Home - Table of Contents

Species Occurrence: What, Why, and Where?

Randall B. Boone and William B. Krohn
Maine Cooperative Fish and Wildlife Research Unit, University of Maine, Orono


What are appropriate scales to use to map species ranges? Why are terrestrial (i.e., non-fish) vertebrates distributed as they are? Where are species likely to occur, and are those occurrences predictable? This is a sample of the questions addressed in a recently completed doctoral dissertation that forms a biogeographical foundation for Maine Gap Analysis. Here we review the research and its utility to GAP. The brevity of this report demands that details be omitted; for more information see Boone (1996) or other publications reporting the research.

Range Mapping and Assessment

Among the many decisions required when mapping species ranges, three are fundamental: 1) the methods used to assess the accuracy of ranges; 2) the size and shape of tessellation (i.e., a regularly- or irregularly-shaped geometric grid) used to map both ranges and distributions; and 3) the tessellation used to map observation data (e.g., atlas data) used in assessments. To address these questions, we defined range boundaries for the 187 bird species that breed in inland Maine, using atlas data, rare occurrence records, literature, and expert review.

We did not use data from the Breeding Bird Survey (BBS) to define ranges, making a decision early on to use BBS information for testing. From the BBS, smoothed relative abundance maps were created for each species for which there was adequate data, using block kriging (following others, e.g., Maurer 1984; Price et al. 1995), based on a 324 km2 grid, and mean numbers of birds observed from 1984-1993 on 189 routes within a 300-km radius of Maine. Of 186 bird species breeding in inland Maine, 80 had range limits in the state. Of these, 47 species had adequate BBS data to yield smoothed abundance maps. Zero contours (i.e., range limits) are difficult to define in relative abundance maps. Instead, we used an algorithm that identified a boundary having the best fit between the species range map defined for Gap Analysis and the BBS abundance map. An iterative ARC/INFO program compared the number of grid cells in disagreement for abundances until an optimum fit was reached.

When avian range maps were compared to relative abundance maps, the ranges compared reasonably well. For species with high-quality kriged maps (n = 18), the median error between generalized ranges and observed data was 8% of the state’s surface area. When the disagreement in area was considered, species with good kriged maps differed by only 3.9% of the state’s area; over all species, the error was 4.5%.

When smooth-lined range maps were compared to ranges generalized to tessellations of various block-sizes and shapes (Fig. 1), the error introduced into ranges was modest for block-sizes < 1,000 km2 (Fig. 2a). As examples, townships (mean size of 93 km2) introduced a maximum error of 1.5%, EMAP hexagons (640 km2) introduced 4%, and counties (mean size 4,900 km2) introduced 20% maximum error.

Figure 1. Tessellations used to map each bird species that had range boundaries in Maine. Shaded areas represent the range of Boreal Chickadee (Parus hudsonicus), remapped to each tessellation using a 50% of-area rule. a) townships, b) 7.5-minute quadrangles, c) 30-minute quadrangles, d) 1-degree quadrangles, e) EMAP hexagons, and f) counties.

Figure 2. a) The error introduced into range maps due only to generalizing to tessellations, relative to smooth range maps, based upon ranges of 47 birds. Error bars report standard deviation; selected tessellations are labeled. b) The relation between observation block size and the proportion of the state considered confirmed breeding.

Observations of species may represent areas larger than the site surveyed (e.g., an observation of a bird in an atlasing effort leads to an entire block of thousands of km2 to be shown as occupied). The size of these observation blocks affects how much of the state will be labeled as "confirmed breeding" in mapping efforts (i.e., the perceived confidence in the statewide observation data). The proportion of the state with "confirmed breeding" rose steeply for block sizes < 1,000 km2 (Fig. 2b). For block sizes over 1,000 km2, perceived confidence did not increase in proportion to block size.

In summary, mapping the ranges of many species requires balancing conflicting utilities—large tessellations used to map ranges increase confidence in the range but introduce error, and large observation blocks increase perceived confidence in species’ range maps but may be misleading because of heterogeneous landscapes. Whatever the outcome, decided upon based on the balance of utilities described above, tessellations or observation blocks > 1,000 km2 should not be used when mapping an area the size of Maine, especially tessellations that are irregularly shaped (e.g., counties in many states).

Biogeographic Relations

We partitioned variation in richness into its components for each terrestrial vertebrate class, i.e., amphibians (n = 17, 6 with range limits in the state), reptiles (16, 13), mammals (56, 20), birds (186, 80), and all species (275, 119). These statewide distributions were compared quantitatively to geomorphology, climate, and woody plant distributions, mapped using a grid of 324 km2 cells. Amphibian and reptile ranges were related positively to productivity (e.g., heat accumulation, maximum temperature, frost-free period) and negatively to average annual snowfall. Seven mammal ranges were related positively to productivity, and six were positively associated with snowfall and elevation. Many bird ranges (n = 47) were positively associated with productivity and negatively with snowfall, but some (n = 29) were related oppositely (Boone 1996). Birds that were classified as forest specialists, and those classified as early successional, were spatially coincident with the north-south and east-west plant transition zones, respectively. Forest generalists and birds classified as using barren/urban or wetland/water habitats were not associated with either plant transition zone. In models describing variation in total species richness, climatic variation was the best descriptor (r2 = 92% in tree regression), followed by woody plant distributions (87%) and geomorphology (87%). Reptiles were highly correlated with environmental variables (93%), followed by amphibians (85%), birds (82%), and mammals (81%).

Are Errors in Species Occurrences Predictable?

GAP researchers assessing their predictions of species distributions have tested their work by comparing species predicted for an area to check lists (e.g., Edwards et al. 1996). Researchers report error rates without further interpretation, or relate a posteriori the error rates of species to their ecological attributes. Instead, we suggest that species should be ranked as to how likely they would be to occur in future surveys, which is closely related to how likely a species’ distribution will be predicted correctly in gap analyses. These rankings become a form of a priori hypothesis regarding the relative accuracy of potential occurrence predictions for groups of species.

We created a method that allowed avian species to be ranked as to how likely they would be to occur in future surveys. Attributes (e.g., population level, niche width, population trend, body weight, aggregation) were used to model the incidence of occurrence within the Maine Breeding Bird Atlas (BBA; Adamus 1987). Likelihood Of Occurrence Ranks, or LOORs, were assigned to each avian species, based upon the modeled incidences, to reflect how likely the species are to be observed in future surveys. To test the utility of the ranks, the occurrence of birds on six areas located throughout Maine, with species checklists and existing vegetation maps or habitat descriptions, were predicted and then compared to the LOORs.

Using the explanatory data in a logistic model, 78.3% of the variation in species incidences within the BBA was explained. Population aggregation, abundance, and species niche width were the three most importance variables in describing incidence. From the model, LOORs were assigned to the 183 bird species, and species were placed into 10 groups based upon LOORs. The number of species correctly modeled using species-habitat associations was highly correlated with grouped LOORs (p 0.68-0.93, P = 0.032)—species judged a priori to be unlikely to be modeled correctly were not (Fig. 3). Sites with checklists from many years (e.g., > 10 years) and from large areas (e.g., > 1,000 ha) yielded the lowest commission error.

Figure 3. For each site tested, the number of species correctly modeled for each group of LOORs. Sites are ordered by size from largest to smallest: a) Acadia National Park (64,235 ha, checklist covering 25 yrs); b) Moosehorn National Wildlife Refuge (9,172 ha; 52 yrs); c) Sunkhaze Meadows National Wildlife Refuge (3,764 ha, 8 yrs); d) northern Maine forests (555 ha, surveys over 2 yrs); e) Holt Research Forest (40 ha, 10 yrs); and f) Nesowadnehunk Field, Baxter State Park (16 ha, 3 surveys over 15 yrs).

These results demonstrate that the confidence levels assigned to Gap Analysis results is dependent on the test sets and on the species included. Small areas or sites with surveys of too few years will yield tests with high commission errors. (Editor’s note: see especially Gibbons et al. 1997.) Direct comparisons between different modeling efforts is not straightforward. For example, areas with a high proportion of rare species (e.g., some islands) are likely to have high commission errors, regardless of the accuracy of models. Researchers testing their own results may ensure a checklist is essentially complete using species-accumulation curves, resampling techniques, or expert opinion. Conducting tests on areas of several sizes is helpful. Researchers interested in comparing the results of several modeling efforts may find standardizing species incidence using the BBS helpful. Omission and commission errors for quantiles of standardized incidences, for example, would be comparable.

Significance to GAP and Future Directions

Assessments of ranges using BBS data suggest that the empirical methods used to define vertebrate ranges, at least for birds, worked reasonably well. GAP projects with ample data for defining ranges (e.g., state atlases) might consider reserving BBS for testing. Regarding tessellation shape and size, under Maine conditions, mapping ranges using tessellations up to the size of EMAP hexagons is reasonable, but using irregular shapes or larger tessellations results in a loss of information. In Maine, we will store ranges in three formats: 1) their original tessellation (townships), 2) as smooth range lines, and 3) using EMAP hexagons.

The accuracy of avian species predictions in Maine Gap Analysis will be assessed under the framework provided by LOORs. By stratifying species into those likely to be predicted correctly and those unlikely to be predicted correctly, our omission and commission errors for sites with checklists will be more informative. The biogeographic analyses have led to another method of assessing GAP results that can be compared to using checklists. The correlations described were done with range maps, not predicted occurrences. After predicting the occurrence of species within Maine, the correlations will be recalculated. If species-habitat associations used in Gap Analysis are indeed useful, we would expect the correlations to improve.

Literature Cited

Adamus, P.R. 1987. Atlas of breeding birds in Maine, 1978-1983. Maine Department of Inland Fisheries and Wildlife, Augusta, Maine.

Boone, R.B. 1996. An assessment of terrestrial vertebrate diversity in Maine. Ph.D. Thesis, University of Maine, Orono.

Edwards, T.C., Jr., E.T. Deshler, D. Foster, and G.G. Moisen. 1996. Adequacy of wildlife habitat relation models for estimating spatial distributions of terrestrial vertebrates. Conservation Biology 10:263-270.

Gibbons, J.W., et al. 1997. Perceptions of species abundance, distribution, and diversity: Lessons from four decades of sampling on a government-managed reserve. Environmental Management 21:259-268.

Maurer, B.A. 1994. Geographical population analysis: tools for the analysis of biodiversity. Blackwell Scientific, Boston, Massachusetts.

Price, J., S. Droege, and A. Price. 1995. The summer atlas of North American birds. Academic Press, Harcourt Brace and Company, New York.

Home - Table of Contents