Assessing range-wide habitat suitability for the Lesser Prairie-Chicken
 
Évaluation de la qualité d'habitat dans l'ensemble de l'aire du Tétras pâle

• Abstract
• Introduction
• Methods
  • Data sets
  • Modeling
• Results
• Discussion
  • Management implications
• Conclusion
• Responses to this article
• Acknowledgments
• Literature Cited

INTRODUCTION

Land-use change resulting in habitat loss is one of the primary factors affecting population declines of many wildlife species (Morrison et al. 2007). Minimization of the effects of land development and land-use change on species’ conservation may be possible if conservation planning can be included in development decisions. However, this is only possible if data to inform conservation needs are available at the time decisions are being made. Models of species distribution and habitat suitability can be created using fewer than 25 locations to successfully guide future field surveys (Pearson et al. 2007) and can inform conservation decisions related to land-use conversion (Marini et al. 2009, Thorn et al. 2009).

In general, logistical constraints and sparse, clustered, or observational (rather than probabilistically sampled) data often limit statistical analyses that can be conducted and the interpretive value of results for rapid conservation management decisions. Thus, decisions are often based on models made with readily available data, ranging from presence-only to binary models (presence-absence) to count-based models (abundance). Models based on nonrandom data may be misleading because of sampling biases, such as nonrandom road surveys (Elith et al. 2011). Nonrandom presence-only data are often all that are available, especially for rare species, and when implemented carefully can lead to useful models as indicated in the application mentioned above. Maxent, a machine learning method that uses the principal of maximum entropy to identify relationships between available environment and observed locations (Phillips et al. 2006, Phillips and Dudik 2008), is part of a suite of techniques known as species distribution modeling, and identifies relationships between observations of species and the available environment to predict habitat suitability at unknown locations (Guisan and Thuiller 2005, Franklin 2010).

The Lesser Prairie-Chicken (Tympanuchus pallidicinctus) is a prairie grouse species distributed on the Great Plains of the United States, including the states of Colorado, Kansas, New Mexico, Oklahoma, and Texas. Lesser Prairie-Chickens have experienced an estimated 90% reduction in range since the 1800s (Taylor and Guthery 1980, Hagen et al. 2004) and evidence suggests population changes vary greatly across the region (Holt 2012, McDonald et al. 2014, Garton et al. 2016). As a result, the Lesser Prairie-Chicken was petitioned for protection under the Endangered Species Act in 1995, and the species was listed as threatened across its range in May 2014 (U.S. Fish and Wildlife Service 2014). Biologists cite the loss of native prairie as the main cause of decline for Lesser Prairie-Chickens. Among other things, losses can occur from grazing practices resulting in reduced vegetation structure and woody vegetation encroachment, which fragment and deteriorate habitat (Hagen and Giesen 2005).

In addition to loss of native habitat, an emerging and potentially critical threat is energy development within the current distribution of the Lesser Prairie-Chicken (Pruett et al. 2009, Hagen et al. 2011, Jarnevich and Laubhan 2011). The North American Great Plains is currently undergoing rapid land-use change as a result of increasing energy development throughout the region (Allred et al. 2015, American Wind Energy Association 2015). This development has the potential to impact the habitat of species in the region, and cause declines or hasten declines of species’ populations. Several recent studies have documented avoidance of anthropogenic structures and human disturbance by prairie grouse species (Centrocercus and Tympanuchus spp.; Holloran 2005, Pitman et al. 2005, Walker et al. 2007, Pruett et al. 2009, Grisham et al. 2014). The number of Greater Sage-Grouse (C. urophasianus) males displaying at leks decreased with increasing natural gas field-related disturbances around leks (Holloran 2005, Walker et al. 2007). Additionally, male attendance at leks and the number of active Sage-Grouse leks declined at a faster rate within natural gas fields compared with areas outside natural gas fields (Holloran 2005, Walker et al. 2007). In Kansas and Texas, Lesser Prairie-Chicken nests were located further than expected from transmission lines, improved roads, and oil or gas wellheads even though otherwise-suitable habitat surrounded these features (Pitman et al. 2005, Grisham et al. 2014). Pruett et al. (2009) examined the avoidance behavior of Lesser and Greater Prairie-Chickens (T. cupido) to power lines and highways in Oklahoma and found birds avoided the power lines and few nests were found within 2 km of the power lines. Other studies in northcentral Kansas, at the center of the Greater Prairie-Chicken extant range, found that although nest site selection and survival were not negatively affected by proximity to wind turbines (McNew et al. 2014), females avoided wind turbines that could lead to local extirpation of the species in proximity to wind turbines (Winder et al. 2014). In Texas, Timmer et al. (2014) found the density of Lesser Prairie-Chicken leks decreased as the density of active oil and gas wells and paved roads increased.

The distribution and population of Lesser Prairie-Chickens have been monitored by state wildlife biologists and managers through lek surveys, but until recently a considerable amount of interstate variation existed in survey methodologies (D. M. Davis, R. E. Horton, E. A. Odell, R. D. Rodgers, and H. A. Whitlaw 2008, unpublished manuscript). Nesting habitat is considered a factor limiting population size (Wisdom and Mills 1997), although nesting site data are only available from a few, localized research studies. On the other hand, landscape features indicative of lekking activity are not thought to limit the population, but lek data is available across much more of the species’ range. Breeding habitats are closely associated with lek sites (Hagen and Giesen 2005, Pitman et al. 2006), and leks have been used as a surrogate for nesting habitat over broad spatial extents in the past (Jarnevich and Laubhan 2011). Given that lek data are the only somewhat consistent data available range wide, we chose to focus our modeling efforts on this single life history event. Even so, the available data for Lesser Prairie-Chicken management decisions vary in terms of survey effort, methodology, and spatial coverage, and many are derived from nonrandom sampling. For example, some states have set routes that are monitored with a set protocol, while others visit historic and currently known lek locations (see Van Pelt et al. 2013 for detailed descriptions of each states’ monitoring efforts). Until recently, surveys also lacked specific absence data or information on survey extents and effort. Therefore, Maxent is an appropriate technique to estimate relationships between environmental characteristics and lek occurrence because much of the available data for Lesser Prairie-Chicken populations consist of presence-only locations from nonrandom samples with an unknown sampling frame.

The objectives of our study were to predict habitat suitability for Lesser Prairie-Chicken leks and to explore relationships of occurrence with landscape characteristics and anthropogenic effects that may influence their distribution. We used Maxent to develop habitat suitability models for Lesser Prairie-Chicken leks using existing data merged from multiple collection efforts throughout the current occupied range. We then applied these models to all counties intersecting the 2012 estimated occupied range of the Lesser Prairie-Chicken (Van Pelt et al. 2013) to identify areas of higher habitat suitability for the species. Our overarching goal in reaching these objectives was to inform management decisions related to identifying locations for habitat conservation and developing a habitat conservation program for Lesser Prairie-Chickens.

METHODS

Our study extent included the 2012 estimated occupied range of the Lesser Prairie-Chicken buffered by 16.1 km as defined by the Lesser Prairie-Chicken Interstate Working Group and mapped on the Southern Great Plains Crucial Habitat Assessment Tool web site (Fig. 1; Van Pelt et al. 2013). We expanded the area to county boundaries to match political jurisdictions for management applications. This resulted in a 28,420,417 ha area, covering 89 counties in Colorado, New Mexico, Kansas, Oklahoma, and Texas.

Data sets

All five states encompassing the historic range cooperated in an effort to compile lek location data from across the species’ range. We selected lek locations from each state that had been observed between 2002 and 2012 to limit issues related to changes in land use since the time of observation. The definition of a lek, i.e., number of birds required to be present at the time of observation, varied by state, and given that some locations were opportunistic and not revisited we did not require leks to be observed in multiple years. We removed lek locations within 100 m of each other from different years and used only the most recent location to avoid pseudo-replication in the data. Because of potential issues related to coordinate accuracy as reported by the states collecting data, we dropped any locations that state representatives, headed by the coauthor from each state agency, believed had a spatial accuracy less than 100 m, i.e., removed leks with location data uncertainty such that it may have been assigned to an incorrect grid cell. We developed the lek model using a grid cell resolution of 210 m. These criteria resulted in 1402 unique locations including 76 in Colorado, 669 in New Mexico, 189 in Kansas, 185 in Oklahoma, and 283 in Texas. The sample size difference by state results from both the actual distribution of leks on the landscape, i.e., real differences in lek occupancy that exist between states, and artifacts from different historical survey methodologies discussed above and availability of data (sampling bias and privacy issues).

Potential predictor variables were chosen based on the life history of the species, and included land-cover metrics, topography, and anthropogenic features for a total of 15 predictors (Table 1). Our models were partially limited by the availability of consistent quality land-cover data throughout the range. We used three different sets of land-cover data based on which land-cover data set was determined as best for each state by the state representatives. We began with National Land Cover Dataset (NLCD) vegetation classes for all states (Fry et al. 2011), focusing on shrubland, grassland/herbaceous, and pasture/hay classes as important, but because of errors in this dataset we replaced it for Texas and New Mexico. For Texas, we used the Texas Ecological Systems Classification Project data set (Elliot et al. 2014); for New Mexico, we used the southwest regional gap analysis project land cover (USGS National Gap Analysis Program 2004). We also included a metric quantifying the amount of land in the conservation reserve program (CRP), because this land-use type has been important in previous models (Jarnevich and Laubhan 2011).

Because of the large number of land-cover classes, the quality of the classifications at a fine thematic scale, and our interest in the spatial context of location, experts from each state classified their land cover into known or suspected used and unused classes based on annual lek surveys and monitoring data (see Appendix 1 for more details and a list of the classes). To explore what landscapes may be important to the occurrence of leks, we created vegetation predictors of percent known used, percent suspected used, and percent CRP summarizing a 1600 m diameter neighborhood and a 5000 m diameter neighborhood around each location. These distances cover the range in reported area around leks used by Lesser Prairie-Chickens (newer estimate around 1.5 km from Pitman et al. 2006, Boal and Pirius 2012, and Grisham et al. 2014 to older estimates of 4.8 km used by Fuhlendorf et al. 2002). Known used consisted of vegetation broadly classified as “shrubland, steppe and savanna systems” while suspected used consisted of vegetation broadly classified as “grassland systems” (Appendix 1). Introduced grasses also contributed to the two categories. We collapsed known used and suspected used into a single predictor after preliminary runs, resulting in a single land-cover variable represented by percent of area within 5000 m with land cover classified as either known used or suspected used (see Appendix 2). We created predictors for well density using counts of active wells in an 800 m and 1600 m diameter neighborhood around the locations, which we later restricted to a 1600 m area based on preliminary runs (Table 1; see Appendix 2). These distances were based on expert opinion and the set back recommendations from anthropogenic features in Hagen et al. (2011). Other anthropogenic variables included distance to active wells, roads, highways, transmission lines, tall structures, and landscape condition (a measure of human impact integrating several anthropogenic factors; Table 1). Additional predictors included topographic ruggedness index (a measure of topographic heterogeneity in the region around a focal cell), state, and enhanced vegetation index (EVI; a spectral vegetation index that calculates photosynthetically active vegetation while accounting for effects from atmospheric and soil influences; Huete et al. 2002). All predictors were created at a 210 m resolution to match decisions regarding the minimum mapping unit for modeling leks as described in Table 1.

Modeling

To develop species distribution models, we used Maxent (version 3.3.3k). Maxent requires presence locations of a species, background or pseudo-absence locations representative of the sampled environment, and environmental predictors. It utilizes the principle of maximum entropy to determine statistical relationships between presence locations and the environment by comparing the environment where an organism is found to the available environment. We implemented Maxent within the Software for Assisted Habitat Modeling (SAHM; version 1.0; Morisette et al. 2013).

Predictor importance and response curves in Maxent are sensitive to cross-correlations, so we removed variables with high correlations of |r|>0.7 using the maximum of the Pearson’s, Spearman’s, and Kendall’s correlation coefficients (Dormann et al. 2013). The three methods use different ways to identify correlations, which is useful given that Maxent utilizes linear and nonlinear relationships. We retained the variable from correlated pairs that was thought to be most directly related to Lesser Prairie-Chicken presence, indicating which predictor was retained and why others were dropped in Table 1.

All five states contributed lek data with varying degrees of completeness because of landowner confidentiality requirements. Colorado and Kansas provided all known lek locations that were collected with a GPS unit, and some locations were omitted for Oklahoma, New Mexico, and Texas. However, we do not know the exact number of locations or the specific areas of omitted data within these states. Thus, we limited the random selection of the 10,000 background locations to within 10 km of a recorded lek so that we did not sample areas outside the known sampled area (a 3,672,555 ha area; Phillips et al. 2009, VanDerWal et al. 2009). We chose 10 km because it extended the available area to several pixels beyond each presence location, allowing for enough environmental variation between presence locations and background locations to produce a meaningful model while still minimizing the inclusion of potentially unsampled areas. The number of background points required to adequately represent the available environment is estimated to be 10,000 (Phillips and Dudik 2008, Barbet-Massin et al. 2012).

We fit the lek model 25 times, withholding a different random 30% of presence locations during each of the 25 replicate model runs. We also set the maximum number of iterations to 5000 to allow models to converge. Otherwise we used the default settings for Maxent. However, in preliminary models we noticed signs of overfitting in the models including very complex response curves and large differences between receiver operating characteristic area under the curve (AUC) values calculated for training and testing data (i.e., > 0.05). We therefore tested alternate values of the regularization parameter that controls model complexity in Maxent and chose a value of two for subsequent runs. We ran models that both included and excluded state as a predictor to account for the potential differences in sample sizes between states mentioned above. Models including state assumed that differences in number of observed leks were an artifact of sampling bias as described above, and the state variable was meant to account for this difference. Models excluding state assumed that these differences reflected actual differences in occupancy, i.e., sample size differences arose because of environmental conditions differing between states. Because truth is probably some combination of these hypotheses, the two model scenarios provided bounds around the expected answer.

We evaluated models by examining the AUC value for the test data, which generally ranges between 0.5 and 1 and is a measure of discrimination ability of the model (Fielding and Bell 1997). Values less than 0.5 are no better than random, values between 0.5 and 0.7 are rather low accuracy, values between 0.7 and 0.9 are useful for some purposes, and values above 0.9 represent high accuracies (Swets 1988). Maxent provides two different evaluations of variable contribution to models. Permutation importance is calculated on the converged model only by examining change in AUC when randomly permuting the values among the presence and background data for each variable while holding the other variables constant. This metric provides a model-independent measure of the relative influence of each predictor in each model. The second contribution criterion is variable contribution, which is calculated based on the additive regularized training gain (positive addition or negative subtraction) at each iteration of the algorithm as it reaches convergence. Maxent also produces a multidimensional environmental similarity surface (MESS) when projecting the model onto new locations, e.g., such as from our constrained background locations to counties encompassing the historic range. The MESS surface includes increasingly negative values as the environmental conditions at the new location depart from those used in developing the model by comparing the range in values for each environmental variable at the locations used to develop the model (presence and background) to the value at the new location (Elith et al. 2010). We used this surface as a measure of uncertainty in predictions, classifying any negative value as a location with novel environmental conditions, i.e., conditions that were outside the range of parameters used to train the model.

To compare model results and provide a simplified product, we discretized the model predictions into suitable and unsuitable classes using three threshold rules produced in Maxent. The rules included the minimum training presence, which determines the minimum predicted value for any presence location used to train the model to use as the threshold; five percentile training presence threshold, which orders the predicted values for the presence locations used to train the model and selects the value that would misclassify the bottom 5%; and the 10 percentile training threshold, which misclassifies the bottom 10%. Because Maxent produces maps with a continuous index of relative habitat suitability rather than a probability, using these thresholds highlight areas in four different classes of relative suitability, with the 10% being the highest class followed in order by 5%, minimum training presence, and unsuitable.

RESULTS

As expected, several predictor variables were correlated. Our reduced, uncorrelated set included 11 predictors (Table 1; Appendix 3). The lek model performed well with an average test AUC value of 0.79 (Table 2). The lek locations used to train the model captured much of the available environment in the region of interest, with only 2.7% of the study area containing novel environments. New Mexico and Oklahoma contributed most to the novel area (Figs. 2 and 3).

For the model including state, relative ranking of the top two important predictors was the same regardless of method used to calculate importance, with state and land cover contributing at least 50% (Table 3). State was the most important predictor in the model, with the New Mexico category, which also had the largest sample size, associated with the greatest lek suitability. The univariate response for the other state categories were similar, but the Colorado response was associated with higher suitability when the marginal effect was calculated (varying one variable while holding all others constant at their mean value). Topographic ruggedness was also a high predictor, ranking third according to permutation importance, and closely followed by average EVI. With removal of state, topographic ruggedness index and landscape condition increased in importance, forming the top three predictors along with land cover and again followed by EVI. In both models there was a general trend of increasing suitability with increasing percentage of known or suspected used land cover, and a decreasing suitability with increasing topographic ruggedness.

Anthropogenic features also contributed to the models. In the model with state, all anthropogenic predictors except landscape condition had a relative contribution of less than 5% based on permutation importance. However, if viewed in aggregate their contribution is 13.2% (sum of highways, transmission lines, Federal Aviation Administration structures, and well density). With removal of state, importance of anthropogenic features besides landscape condition increased, with an additional four features having contributions > 5% and a total sum of 27.9%. Of the anthropogenic features, landscape condition was most important, followed by transmission lines. Examining the response curves (data not shown), landscape condition had a sharp increase in suitability followed by a slow, noisy decrease. Distance from transmission lines, highways, secondary roads, and Federal Aviation Administration structures followed a similar pattern, with a sharp increase in suitability as distance increased (up to roughly 2.5 km [with a noticeable drop in steepness around 150 m], 1.2 km, 3 km, and 23 km, respectively) followed by a more variable pattern dependent on the specific variable. Active well density had a negative relationship with suitability, with a small decrease with even one well followed by a very precipitous decline in suitability with more than 5 wells in the 1600 m area around a lek. Variable importance as measured by a change in training gain using a jackknife test was similar for both training and testing data.

The habitat suitability index for the study area discriminated areas within each state as having relatively high suitability compared with other areas within our study region (Fig. 2a). As expected, the three different thresholds we examined to discretize the continuous habitat suitability index to suitable and unsuitable habitat classifications differed. The minimum training presence classified 86% of the study area as suitable, while the 10 percentile threshold decreased this amount to 28% of the study area (Tables 2 and 4, Figs. 2b and 3b). There were minimal differences in spatial predictions when state was removed (Tables 2 and 4, Figs. 2a and 3a).

Examining the model predictions for each state, Colorado and Oklahoma always had the lowest area of predicted suitable habitat, but the percent of the study area for the state that was suitable was high (93 and 90%, respectively; Table 4). With the more conservative threshold values (5 and 10 percentile), Texas and New Mexico had the largest amount of suitable habitat. Comparing the two models, Colorado, Kansas, and Texas all had increased area predicted as suitable when state was removed (Appendix 4).

DISCUSSION

Given the threats facing declining Lesser Prairie-Chicken populations (U.S. Fish and Wildlife Service 2014), there is a recent interest in the use of spatial models to relate landscape features with density or occurrence or to identify suitable habitat (Gregory et al. 2011, Jarnevich and Laubhan 2011, Timmer et al. 2014). These models are particularly useful when there is a need for science-based decisions to balance energy development and habitat requirements for species of conservation concern, such as Lesser Prairie-Chickens (Jarnevich and Laubhan 2011). Maximum entropy models estimate the statistical relationship between the environment where a species occurs, i.e., presence-only data, and the available environment (Elith et al. 2011). Gregory et al. (2011) developed hierarchical entropy models to categorize Greater Prairie-Chicken lek suitability in eastern Kansas to target areas for conservation efforts. Jarnevich and Laubhan (2011) also developed Maxent models of environmental and anthropogenic features to predict relative habitat suitability for Lesser Prairie-Chicken leks in Kansas to guide energy development. Timmer et al. (2014) developed spatially-explicit models, based on hierarchical distance sampling, to relate Lesser Prairie-Chicken lek density with anthropogenic and vegetative features in Texas. Previous models of prairie-chicken lek habitat have been built for a subset of the entire range (e.g., Jarnevich and Laubhan 2011, Timmer et al. 2014), but produced similar results to our range-wide assessment. All models highlighted the importance of anthropogenic features in predicting habitat suitability for Lesser Prairie-Chicken leks. Further, range-wide models such as ours identify species relationships with biotic and abiotic variables and their response to disturbance across the entire range, providing important information of range-wide patterns for conservation and management.

Although state was the most important predictor when included, our results suggest vegetation type (percent used or suspected used) was important as expected in the lek suitability model. Including state as a predictor did not greatly alter model results, and its importance was likely partly driven by sampling bias. These results are not surprising because Lesser Prairie-Chickens conduct most of their daily activities and complete their life cycle within 1.5 km of known leks (Pitman et al. 2006, Boal and Pirius 2012, Grisham et al. 2014). Therefore, lek suitability should increase in landscapes that contain high amounts of vegetation classes known to be used by Lesser Prairie-Chickens, i.e., shrubland or grassland systems, for other life stages such as those described by Hagen et al. (2013). Woodward et al. (2001) found suitable composition of vegetation in a 4.8 km area around leks varied between New Mexico, Texas, and Oklahoma for Lesser Prairie-Chicken populations, but had similar patterns of vegetation and land use. Here, we found that suitability increased with increasing amounts of known or suspected used land cover within a 5000 m around a lek. Landscapes, on average, comprised 86.5% native vegetation, which generally consisted of shrubland though the amount varied by state (Woodward et al. 2001). In Texas, Timmer et al. (2014) found percent grassland, total percent of grassland and shrubland, paved road density, and active oil and gas well density were the best predictors of lek density. They observed an inverse relationship between the anthropogenic variables and lek density. However, more complex relationships were observed for the vegetative variables, e.g., a quadratic relationship for percent grassland that varied by region. In Cochran County, Texas, lek density was highest in native shinnery oak (Quercus havardii) rangeland interspersed with some cultivated land (5-37% grain sorghum fields; Crawford and Bolen 1976).

Although avoidance of anthropogenic features has been demonstrated in Kansas, Oklahoma, and Texas (Robel et al. 2004, Pitman et al. 2005, Pruett et al. 2009, Hagen et al. 2011, Grisham et al. 2014), the relative impact of these features remains unknown across the remainder of the Lesser Prairie-Chicken’s distribution. In Kansas, nest sites were located further from utility lines, buildings, and improved roads than expected at random (Pitman et al. 2006). Pruett et al. (2009) found few Lesser Prairie-Chicken nests within 2 km of a power line in an Oklahoma study and only one Greater Prairie-Chicken nest within 2 km of the power line. Hagen et al. (2011) found a general pattern of avoidance to anthropogenic features in monthly home ranges in Kansas, and a before-after-control-impact design revealed that Lesser Prairie-Chicken monthly use areas were less likely to include utility lines. Although the causative agents behind avoidance of anthropogenic features remain unknown, it may be due to the functional elimination of suitable habitat (Robel et al. 2004) through general avoidance of noise, the potential for predators to perch on features, or a neophobic response to these features from evolving on a landscape devoid of tall structures. Our models provide evidence that these factors are important across the entire range of Lesser Prairie-Chickens, with a positive relationship between suitability and distance to features over distances examined in the above studies.

The Lesser Prairie-Chicken data available to create the models had several biases related to sampling. We attempted to control for some of the problems by limiting selection of background points, but the bias may still have affected model results. By limiting background point selection to areas we knew were sampled, we left out areas on the edges of the historic range that could still potentially provide some habitat. Inclusion of these other areas could potentially alter model results and variable relationships. Several different methods of assessing uncertainty in the model can be used to guide model usage. Locations where the two models did not overlap, had high standard deviation among replicate runs, or had novel environments, e.g., portions of New Mexico and Oklahoma, would be good places to target sampling to improve future modeling efforts (Crall et al. 2013). Predictor variables also had uncertainty, such as the three land cover data sets that were merged to create the land cover predictor. These were created at different points in time, which may affect model results. However, the NLCD evaluation, visualization and analysis tool indicated minimal land cover change for each state within our study region for 2001 to 2011 (0.46% of Kansas, 3.62% of Texas, 2.12% of Oklahoma, 1.34% of Colorado, and 0.95% of New Mexico; https://www.sciencebase.gov/catalog/item/541369e9e4b0239f1986bcc6). By collapsing to two categories (known or suspected used versus not used), we may have minimized some of these land cover differences. All three data sets were created during our sampling time frame.

Recently a range-wide aerial survey for Lesser Prairie-Chickens has been developed and is now operational (McDonald et al. 2014). As these data continue to be collected, they could provide the basis for better species distribution models which could incorporate abundance information to move beyond simple habitat suitability models. The data have already been used to estimate probability of occupancy range-wide (L. McDonald, F. Hornsby, T. Ritz, and G. Gardner 2013, unpublished data). Additionally, some nest data have been aggregated across the five state range and could be used to develop models for nesting habitat suitability. Adding another life history event would provide another layer of information that managers could use to inform decision making.

Management implications

Loiselle et al. (2003) cautioned against overpredicting suitable habitat because this may misguide conservation efforts. Using a higher threshold value such as 5 or 10 percentile training presence could minimize false-positive errors by limiting locations in less optimal habitat. However, these thresholds misclassify some known leks as being in unsuitable habitat. These models can then be used for targeting of conservation actions such as easements and management contracts. Using the moderate 5 percentile threshold, more than 50% of the study area was classified as unsuitable and focusing development within these areas could potentially minimize impact to Lesser Prairie-Chickens. The three thresholds presented provide a way to rank habitat suitability depending on management objectives, with the 10 percentile threshold highlighting areas with the highest relative suitability compared to the others.

These results have been integrated with other information regarding Lesser Prairie-Chicken habitat including focal areas determined by teams in each state as priority locations with intact suitable habitat (with suitability based on the Maxent models), habitat corridors, and the estimated occupied range based on state surveys to produce a coarse scale map to classify the historic range into four classes of conservation priority. This tool, the Southern Great Plains Crucial Habitat Assessment Tool (http://kars.ku.edu/maps/sgpchat/), can be used to guide siting of development and targeting of conservation. This tool is a component of the Western Association of Fish and Wildlife Agencies mitigation framework as described in the Lesser Prairie-Chicken Range-wide Conservation Plan (Van Pelt et al. 2013).

CONCLUSION

These models represent a collaborative effort across management agencies to work toward the conservation of a threatened species. We used existing data for a single, well-observed range-wide habitat use, lekking, to make inferences about general habitat requirements for a species. These results provide information that can be used to meet management objectives and potentially guide further sampling efforts in the absence of other information. Information on habitat use for other life stages could be useful in better quantifying seasonal habitat needs (D’Elia et al. 2015). These analyses represent a means to utilize existing data from multiple sources for a single use to make inferences regarding locating energy development and conservation efforts.

LITERATURE CITED

Allred, B. W., W. K. Smith, D. Twidwell, J. H. Haggerty, S. W. Running, D. E. Naugle, and S. D. Fuhlendorf. 2015. Ecosystem services lost to oil and gas in North America. Science 348:401-402. http://dx.doi.org/10.1126/science.aaa4785

American Wind Energy Association. 2015. U.S. Wind industry fourth quarter 2014 market report. American Wind Energy Association, Washington, D.C., USA. http://awea.files.cms-plus.com/4Q2014%20AWEA%20Market%20Report%20Public%20Version.pdf

Barbet-Massin, M., F. Jiguet, C. H. Albert, and W. Thuiller. 2012. Selecting pseudo-absences for species distribution models: how, where and how many? Methods in Ecology and Evolution 3:327-338. http://dx.doi.org/10.1111/j.2041-210X.2011.00172.x

Boal, C. W., and N. E. Pirius. 2012. Winter ecology and habitat use of Lesser Prairie-Chickens in west Texas, 2008-11. Open-File Report 2012-1073. U.S. Geological Survey Reston, Virginia, USA.

Crall, A. W., C. S. Jarnevich, B. Panke, N. Young, M. Renz, and J. Morisette. 2013. Using habitat suitability models to target invasive plant species surveys. Ecological Applications 23:60-72. http://dx.doi.org/10.1890/12-0465.1

Crawford, J. A., and E. G. Bolen. 1976. Effects of land use on Lesser Prairie Chickens in Texas. Journal of Wildlife Management 40:96-104. http://dx.doi.org/10.2307/3800160

D’Elia, J., S. M. Haig, M. Johnson, B. G. Marcot, and R. Young. 2015. Activity-specific ecological niche models for planning reintroductions of California Condors (Gymnogyps californianus). Biological Conservation 184:90-99. http://dx.doi.org/10.1016/j.biocon.2015.01.002

Dormann, C. F., J. Elith, S. Bacher, C. Buchmann, G. Carl, G. Carré, J. R. G. Marquéz, B. Gruber, B. Lafourcade, P. J. Leitão, T. Münkemüller, C. McClean, P. E. Osborne, B. Reineking, B. Schröder, A. K. Skidmore, D. Zurell, and S. Lautenbach. 2013. Collinearity: a review of methods to deal with it and a simulation study evaluating their performance. Ecography 36:27-46. http://dx.doi.org/10.1111/j.1600-0587.2012.07348.x

Elith, J., M. Kearney, and S. Phillips. 2010. The art of modelling range-shifting species. Methods in Ecology and Evolution 1:330-342. http://dx.doi.org/10.1111/j.2041-210X.2010.00036.x

Elith, J., S. J. Phillips, T. Hastie, M. Dudík, Y. E. Chee, and C. J. Yates. 2011. A statistical explanation of MaxEnt for ecologists. Diversity and Distributions 17:43-57. http://dx.doi.org/10.1111/j.1472-4642.2010.00725.x

Elliot, L. F., D. D. Diamond, C. D. True, C. F. Blodgett, D. Pursell, D. German, and A. Treuer-Kuehn. 2014. Ecological mapping systems of Texas: summary report. Texas Parks and Wildlife Department, Austin, Texas.

Fielding, A. H., and J. F. Bell. 1997. A review of methods for the assessment of prediction errors in conservation presence/absence models. Environmental Conservation 24:38-49. http://dx.doi.org/10.1017/S0376892997000088

Franklin, J. 2010. Mapping species distributions: spatial inference and prediction. Cambridge University Press, New York, New York, USA. http://dx.doi.org/10.1017/cbo9780511810602

Fry, J., G. Xian, S. Jin, J. Dewitz, C. Homer, L. Yang, C. Barnes, N. Herold, and J. Wickham. 2011. Completion of the 2006 National Land Cover Database for the Conterminous United States. Photogrammetric Engineering and Remote Sensing 77:858-864.

Fuhlendorf, S. D., A. J. W. Woodward, D. M. Leslie, and J. S. Shackford. 2002. Multi-scale effects of habitat loss and fragmentation on Lesser Prairie-Chicken populations of the U.S. Southern Great Plains. Landscape Ecology 17:617-628. http://dx.doi.org/10.1023/A:1021592817039

Garton, E. O., C. A. Hagen, G. M. Beauprez, S. C. Kyle, J. C. Pitman, D. S. Schoeling, and W. E. Van Pelt. 2016. Population dynamics of the Lesser Prairie-Chicken. Pages 49-76 in D. A. Haukos, and C. W. Boal, editors. Ecology and conservation of Lesser Prairie-Chicken. CRC, Boca Raton, Florida, USA.

Gregory, A., L. B. McNew, T. J. Prebyl, B. K. Sandercock, and S. M. Wisely. 2011. Hierarchical modeling of lek habitats of Greater Prairie-Chickens. Pages 21-32 in B. K. Sandercock, K. Martin, and G. Segelbacher, editors. Ecology, conservation, and management of grouse. University of California Press, Berkeley, California, USA.

Grisham, B. A., P. K. Borsdorf, C. W. Boal, and K. K. Boydston. 2014. Nesting ecology and nest survival of Lesser Prairie-Chickens on the southern high plains of Texas. Journal of Wildlife Management 78:857-866. http://dx.doi.org/10.1002/jwmg.716

Guisan, A., and W. Thuiller. 2005. Predicting species distribution: offering more than simple habitat models. Ecology Letters 8:993-1009. http://dx.doi.org/10.1111/j.1461-0248.2005.00792.x

Hagen, C. A., and K. M. Giesen. 2005. Lesser Prairie-Chicken (Tympanuchus pallidicinctus). In A. Poole, editor. The birds of North America online. Cornell Lab of Ornithology, Ithaca, New York, USA. [online] URL: http://bna.birds.cornell.edu/review/species/364

Hagen, C. A., B. A. Grisham, C. W. Boal, and D. A. Haukos. 2013. A meta-analysis of Lesser Prairie-Chicken nesting and brood-rearing habitats: implications for habitat management. Wildlife Society Bulletin 37:750-758. http://dx.doi.org/10.1002/wsb.313

Hagen, C. A., B. E. Jamison, K. M. Giesen, and T. Z. Riley. 2004. Guidelines for managing Lesser Prairie-Chicken populations and their habitats. Wildlife Society Bulletin 32:69-82. http://dx.doi.org/10.2193/0091-7648(2004)32[69:GFMLPP]2.0.CO;2

Hagen, C. A., J. C. Pitman, T. M. Loughin, B. K. Sandercock, and R. J. Robel. 2011. Impacts of anthropogenic features on Lesser Prairie-Chicken habitat use. Pages 63-75 in B. K. Sandercock, K. Martin, and G. Segelbacher, editors. Ecology, conservation, and management of grouse. University of California Press, Berkeley, California, USA.

Holloran, M. 2005. Greater Sage-Grouse (Centrocercus urophasianus) population response to natural gas field development in western Wyoming. University of Wyoming, Laramie, Wyoming, USA.

Holt, R. D. 2012. Breading season demographics of a Lesser Prairie-Chicken (Tympanuchus pallidicinctus) population in the northeastern Texas panhandle. Texas Tech University, Lubbock, Texas, USA.

Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao, and L. G. Ferreira. 2002. Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sensing of Environment 83:195-213. http://dx.doi.org/10.1016/S0034-4257(02)00096-2

Jarnevich, C. S., and M. K. Laubhan. 2011. Balancing energy development and conservation: a method utilizing species distribution models. Environmental Management 47:926-936. http://dx.doi.org/10.1007/s00267-011-9651-2

Loiselle, B. A., C. A. Howell, C. H. Graham, J. M. Goerck, T. Brooks, K. G. Smith, and P. H. Williams. 2003. Avoiding pitfalls of using species distribution models in conservation planning. Conservation Biology 17:1591-1600. http://dx.doi.org/10.1111/j.1523-1739.2003.00233.x

Marini, M. Â., M. Barbet-Massin, L. E. Lopes, and F. Jiguet. 2009. Major current and future gaps of Brazilian reserves to protect Neotropical savanna birds. Biological Conservation 142:3039-3050. http://dx.doi.org/10.1016/j.biocon.2009.08.002

McDonald, L., G. Beauprez, G. Gardner, J. Griswold, C. Hagen, F. Hornsby, D. Klute, S. Kyle, J. Pitman, T. Rintz, D. Schoeling, and B. Van Pelt. 2014. Range-wide population size of the Lesser Prairie-Chicken: 2012 and 2013. Wildlife Society Bulletin 38:536-546. http://dx.doi.org/10.1002/wsb.417

McNew, L. B., L. M. Hunt, A. J. Gregory, S. M. Wisely, and B. K. Sandercock. 2014. Effects of wind energy development on nesting ecology of Greater Prairie-Chickens in fragmented grasslands. Conservation Biology 28:1089-1099. http://dx.doi.org/10.1111/cobi.12258

Morisette, J. T., C. S. Jarnevich, T. R. Holcombe, C. B. Talbert, D. Ignizio, M. K. Talbert, C. Silva, D. Koop, A. Swanson, and N. E. Young. 2013. VisTrails SAHM: visualization and workflow management for species habitat modeling. Ecography 36:129-135. http://dx.doi.org/10.1111/j.1600-0587.2012.07815.x

Morrison, J. C., W. Sechrest, E. Dinerstein, D. S. Wilcove, and J. F. Lamoreux. 2007. Persistence of large mammal faunas as indicators of global human impacts. Journal of Mammalogy 88:1363-1380. http://dx.doi.org/10.1644/06-MAMM-A-124R2.1

Pearson, R. G., C. J. Raxworthy, M. Nakamura, and A. T. Peterson. 2007. Predicting species distributions from small numbers of occurrence records: a test case using cryptic geckos in Madagascar. Journal of Biogeography 34:102-117. http://dx.doi.org/10.1111/j.1365-2699.2006.01594.x

Phillips, S. J., R. P. Anderson, and R. E. Schapire. 2006. Maximum entropy modeling of species geographic distributions. Ecological Modelling 190:231-259. http://dx.doi.org/10.1016/j.ecolmodel.2005.03.026

Phillips, S. J., and M. Dudik. 2008. Modeling of species distributions with Maxent: new extensions and a comprehensive evaluation. Ecography 31:161-175. http://dx.doi.org/10.1111/j.0906-7590.2008.5203.x

Phillips, S. J., M. Dudik, J. Elith, C. H. Graham, A. Lehmann, J. Leathwick, and S. Ferrier. 2009. Sample selection bias and presence-only distribution models: implications for background and pseudo-absence data. Ecological Applications 19:181-197. http://dx.doi.org/10.1890/07-2153.1

Pitman, J. C., C. A. Hagen, B. E. Jamison, R. J. Robel, T. M. Loughin, and R. D. Applegate. 2006. Nesting ecology of Lesser Prairie-Chickens in sand sagebrush prairie of southwestern Kansas. Wilson Journal of Ornithology 118:23-35. http://dx.doi.org/10.1676/1559-4491(2006)118[0023:NEOLPI]2.0.CO;2

Pitman, J. C., C. A. Hagen, R. J. Robel, T. M. Loughin, and R. D. Applegate. 2005. Location and success of Lesser Prairie-Chicken nests in relation to vegetation and human disturbance. Journal of Wildlife Management 69:1259-1269. http://dx.doi.org/10.2193/0022-541X(2005)069[1259:LASOLP]2.0.CO;2

Pruett, C. L., M. A. Patten, and D. H. Wolfe. 2009. It’s not easy being green: wind energy and a declining grassland bird. Bioscience 59:257-262. http://dx.doi.org/10.1525/bio.2009.59.3.10

Riley, S. J., S. D. DeGloria, and R. Elliot. 1999. A terrain ruggedness index that quantifies topographic heterogeneity. Intermountain Journal of Sciences 5:23-27.

Robel, R. J., J. A. Harrington, C. A. Hagen, J. C. Pitman, and R. R. Reker. 2004. Effect of energy development and human activity on the use of sand sagebrush habitat by Lesser Prairie Chickens in southwestern Kansas. Transactions of the Sixty-Ninth North American Wildlife and Natural Resources Conference 69:251-266.

Swets, J. A. 1988. Measuring the accuracy of diagnostic systems. Science 240:1285-1293. http://dx.doi.org/10.1126/science.3287615

Taylor, M. A., and F. S. Guthery. 1980. Status, ecology, and management of the Lesser Prairie-Chicken. in. General Technical Report RM77. U.S. Forest Service Rocky Mountain Forest and Range Experiment Station, Fort Collins, Colorado, USA.

Thorn, J. S., V. Nijman, D. Smith, and K. A. I. Nekaris. 2009. Ecological niche modelling as a technique for assessing threats and setting conservation priorities for Asian slow lorises (Primates: Nycticebus). Diversity and Distributions 15:289-298. http://dx.doi.org/10.1111/j.1472-4642.2008.00535.x

Timmer, J. M., M. J. Butler, W. B. Ballard, C. W. Boal, and H. A. Whitlaw. 2014. Spatially explicit modeling of Lesser Prairie-Chicken lek density in Texas. Journal of Wildlife Management 78:142-152. http://dx.doi.org/10.1002/jwmg.646

U.S. Fish and Wildlife Service. 2014. Endangered and threatened wildlife and plants; determination of threatened status for the Lesser Prairie-Chicken. Federal Register 79:19974-20071.

USGS National Gap Analysis Program. 2004. Provisional digital land cover map for the southwestern United States. Version 1.0. RS/GIS Laboratory, College of Natural Resources, Utah State University, Utah, USA.

Van Pelt, W. E., S. Kyle, J. Pitman, D. Klute, G. Beauprez, D. Schoeling, A. Janus, and J. Haufler. 2013. The Lesser Prairie-Chicken range-wide conservation plan. Western Association of Fish and Wildlife Agencies Cheyenne, Wyoming, USA.

VanDerWal, J., L. P. Shoo, C. Graham, and S. E. William. 2009. Selecting pseudo-absence data for presence-only distribution modeling: How far should you stray from what you know? Ecological Modelling 220:589-594. http://dx.doi.org/10.1016/j.ecolmodel.2008.11.010

Walker, B. L., D. E. Naugle, and K. E. Doherty. 2007. Greater Sage-Grouse population response to energy development and habitat loss. Journal of Wildlife Management 71:2644-2654. http://dx.doi.org/10.2193/2006-529

Winder, V. L., L. B. McNew, A. J. Gregory, L. M. Hunt, S. M. Wisely, and B. K. Sandercock. 2014. Space use by female Greater Prairie-Chickens in response to wind energy development. Ecosphere 5:1–17. http://dx.doi.org/10.1890/es13-00206.1

Wisdom, M. J., and L. S. Mills. 1997. Sensitivity analysis to guide population recovery: prairie-chickens as an example. Journal of Wildlife Management 61:302-312. http://dx.doi.org/10.2307/3802585

Woodward, A. J., S. D. Fuhlendorf, D. M. Leslie, and J. Shackford. 2001. Influence of landscape composition and change on Lesser Prairie Chicken (Tympanuchus pallidicinctus) populations. American Midland Naturalist 145:261-274. http://dx.doi.org/10.1674/0003-0031(2001)145[0261:IOLCAC]2.0.CO;2