Martha Grinder, Wolfgang Grunberg, D. Phillip Guertin, Paul R. Krausman.
ABSTRACT: Coyotes (Canis latrans) are becoming common in urban areas throughout western North America. Our goals are to characterize the areas used by coyotes in Tucson, Arizona, and to create a predictive model to determine the presence or absence of coyotes in all land-use categories throughout Tucson. We used location data from 13 radiocollared coyotes, combined with a previously-existing ArcInfo database of land-use categories in Tucson, to create the model. In ArcInfo we created nine grids of habitat variables. These variables were the distances from coyote locations and from 5,000 random points, to each of nine land-use categories. We then used stepwise logistic regression to choose the seven variables that differentiated between coyote locations and random variables. We created a logistic regression model with these seven distance variables. We multiplied the regression coefficients for each variable by the respective grid, and added the grids to create a probability grid for locating coyotes in Tucson. The current model indicates that most areas in Tucson are likely to be used by coyotes. The model may be refined by adding some new variables and by creating the model at a different scale.
Coyotes (Canis latrans) are one of the least specialized, most primitive living members of their genus (Nowak 1978). They are opportunistic feeders (Bekoff and Wells 1980), have a flexible social structure (Andelt 1985), and can be active at night and during the day (Shargo 1988). These traits have enabled the coyote to adapt to diverse environments throughout North America (Bekoff 1982).
As urban centers throughout North America expand, coyotes are also adapting to life in the city (Howell 1982, Quinn 1991), which makes them ideal subjects for study of the affects of urbanization on wildlife (Bekoff and Wells 1986). To date, however, no studies have thoroughly investigated the response of coyotes to increasing urbanization. Some studies have investigated coyotes on the outskirts of urban areas (Shargo 1988, Atkinson and Shackleton 1991, Bounds 1993, McClure 1993), but have not concentrated on their use of urban centers. Researchers have recorded the presence of coyotes in Los Angeles, California (Gill and Bonnett 1973, Howell 1982) and Seattle, Washington (Quinn 1991, 1995) but because of experimental design limitations, were not able to adequately quantify habitat use and movement patterns.
Although many aspects of coyote ecology along a rural-urban gradient are difficult to predict, one would expect coyotes to inhabit places that provide food, cover, water, and fulfill any other special needs they may have, such as holes for den sites during the breeding season. Land-use categories in Tucson, that might provide for such needs, include parks, areas with natural vegetation, washes, and some residential areas (Shaw et al. 1996). Our goals are to characterize the land-use categories that are used by coyotes in Tucson, and to create a model that predicts the potential use of different areas throughout Tucson.
The area in which we captured and radiocollared coyotes encompassed most of the city of Tucson, Arizona, and a few urbanized areas directly outside of the city limits (Figure 1). Tucson, in eastern Pima county, encompasses 493 km2, with an estimated population of 456,100 (Tucson Planning Department, unpubl. data). Tucson is situated in the Sonoran Desert, the most varied and the hottest of the North American deserts. The elevation is 745 m in midtown Tucson, and increases toward the foothills of the Santa Catalina Mountains to the north, the Tucson Mountains to the west, and the Rincon Mountains to the east. The climate in Tucson is characterized by low, unevenly distributed rainfall (about 28 cm annually; Sellers and Hill 1974), low humidity, high air temperatures and periodic strong winds (Hastings and Turner 1965).
The size of coyote home ranges and the land-use categories used by coyotes in Tucson was variable. To account for these differences among coyotes, we created a study area from the combined home ranges. We buffered each home range with one-half the long axis of that home range, and merged all buffered home ranges into one primary study area; one buffered home range was disjunct from the others and created a small satellite to the study area (Figure 1).
We used a previously created ArcInfo vector database (Shaw et al. 1996) as a basis for creating land-use categories. The original database (Shaw et al. 1996) assigned all of eastern Pima county, including Tucson, to one of 33 land-use categories at a resolution of one acre (0.4047 ha). We collapsed the original land-use categories into seven land-use categories. We formed the new land-use categories based information from Shaw et al. (1996) on the amount of native and non-native vegetation present, the amount of human activity present, and on obvious structural differences and similarities among the original land-use categories. To supplement the original database, we also used two Pima county vector coverages (ALRIS database), for washes and streets.
The natural category included residential areas with low-density housing (<1 house/0.4 ha), state and federal parks, privately-owned natural open space, and cropland. The commercial category included industrial areas, malls and other shopping centers, public buildings, and office buildings. The park category included schools, military grounds, cemeteries, zoos, golf courses, neighborhood, district and regional parks, and stables or pens with horses or cows. The vacant category included mines, landfills, graded vacant land, abandoned agricultural lands, and railway yards. The residential category included neighborhoods with > one house/0.4 ha. The wash category included major and minor rivers and washes. The road category included only roads with > four lanes; smaller roads were incorporated into the surrounding land-use categories.
We trapped coyotes using padded leg-hold traps (# 3 Victor Softcatch Coilspring, Lititz, Pa). We trapped and radiocollared 14 coyotes from October 1996 through March 1997, and five coyotes from December 1997 through January 1998. We will use the data from 13 coyotes. We tried to trap in locations that represented a variety of areas and human population densities within Tucson. We were not able to trap everywhere we chose because some landowners did not allow us to trap on their property; other areas were too often traversed by dogs and people that might step in a trap. We closed the traps at dawn and opened them at dusk daily to minimize the time that a coyote spent in a trap, and to minimize the chances of trapping non-target animals. We immobilized each trapped coyote with a noose rod, muzzle, and nylon stockings to tie its legs (Woolsey 1985). We then fit the coyote with a radiocollar (Telonics Inc., Mesa, Ariz.). We weighed each trapped coyote, determined the coyote’s sex and reproductive condition, and approximated its age (< 12 months, > 12 months, > 24 months) by looking at tooth wear (Gier 1968) and, for young of the year, by looking at the condition of the coat and tail. Finally, we evaluated the animal’s general health by checking for external parasites, wounds, or other signs of ill heath, and released it at the trapping site.
We located coyotes by homing with hand-held Yagi antennas (White and Garrott 1990:42). We attempted to visually locate animals, if we could do so without trespassing or disturbing the animal. We tested each technician’s error in locating coyotes by placing radiocollars at locations, known to the tester, in various habitats and having technicians estimate collar locations via their usual homing procedure (Litvaitis and Shaw 1980, Bounds 1993).
We collected data throughout the year. We located each coyote > two times/week, once during the day and once at night. We broke both day and night into two, six-hour blocks, and made an equal number of locations during each block. We recorded coyote locations on enlarged sections of a Tucson street map. We then entered the locations for each coyote into Arc/View as shapefiles. Each location was coded to an attribute table that contained the date, time of day, observer, and comments. Finally, we converted the coyote locations into ArcInfo coverages; this allowed us to assign the locations UTM coordinates.
We imported the ArcInfo coverages of coyote locations to the home-range package RANGES V to obtain the 95% adaptive kernel home-range estimates that we used to create our study area. The adaptive kernel method is a nonparametric method of estimating home-range size that allows one to determine core areas of activity, an important factor in urban areas where coyotes habitat may be fragmented.
We used distances to land-use categories as the independent variables from which we built the model to predict coyote use of land-use categories in Tucson. To predict coyote use of different land-use categories we compared coyote locations to 5,000 randomly chosen points (Appendix) relative to these distance variables.
Using ArcInfo, we rasterized the land-use categories in the study area at a resolution of 10 m, and separated the categories into nine different grids. We created one grid per land-use category created from the original Tucson database; we also created grids from a county coverage of washes and a county coverage of streets. The county wash coverage was more accurate and complete than the Tucson coverage; the county streets coverage included all types of streets, whereas the Tucson coverage only included streets with > four lanes.
Using the GRID command EUCDISTANCE, we created distance grids for each land-use category so that all cells in the study area were given a distance value to the nearest cell containing that particular land-use category. Using the GRID command SAMPLE, we then obtained text files containing the distance of each coyote location, and each random point, to each land-use category.
We created a single text file containing coyote locations with their associated distance variables and random points with their associated distance variables. We coded the coyote locations as ones and the random points as zeros. We then performed stepwise logistic regression (Afifi and Clark 1984) to determine which distance variables to include in our model of coyote habitat.
We performed logistic regression on the distance variables that we found to be indicative of coyote locations in the study area. We then used these variables to create probability model (Christopherson et al. 1996, Figure 2). We multiplied the distance variable in each cell in the study area by the logistic regression coefficient for that variable, and summed the weighted variables and the corrected y-intercept (Warren 1990, Appendix), to create a probability of that cell being a coyote location. To clarify the results and facilitate comparisons, we logistically transformed (Appendix) each cell’s probability value, re-scaling its value to a probability score between zero and one. Finally, we used the RECLASSIFY command in GRID to change the grid’s cell size to one acre (0.4 ha). We did this because our land-use categories came from a database that was at this resolution. Our end product was a map of the study area with a continuously changing surface, corresponding to the area’s probability of containing a coyote location.
We then queried the grid in two ways, using the GRID command SAMPLE, to see how well the model predicted our coyote locations. First, we queried the original probability grid to determine the average probability values for coyote locations versus random points. If the model accurately differentiated between coyote locations and random points, the average coyote location should have a probability score near to one and the average random point should have a probability score near to zero. Second, we created a grid in which all probability values between 0.0000 and 0.4999 were given a value of zero and all probability values between 5.0000 and 1.0000 were given a value of one. We then tallied the number of ones and zeros included in the coyote locations versus the random points. Coyote locations should have mostly ones; random locations should have mostly zeros.
We chose seven variables for inclusion in the regression model: distance to county washes, distance to natural areas, distance to parks, distance to residential areas, distance to roads, distance to vacant areas, and distance to the Tucson washes. All of these variables were highly significant (P < 0.001) coefficients in the regression model. The regression coefficients for distance to county washes and distance to vacant areas were negative; all other regression coefficients were positive (Figure 2).
The probability grid based on these seven variables contains a large number of probabilities that are close to one (mean = 0.875 SD = 0.157, Figure 3). This indicates that, based on the variables we used to construct the model, most areas in Tucson are suitable for coyotes. The mean probability for coyote locations was 0.7287 (SD = 0.2040). The mean probability value for random points was 0.8899 (SD = 0.1366). When the probability grid was converted to a grid with zeros and ones, 89% of the coyote locations had values of one and 98% of the random points had values of one. An overlay of the model’s probabilities (in Arc/View, as a hillshade) on the original map of the study area (Figure 1) indicated that areas near the commercial center of town, especially around washes, had the lowest probabilities of being used by coyotes.
Our model is a good first step toward being able to locate the areas in Tucson that are used by coyotes. It tells us that most of Tucson is has the potential to be used, at least transiently, by coyotes. As coyotes have been observed in almost every location in Tucson, this is probably a true statement. We need, however, to make some refinements to the model to precisely locate areas that are used by coyotes to fulfill their habitat requirements.
Our model was based on the technique used by Christopherson et al. (1996) to determine the probability of finding Iron Age I and Iron Age II archeological sites in Jordon. The model built by Christopherson et al. (1996) accurately predicted the presence of both types of sites. Archeological sites, unlike coyote locations, are evidence that the inhabitants of interest invested time and energy in a particular site, and chose it over other sites. By contrast, a coyote location may be obtained for an animal that is traveling on its way to an unknown destination. In addition, our random points were not necessarily locations without coyotes. A random location could have been a point where a collared coyote was located, or any coyote had been at one time without being located. Therefore, coyote locations and random points are not indicative of the same investment of time and energy at a spot as is the location of an archeological site or a non-archeological site. This may, in part, be the reason why we were not able to accurately predict land-use categories in Tucson that would be used by coyotes. We are currently following coyotes for up to six hours at a time to determine how much time they spend in various land-use categories. These locations may be more useful in constructing a model of the type described above.
In addition, the scale of resolution we used to create grids of land-use categories may have been inappropriately small. Coyotes probably choose to remain in a location based on more than just the habitat within a 10 x 10 m area. Mobile animals, coyotes may consider the habitat features within 100 m or more of their present location when choosing places to inhabit. To address the scale at which coyotes may be making these decisions, we will consider re-analyzing our current data using a larger grid size.
Finally, coyotes may not be choosing to inhabit particular locations based on the distance to certain land-use categories. A better way to evaluate coyote use of land-use categories in Tucson may be to create a series of grid layers, each of which has cells that contain information on the percentage of a particular land-use category that is contained within the cell. We will need to collect additional data on the number and distribution of such landscape features as ponds and open water to include them in the model; coyotes in the desert southwest may be limited in the habitat they can use by their proximity to open water. We may also find it necessary to evaluate the abundance of prey items (i.e., birds, ground squirrels) in various parts of Tucson, and add this information to the model.
We thank G. Christopherson for leading us to the formulas listed in the Appendix, and for helpful comments during the construction of the model. We also acknowledge the logistic support of the Advanced Resources Technology Group, The University of Arizona. Finally, we appreciate K. Barnes’, A. Remen’s, and B. Pearson’s hard work in the field. This project was funded by the Arizona Game and Fish Deparment Heritage Program and The University of Arizona, College of Agriculture.
Afifi, A. A., and V. Clark. 1984. Computer-aided multivariate analysis. Lifetime Learning Publications, Belmont, California.
Andelt, W. F. 1985. Behavioral ecology of coyotes in south Texas. Wildlife Monographs 94:1-45.
Atkinson, K. T., and D. M. Shackleton. 1991. Coyote, Canis latrans, ecology in a rural-urban environment. Canadian Field-Naturalist 105:49-54.
Bekoff, M. 1982. Coyote. Pages 447-459 in J. E. Chapman and g. A. Feldhammer, eds. Wild mammals of North America. Johns Hopkins University Press, Baltimore, Maryland.
______, and M. C. Wells. 1980. The social ecology of coyotes. Scientific American 242:130-148.
______, and ______. 1986. Social ecology and behavior of coyotes. Pages 252-338 in J. S. Rosenblatt, C. Beer, M. Busnel and P. Slater, eds. Advances in the study of behavior. Academic Press, Inc. Orlando, Florida.
Bounds, D. L. 1993. Coyotes and national park units and home ranges areas and daily movements of coyotes near Saguaro National Monument. M.S. Thesis, The University of Arizona, Tucson.
Christopherson, G. L., D. P. Guertin, and K. A. Borstad. 1996. GIS and Archaeology: Using ArcInfo to increase our understanding of ancient Jordan. Proceedings, Esri Conference. Http://www.Esri.com/base/common/userconf/proc96/TO150/PAP119/P119.HTM.
Gier, H. T. 1968. Coyotes in Kansas. Kansas Agricultural Experiment Station Bulletin Number 393. Revised.
Gill, D., and P. Bonnett. 1973. Nature in the urban landscape: a study of city ecosystems. York Press, Baltimore, Maryland.
Hastings, J. R., and R. M. Turner. 1965. The changing mile. The University of Arizona Press, Tucson, Arizona.
Howell, R. G. 1982. The urban coyote problem in Los Angeles county. Vertebrate Pest Conference. 10:21-23.
Litvaitis, J. A., and J. H. Shaw. 1980. Coyote movements, habitat use, and food habits in southwestern Oklahoma. Journal Wildlife Management 44:62-68.
McClure, M. F. 1993. Densities and diets of coyotes near Saguaro National Monument. M.S. Thesis, The University of Arizona, Tucson.
Nowak, R. M. 1978. Evolution and taxonomy of coyotes and related Canis. Pages 3-16 in M. Bekoff, ed. Coyotes: biology, behavior, and management. Academic Press, New York, New York.
Quinn, T. 1991. Distribution and habitat associations of coyotes in Seattle, Washington. Wildlife Conservation in Metropolitan Environments Symposium Series 2:48-51.
_____. 1995. Using public sighting information to investigate coyote use of urban habitat. Journal of Wildlife Management 59:238-245.
Sellers, W. D., and R. H. Hill. 1974. Arizona Climate 1931-1972. The University of Arizona Press, Tucson, Arizona.
Shargo, E. S. 1988. Home range, movements, and activity patterns of coyotes (Canis latrans) in Los Angeles suburbs. Ph.D. Thesis. University of California, Los Angeles, California.
Shaw, W. W., L. K. Harris, M. Livingston, J. Charpentier, C. Wissler. 1996. Pima County Habitat Inventory Phase II. Final Report. The University of Arizona, School of Renewable Natural Resources, Tucson, Arizona.
Warren, R. E. 1990. Predictive modeling of archeological site location: a case study in the midwest. Pages 210 - 215 in Interpreting space: GIS and archaeology. London, Taylor and Francis.
White, G. C. and R. A. Garrott. 1990. Analysis of wildlife radio-tracking data. Academic Press, Inc., San Diego, California.
Woolsey, N. G. 1985. Coyote field guide. Special Report 15, Arizona Game and Fish Department, Phoenix, Arizona.
MARTHA GRINDER, Wildlife and Fisheries Sciences, School of Renewable Natural Resources, The University of Arizona, Tucson, Arizona 85721, USA. mic@u.arizona.edu
WOLFGANG GRUNBERG, Renewable Natural Resource Sciences, School of Renewable Natural Resources, The University of Arizona, Tucson, Arizona 85721, USA. Grunberg@u.arizona.edu
D. PHILLIP GUERTIN, Landscape Studies, School of Renewable Natural Resources, The University of Arizona, Tucson, Arizona 85721, USA.
PAUL R. KRAUSMAN, Wildlife and Fisheries Sciences, School of Renewable Natural Resources, The University of Arizona, Tucson, Arizona 85721, USA.