Great Basin Birds10, 2008, pp.16-25 © 2008 by the Great Basin Bird Observatory

Swainson’s Hawk nesting habitat and patterns of expected land-use change Jessi L. Brown1,3,Chris W. Briggs1 and Michael W. Collopy2 1Department

of Natural Resources and Environmental Science and Program in Ecology, Evolution, and Conservation Biology, 1000 Valley Rd., University of Nevada, Reno, NV 89512; 3E-mail: [email protected] 2Academy for the Environment, 108 Mackay Science Building, University of Nevada, Reno, NV 89557

INTRODUCTION Loss and fragmentation of suitable breeding habitat is an important factor cited as the driving force behind recent declines in global raptor populations (Wilcove et al. 1998). Although many of the larger raptors are relatively longlived so that population levels may be maintained with low levels of successful reproduction, in many cases the shortage of quality nesting sites is great enough for mortality to outpace replacement of the breeding population. Therefore, the identification of high quality breeding habitat coupled with measures to conserve those areas of critical habitat is rapidly becoming a major priority of those entities tasked with managing imperiled species. Quantitative assessments of habitat quality produced by models that are both scientifically defensible yet comprehensible by policymakers are an essential component of this approach to species conservation (King and Kraemer 1993). Such models could be applied to better understand the situation of the Swainson’s Hawk (Buteo swainsoni), a raptor listed in 1983 as threatened by the State of California in response to indications that the population had declined by as much as 90% since the 1940s (Bloom 1980). Despite this change in the bird’s legal status, the species’ long-term persistence is at-risk due to ongoing loss of nesting and foraging habitat (Estep 1989). A sizable population of Swainson’s Hawks currently inhabits the Butte Valley of extreme northern California, where cultivation of alfalfa (Medicago sativa) subsidizes large numbers of a favored prey animal, the Belding’s ground squirrel (Spermophilius belding; Whisson et al. 1999). The current pattern of land use may soon change, however, as farmers react to economic pressures by converting alfalfa fields to more profitable row crops, particularly strawberries. These row crops support low prey density, and are generally unsuitable for foraging by Swainson’s Hawks due to a lack of prey and inability to observe prey because of high vegetation density (Estep 1989). In contrast, alfalfa has been shown to be positively correlated with reproductive success of Swainson’s Hawks in Butte Valley, likely because it provides an almost ideal foraging substrate (Woodbridge 1991). Therefore, broad-scale conversion of alfalfa to row crop agriculture is expected to negatively impact the local Swainson’s Hawk population, but the magnitude of this effect is unclear. Theories of animal distribution predict that individuals should choose nest sites 16

Vol. 10 brown, briggs, and collopy

17

located close to their primary foraging areas, and nests further from foraging areas will entail increases in reproductive costs, and thus decreases in brood size (Rosenzweig 1981). For Swainson’s Hawks, Briggs (2007) did not find a strong correlation between nest success and distance to agriculture. However, Woodbridge (1991) found that when nests that failed due to disturbance or predation were excluded form the analysis, distance to alfalfa was a significant predictor of nest success. The amount of available foraging habitat also may influence reproductive success, as greater foraging area around a nest site may increase prey abundance, and the ability to provide for young (Lack 1966). We quantified the potential effects of landscape-level agricultural pattern change scenarios on the amount of Swainson’s Hawk nesting habitat in the Butte Valley. To do so, we built models assessing probability of nest site occurrence based on observations of nests gathered as part of a long-term study of the population dynamics of local Swainson’s Hawks. Locations of known nest sites were used in conjunction with remote-sensed spatial datasets to assess probability of nesting across the entire landscape. Once validated, these models were used to evaluate the effects of two potential scenarios of changes in land use. In one scenario, all alfalfa was converted to row crops, except for those holdings belonging to the two largest privately owned ranches. The second scenario converted a similar number of fields, but existing alfalfa fields were randomly chosen for conversion, resulting in a more spatially diffuse pattern of land-use change across the landscape. METHODS Study Area. We conducted our study in the Butte Valley of extreme northcentral California (49° 41’ N, 122° 00’ W), at the northwest edge of the Great Basin. Over the course of this study, the Butte Valley consisted of roughly 21% sagebrush steppe (Artemisia spp.), 22% juniper woodland (Juniperus occidentalis), and 55% agricultural land (Woodbridge 1991). The primary crop type was irrigated alfalfa, which tended to support large numbers of rodents, primarily Belding’s ground squirrels and voles (Microtus spp.). The remainder of the crops farmed included cereal grains, potatoes, strawberries, carrots, and green onions. Field Methods. Surveys for the presence of Swainson’s Hawks first began in 1979, with the goal of assessing hawk populations statewide. The population was monitored more intensively and valley-wide beginning in the 1990s. Adult hawks were monitored during the pre-breeding season (April and May) for signs of breeding behavior, such as undulating flight, copulation, and nestbuilding activity. Once pairs were identified, they were observed until precise nest sites were located. As Swainson’s Hawks build large, conspicuous stick nests, nests were generally straightforward to find. We used 1-m resolution digital ortho-quarter quadrangles (DOQQs) to plot nest site locations, and then imported the locations into ArcGIS 9.1 (ESRI 2005). Modeling Procedures. Nest site locations from 2004 (N=54) were used as training points to develop models of the probability of nest occurrence with

18 swainson's Hawk

september 2008

the ArcGIS extension ArcSDM 3.1 (Sawatsky et al. 2004). We chose a variety of spatial datasets as potential predictors for the models. Slope of the landscape was considered important, as few hawks were encountered away from the flatter valley areas, and was calculated from a digital elevation map (DEM) of the study area. Juniper tree locations were determined by using Feature Analyst to generate points representing individual juniper trees based on grayscale 1-m resolution DOQQs (Visual Learning Systems 2002). A 10-m moving window filter was then applied to produce a grid surface of juniper tree density. Agricultural fields were manually digitized from 1-m resolution DOQQs, and later classified into alfalfa or row crop fields. Raster files of distance to alfalfa and row crops were then generated containing 100-m interval classes. Land cover data (California Department of Forestry and Fire Protection 2002) at 100-m resolution was reclassified into eight categories based on dominant life-form (i.e. Agriculture, Barren/other, Conifer, Hardwood, Herbaceous, Shrub, Urban, and Water). Model Comparison and Validation. Maps of nesting habitat quality based on current conditions were compared to each other using the spatial correlation measures of Cramer’s coefficient V, the contingency coefficient C, and Kappa coefficient (Bonham-Carter 1996). We validated the models describing the current scenario of land-use by evaluating the efficiency of classification (i.e. success-rate curve or SRC) of the original 54 training points (Chung and Fabbri 2003). We then evaluated the models’ efficiency of prediction (prediction-rate curve or PRC) considering 534 nest site locations identified during the period 1992 to 2003, which had not been used to generate the models. Weights of Evidence Modeling. The weights-of-evidence methodology for analyzing interactions between training points (in this case, nest site locations) and spatial data is based on the application of Bayes’ Rule of Probability (Raines et al. 2000). Using Bayes’ Rule, a pair of weights corresponding to the presence and absence of training points is calculated for each spatial dataset. An overall measure of the importance of the spatial dataset to the model in question can be expressed as the range-in-weight values (i.e., the contrast) for that particular dataset. Confidence that the contrast deviates significantly from zero can be determined by examining the Studentized value of contrast (i.e., the ratio of the contrast to its standard deviation), following the general guideline that Studentized contrast values greater than 2.0 are significant. The weights of the spatial layers were then combined to form a response layer in which each cell contained the posterior probability of occurrence of a Swainson’s Hawk nest. All models were assessed for conditional independence by using the AgterbergCheng Conditional Independence Test (Agterberg and Cheng 2002). Expert Weights of Evidence Modeling. After a suite of weights-of-evidence models were generated using the spatial data describing the current situation in the Butte Valley, we used information from this modeling process to evaluate changes in nesting habitat quality under two different scenarios of land-use change. The weights previously calculated from one of the data-driven weightsof-evidence models were used to generate new models of nesting habitat quality in the “expert weights-of-evidence” mode (Sawatsky et al. 2004). In one scenario,

Vol. 10 brown, briggs, and collopy

19

all of the 44 alfalfa fields not controlled by the two primary landowners in Butte Valley were converted to row crop agriculture (hereafter, the Major Ranches scenario). In the other, the same number of alfalfa fields was converted to row crops, but in a randomly dispersed pattern (Dispersed scenario). RESULTS Modeling Current Conditions. We considered four candidate weights-ofevidence models (Table 1). In general, output from the multi-class models appeared more plausible than the binary model (Fig. 1). Our correlation measures showed that the multi-class models were all quite similar, as the Cramer’s V, Contingency C, and Kappa coefficients tended to be close to 1.0, whereas the binary weights-of-evidence model diverged greatly from the multi-class models (Cramer’s C varied from 0.231 to 0.420, Contingency C varied from 0.311 to 0.511, and Kappa from -0.043 to 0.012). When we evaluated model performance by examining model classification efficiency, we found in general that model efficiency did not differ between training and validation points (Fig. 2). However, the multi-class weights-of-evidence models were more efficient at classifying training points than the binary weights-of-evidence model. Projecting Land-Use Change Effects. Due to the high classification efficiency noted for the weights-of-evidence model WofE5, we chose this model as the basis for the projection of land-use change effects. Considering the Dispersed scenario, the output from the expert weights-of-evidence was very similar to the current scenario (Cramer’s V=0.987, Contingency C=0.785, and Kappa=0.877; Fig. 3). However, the map resulting from the Major Ranches scenario appeared very different from the current scenario, and was not spatially correlated with the original map (Cramer’s V=0.557, Contingency C=0.619, Kappa=0.423; Fig. 3). Under the Major Ranches scenario, the area of Likely nesting habitat decreased markedly (Fig. 4). The projected percent loss of habitat was about 14% loss under the Dispersed scenario and 57% loss under the Major Ranches scenario. Table 1. Candidate weights-of-evidence (WofE) models describing Swainson Hawk nesting habitat in Butte Valley, CA. C.I. Test Results are the Agterberg and Cheng Conditional Independence Test Probability that the model is not conditionally independent. Probability values greater than 95% or 99% indicate that the hypothesis of CI should be rejected, but any value greater than 50% indicates that some conditional dependence occurs. Model Name WofE 1 WofE 3 WofE 5 WofE Binary

Input Files Juniper, slope, and land cover Juniper, slope, land cover, and distance to alfalfa Juniper, slope, and distance to alfalfa Slope and distance to alfalfa

C.I. Test Results 58.4 67.5 57.9 50.7

DISCUSSION Projecting Land-Use Change Effects. Although conventional wisdom recognizes habitat fragmentation to be particularly detrimental to habitat quality (Fahrig 2003), our results suggest that fragmentation of Swainson’s Hawk foraging habitat at a certain scale may not substantially reduce nesting habitat quality. These results are supported by the results of Briggs (2007),

20 swainson's Hawk

september 2008

Vol. 10 brown, briggs, and collopy

who showed that reproductive success of Swainson’s Hawks was not correlated with distance to or amount of agriculture within 500 m of the nest site. With individuals traveling up to several kilometers to forage (Woodbridge 1991), it seems likely that slight fragmentation would have relatively little effect on the amount of likely nesting habitat existing in Butte Valley. The scenario of Dispersed alfalfa conversion presented here is projected to have little impact

Figure 2. Success-rate (top panel) and prediction-rate (bottom panel) curves (SRC and PRC) for weights-of-evidence models of Swainson’s Hawk nesting habitat. See Table 1 for model input parameters.

Figure 1. Weights-of-evidence models, A) WofE 1, B) WofE 3, C) WofE 5, and D) WofE Binary, designed to model occurrence of Swainson’s Hawks’ nests under current conditions. See Table 1 for model input parameters. Models classified into three classes by “natural breaks” option in ArcMap.

21

22 swainson's Hawk

september 2008

Figure 3. Projected change in Swainson’s Hawk nest occurrence according to expert weights-ofevidence models, considering juniper density, slope, and distance to alfalfa. A) Data-driven WofE 5, showing predicted occurrence of nests under current conditions. B) Expert WofE model projecting nesting habitat under Dispersed scenario. C) Expert WofE model projecting nesting habitat under Major Ranches scenario.

Vol. 10 brown, briggs, and collopy

23

on the proportion of area considered to be Likely to support hawk nests. This suggests that the remaining alfalfa fields in this scenario are easily accessible by Swainson’s Hawks due to their ability to forage across large areas of the landscape (Wiens 2000). In contrast, preserving alfalfa fields in a spatially clustered area is projected to greatly reduce the area of Likely nesting habitat. However, this may not account for the importance of intra-annual temporal variation in cutting fields to allow small mammal prey to be accessible by Swainson’s Hawks. Currently, there is relatively little fragmentation of suitable foraging habitat, with much of the cultivated valley floor growing alfalfa or lying fallow. This makes extrapolations to scenarios of increased fragmentation difficult. Some Swainson’s Hawks travel several kilometers to reach suitable alfalfa fields (Woodbridge 1991), and under the Dispersed scenario this would likely be little changed. Our analysis suggests that some loss of alfalfa fields, depending on landscape pattern, might not negatively affect the quality and spatial distribution of Swainson’s Hawk nesting habitat in the Butte Valley. However, these results should be considered preliminary, and need to be validated. Loss of alfalfa fields and their associated abundant rodent fauna may have unforeseen consequences on the interactions between competing predators. Larger raptors, such as Bald Eagles (Haliaeetus leucocephalus), Golden Eagles (Aquila chrysaetos), and Red-tailed Hawks (Buteo jamaicensis), also take advantage of the unusually high densities of prey in the Butte Valley. Reduction of this resource may stimulate greater interspecific and intraspecific competition for food, causing greater reductions in the number of Swainson’s Hawks supported. Therefore, this model may be biased by assessing nesting habitat instead of directly examining the variables that may be of the greatest concern; foraging habitat. Our models may underestimate the negative influence of row crops on Swainson’s Hawks, as row crops comprise a small proportion of the current landscape. The models also ignore several important aspects of Swainson’s Hawk ecology (e.g. density dependant effects) that could lead to decreased overall abundance even if the amount of nesting habitat remained relatively unchanged. Finally, we cannot account for potential legacy effects, in which individuals cannot track habitat change over time and remain in areas surrounded by row crops despite it being poor habitat. Future modeling efforts should take food abundance and density dependant effects into account to validate models and predict if overall hawk abundance and reproductive success would be affected. We caution against using these results to extrapolate habitat changes to the population level. ACKNOWLEDGMENTS

Figure 4. Proportion of study area designated as Likely, Neutral, or Unlikely to contain Swainson’s Hawks’ nests according to modeling scenario. Weights-of-evidence (WofE) models were classified using natural breaks.

We would like to thank Brian Woodbridge for assistance in this project and for validation points to help validate models. Other assistance was provided by Christy Cheyne and the U.S. Forest Service. This project would not have been possible without the effort of a number of volunteers, especially Brian Smucker and Stephen Wilson. Finally a great thanks to the farmers and ranchers of Butte Valley, who allowed us to search their lands for nesting hawks, particularly

24 swainson's Hawk

september 2008

Prather Ranch and DonLo Ranch. Research was supported in part by Nevada Agricultural Experiment Station, publication # 52087100.

Literature Cited Agee, J.K. 1993. Fire ecology of Pacific Northwest Forests. Island Press, Washington, D.C. Agterberg, F.P., and Q. Cheng. 2002. Conditional independence test for weights-of-evidence modeling. Natural Resources Research 11: 249-255. Bloom, P.H. 1980. The status of Swainson’s Hawks in California. California Department of Fish and Game Federal Aid in Wildlife Restoration Project. W-54-R12. Bonham-Carter, G.F. 1996. Geographic information systems for geoscientists – modeling in GIS. Elsevier Science Inc., New York, NY. Briggs, C.W. 2007. Survival and reproductive success of Swainson’s Hawks in Butte Valley, California. M.Sc. Thesis, University of Nevada, Reno, Reno, NV. California Department of Forestry and Fire Protection. 2002. Multi-source land cover data (Version 2002 v1) [Data file]. Retrieved October 15, 2006, from CDF-FRAP Web site: http://frap.cdf.ca.gov/data.html. Chung, C.F., and A.G. Fabbri. 2003. Validation of spatial prediction models for landslide hazard mapping. Natural Hazards 20: 451-472. ESRI. 2005. ArcGIS. Version 9.1. Environmental Systems Research Institute, Inc. Redlands, CA. Estep, J.A. 1989. Biology, movements, and habitat relationships of the Swainson’s Hawk in the Central Valley of California, 1986-87. California Department of Fish and Game, Nongame Bird and Mammal Section Report. Fahrig, L. 2003. Effects of habitat fragmentation on biodiversity. Annual Review of Ecological and Evolutionary Systems 34:487-515. King, J.L. and K.L. Kraemer. 1993. Models, facts, and the policy process: the political ecology of estimated truth, p. 353-360. In Environmental modeling with GIS (M. F. Goodchild, B. O. Parks and L. T. Steyaert, Eds.). Oxford University Press, New York, NY.

Vol. 10 brown, briggs, and collopy

25

Lack, D. 1966. Population studies of birds. Oxford, Oxford University Press. Raines, G.L., G.F. Bonham-Carter and L. Kemp. 2000. Predictive probabilistic  modeling using ArcView GIS [Data file]. Retrieved October 15, 2006, from ArcUser Web site: http://www.esri.com/news/ arcuser/0400/files/wofe.pdf Rosenzweig, M.L. 1981. A theory of habitat selection. Ecology 62:327-335. Sawatzky, D.L., G.L. Raines, G.F. Bonham-Carter and C.G. Looney. 2004. ARCSDM3.1: ArcMAP extension for spatial data modeling using weights of evidence, logistic regression, fuzzy logic and neural network analysis. http://www.ige.unicamp.br/sdm/ARCSDM31/. Visual Learning Systems. 2002. User manual, Feature Analyst extension for ArcView 3.2. Visual Learning Systems, Inc. Missoula, MT. Whisson, D.A., S.B. Orloff, and D.L. Lancaster. 1999. Alfalfa yield loss from Belding’s ground squirrels in northeastern California. Wildlife Society Bulletin 27: 178-183. Wiens, J.A. 2000. Ecological heterogeneity: an ontogeny of concepts and approaches, p. 9-31. In The ecological consequences of heterogeneity (M. J. Hutchin, E. A. John, and A. J. A. Stewart Eds.). Blackwell Science. Oxford, U.K. Wilcove, D.S., D. Rothstein, J. Dubow, A. Phillips, and E. Losos. 1998. Quantifying threats to imperiled species in the United States. BioScience 48:607-615. Woodbridge, B. 1991. Habitat selection by nesting Swainson’s hawk: a hierarchical approach. M. Sc. Thesis, Oregon State Univ., Corvallis, OR.