcell obtains throughout all 52 weeks of the year.

Maximum relative abundance can be a bit tricky to understand, so I’ll take a few moments to explain. To show why we produce these maps in the first place, let’s look at a map of breeding season relative abundance for Bay-breasted Warblers in New York State. You can see that it barely shows any breeding locations for this species. While this is true, it neglects to show that huge numbers of Bay-breasted Warblers migrate through the state each year, moving through almost every grid cell in the state. This is where maximum relative abundance maps come in.

So, to understand how these maps are made, first let’s look at the relative abundance map for each week throughout the year. This animation cycles through 52 weeks, and you can see the warblers migrating north in the spring, and then south again in the fall. The maximum relative abundance maps are the equivalent of pressing each of these weekly relative abundance cells into one of those impression boards that you can see pictured here. Each week presses into the board and it stores the maximum relative abundance value at each cell throughout the year.

Looking at this in the form of an animation, we can watch the image on the right develop as the year progresses. In the spring most cells light up and hold their values through the summer, and then in the fall some of those cells get even darker. In this way, maximum relative abundance distills the year-round distribution of the bird into a single layer. The range-wide raster data on maximum relative abundance are also available for Download on the eBird Status and Trends website.

To help visualize these layers, we’ve provided plotted images of maximum relative abundance in PNG format. You can download by clicking the link labeled “full-year maximum abundance maps.”

The maps show a color scale ranging from low, in yellow, to high, in purple and blue.

In summary, we’ve gone over the folder of .tif files with breeding season relative abundance, the folder with .tif files showing full-year maximum relative abundance, and the folder with maximum relative abundance images. We hope that this information is useful to you, and thank you for tuning in.

End of Transcript

Welcome,  

This video is about the state-level Shared Stewardship and Shared Stewardship Uniqueness maps that are provided on this website. To access these data products, scroll to the links at the bottom of the page. Find your state and click the “Shared Stewardship Maps” link to download a .zip file that contains these maps and the underlying data. These maps come in both .png images and .tif file formats, which you can easily load into your GIS software.  

Shared Stewardship maps (shown on the left here) show where the species that breed in a state are concentrated during the nonbreeding season. Because many species winter in the same regions, these maps may look similar between neighboring states. Since we wish to emphasize uniqueness of these connected geographies as well, we provide Shared Stewardship Uniqueness maps (shown on the right), which emphasize a given state’s ability to uniquely contribute to certain areas. These maps are best used in tandem. Our purpose in providing these maps is to support full annual cycle conservation of migratory birds and inform corresponding stewardship actions.  

When you download the folder for a state, you’ll see 6 different files inside. For both the Shared Stewardship and Shared Stewardship Uniqueness maps, you’ll see a plotted map in PNG format, as well as a .tif raster file with the spatial data behind each plot. There is also a CSV with the migratory species used in the analysis, and terms of use document.  

Here’s how we select the species to put in these maps: We start with migratory species that have high quality eBird relative abundance data for both breeding and nonbreeding seasons. Out of these, we select species for which at least 1% of the species’ total breeding population breeds within a given state and at least 85% of its breeding population migrates out during the nonbreeding season. You can find a complete list of species used to create these maps for your state in the in the same .zip file as the maps themselves. The species list is a CSV file, which you can open in Excel.  

Now, we’ll demonstrate how we create these maps so you can get a better understanding of what they represent. Shared Stewardship maps are essentially weighted sums of species’ nonbreeding populations, with the percentage of a species’ breeding population in a given state as the weight. For example, for the state of Missouri, we would multiply the summed percent breeding population of Wood Ducks in Missouri with its percent of population map during the nonbreeding season, then multiply the summed percent breeding population of Gray Catbirds with its map nonbreeding percent of population. We do the same thing for all other species on Missouri’s species list, and sum them all to create the map on the right here. So the values on this map are an index giving the relative strength of connection. If a species has a higher percent breeding population in a state, its nonbreeding map will contribute more to that state’s shared stewardship.  

It is important to note, however, that the nonbreeding connections shown in these maps are at the species-level, not the population or individual level. In other words, the individual birds that winter in the areas highlighted on the map here might not necessarily be the same ones that breed in Missouri. They are, however, members of the same species. So, this map tells us that the nonbreeding distributions of species that breed in Missouri, weighed by their Missouri breeding percent, is highest in areas such as the Yucatan peninsula, Central America, and Cuba. This information can be supplemented by on-the-ground migration tracking data, if it’s available.  

Shared Stewardship Uniqueness maps represent a given state’s proportion of stewardship connections relative to all U.S. states. To produce these maps, we first calculate a total by summing the Shared Stewardship maps of all U.S. states. We add Alaska’s stewardship map to Arkansas’, and so on. Then, for a given state (Missouri in this case), we divide its Shared Stewardship map by the total to generate its Shared Stewardship Uniqueness map, which indicates areas where Missouri has a high proportion of stewardship connections relative to other states. At first sight, this map may look similar to the Shared Stewardship map shown earlier. To clearly see what each map emphasizes, it is best to compare them side-by-side.  

As we can see, areas where Missouri has strong stewardship connections, like Cuba, do not always have the highest uniqueness values. This is because many other states also have strong connections to Cuba. On the other hand, areas where Missouri has moderate shared stewardship connections might be highlighted more in the uniqueness map if these same connections are not common in other states.  

Again, it is important to note that these maps are at the species and not population level. Ultimately, they are meant to be used for reference in conjunction with other resources when making decisions for conservation across the full annual cycle of birds.  

Thanks for watching.

End of Transcript