Washington Breeding Bird Atlas: Methods
Overview: Models were developed by using known locations to delineate range limits, and using known locations, literature review, and expert consultation to develop habitat associations. Predicted distributions were created by selecting appropriate habitats in the land cover map within each species' range limits. NOTE that the locations on the map are not points, they represent the centroid of the 3 mile x 3 mile Breeding Bird Atlas block.
Record Collection: We assembled a database of over 117,000 records of breeding birds in Washington. All records were collected after 1986. Three major databases contributed: the Washington Breeding Bird Atlas project (95,417), Fred Dobler's shrub-steppe bird records (15,817), and the WDFW Natural Heritage Database (6,193). The mapped distributions of the points from the first two were scrutinized and error-checked by the first and second authors. Data points from the Natural Heritage Database were checked by the senior author, and generalized from point locations up to the BBA 3mi x 3mi block. Collection of data varied for each database; methods are summarized below.
The Breeding Bird Atlas data were collected by volunteers within quarter-township (9 sq. miles) 'blocks' throughout the state. Data were supplied from 2,312 of the 7,912 available blocks. The project was started in 1987 and data were collected through 1995. Birds encountered by volunteers within the blocks were recorded and assigned to one of four evidence categories, with evidence noted on the card as follows:
Species observed, but no evidence of breeding
Species in suitable habitat during nesting season
Singing male present in suitable habitat
Multiple singing males (7 or more) found during one visit
Pair observed in suitable habitat
Courtship behavior, copulation, cloacal protuberance
Visiting probable nest site
Agitated behavior from adults
Nest building or excavation
Brood patch or egg in oviduct
Used nest or eggshells of positive identity
Recently fledged young incapable of sustained flight
Occupied nest: adults entering, leaving, or incubating, but nest contents unseen
Adult bringing food to a nest
Adult removing fecal sac from nest
Nest with eggs or young found, seen or heard
These data were entered by volunteers, and imported into a geographic information system from an Ingres database. The GIS coverage of BBA blocks was created by appending the Public Land Survey (PLS) coverages of the Census Bureau's TIGER files, and then dissolving section lines to create a grid of quarter-townships. Water bodies were removed from the grid to maintain a continuous grid. This coverage had an attribute table with a record for each species for every block, with a value of 0 - 5 as shown below:
0 = no data
1 = 'Observed' BBA status
2 = 'Possible' BBA status
3 = 'Probable' BBA status
4 = 'Confirmed' BBA status
5 = confirmed nesting evidence known prior to 1987, but after 1980.
Where blocks had multiple data points, the highest number between 0-4 was retained as the value. If no data were available except a class '5' data point, then 5 was assigned as the value.
The Washington Department of Wildlife shrub-steppe bird data did not contain enough information to yield a 'probable' or 'confirmed' breeding status, so all these data points were added as 'possible' breeders in the case of breeding birds, or 'observed' in the case of known migrants (such as White-crowned Sparrow or Dark-eyed Junco). These data were collected at multiple sites throughout the Columbia Basin in habitats dominated by Artemisia species, and were made available by Fred Dobler.
The WDFW Heritage Database contained records of variable quality. Since some records had a high degree of confidence, while others seemed highly improbable, each record (n=6193) was reviewed by the senior author and assigned to a value as described above. If a record had any degree of uncertainty, it was interpreted conservatively, with many records being wholly removed from the database.
Habitat Association: Literature review, record location and consultation with experts were used to develop habitat associations. The availability of habitat association information varied widely among species, from exceedingly poor to more than we could use. Our challenge with all species was to translate the information from its reported scale and description to our (usually coarser) scale and descriptions. For example, an animal reported to be primarily associated with small clearings in conifer forest but not with clear-cuts would be associated in our models with conifer forest, not non-forested cover, because small forest openings would be well below our minimum mapping unit of 100 ha. Our models relied heavily on species' association with vegetation zones (e.g., the Ponderosa Pine zone or the Interior Douglas-fir zone). Vegetation zones are a surrogate for the combination of interacting environmental parameters (elevation, precipitation, aspect, latitude, etc.) that create conditions that favor dominance of a set of plant species. Animals often respond similarly to the collection of environmental parameters that determine a zone. Location information was most useful in establishing a species' presence in a vegetation zone. Detailed habitat descriptions were also helpful. On the other hand, reports of elevational limits without information about location or habitat were of limited value because a species' elevational limits are usually dependent on a combination of environmental parameters.
Habitat Quality Coding: For the actual modeling, we created a habitat coding matrix with all possible habitats forming the rows and the species forming the columns. The habitats we used for modeling had three parts: ecoregion, vegetation zone and actual cover. The suitability of each habitat for each species was assigned to one of eight categories:
0 - Not suitable.
1 - Present, but habitat quality unknown. (this code was rarely applied.)
2 - Good habitat in a vegetation zone core for that species.
3 - Adequate habitat in a vegetation zone core for that species.
4 - Land cover in core zone contingently suitable with suitable microhabitats.
5 - Good habitat in a vegetation zone peripheral for that species.
6 - Adequate habitat in a vegetation zone peripheral for that species.
7 - Land cover in periperhal zone contingently suitable with suitable microhabitats.
Good habitat would be expected to be more suitable than adequate habitat. For a habitat to be labeled 'contingently suitable,' there had to be the possibility that imbedded suitable habitats below the minimum mapping unit were present, but these imbedded suitable habitats were not expected to occur in all or most of the larger habitats. For example, small clearings created by the death of a few large trees would be expected to occur in almost any 100 ha forest stand, but imbedded habitats like ponds, talus slopes, caves, etc. would be less likely. For an animal associated with small clearings, forest would be good or adequate habitat (depending on the species), but for an animal associated with small ponds within forest, forest would be contingently suitable.
Assuming appropriate habitat, if a species was believed to be present in a zone chiefly because of overflow from an adjacent zone (i.e., the zone was likely to be a population sink) or the population density of a species within a zone was believed to be low compared to other zones, then the zone was labeled peripheral.
The Subjective Nature of Habitat Quality Coding: Assignment of habitat quality codes was obviously subjective. We expected that assigning codes would be difficult and contentious, and initially planned to base our models on presence alone. We moved to habitat quality coding because the ultimate use of the maps was as an aid in biodiversity management. Models based on presence alone imply that habitats of obviously different quality are equally important in management. In practice, we found that habitat quality coding was usually easy. Lack of confidence in assigned codes was more likely to be based on lack of knowledge rather than the subjective nature of code assignment. We attempted to address coding uncertainties for a species in the Comments section included in each species account.
The Effects of Land Cover Mapping Error and Resolution on Vertebrate Models: Error propagation through layers is a serious problem in Geographic Information System manipulation of data. However, the presumption that error in a cover is equal to the product of the errors in each layer comprising that cover is an over-simplification. The effect of an error in a source layer upon a derived cover depends on what is being derived. The effect of errors in the land cover map on the final error of the vertebrate map is particularly dependent upon the species being modeled. For species that thrive in a wide variety of habitats and zones (e.g., Common Raven - Corvus corax), only the most severe of errors in the land cover map (e.g., mistaking ocean for land) would produce a serious error in the predicted distributions. Other species, however, are dependent on habitat parameters that we could not map (e.g. snag density or the distribution of a particular species of shrub). For these species, we could expect to correctly exclude clearly unsuitable habitats, but the remaining included habitat will likely be more extensive than the actual distribution. For some species, vegetation zone is more important to their distribution than actual cover. In these cases, the accuracy of the modeled distribution will depend on how well we mapped the zone boundaries rather than inaccurate interpretation of the actual cover. We addressed modeling uncertainties and problems in the Comments sections of the written species' accounts (available only in print).