Occurrence Maps

We are excited to display the preliminary results of our modeling research using eBird data. These maps, which are called STEM (Spatio-Temporal Exploratory Model) maps, use eBird stationary and traveling count checklists that report all species. The location of each checklist is associated with remotely-sensed information on habitat, climate, human population, and demographics generating a suite of approximately 60 variables describing the environment where eBird searches take place. By relating these environmental variables to observed occurrences, STEM is used to make predictions at unsampled locations and times.  Models are trained one species at a time. Following model training, the expected occurrence for that species is predicted on each of 52 days, one per week throughout 2009, at some 130,000 locations sampled throughout the conterminous US. This massive volume of information is then summarized on maps, which in many cases reveal novel information about the annual cycles of North American birds. These maps showcase the power of eBird – year-round, continental-scale monitoring of all species.

Obviously, these maps show only the Lower 48 United States. This is because the landscape and climatic variables used to model the bird occurrence are only available for the Lower 48. The Lab and its partners are engaged in ongoing research to find and incorporate satellite-sensed data that will be applicable worldwide and at least allow us to expand these maps to Canada, Central America, and South America.

Each species map is displayed with a text overview of the broad-scale migration patterns, along with an interesting biological story to consider. Of course, every map has many more stories to tell, and we invite you to provide your comments and reactions on the eBird blog.

Recent Additions: February 14, 2012

Provide comments on any of these maps on the eBird 'Chip Notes' blog. Note that the below links are arranged in order of release, with the most recent ones on top. The links at the right are arranged in taxonomic order, which should make it easier for find a specific species.

*NEW* Upland Sandpiper (Bartramia longicauda)

Year-round animation.

*NEW* Yellow-throated Vireo (Vireo flavescens)

Year-round animation.

*NEW* Louisiana Waterthrush (Parkesia moticilla)

Year-round animation.

*NEW* Cerulean Warbler (Setophaga cerulea)

Year-round animation.

Year-round animation.

*NEW* Rose-breasted Grosbeak (Pheucticus ludovicianus)

Year-round animation.

*NEW* Lazuli Bunting (Passerina amoena)

Year-round animation.

Year-round animation.

Dusky Flycatcher (Empidonax oberholseri)

Year-round animation.

Eastern Bluebird (Sialia sialis)

Year-round animation

White-throated Sparrow (Zonotrichia albicollis)

Year-round animation.

Blackpoll Warbler (Dendroica striata)

Year-round animation.

Willow Flycatcher (Empidonax traillii)

Year-round animation.

Western Meadowlark (Sturnella neglecta)

Year-round animation.

Swainson's Hawk (Buteo swainsoni)

Year-round animation

Year-round animation.

Year-round animation.

Brown-headed Nuthatch (Sitta pusilla)

Static map for 24 May.

Wood Thrush (Hylocichla mustelina)

Year-round animation.

American Pipit (Anthus rubescens)

Year-round animation.

Grasshopper Sparrow (Ammodramus savannarum)

Year-round animation.

Dickcissel (Spiza americana)

Year-round animation

Year-round animation.

Western Tanager (Piranga ludoviciana)

Year-round animation.

While some of these maps match the known distribution of birds very well, some maps extrapolate into areas where we know the species does not occur. Often this is caused by regions of sparse eBird data, such as northern Minnesota, northern Maine, much of Nevada, sparsely-settled regions in the upper Great Plains, Montana, and elsewhere. In some other cases (south Florida for example), the habitat information seems to be insufficient to understand the landscape as it relates to bird occurrence. In all of these cases, however, we believe that more eBird checklists from these regions will improve the model’s ability to understand bird occurrence. So we strongly encourage you to check out our story that discusses the weaknesses in our eBird coverage in the United States, and to contribute any checklists you have from these regions.

Please remember that these maps tend to focus attention on areas where the species occurs at high frequency. Birders are very tuned in to rare birds at the fringes of their ranges. For example, we tend to consider south Texas to be the northern limit for Hook-billed Kite and Brown Jay (even though fewer than ten pairs of each occur along a 100-mile stretch of the Rio Grande); in fact, Brown Jay may no longer breed on the United States side of the river. Similar examples of isolated and very small populations at the extreme fringes of their ranges (e.g., Cerulean Warbler in Massachusetts and New Hampshire, Hooded Warbler in Minnesota, Bobolink in Nevada, and American Redstart in California) are reflected in field guide range maps, but in reality reflect extremely small populations that may consist of only a few pairs of birds. While birders may consider these to be "within the normal range of a species," in reality they are extremely localized exceptions, and these very faint signals are typically not shown on the STEM maps. STEM is fundamentally showing the probability of encountering the species at a randomly selected point on the landscape, so these locally isolated populations really should not be shown on these maps in many cases. Please keep in mind the occurrence scale (see the scale on the right -- best visible in the 'large' versions of theses maps), try to consider the probability of encountering the species at random, and we think you will find that these maps are very accurate.

We do invite comment. The maps are not perfect and it is an ongoing research project to improve them. We are currently incorporating additional landscape variables, including hydrology and satellite 'greening' data, which we hope will further improve results. In addition, more eBird checklists from more diverse locations really help these models perform and the exponential growth in eBird checklist volume will pay great dividends for these results. Please drop in to the eBird 'Chip Notes' blog to share your thoughts on these maps or comments on the analysis.

Support for the development of these maps comes from the Leon Levy Foundation, the Institute of Computational Sustainability at Cornell University, DataONE, Oak Ridge National Laboratory Distributed Active Archive Center (ORNL DAAC), and TeraGrid.