Using Satellite Data to Support Fieldwork

Can speczes distributions be predicted?

by Diane Debinski

Although species extinction has become a global concern during the last decade, our knowledge of species distribution patterns remains limited. If we don't know where a species has existed historically, we cannot determine if its range is contracting or expanding. This can make it difficult to identify a species as endangered until it is close to extinction. One way to address this problem is to try to predict which species m';ly be at risk based on their habitat distributions. A variety of on-the-ground techniques has been developed for monitoring species distribution patterns, but they are labor-intensive and cost! y. After conduct- ing a three-year biodiversity inventory of birds and butterflies throughout the wide array of habitats in Glacier National Park, I became interested in developing methods to make field surveys more efficient remotely sensed habitat types to predict Above: The author, Camille King, Katie and cost effective. Satellite data can be animal species distribution is still unre- Horst, and Lies/ Kelly discuss the results used to identify remotely sensed habitat fined. The Environmental Protection of their morning bird census at Twin types based on vegetation density, mois- Agency's biodiversity and habitat initia- Cabin Creek. Photos by lames Pritchard. ture content, and species composition. I tive is investigating the use of low-cost decided to test the use of satellite data in satellite data as an alternative to ground- technique used to compare locations of predicting plant and animal species dis- based habitat assessment. plant and animal habitats at a study site to tribution patterns. Our ability to distinguish different veg- those in existing preserves) use Although vertebrate biologists have etation types using only satellite data LANDSAT Thematic Mapper (TM) im- long used knowledge of an animal's may be limited. Spectral reflectance pat- agery to deterntine the boundaries of veg- habitat to predict its presence, and more terns are influenced by a combination of etation types. Then other data is incorpo- recently used satellite data to identify topography, moisture, elevation, and veg- · rated (e.g., aerial and high-altitude pho- species-specific habitat sites, the use of elation. Proponents of gap analysis (a tography and ground-based vegetation 2 Yellowstone Science maps and field surveys) to label the veg- the crest of the Madison Range to the suited in a merging of the 50 classes to etation types to series level. A major crest of the Gallatin Range (east/west). create five forest habitat types and six criticism of this approach has been that This area was chosen to develop our meadow habitat types. Mapwork and field gap analysis does not involve enough· model because it includes a wide range of surveys were then used to identify five ground-truthing of information. Even if elevation and moisture gradients and the spatially distinct examples of the three a habitat appears suitable, we do not patchiness of post-fire successional habi- mixed conifer-forest and six meadow know how often a species actually occurs tats, and because bird and butterfly spe- types, and 100 x 100m plots were staked at the predicted site. cies lists, including more than 100 spe- out at each of the 45 sites. cies of each, are available for the eco- Habitat types were based on remote Research Objectives system. sensing cluster analysis, followed by Birds were used to test the hypothesis ground-truthing with USFS stand-survey My own goal was not to do away with because they are conspicuous, ecologi- maps and aerial photos. local field sampling, but to use remotely cally diverse and use a wide variety of During the summers of 1993-1995, we sensed habitat types and geographic in- food and other resources, and are often inventoried each site for vegetation, but- formation systems (GIS) analysis to pre- more sensitive to environmental change terflies, and birds. dict species distribution patterns so that than other vertebrates. Butterflies were • Trees were sampled by establishing a fieldwork could be focused on specific chosen because they are well known taxo- 100-m transect on one side of each plot sites within the study area. If remotely- nomically, easily identified in the field, and surveying every tree within 3 m on sensed habitat types prove to be good and their diversity is correlated with un- either side of the transect line for species predictors of species assemblages, this derlying plant diversity. Birds and butter- and diameter at breast height. • could provide a more cost-effective tech- flies made a good combination because Meadows were sampled by estimating nique for monitoring biodiversity than they are active at different times of the total cover for each plant class (forbs, 2 ground-based field work alone. To test day. Birds were surveyed in the early grasses, and shrubs) within 25 one-m this hypothesis, I needed to: morning, and butterflies from mid-morn- plots, placed evenly along a 100-m • Determine the extent of the relation- ing through the afternoon. transect. For comparison purposes, cov- ship between remotely sensed habitat erage estimates were also made for each types and plant and animal species distri- Species and Habitat Characterization class using a 100 x 100m plot at each site. bution; and • Birds were surveyed in 35 plots com- • Test the predictability of species as- The remotely sensed data were clus- prising three forest types (Fl-F3) and semblages based on knowledge of this tered into 50 spectrally distinct classes five meadow types (M2-M6). We con- relationship. that were evaluated using U.S. Forest ducted aural and visual surveys using two The presence of a particular plant spe- Service stand survey maps, aerial pho- groups of two observers moving system- atically through the plots for 45 minutes. cies at a specific site can be highly indica- tography, and personal knowledge of the tive of the particular microhabitat of that study area. A preliminary analysis re- Sampling was repeated three times in site. Because the plant species that provide dominant cover play a major role in REMOTELY SENSED HABITAT TYPES determining spectral reflectance patterns, we needed to test the relationship between remotely sensed habitat types and the actual plant community. If plant spe- Mixed conifer forest: lodgepole pine (Pinus contorta), cies distribution could not be predicted spruce (Picea englemanni), and Douglas-fir (Pseudotsuga menzezii) using remotely sensed data, relationships F3 High density between remotely sensed data and ani- F2 Lower density mal taxa would be highly unlikely. Thus, Fl Fairly sparse a plant survey is the critical link between remotely sensed data, habitat, and other DF Douglas-fir species distribution patterns. WB Whitebark pine (Pinus albicalus) Research Desig'n

My colleagues and I initiated this research to link habitat components (e.g., Ml Hydric/lush meadow grasses, forbs, shrubs, and trees) with M2-M4 Decreasing moisture gradient birds and butterflies. Our study area was MS Moist sagebrush/cinquefoil meadow the northwest corner of the Yellowstone M6 Xeric, mostly dry sagebrush shrubland ecosystem, from Porcupine Creek to Ba- con Rind Creek (north/south), and from Summer 1996 3 each plot during the summer of 1993. types (wet meadows). The Yellowstone • Butterflies were surveyed in 30 meadow Checkerspot(Euphydryas gillettii), a spe- sites (five of each of the Ml-M6 types). cies that typically prefers wetsedgemead- Three people netted and released butter- ows, was found only in Ml meadows (see flies for 20 minutes in three randomly cover photo). This butterfly lays its eggs selected 50 x 50 m subplots. Sampling only on a shrub called black twinberry was repeated two or three times in each (Lonicera involucrata), and only if the subplot during the summers of 1993 and shrub is in a wet meadows. Checkerspot 1995. Butterflies were not surveyed in populations appear to be declining, and forests due to their low density there and may become a species of concern in fu- the difficulty of maneuvering with nets. ture years. Other moisture"loving spe" cies that showed significant preferences Comparing Satellite Data to Field for Ml and M2 meadows included the Observations greenish clover blue (Plebejus saepiolus) and four medium-size orange butterflies There appeared to be significant rela- in the family Nymphalidae: the western tionships between remotely sensed data meadow fritillary (Bolo ria epithore), the and the vegetation we found in our field silver meadow fritillary (Bolo ria selene), observations, and these impressions were Butterfly surveys can be an aerobic sport. the bog fritillary (Boloria frigga), and quantified. through statistical analysis, Here Camille King goes for the gusto in the painted lady (Vanessa c01·dui). The which showed several important rela- chasing down a vigorous swallowtail. host plants of the fritillaries and the blue tionships between satellite data and spe- include willows, violets, legumes; the cies distribution patterns of vegetation, dwellers. When habitats were clumped painted lady is more of a generalist. birds, and butterflies. into broad categories, preferences were • Four species preferred M4-M6 habitat Vegetation. Field surveys in 1993 as follows: types (dry meadows). Species such as the validated the vegetation density, compo- • The mountain chickadee, which is usu- lupine blue (Plebejus icariodes), the ring- sition, and moisture gradients expected ally found in coniferous forests, preferred let (Coenonympha inornata), and the from the satellite data. For example, the forest habitat types to the meadow Mormon fritillary (Speyeria mormonia) forest density decreased from Fl to F3 habitat types. showed preferences for M5 and M6 mead- forests, and F3 forests tended to be lo- • The song sparrow and rufous-sided ows. These species are typically found in cated on steep, north-facing slopes. Ml towhee preferred Ml-M2, the wet willow and M2 meadows were characterized by meadows. sedges and grasses that prefer wet sites, • The dark-eyedjunco, which is usually and M3 meadows by willow thickets and found in forest or forest-edge habitats, flowering vegetation, while M4, M5, and preferred the forest habitat types to the M6 meadows were characterized by a meadow habitat types. progression of plants that tended towards • The violet-green swallow, which is the drier end of the moisture spectrum found in a variety of habitats including (e.g., sagebrush, fescue, brorne, sedum, towns, preferred M5-M6 meadows. and aster). I will focus here, however, on • The hairy woodpecker, which is often the results of the animal data. found in medium moisture meadows with Birds. A total of74 bird species and42 aspen stands or dead standing timber, butterfly species were observed during preferred medium to drier meadows (M3- the surveys. Several species of birds M4). exhibited a preference for one or more of • The American robin and red-breasted the remotely-sensed habitat types. The nuthatch preferred F3, the denser forest, frequency of seven bird species (moun- while theruby-crowned kinglet preferred tain chickadee, brown creeper, American Fl, the more open forest. crow, orange-crowned warbler, hermit Butterflies. Ten species were found in thrush, American robin, and song spar- all meadow habitat types, while another row) was significantly different in the ten species were each found in only one forest versus meadow habitat types. Ex- meadow type. Thirteen species showed cept for the song sparrow, all of these significant preferences for one or more species preferred the forest habitat types. specific habitat types, including several This is what we would have predicted, for species that were found only in extremely Some butte1f!ies are more reclusive than song sparrows are typically found in wet wet or dry meadows. others. Shown here is Lyceana mari- meadows, while the other birds are forest • Five species preferred Ml-M3 habitat posa, the Mariposa Copper.

4 Yellowstone Science enough and its habitat specific enough to exhibit asignificantrelationship with one or more remotely sensed habitat types. Our next step will be to use the models developed in Yellowstone to predict the species distribution patterns in Grand Teton National Park. The test of our predictive models will begin during the 1996 field season. In summary, remote sensing will be helpful in locating possible sites for rare Diane Debinski carefully examines Katie Horst measures Ollt the] OOx 1OOm species with known habitat associations a butterfly for identification prior to plot and stakes flags preparing for a (e.g., a species restricted to dense forest releasing it. census at Twin Cabin Creek. or wet meadows) by identifying the por- tions of an ecosystem in which the spe- sagebrush habitats and span a range of preferred a broad category of habitats cies' habitat is likely to be found. This colors (tan, orange, and blue) and sizes. (e.g., forest vs. meadow). In other cases, technique could also be valuable in set- Their host. plants include legumes, vio- species distinguished among finer grada- ting up programs to monitor the effects of lets, and grasses. tions of habitats (e.g., wet meadows vs. global climate change on the distribution • Four species preferred M3 or M4 habi- dry meadows). Several species, includ- of certain species. However, the limita- tat types (intermediate-moisture mead- ing theYellowstone Checkerspot, showed tions of this technique must be recog- ows), which were characterized by di- a preference for one specific kind of nized; extremely rare species and species verse flowering plants such as wild gera- meadow or forest. All of the species- that are not habitat-specific will continue nium, strawberry, prairie smoke and habitat relationships we observed make to require monitoring through more field- cinquefoil, but sometimes had a few sage- sense given known species-habitat pref- intensive methods. brush or willow as well. Butterflies found erences. in these meadows included Sara's orange Plants were much more highly corre- Acknowledgements tip (Anthocharis sara), the orange sulfur lated with remotely sensed habitat types (Colias eurytheme), the small wood than were animals. This can be explained This research was supported by grants nymph ( Cercyonis oetus), and the dappled by severalfactors: (1) the remote sensing from the University of Wyoming Na- marble (Euchloe ausonides). All of these image is reflecting the actual presence of tional Park Service Research Center and species are medium in body size and light these plants on the ground; (2) plant data the University of Kansas' General Re- in color. Their host plants include le- are measured in terms of coVerage while search Fund and Panorama Society. gumes, mustards, and grasses. animal data are measured as presence or Evelyn Merrill, Alan Vandiver, and Ron absence; and (3) plants are stationary on Krager provided baseline data. I would How Effective is the Use of Satellite the landscape, whereas animals move like to thank Mark Jakubauskas for his Data? through the landscape and may or may help in creation of the maps, Kelly not be present when the data are col- Kindscher and Alexandria Fraser for their We expected satellite data to be a rela- lected. botanical expertise, and Susan White for tively good predictor of vegetation, and Many factors besides vegetation type her assistance in data analysis. Camille this expectation was met by our data affect species presence and can cloud the King, Lies! Kelly, Katie Horst, Rebecca analysis. All of the plant species with observed relationship between species Hurst, Amy Trainer, Paul Rich, and minimum cover (at least 5 percent in at and vegetation. Even if a habitat appears Michelle Wieland assisted in the field least one plot) showed a statistically sig- suitable for it, a species may not be present work. The Gallatin National Forest and nificant difference among the remotely- because of historical factors (e.g., hunt- the Montana Department of Fish, Wild- sensed meadow habitat types. Forest ing) or competition with other species. In life and Parks provided us with housing. habitat types were significantly different addition, species found in a wide range of Finally, I would like to thank Phil with respect to both density (Fl being habitat types will not demonstrate a sta- Humphrey and James Pritchard for their lowest and F3 being highest) and species tistical correlation with one specific habi- encouragement. composition. tat type. Many bird and butterfly species The next step was to determine whether fit into this "generalist" category. Spe- satellite data could be correlated with cies that were found in only a few sites, on distributions of selected animal taxa. the other hand, do not provide enough Diane Debinski is an assistant professor Several species of birds and butterflies data for rigorous statistical relationships. in the Department ofAnimal Ecology at were associated with one or more re- Thus, in order to predict where one might Iowa State University. This article remotely sensed habitat types. In some find a species using remote sensing and flects her special interest in conservation cases, our data showed that a species GIS methods, a species must be common biology and biodiversity assessment. Summer 1996 5