Ensembled evaluations of habitat suitability for prioritizing lesser prairie‐chicken conservation

Abstract

Populations of lesser prairie‐chickens (Tympanuchus pallidicinctus, hereafter “prairie‐chickens”) in the Mixed‐Grass Prairie Ecoregion of the southern Great Plains have been projected to go extinct in the next 100 years unless targeted conservation efforts are implemented to increase the size and connectivity of subpopulations through either translocation or habitat restoration. To expand on current conservation efforts, we used ensemble approaches to identify potential habitat and assist managers in accurately prescribing management actions for prairie‐chicken conservation. We developed lek‐based relative habitat suitability models within the Mixed‐Grass Prairie Ecoregion using both resource selection function and Random Forest classification trees and calculated ensembled predictions of relative habitat suitability across all models. Next, we conducted a least‐cost path analysis to identify potential corridors connecting potentially suitable, unoccupied habitat to current subpopulations. Ensembled predictions identified 4575 km² of potential prairie‐chicken habitat both occupied and unoccupied. We identified three contiguous areas of potentially suitable and unoccupied habitat (28–74 km²) that could potentially harbor a self‐sustaining population. Ensembled predictions can be used to strategically implement restoration projects to enhance the quality and connectivity of habitat within and adjacent to the species' current distribution. Least‐cost path analyses revealed a low degree of connectivity between areas of occupied and unoccupied habitat, highlighting the importance of implementing habitat improvement projects to increase connectivity for prairie‐chicken persistence. Our results provide information that professionals may use to prioritize conservation delivery for prairie‐chickens in the Mixed‐Grass Prairie Ecoregion.

INTRODUCTION

Populations of lesser prairie-chickens (Tympanuchus pallidicinctus; hereafter “prairie-chickens”) in the southern Great Plains have declined by an estimated 85% over the past century (Garton et al., 2016; Taylor & Guthery, 1980) and are currently listed as endangered or threatened under the Endangered Species Act (USFWS, 2016, 2023). Historically, prairie-chickens occurred in large swaths of grass- and shrubland habitat in Texas, New Mexico, Colorado, Kansas, and Oklahoma, with range-wide populations speculated to be as high as 2 million birds (Garton et al., 2016; Hagen et al., 2004; Taylor & Guthery, 1980). However, large-scale changes in landscape composition and land use following increased human settlement in the early 1900s greatly reduced the amount and connectivity of available habitat and constrained the species' range to four disjunct ecoregions: Shinnery Oak, Sand Sagebrush, Short-Grass/Conservation Reserve Program (CRP), and Mixed-Grass Prairie (Figure 1; McDonald et al., 2014; Taylor & Guthery, 1980). In addition, prolonged and more frequent droughts have negatively impacted the species, with range-wide populations declining by ~45% following a severe drought between the years 2011 and 2013 (Garton et al., 2016; McDonald et al., 2014). Fortunately, recent efforts aimed at restoring prairie-chicken habitat seem to have had a positive effect on the species overall; however, increases in range-wide numbers are largely attributed to an increase in population sizes in the Short-Grass/CRP Ecoregion (Nasman et al., 2022) where large efforts have been made to reseed marginal cropland back to native vegetation through the CRP (Dahlgren et al., 2016; Spencer et al., 2017; Sullins et al., 2018). Populations throughout the rest of the species distribution, including the Mixed-Grass Prairie, have seen very little change in population size since 2016 (Nasman et al., 2022).

[IMAGE OMITTED. SEE PDF]

The Mixed-Grass Prairie Ecoregion is at the geographical center of the extant distribution of prairie-chickens and is historically believed to have supported the highest densities (McDonald et al., 2014; Van Pelt et al., 2013; Wolfe et al., 2016) suggesting significant potential for population increases in the region. Consequently, state, federal, and private stakeholders have collaborated extensively to develop management strategies aimed at increasing the number and distribution of prairie-chickens in the area. Despite these efforts, prioritization and strategic implementation of conservation measures remain critical needs (Sullins et al., 2019). Recent analyses report that prairie-chicken populations in the Mixed-Grass Prairie declined 1%–2.3% annually during 2005–2022 (Garton et al., 2016; Hagen et al., 2017; Nasman et al., 2022). Local populations showed signs of recovery following the drought in 2011–2013 but have recently declined again, with a population estimate of 4512 birds in 2022. Additionally, estimates of long-term persistence in the Mixed-Grass Prairie Ecoregion are pessimistic due to projected declines in carrying capacity (Garton et al., 2016; Hagen et al., 2017) resulting from continued habitat loss to cultivation (Woodward et al., 2001), energy development (Hagen et al., 2011; Peterson et al., 2020; Plumb et al., 2019; Sullins et al., 2019), and the expansion of eastern red cedar (Juniperus virginiana; Lautenbach, 2017; Lawrence et al., 2022). Prairie-chicken populations in the Mixed-Grass Prairie Ecoregion are currently listed as threatened under the Endangered Species Act (USFWS, 2023).

Prairie-chickens are highly sensitive to changes in landscape composition that reduce the amount and connectivity of available habitat (Woodward et al., 2001), and this sensitivity is influenced by several biological traits that affect prairie-chicken space use (Niemuth, 2011). For example, prairie-chickens tend to avoid tall features likely due to a perceived increase in predation risk by raptors (LeBeau et al., 2023; Plumb et al., 2019; Sullins et al., 2018). Thus, features such as trees and powerlines can cause barriers to movement and render otherwise suitable habitat unusable and concentrate prairie-chickens into smaller areas (Hagen et al., 2011; LeBeau et al., 2023; Londe et al., 2022; Peterson et al., 2020; Plumb et al., 2019). In addition, the average dispersal distance for prairie-chickens is 16–17 km, and mean daily movements average between 0.5 and 2 km, making potential areas of high-quality habitat separated by large expanses of non-habitat (e.g., areas of high cropland or tree cover) inaccessible (Boal & Haukos, 2016; Earl et al., 2016; Gulick, 2019; Peterson et al., 2020). Because lesser prairie-chickens have limited dispersal capabilities, broad-scale habitat fragmentation can reduce connectivity between subpopulations and thus the potential for demographic rescue and maintenance of genetic diversity through the interchange of individuals or natural recolonization. Consequently, restoring and conserving core areas of grassland that are free of anthropogenic features and connected by patches of habitat that facilitate movement is essential to maintaining the long-term sustainability of prairie-chicken populations (Costanzi & Steifetten, 2019; DeYoung & Williford, 2016; Samson, 1980).

Spatially explicit habitat models are valuable tools for relating large-scale habitat conditions to species occurrence and are commonly used to guide landscape-level management and conservation decisions (Clevenger et al., 2002; Jarnevich et al., 2016; Zeigenfuss et al., 2000). Habitat models relating prairie grouse (Tympanuchus spp.) occurrence to landscape-scale habitat conditions have been developed using GPS data collected from marked birds (LeBeau et al., 2023; Sullins et al., 2019) or lek location data (Burda et al., 2022; Doherty et al., 2018; Garton et al., 2011; Hovick et al., 2015; Jarnevich et al., 2016). For example, Jarnevich et al. (2016) used machine-learning niche models to predict prairie-chicken habitat by comparing habitat conditions at known leks with a set of pseudo-random points. Results from this study were integrated into the Southern Great Plains Crucial Habitat Assessment Tool (CHAT), an online decision support tool used to identify high-priority areas for prairie-chicken habitat conservation and low impact areas for responsible energy development (WAFWA GIS Services, 2013; ). A more recent study used telemetry data to predict prairie-chicken habitat by comparing habitat conditions at a subset of prairie-chicken GPS locations versus a set of pseudo-random points using a machine-learning model (Sullins et al., 2019). Predictions from this model informed two large-scale conservation recommendations: (1) targeted tree removal and (2) enrollment in the CRP. While slightly different in scale and purpose, both habitat models have been crucial in advancing prairie-chicken conservation efforts.

Cartographical depictions of relative habitat use are a useful tool in wildlife management; however, they can also provide a false sense of certainty when making management decisions because predictions and the associated error depend on the model structure and the data used. As such, predictions derived across multiple models and datasets can be highly variable and depict different delineations of potential habitat, which may make it difficult to accurately prescribe management actions for species conservation (Lawler et al., 2006; Pearson et al., 2006). To deal with this uncertainty, recent research has employed ensemble approaches where multiple discrete and independent models are developed and predictions from each are compared and combined into one averaged prediction, reducing the error associated with any one model on its own; the result is more robust predictions of habitat suitability (Araújo & New, 2007; Kotu & Deshpande, 2014; Marmion et al., 2009).

To enhance predictions of habitat quality, occurrence information could be combined with a restricted dataset of habitat use in which species have consistently been observed through time. For example, current habitat models for prairie-chickens incorporated a single model structure using all known or a subset of all known prairie-chicken locations (lek locations or GPS locations) without considering whether the use of a particular area was persistent over time (Jarnevich et al., 2016; Sullins et al., 2019). Because prairie-chickens are highly sensitive to interannual changes in habitat conditions, it may be more effective to focus habitat models on datasets that index stable habitat use. For example, stable leks, or leks that have persisted at a site through time, likely correspond with higher quality prairie-chicken habitat than non-stable leks, and thus are more likely to persist throughout annual variations in environmental conditions (e.g., drought) and anthropogenic use. Moreover, as females generally prefer stable leks over newly established leks (Haukos & Smith, 1999), stable lek locations likely play a crucial role in prairie-chicken demography and may provide more accurate depictions of high-quality habitat.

Under current restoration efforts, prairie-chicken habitat is projected to decline by 13%–35% by ~2050 (USFWS, 2022). Developing additional tools and expanding methods to identify prairie-chicken habitat and provide information regarding the value of potential habitat restoration activities for increasing connectivity is essential to prairie-chicken recovery. Our goal was to expand on previous analyses (Jarnevich et al., 2016; Sullins et al., 2019) identifying prairie-chicken habitat by using ensemble approaches and a set of more restrictive habitat use criteria to not only identify core areas for focusing potential prairie-chicken conservation, but to identify habitat conditions associated with stable lek locations that we expected to more accurately represent habitats of higher quality. To do this, we developed lek-based models of relative habitat suitability within the Mixed-Grass Prairie Ecoregion using both resource selection function (RSF) and Random Forest classification trees and calculated ensembled predictions of relative habitat suitability across all models. For each approach, we developed two predictive models: one based on all known lek locations identified by cooperating state wildlife agencies and a more restrictive model based upon leks classified as stable. Specifically, our objectives were to (1) use an ensemble of models to develop spatially explicit predictions of prairie-chicken habitat within the Mixed-Grass Prairie Ecoregion, (2) identify core areas of potentially suitable habitat, both occupied and unoccupied, (3) use ensemble predictions and least-cost path analyses to assess connectivity of identified unoccupied habitat to current self-sustaining subpopulations, and (4) provide information for identifying potential areas for targeted habitat restoration efforts.

METHODS

Study area

Our study area included the Mixed-Grass Prairie Ecoregion within southcentral Kansas, northwestern Oklahoma, and the northeastern portion of the Texas panhandle (McDonald et al., 2014; Van Pelt et al., 2013; Wolfe et al., 2016). We buffered the Ecoregion polygon to 16 km, which corresponds to the average dispersal distance for a prairie-chicken (Figure 1; Earl et al., 2016; Peterson et al., 2020; Van Pelt et al., 2013); this overlapped a small portion of the Short-Grass/CRP Ecoregion to the south. However, delineations of ecoregions are human constructs, and transitions between ecoregion types are better represented as gradients. The total area within the extent of our analyses was ~66,000 km² (~16.3 million acres); ~40,280 km² (~10 million acres) of the study area is in grass- or shrubland cover (LANDFIRE, 2020). Vegetation within the Mixed-Grass Prairie Ecoregion is a mixture of sand sagebrush (Artemisia filifolia) and mid-height perennial grasses. Woody vegetation includes sand plum (Prunus spp.), cottonwood (Populus deltoides), and eastern red cedar. Upland soils are typically deep, loamy sands, and precipitation ranges between 40 and 75 cm annually. The primary land use for the area is livestock grazing (USDA, Natural Resource Conservation Service, ).

Lek data

We obtained prairie-chicken lek location and survey data for the years 2010–2019 for the Mixed-Grass Prairie Ecoregion collected by the Kansas Department of Wildlife and Parks (KDWP) and the Oklahoma Department of Wildlife Conservation (ODWC; Figure 1). Lek data from Texas were unavailable due to landowner privacy policies. Lek survey data for prairie-chicken lek locations were gathered by KDWP and ODWC on annual spring surveys. Aerial and road count data, along with opportunistic lek counts, were included in this analysis.

Landscapes surrounding lek locations (i.e., within 5 km) were assumed to represent habitat conditions that support populations of prairie-chickens (Gehrt et al., 2020; Sullins et al., 2019; Winder et al., 2015). Leks occurring within 300 m of one another across multiple survey years were considered the same lek and only used once (Hovick et al., 2015). We classified leks into two categories: (1) leks where birds were counted in at least one year during 2015–2019 and (2) leks where birds were only counted in years 2010–2014 and not 2015–2019. We then developed a set of criteria to classify a subset of known lek locations as being “stable,” where stable lek locations were those that had persisted (i.e., detected 3 out of 5 consecutive years) or had ≥10 birds within an individual year and were within 2 km of other stable lek locations which is the recommended maximum distance between leks in a complex (Applegate & Riley, 1998; Haukos & Zavaleta, 2016; Wolfe et al., 2016). In addition, select lek locations were added to the stable lek data after confirming their stability with local researchers (N. Parker and D. Sullins, Kansas State University, personal communication). Leks classified as stable were used to identify habitat conditions associated with lek locations that have persisted. We then generated a set of 20 random background points for every lek location (Milligan et al., 2020; Northrup et al., 2013) within terrestrial areas in portions of Kansas and Oklahoma within our study area buffer (e.g., not located within water bodies) and ≥5 km from towns (Hagen et al., 2011; Plumb et al., 2019).

Habitat data

We collated 25 geospatial layers representing habitat conditions known to affect prairie-chicken recruitment, survival, and lek persistence (Appendix S1: Table S1). All geospatial layers were imported into ArcGIS Pro (ESRI, Redlands, CA), resampled to a ~30-m resolution, and clipped to the study area (Earl et al., 2016; Peterson et al., 2020; Van Pelt et al., 2013). We conducted a circular moving-window analysis using tools in ArcGIS Pro to quantify habitat conditions across the entire Mixed-Grass Prairie. All analyses were bounded to a 5-km radius of each 30-m cell to account for known avoidance distances to anthropogenic features (0.5–2 km; Hagen et al., 2011; Plumb et al., 2019), demographic influences (3 km; Ross et al., 2016), distance of nest site selection from an active lek (4.8 km; Giesen, 1994; Hagen et al., 2003), and the importance of grassland composition at the 5-km scale (Sullins et al., 2019). We then used the “raster” and “rgdal” package in program R to import spatial files and extract values of habitat covariates for each lek location and random point (Bivand et al., 2013; Hijmans & Etten, 2013). To ensure temporal consistency in habitat conditions, we extracted habitat covariate values for lek occurrences and their associated random points in the years 2010–2014 from geospatial layers from 2014 and habitat covariate values for leks that occurred in 2015–2019 from 2019 geospatial layers. For non-annually updated covariates (e.g., roadways), we use the most recent available data relative to these years (Table 1; Appendix S1).

TABLE 1 Coefficients and SEs for our two resource selection function models predicting the relative probability of a lek occurring in the Mixed-Grass Prairie Ecoregion of Kansas, Oklahoma, and Texas.

Coefficient	B	SE
RSF—all leks
Ave. tree	−0.335	0.064
Ave. PFG	−0.074	0.077
Ave. PFG2	0.00029	0.0007
Ave. AFG	−0.119	0.033
Ave. shrub	0.649	0.177
Ave. shrub2	0.016	0.005
Ave. cropland	−2.518	1.755
Ave. cropland2	−4.990	2.339
Ave. summer temp.	11.835	24.784
Ave. summer temp2	−0.254	0.470
Ruggedness	0.006	0.067
Ruggedness2	−0.003	0.002
Distance to trans. lines	0.1034	0.036
Distance to trans. lines2	−0.003	0.001
Distance to highway	0.213	0.057
Distance to highway2	−0.009	0.004
Distance to wind turbine	0.053	0.024
Distance to wind turbine2	−0.0001	0.00
Density of oil wells	−0.039	0.007
Density of windmills	−0.028	0.019
Variation in PFG	0.076	0.143
Variation in PFG2	0.0013	0.003
Variation in BG	−0.16	0.059
Log (variation in shrub)	0.878	0.326
Constant	−131.2	326.4
RSF—stable leks only
Ave. tree	−0.18	0.13
Ave. PFG	−0.36	0.11
Ave PFG2	0.005	0.001
log (ave. AFG)	4.49	1.35
log (ave. litter)	3.96	1.53
Ave. cropland	−0.90	3.47
Ave. cropland2	−4.87	4.21
Ave. summer temp	70.80	49.86
Ave. summer temp2	−1.38	0.95
Ruggedness	0.08	0.13
Ruggedness2	−0.007	0.004
Distance to highway	0.06	0.03
Density of roads	0.009	0.0008
Density of roads2	−0.00046	0.00002
Density oil wells	−0.05	0.01
Density wind turbines	−0.09	0.07
Variation in AFG	−0.28	0.10
Variation in PFG	0.16	0.06
Constant	−924.95	653.66

Model development

Prior to developing our models, we tested for potential spatial autocorrelation among lek locations with a Moran's I test available in the “dharma” package in program R (Hartig & Hartig, 2017; Moran, 1950). We then used two modeling techniques and all lek location data to develop two habitat suitability models to model the effects of habitat conditions within a 5-km radius and predict the relative probability of prairie-chicken lek occurrence across the entire Mixed-Grass Prairie Ecoregion. Modeling techniques used included RSF and Random Forests (Boyce et al., 2002; Breiman, 2001; Evans et al., 2010; Manly et al., 2002). We then developed a second set of models using only leks classified as stable. All RSF and Random Forest models were developed under the assumption that lek occurrence indicates prairie-chicken occurrence; thus, all predictions surrounding predicted lek locations represent suitable prairie-chicken habitat at the 5-km scale.

Resource selection functions

To evaluate potential nonlinear responses for each habitat covariate, we used single variable generalized additive models (GAM) with lek locations and random points as binary responses (Crawley, 2007; Wood, 2017). GAM spline multiple smoothing functions together and are better at capturing nonlinear and complex relationships relative to other univariate regression models. We evaluated linear, quadratic, and natural log (i.e., ln[x + 0.001]) threshold responses for each habitat covariate used in our analysis by examining plots of predicted relationships and the partial residuals (Appendix S2: Figure S1). Following patterns observed in our GAM analyses, we fit a fully parameterized RSF and estimated mean coefficients of each habitat covariate using a generalized linear model (GLM) with a binomial error structure and a logistic link function (Boyce & McDonald, 1999; Manly et al., 2002). To prevent our model from overfitting our data and limiting predictability to potential new areas, we used backward stepwise selection and Akaike information criterion (AIC) to iteratively remove covariates that contributed little to explaining the variation in our data (p-value ≥0.05; Akaike, 1987; Arnold, 2010; Burnham & Anderson, 2002).

Random Forest classification trees

Random Forest classification trees are highly sensitive to imbalanced datasets in which the minority class (i.e., represents only a small percentage of the entire dataset) has a much lower number of observations than the majority class (i.e., represents a large percentage of the dataset; Evans & Cushman, 2009). Therefore, we reduced observations of random points (majority class) to equal the number of observations of leks (minority class) because this provided the lowest error rate for our lek locations without drastically compromising the model's ability to correctly classify random points (Evans & Cushman, 2009). Next, we developed Random Forest models where we again compared habitat conditions at lek locations to those at random points (Breiman, 2001; Evans et al., 2010). We conducted Random Forest analyses using the “caret” package in program R to increase model performance by refining model training parameters such as the number of branches that will grow at each split (Appendix S2: Figure S3) and the number of randomly sampled habitat covariates at each node in a classification tree (Appendix S2: Figure S4; Kuhn, 2008). After fitting our Random Forest models, we evaluated the importance each habitat covariate had in predicting lek occurrence using standard measures of variable importance (Breiman, 2001; Evans et al., 2010). Variable importance was estimated by quantifying the difference in predictive accuracy when a variable (i.e., habitat covariate) was included in a test dataset versus if it was not included for each tree in a Random Forest. The differences are then averaged across all trees to represent a relative importance value for each habitat covariate (Evans et al., 2010). Predictions from all models were rescaled and scored between 0 and 1, where 0 indicated relatively low habitat suitability and 1 indicated high habitat suitability.

Model validation

Following Boyce et al. (2002), we assessed the predictive capabilities of our RSF models using a 10-fold cross validation and a Spearman rank correlation (r_s) analysis to compare the frequency of predicted values of the test dataset within 10 bins (bin 1 indicates low habitat suitability and bin 10 indicates high habitat suitability) to the associated RSF score for each individual bin. Analysis was conducted over 500 iterations and averaged. A predictive model will have a Spearman rank statistic ≥0.70, indicating a strong positive correlation and more use data (lek locations) is falling into higher RSF bins (Boyce et al., 2002).

To validate our Random Forest models, we used a receiver operating characteristic combined with an area under the curve (ROC-AUC) analysis where we withheld 20% of the original data from model development to use in evaluating our Random Forest models predictive performances over 500 iterations. Area under the curve scores were then averaged to provide an overall estimate of predictive performance (Boyce et al., 2002; Fielding & Bell, 1997). A predictive model will have a ROC-AUC score of 0.70 or higher, where a score of 0.50 indicates the model does no better at predicting lek and random locations than random chance, and a score of 1.0 indicates the model classified leks and random points perfectly (DeLeo, 1993; Fielding & Bell, 1997).

Ensembled predictions

We calculated the average habitat suitability scores across our ensemble of models to reduce the uncertainty associated with each model and to make more unbiased predictions of potential prairie-chicken habitat (Araújo & New, 2007; Hao et al., 2020; Han et al., 2021). Averaged predictions were then used to create our final map depicting the relative probability of a prairie-chicken lek occurring across the Mixed-Grass Prairie Ecoregion.

Identifying potential prairie-chicken habitat

To quantify the total number of square kilometers classified as habitat potentially suitable to support prairie-chickens, we multiplied the total number of 30-m² cells that had a habitat suitability score ≥ the mean value for lek locations by the resolution of our raster (30 m²). We then converted square meters to square kilometers (Equation 1). 1 $$$ \hat{K}={10}^{-6}\left(S\times 30\ {\mathrm{m}}^2\right). $$$

We removed all large water bodies, major cities, and towns because areas within water or areas associated with exurban development are unavailable for prairie-chicken use (i.e., non-habitat; Hagen et al., 2011; Plumb et al., 2019). While it is difficult to define the minimum area of habitat needed to support a viable prairie-chicken population due to differences in the availability and configuration of habitat conditions, recommended space needs have ranged between approximately 25–200 km² (Bidwell et al., 2003; Davis, 2005; Haufler et al., 2012). Thus, we defined contiguous areas of identified occupied and unoccupied habitat that were ≥25 km² as potentially suitable. Areas of high prediction uncertainty (>0.1 SE) with our RSF models (Appendix S2: Figure S2) were masked out and subtracted from the total amount of square kilometers identified as potentially suitable habitat for prairie-chickens. Areas that had a SE > 0.1 were seldom located in areas of high probability of lek occurrence, so it did not substantially change the estimated number of square kilometers of prairie-chicken habitat. We did not model uncertainty within our predictions of habitat suitability for our Random Forest classification tree models. Quantifying prediction uncertainty with Random Forest classification tree models is problematic; a Random Forest classification model consists of a large number of deep trees where each tree is trained on a random selection of covariates, making it difficult to interpret and depict uncertainty in a model with a large number of covariates (Evans et al., 2010; Hao et al., 2020).

Evaluating habitat connectivity

We identified potential corridors and evaluated the relative connectivity of identified habitat patches using a least-cost path analysis. To do this, we used the inverse of predictions from our ensembled map as a cost path function and assumed the cost for movement is negatively related to the habitat suitability score derived from our ensembled predictions (Chetkiewicz & Boyce, 2009). We then used the least-cost path analysis tool in ArcGIS Pro 2.9 (ESRI, 2020) to find the path of least resistance between identified patches of occupied habitat and potentially suitable, but unoccupied habitat.

Current goals for connectivity zones between contiguous (i.e., focal) areas of prairie-chicken habitat are: (1) at least 40% good-to-high-quality habitat, (2) no greater than 3 km between focal areas, and (3) a minimum of 8 km in width (Van Pelt et al., 2013). Thus, we buffered paths of least resistance on each side of identified least-cost paths to 4 km to identify potential areas for targeted habitat restoration efforts to improve habitat connectivity. We then evaluated the relative connectivity of contiguous areas of suitable prairie-chicken habitat ≥25 km² based on the distance to currently occupied areas of prairie-chicken habitat and the relative quality of habitat between identified areas of potential prairie-chicken habitat (Van Pelt et al., 2013).

All data and code are provided in Solomon et al. (2025).

RESULTS

We identified 272 lek locations in the Mixed-Grass Prairie Ecoregion within Kansas and Oklahoma (Figure 1) and classified 88 of those lek locations as stable during 2010–2019. We found no evidence for spatial autocorrelation among lek locations (Morans we ≤0.01, p = 0.66). The majority of habitat covariates differed substantially between lek locations and random points (Appendix S1: Table S2). All models exhibited a strong positive relationship with average perennial forb and grass cover and distance to highway and transmission lines, and a strong negative relationship with average cropland cover, average annual forb and grass cover, average tree cover, and density of oil wells (Tables 1 and 2; Appendix S2: Figures S5 and S6). Habitat covariates of less importance included variation in bare ground, litter, and shrub cover, distance to oil wells, wind turbines, roadways, and density of wind turbines. All RSF and Random Forest models exhibited high predictive accuracy with cross-validated ROC-AUC scores and Spearman rank coefficients ≥0.80 (Table 3).

TABLE 2 Top 10 variables selected for Random Forest (RF) classification models predicting lesser prairie-chicken lek occurrence across the Mixed-Grass Prairie Ecoregion in Kansas, Oklahoma, and Texas.

Variable	Importance
RF—all leks
Ave. tree	100
Ave. PFG	84.87
Ave. annual precipitation	75.31
Density of oil wells	69.50
Ave. cropland	68.94
Distance to highway	65.98
Ave. AFG	57.48
Distance to trans. lines	51.28
Ave. summer temperatures	50.42
Ruggedness	47.61
RF—stable leks only
Ave. tree	100
Ave. PFG	80.09
Distance to highway	76.92
Density of oil wells	68.89
Ave. cropland	65.11
Distance to trans. lines	59.79
Density of roadways	59.70
Ave. annual precipitation	55.17
Ave. shrub	55.12
Ruggedness	52.07

TABLE 3 Cross-validated Spearman rank correlation coefficients and area under the curve (AUC) scores and 95% CIs, mean and 95% CIs for habitat suitability scores for lek locations, and total area predicted as potentially suitable habitat for lesser prairie-chickens for each resource selection function (RSF) and Random Forest (RF) classification tree model and our ensembled model predicting the relative probability of a lek occurring.

Modeling technique	Spearman rank (RSF) or AUC-ROC (RF)	Mean habitat suitability score	95% CI	Area predicted as suitable prairie-chicken habitat (km2)
Score	95% CI
RSF
All leks	0.86	0.69–0.96	0.74	0.69–0.87	6180
Stable leks only	0.84	0.57–0.94	0.78	0.46–0.95	4914
RF
All leks	0.90	0.84–0.95	0.76	0.37–0.94	4526
Stable leks only	0.86	0.74–0.94	0.87	0.43–0.98	7728

Mean habitat suitability scores for lek locations in all models ranged between 0.74 and 0.87 (Table 3). Total area identified as potentially suitable prairie-chicken habitat ranged between 4526 and 7728 km² (6900–18,000 acres) across models, with a mean of ~6107 km² (1.5 million acres; Table 3; Figure 2); however, in general, areas identified as having high suitability remained relatively consistent across all individual models (Figure 2). The Random Forest model developed using all lek location data had the most conservative estimates of prairie-chicken habitat (4526 km²) while the Random Forest model developed with stable lek-only data had the most liberal predictions of prairie-chicken habitat (7728 km²; Table 3; Figure 2).

[IMAGE OMITTED. SEE PDF]

Mean habitat suitability score for lek locations using our ensembled predictions was 0.78 (CI: 0.75–0.89; Figure 3) and total number of square kilometers identified as potentially suitable prairie-chicken habitat was ~4576 km² (1.1 million acres), both occupied and unoccupied (Table 3; Figure 4). Primary areas identified as suitable habitat in Kansas and Oklahoma are already occupied by prairie-chickens (i.e., areas are in close proximity to currently active lek locations). However, we identified three areas between ~28 and ~73 km² of suitable but potentially unoccupied habitat within Seward county in Kansas, and Beaver, Ellis, and Woodward counties in Oklahoma (Table 4; Figure 5). We also identified large areas of potentially suitable habitat for prairie-chickens in Texas (Table 4; Figure 5), but as we were unable to obtain lek survey data from Texas state agencies, it is unknown whether identified areas are currently occupied by prairie-chickens at this time. In addition, as models were developed using data solely from Kansas and Oklahoma, there is a higher degree of uncertainty associated with predictions of habitat suitability in Texas.

[IMAGE OMITTED. SEE PDF]

[IMAGE OMITTED. SEE PDF]

TABLE 4 Counties and state, the number of contiguous square kilometers, and distance to the nearest subpopulations of lesser prairie-chickens for each area identified as potentially suitable but potentially unoccupied lesser prairie-chicken habitat in the Mixed-Grass Prairie Ecoregion in Kansas, Oklahoma, and Texas.

Identified area/color on map	Counties and state	Area predicted as suitable prairie-chicken habitat (km2)	Distance to nearest subpopulation (km)
1/purple	Ellis and Woodward, OK	74	15.0
2/yellow	Seward, KS; and Beaver, OK	46	5.0
3/orange	Ellis, OK	28	4.0
4/blue	Ochiltree and Lipscomb, TX	40	NA
5/green	Roberts and Gray, TX	97	NA

[IMAGE OMITTED. SEE PDF]

We found low-to-moderate potential for natural recolonization between areas identified as potentially unoccupied and suitable for prairie-chickens and areas currently occupied by prairie-chickens (Appendix S2: Figure S7). Euclidean distances between areas of occupied habitat and potentially unoccupied habitat were greater than 3 km and <1% of raster cells were classified as potentially suitable for prairie-chickens in between (Table 4; Figure 6). Areas between occupied and potentially unoccupied habitat had an average habitat suitability score of 0.58 (SD: 0.13).

[IMAGE OMITTED. SEE PDF]

DISCUSSION

Assessing the quality and connectivity of potential habitats is the logical first step toward the implementation of effective conservation practices. Our goal was to build on previous research and develop a highly predictive habitat model by combining predictions across multiple modeling techniques and response types (all leks vs. stable leks) to provide rigorous estimates of habitat suitability for prairie-chicken conservation. Habitat conditions defining predicted prairie-chicken habitat suitability within models are consistent with previous research and include low proportions of cropland, annual grass, and tree cover, lower density of oil wells, greater proportions of perennial grass and forb cover, and greater relative distances to highways and transmission lines (Fuhlendorf et al., 2002; Hagen et al., 2011; Lautenbach, 2017; Lawrence et al., 2022; Peterson et al., 2020; Plumb et al., 2019; Woodward et al., 2001). We identified three smaller areas of potentially suitable and unoccupied habitat within Kansas and Oklahoma that may benefit from directing conservation efforts toward strategically implementing habitat restoration projects within and adjacent to identified areas. Least-cost path analyses revealed a low degree of connectivity between areas of occupied and unoccupied habitat, highlighting the importance of implementing habitat improvement projects to increase connectivity for prairie-chicken persistence. Our results provide information that professionals may use in conjunction with current research and management tools (Jarnevich et al., 2016; Schindler et al., 2020; Sullins et al., 2019; USFWS, 2022) to prioritize conservation delivery for prairie-chickens in the Mixed-Grass Prairie Ecoregion.

Model evaluations

All four RSF and Random Forest classification tree models had high predictive accuracy. Core areas identified as potential prairie-chicken habitat remained relatively consistent across all four models, supporting our interpretation of high habitat suitability within these areas. High agreement of predictions across models is not unexpected given the restrictive habitat requirements of prairie-chickens. Habitat models for habitat specialists, like prairie-chickens, are typified by higher predictive accuracy than those of habitat generalists (Grenouillet et al., 2011; McPherson & Jetz, 2007; Segurado & Araujo, 2004). Similarities in predictions of core habitat between all four of our models could also be a product of consistent behavioral traits of prairie-chickens (e.g., avoidance of human altered landscapes). In addition, some research indicates models developed for residential species have higher accuracy than models developed for migratory or nomadic species whose occurrence depends on seasonal or annual resource availability (McPherson & Jetz, 2007). Nevertheless, we did find some differences in habitat predictions among models. Predicted total potential habitat varied from 4526 to 7728 km² (1.1–1.9 million acres) across models, highlighting the uncertainty that exists among modeled predictions of habitat suitability (Lawler et al., 2006; Pearson et al., 2006). To address this uncertainty, we employed an ensemble approach to improve certainty in our estimates of habitat suitability and provide more rigorous inferences regarding prairie-chicken conservation (Marmion et al., 2009). Generally, ensemble approaches reduce the probability of wrongfully predicting potentially suitable habitat in areas of non-habitat or vice versa (i.e., lowers the risk of a Type I/II error; Araújo & New, 2007; Hao et al., 2020; Marmion et al., 2009; Ramirez-Reyes et al., 2021).

Our ensembled predictions provided the second most conservative delineation of habitat suitability, identifying a total of ~4,575 km² (1.1 million acres) of potential prairie-chicken habitat in Kansas, Oklahoma, and Texas, and were very similar to predictions made with our Random Forest model developed using all lek data, which identified 4526 km² (1.1 million acres) of potentially suitable habitat. High similarity could be because Random Forest models are themselves an ensemble learning algorithm and thus better capture data complexities relative to many regression-based or decision tree models (Ali et al., 2012; Evans et al., 2010). Additionally, some research suggests that models allowing users to control model parameters (e.g., the number of branches that grow in each tree), such as our Random Forest models, perform just as well as ensembled predictions from multiple models (Hao et al., 2020).

We developed stable lek models with the intention of identifying habitat conditions that support viable populations through time. Predictions of available habitat in our RSF model using stable lek-only data was more restrictive than the RSF model using all lek data, indicating the stable lek-only model may be focusing in on areas of the highest quality prairie-chicken habitat. However, in contrast to the RSF model developed using stable lek-only data, the stable lek model developed using Random Forest classification trees had substantially more liberal predictions of habitat suitability than all other models, predicting the largest area of suitable habitat for prairie-chickens of all our models. Random Forest models developed with small datasets, such as the stable lek model, may have a higher degree of uncertainty due to the low number of observations within the dataset (Ali et al., 2012; Hao et al., 2020). Therefore, the stable lek model developed using Random Forest classification trees may not perform as well as other models when generalizing to new areas. This result further supports the use of ensemble approaches not only in reducing the risk of committing a Type I/II error, but also making more accurate inferences and thus, sound management decisions.

In total, 88/272 lek locations were identified as “stable,” suggesting only about ~33% of leks are located in areas that can support prairie-chickens through fluctuations in environmental and demographic conditions. While potentially true to an extent, a low number of leks identified as stable may also be a result of the inherent nature of lek survey data. Ideally, leks should be located and surveyed consistently across multiple years; however, due to capacity, some lek survey routes may not get surveyed every season. Leks identified opportunistically (i.e., not located on established survey routes) may only be periodically surveyed. While our stable lek criteria attempted to remedy potential lag time in survey years by capping the number of instances a lek had to be surveyed and have birds across consecutive years (3 out of 5 years), there is a high likelihood that some stable leks were not identified during our analysis. Additionally, some leks have never been surveyed due to limited access to private land. This may have limited our dataset and potentially created uncertainty within all models.

Prioritizing conservation actions

Future projections of habitat loss in the Mixed-Grass Prairie Ecoregion indicate a net loss of 2%–37% of available prairie-chicken habitat over the next 25 years depending on the level of future restoration efforts (USFWS, 2022), emphasizing the crucial need for thoughtful prioritization of conservation actions. Long-term sustainability of prairie-chicken populations is largely dependent on having large areas of connected, high-quality habitat. As such, while increasing habitat quality in small areas may temporarily benefit local prairie-chicken populations, it is imperative managers find ways to increase the overall size and connectivity of currently available habitat. Our models identified larger patches of suitable habitat, both occupied and potentially unoccupied, and can assist managers in identifying adjacent areas for targeted restoration efforts and expanding occupied habitat. In addition to expanding occupied areas, smaller areas of identified unoccupied habitat may also serve as an excellent anchor for prioritizing habitat restoration efforts to not only increase connectivity but also serve as a catalyst to increase the overall number of acres of usable habitat (NRCS, 2021). For example, we identified three smaller contiguous areas of potentially unoccupied habitat in Kansas and Oklahoma ranging in size from 28 to 73 km² (6900–18,000 acres). Targeting habitat projects between identified occupied and potentially unoccupied habitat will increase habitat connectivity and substantially increase the number of acres of prairie-chicken habitat. Two areas of potentially unoccupied habitat were also identified in Texas. However, because our models were not developed using data from Texas (i.e., no lek location or random point data), predictions derived within Texas have a high degree of uncertainty and should be interpreted with caution.

Increasing connectivity of habitat to facilitate movement between subpopulations and areas of core habitat is likely the best way to ensure the long-term persistence of the prairie-chicken population in the Mixed-Grass Prairie Ecoregion (DeYoung & Williford, 2016; Garton et al., 2016). Our least-cost path analyses provide a starting point for identifying appropriate conservation actions to increase connectivity. We identified least-cost paths between occupied and unoccupied, but potentially suitable habitat and found low-to-moderate potential for natural recolonization. Long-distance dispersal has been reported as being an infrequent occurrence in established populations of prairie-chickens, with one study reporting only 28% of females and 9% of males attempting long-distance movements (i.e., ≥5 km net displacement from within an individuals established home range) within a summer (Earl et al., 2016; Peterson et al., 2020). In addition, current conservation objectives are that prairie-chicken connectivity zones (1) have 40% good-to-high-quality habitat, (2) are within 3 km of one another, and (3) be 8-km wide (Van Pelt et al., 2013). While identified areas of unoccupied habitat were within 16 km (average dispersal distance of a prairie-chicken) of occupied habitat, none are within 3 km of one another and there is relatively little habitat in between that could provide stopover sites and facilitate movement of individuals from nearby subpopulations. Nevertheless, areas surrounding least-cost paths that have lower habitat suitability scores due to increased tree cover, for example, may be appropriate areas to focus habitat restoration efforts (Sullins et al., 2019). Future research in the Mixed-Grass Prairie should further explore connectivity and dispersal patterns of prairie-chickens to better evaluate the effects of landscape composition on the colonization of unoccupied areas. For example, recent research in the Sand Shinnery Oak Ecoregion has used a combination of graph and circuit theory to quantify lek connectivity and identify populations vulnerable to isolation (McRae & Kavanagh, 2011; Schilder et al., 2022). Other tools such as the UNICOR program have also been useful in identifying connectivity corridors (Cushman et al., 2013); UNICOR applies a modification of Dijkstra's shortest path algorithm (Dijkstra, 1959) to find all shortest paths between two points of interest (e.g., occupied and unoccupied habitat) where the combination of all paths creates a path density map and can be used to identify areas with the greatest potential for species movement (Landguth et al., 2012).

Habitat models are inherently limited by temporal constraints and require regular updates to remain accurate. However, several challenges often hinder the ability to update these assessments over time. For instance, covariates critical to species' habitat use (e.g., maps of roads and energy transmission lines) are frequently updated inconsistently and rarely on an annual basis. This inconsistency makes it difficult to directly compare predictions of habitat suitability across successive model iterations, as it is unclear whether changes in predicted habitat stem from differences in model inputs or actual shifts in habitat conditions (Jarnevich et al., 2021). To effectively prioritize conservation efforts, managers should integrate multiple datasets at varying spatial and temporal scales rather than relying solely on a single model to guide management decisions.

In conjunction with current research and management tools (Jarnevich et al., 2016; Schindler et al., 2020; Sullins et al., 2019; USFWS, 2022), our ensembled predictions and delineations of core habitat can be used to strategically implement management practices for prairie-chicken conservation at the broad scale. In particular, past research by Sullins et al. (2019) may be helpful in prioritizing targeted conservation efforts to increase the size and connectivity of identified habitat at the large scale. Sullins et al. (2019) used telemetry data to predict prairie-chicken habitat across the Short-Grass Prairie Ecoregion, Sand-Sage Ecoregion, and parts of the Mixed-Grass Prairie Ecoregion. Predictions from their model were overlaid with tree density and CRP/tillage risk layers to identify areas for two types of large-scale conservation measures: (1) tree removal and (2) CRP enrollment. The loss and fragmentation of habitat resulting from encroaching eastern red cedar—along with energy development—are likely the top threats to prairie-chickens in the Mixed-Grass Prairie Ecoregion. Overlaying delineations of core habitat and potential connectivity zones between occupied and potentially unoccupied habitat from our model with Sullins et al. (2019) habitat projections and tree density layers provides managers with a robust process for prioritizing tree removal within areas that would maximize benefits for prairie-chickens. While Sullins et al. (2019) model was slightly different in scale (e.g., scale of moving-window analysis for anthropogenic features) and purpose, validation of our lek-based model using Sullins et al.'s telemetry-based habitat model and vice versa may provide insight into potential differences in delineations of identified habitat and potential opportunities for conservation work.

CONCLUSION

Broad-scale habitat analyses represent a crucial first step in prioritizing targeted conservation actions for species recovery. Our ensembled predictions in combination with other studies (e.g., Sullins et al., 2019; USFWS, 2022) provide managers with valuable tools to identify conservation strategies that maximize gains in grassland acres on a large scale. Conservation efforts should prioritize areas where there is a strong, long-term commitment and capacity (e.g., funding and implementation) from stakeholders to maintain prairie-chicken habitats in perpetuity. Identifying site-specific conservation actions within and adjacent to identified prairie-chicken habitat will result in increased connectivity of subpopulations and greater availability of grassland habitat (USFWS, 2022). In the short term, tree removal is likely the best option for maximizing the number of grassland acres in the Mixed-Grass Prairie (Lautenbach, 2017; Sullins et al., 2019; Wolfe et al., 2016). However, in the long term, efforts should focus on actions such as developing cattle grazing strategies and prescribed fire plans to maintain heterogeneity that satisfies necessary habitat requirements needed for all life history stages (Lautenbach et al., 2021).

ACKNOWLEDGMENTS

We would like to thank the Kansas Department of Wildlife and Parks and the Oklahoma Department of Wildlife Conservation for providing lesser prairie-chicken lek survey data. We would also like to thank Turner Enterprises, Inc., for funding this research.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

DATA AVAILABILITY STATEMENT

Data (Solomon et al., 2025) are available from Zenodo: .

References

Akaike, H. 1987. “Factor Analysis and AIC.” Psychometrika 52: 317–332.

Ali, J., R. Khan, N. Ahmad, and I. Maqsood. 2012. “Random Forests and Decision Trees.” International Journal of Computer Science Issues 9: 272–287.

Applegate, R. D., and T. Z. Riley. 1998. “Lesser Prairie‐Chicken Management.” Rangelands Archives 20: 13–15.

Araújo, M. B., and M. New. 2007. “Ensemble Forecasting of Species Distributions.” Trends in Ecology & Evolution 22: 42–47.

Arnold, T. W. 2010. “Uninformative Parameters and Model Selection Using Akaike's Information Criterion.” The Journal of Wildlife Management 74: 1175–1178.

Bidwell, T. G., S. Fuhlendorf, B. Gillen, S. Harmon, R. Horton, R. Manes, R. Rodgers, S. Sherrod, and D. Wolfe. 2003. Ecology and Management of the Lesser Prairie‐Chicken. Stillwater, OK: Oklahoma State University. Extension Circular E‐970.

Bivand, R., T. Keitt, and R. Barry. 2013. “rgdal: Bindings for the Geospatial Data Abstraction Library.” R Package Version 0.8‐9. http://CRAN.R-project.org/package=rgdal.

Boal, C., and D. A. Haukos. 2016. “The Lesser Prairie Chicken: A Brief Introduction to the Grouse of the Southern Great Plains.” In Ecology and Conservation of Lesser Prairie‐Chickens, edited by D. A. Haukos and C. W. Boa, 1–14. Boca Raton, FL: CRC Press.

Boyce, M. S., and L. L. McDonald. 1999. “Relating Populations to Habitats Using Resource Selection Functions.” Trends in Ecology & Evolution 14: 268–272.

Boyce, M. S., P. R. Vernier, S. E. Nielsen, and F. K. A. Schmiegelow. 2002. “Evaluating Resource Selection Functions.” Ecological Modelling 157: 281–300.

Breiman, L. 2001. “Random Forests.” Machine Learning 45: 5–32.

Burda, B. C., M. Somers, K. Conkin, and R. J. Fisher. 2022. “Lek Habitat Suitability for the Sharp‐Tailed Grouse (Typmanuchus phasianellus jamesi) on the Norther Great Plains.” Plos One 17(4): e0265316.

Burnham, K. P., and D. R. Anderson. 2002. In Model Selection and Multimodel Inference: A Practical Information‐Theoretic Approach, edited by D. R. Anderson, 76. New York: Springer.

Chetkiewicz, C. B., and M. S. Boyce. 2009. “Use of Resource Selection Functions to Identify Conservation Corridors.” Journal of Applied Ecology 46: 1036–1047.

Clevenger, A. P., J. Wierzchowski, B. Chruszcz, and K. Gunson. 2002. “GIS‐Generated, Expert‐Based Models for Identifying Wildlife Habitat Linkages and Planning Mitigation Passages.” Conservation Biology 16: 503–514.

Costanzi, J., and Ø. Steifetten. 2019. “Island Biogeography Theory Explains the Genetic Diversity of a Fragmented Rock Ptarmigan (Lagopus muta) Population.” Ecology and Evolution 9: 3837–3849.

Crawley, M. J. 2007. The R book. West Sussex: Wiley.

Cushman, S. A., B. McRae, F. Adriaensen, P. Beier, M. Shirley, and K. Zeller. 2013. “Biological Corridors and Connectivity.” Key Topics in Conservation Biology 2: 384–404.

Dahlgren, D. K., R. D. Rodgers, R. D. Elmore, and M. R. Bain. 2016. “Grasslands of Western Kansas, North of the Arkansas River.” In Ecology and Conservation of Lesser Prairie‐Chickens, edited by D. A. Haukos and C. W. Boa, 259–279. Boca Raton, FL: CRC Press.

Davis, D. 2005. “Status of the Lesser Prairie‐Chicken in New Mexico: Recommendation to not List the Species as Threatened under the New Mexico Wildlife Conservation Act.” Final Investigation Report, New Mexico Department of Game and Fish, Santa Fe, NM. http://www.fws.gov/southwest/es/documents/R2ES/LitCited/LPC_2012/Davis_2005.pdf.

DeLeo, J. M. 1993. “Receiver Operating Characteristic Laboratory (ROCLAB): Software for Developing Decision Strategies that Account for Uncertainty.” In International Symposium on Uncertainty Modeling and Analysis, 318–325. IEEE.

DeYoung, R. W., and D. L. Williford. 2016. “Genetic Variation and Population Structure in the Prairie Grouse.” In Ecology and Conservation of Lesser Prairie‐Chickens, edited by D. A. Haukos and C. W. Boa, 77–98. Boca Raton, FL: CRC Press.

Dijkstra, W. E. 1959. “A Note on Two Problems in Connexion with Graphs.” Numerische Mathematik 1: 269–271.

Doherty, K. E., J. D. Hennig, J. B. Dinkins, K. A. Griffin, A. A. Cook, J. D. Maestas, D. E. Naugle, and J. L. Beck. 2018. “Understanding Biological Effectiveness before Scaling Up Range‐Wide Restoration Investments for Gunnison Sage‐Grouse.” Ecosphere 9: e02144.

Earl, J. E., S. D. Fuhlendorf, D. Haukos, A. M. Tanner, D. Elmore, and S. A. Carleton. 2016. “Characteristics of Lesser Prairie‐Chicken (Tympanuchus pallidicinctus) Long‐Distance Movements across Their Distribution.” Ecosphere 7(8): e01441.

ESRI. 2020. ArcGIS Pro. Release 10. Version 2.7.2. Redlands, CA: Environmental Systems Research Institute.

Evans, J. S., and S. A. Cushman. 2009. “Gradient Modeling of Conifer Species Using Random Forests.” Landscape Ecology 24: 673–683.

Evans, J. S., M. A. Murphy, Z. A. Holden, and S. A. Cushman. 2010. “Modeling Species Distribution and Change Using Random Forest.” In Predictive Species and Habitat Modeling in Landscape Ecology: Concepts and Applications 139–159. New York: Springer New York.

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.

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 US Southern Great Plains.” Landscape Ecology 17: 617–628.

Garton, E. O., C. A. Hagen, G. M. Beauprez, S. C. Kyle, J. C. Pitman, D. D. Schoeling, and W. E. Van Pelt. 2016. “Population Dynamics of the Lesser Prairie‐Chicken.” In Ecology and Conservation of Lesser Prairie‐Chickens, edited by D. A. Haukos and C. W. Boa, 49–76. Boca Raton, FL: CRC Press.

Garton, E. O., J. W. Connelly, J. S. Horne, C. A. Hagen, A. Moser, and M. A. Schroeder. 2011. “Greater Sage‐Grouse Population Dynamics and Probability of Persistence.” In Ecology, Conservation, and Management of Grouse: Published for the Cooper Ornithological Society, edited by B. K. Sandercock, K. Martin, and G. Segelbacher, 293–381. Berkeley, CA: University of California Press.

Gehrt, J. M., D. S. Sullins, and D. A. Haukos. 2020. “Looking at the Bigger Picture: How Abundance of Nesting and Brooding Habitat Influences Lek‐Site Selection by Lesser Prairie‐Chickens.” The American Midland Naturalist 183: 52–77.

Giesen, K. M. 1994. “Movements and Nesting Habitat of Lesser Prairie‐Chicken Hens in Colorado.” The Southwestern Naturalist 39(1): 96–98.

Grenouillet, G., L. Buisson, N. Casajus, and S. Lek. 2011. “Ensemble Modelling of Species Distribution: The Effects of Geographical and Environmental Ranges.” Ecography 34: 9–17.

Gulick, C. K. 2019. “Effects of Working Grassland Management on Lesser Prairie‐Chicken Resource Selection within Home Ranges and During Dispersal Events.” Doctoral diss., Kansas State University.

Hagen, C. A. 2003. “A Demographic Analysis of Lesser Prairie‐Chicken Populations in Southwestern Kansas: Survival, Population Viability, and Habitat Use.” Thesis, Kansas State University.

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.

Hagen, C. A., E. O. Garton, G. Beauprez, B. S. Cooper, K. A. Fricke, and B. Simpson. 2017. “Lesser Prairie‐Chicken Population Forecasts and Extinction Risks: An Evaluation 5 Years Post–Catastrophic Drought.” Wildlife Society Bulletin 41: 624–638.

Hagen, C. A., J. C. Pitman, T. M. Loughin, B. K. Sandercock, R. J. Robel, and R. D. Applegate. 2011. “Impacts of Anthropogenic Features on Habitat Use by Lesser Prairie‐Chickens.” In Ecology, Conservation, and Management of Grouse: Pulished for the Cooper Ornithological Society, edited by B. K. Sandercock, K. Martin, and G. Segelbacher, 63–75. Berkeley, CA: University of California Press.

Han, S., B. D. Williamson, and Y. Fong. 2021. “Improving Random Forest Predictions in Small Datasets from Two‐Phase Sampling Designs.” BMC Medical Informatics and Decision Making 21: 1–9.

Hao, T., J. Elith, J. J. Lahoz‐Monfort, and G. Guillera‐Arroita. 2020. “Testing whether Ensemble Modelling Is Advantageous for Maximising Predictive Performance of Species Distribution Models.” Ecography 43: 549–558.

Hartig, F., and M. F. Hartig. 2017. Package ‘DHARMa’. Vienna: R Development Core Team. http://CRAN.R-project.org/package=dharma.

Haufler, J., D. Davis, and J. Caulfield. 2012. Okalahoma Lesser Prairie‐Chicken Conservation Plan: A Strategy for Species Conservation. Seeley Lake, MT: Ecosystem Management Research Institute. http://www.emri.org/PDF%20Docs/Adobe%20files/Final_OK_LEPC_Mgmt_Plan_23Oct2012.pdf.

Haukos, D. A., and J. C. Zavaleta. 2016. “Habitat.” In Ecology and Conservation of Lesser Prairie‐Chickens, edited by D. A. Haukos and C. W. Boa, 99–132. Boca Raton, FL: CRC Press.

Haukos, D. A., and L. M. Smith. 1999. “Effects of Lek Age on Age Structure and Attendance of Lesser Prairie‐Chickens (Tympanuchus pallidicinctus).” The American Midland Naturalist 142: 415–421.

Hijmans, R., and J. Etten. 2013. “raster: Geographic Data Analysis and Modeling.” R Package Versions 2.1‐25. http://CRAN.R-project.org/package=raster.

Hovick, T. J., B. W. Allred, R. D. Elmore, S. D. Fuhlendorf, R. G. Hamilton, and A. Breland. 2015. “Dynamic Disturbance Processes Create Dynamic Lek Site Selection in a Prairie Grouse.” PLoS One 10(9): e0137882.

Jarnevich, C., P. N. Belamaric, K. Fricke, M. Houts, L. Rossi, G. Beauprez, B. Cooper, and R. Martin. 2021. “Challenges in Updating Habitat Suitability Models: An Example with the Lesser Prairie‐Chicken.” PLoS One 16(9): e0256633.

Jarnevich, C., T. Holcombe, B. Grisham, J. Timmer, C. Boal, M. Butler, J. Pitman, S. Kyle, D. Klute, and G. Beauprez. 2016. “Assessing Range‐Wide Habitat Suitability for the Lesser Prairie‐Chicken.” Avian Conservation and Ecology 11: 1.

Kotu, V., and B. Deshpande. 2014. Predictive Analytics and Data Mining: Concepts and Practice with Rapidminer 148–165. San Francisco, CA: Morgan Kaufmann.

Kuhn, M. 2008. “Caret Package.” Jounal of Statistical Software 28: 1–26.

LANDFIRE: LANDFIRE Existing Vegetation Type layer. (2020, June ‐ last update). U.S. Department of Interior, Geological Survey, and U.S. Department of Agriculture. https://landfire.gov/viewer/.

Landguth, E. L., B. K. Hand, J. Glassy, S. A. Cushman, and M. A. Sawaya. 2012. “UNICOR: A Species Connectivity and Corridor Network Simulator.” Ecography 35: 9–14.

Lautenbach, J. D., D. A. Haukos, J. M. Lautenbach, and C. A. Hagen. 2021. “Ecological Disturbance through Patch‐Burn Grazing Influences Lesser Prairie‐Chicken Space Use.” The Journal of Wildlife Management 85: 1699–1710.

Lautenbach, J. M., R. T. Plumb, S. G. Robinson, C. A. Hagen, D. A. Haukos, and J. C. Pitman. 2017. “Lesser Prairie‐Chick Avoidance of Trees in a Grassland Landscape.” Rangeland Ecology & Management 70: 78–86.

Lawler, J. J., D. White, R. P. Neilson, and A. R. Blaustein. 2006. “Predicting Climate‐Induced Range Shifts: Model Differences and Model Reliability.” Global Change Biology 12: 1568–1584.

Lawrence, A. J., M. A. Boggie, W. R. Gould, S. A. Carleton, and C. T. Nichols. 2022. “Differential Seasonal Avoidance of Anthropogenic Features and Woody Vegetation by Lesser Prairie‐Chicken.” Ornithological Applications 124: 1–15.

LeBeau, C., K. Smith, and K. Kosciuch. 2023. “Lesser Prairie‐Chicken Habitat Selection and Survival Relative to a Wind Energy Facility Located in a Fragmented Landscape.” Wildlife Biology 2023(4): e01091.

Londe, D. W., R. D. Elmore, C. A. Davis, T. J. Hovick, S. D. Fuhlendorf, and J. Rutledge. 2022. “Why Did the Chicken Not Cross the Road? Anthropogenic Development Influences the Movement of a Grassland Bird.” Ecological Applications 32: e2543.

Manly, B. J., L. L. McDonald, D. L. Thomas, T. L. McDonald, and W. P. Erickson. 2002. Resource Selection by Animals: Statistical Design and Analysis for Field Studies. New York, NY: Springer Science & Business Media.

Marmion, M., M. Parviainen, M. Luoto, R. K. Heikkinen, and W. Thuiller. 2009. “Evaluation of Consensus Methods in Predictive Species Distribution Modelling.” Diversity and Distributions 15: 59–69.

McDonald, L., G. Beauprez, G. Gardner, J. Griswold, C. A. Hagen, F. Hornsby, D. Klute, et al. 2014. “Range‐Wide Population Size of the Lesser Prairie‐Chicken: 2012 and 2013.” Wildlife Society Bulletin 38: 536–546.

McPherson, J., and W. Jetz. 2007. “Effects of Species' Ecology on the Accuracy of Distribution Models.” Ecography 30: 135–151.

McRae, B. H., and D. M. Kavanagh. 2011. Linkage Mapper Connectivity Analysis Software. Seattle, WA: The Nature Conservancy.

Milligan, M. C., S. L. Wells, and L. B. McNew. 2020. “Habitat Selection of Female Sharp‐Tailed Grouse in Grassland Manage for Livestock Production.” PLoS One 15(6): e0233756.

Moran, P. A. P. 1950. “Notes on Continuous Stochastic Phenomena.” Biometrika 37: 17–23.

Nasman, K., T. Rintz, D. Pham, and L. L. McDonald. 2022. “Range‐Wide Population Size for the Lesser Prairie‐Chicken 2012–2022.” Western EcoSystems Technology, Inc. Prepared for Western Association of Fish and Wildlife Agencies.

Niemuth, N. D. 2011. “Spatially Explicit Habitat Models for Prairie Grouse.” In Ecology, Conservation, and Management of Grouse: Pulished for the Cooper Ornithological Society, edited by B. K. Sandercock, K. Martin, and G. Segelbacher, 3–20. Berkeley, CA: University of California Press.

Northrup, J. M., M. B. Hooten, C. R. Anderson, Jr., and G. Wittemyer. 2013. “Practical Guidance on Characterizing Availability in Resource Selection Functions under a Use–Availability Design.” Ecology 94: 1456–1463.

NRCS [Natural Resources Conservation Service]. 2021. “A Framework for Conservation Action in the Great Plains Grasslands Biome.” Working Lands for Wildlife, USDA‐NRCS. Washington, D.C. https://wlfw.rangelands.app.

Pearson, R. G., W. Thuiller, M. B. Araújo, E. Martinez‐Meyer, L. Brotons, C. McClean, L. Miles, P. Segurado, T. P. Dawson, and D. C. Lees. 2006. “Model‐Based Uncertainty in Species Range Prediction.” Journal of Biogeography 33: 1704–1711.

Peterson, J. M., J. E. Earl, S. D. Fuhlendorf, R. D. Elmore, D. A. Haukos, A. M. Tanner, and S. A. Carleton. 2020. “Estimating Response Distances of Lesser Prairie‐Chickens to Anthropogenic Features during Long‐Distance Movements.” Ecosphere 11: e03202.

Plumb, R. T., J. M. Lautenbach, S. G. Robinson, D. A. Haukos, V. L. Winder, C. A. Hagen, D. S. Sullins, J. C. Pitman, and D. K. Dahlgren. 2019. “Lesser Prairie‐Chicken Space Use in Relation to Anthropogenic Structures.” The Journal of Wildlife Management 83: 216–230.

Ramirez‐Reyes, C., G. Street, F. J. Vilella, D. T. Jones‐Farrand, M. S. Wiggers, and K. O. Evans. 2021. “Ensemble Species Distribution Model Identifies Survey Opportunities for at‐Risk Bearded Beaksedge (Rhynchospora crinipes) in the Southeastern United States.” Natural Areas Journal 41: 55–63.

Ross, B. E., D. A. Haukos, C. A. Hagen, and J. C. Pitman. 2016. “Landscape Composition Creates a Threshold Influencing Lesser Prairie‐Chicken Population Resilience to Extreme Drought.” Global Ecology and Conservation 6: 179–188.

Samson, F. B. 1980. “Island Biogeography and the Conservation of Prairie Birds.” Proceedings of the North American Prairie Conference 7(2): 293–305.

Schilder, L. J., L. J. Heintzman, N. E. McIntyre, S. Harryman, C. A. Hagen, R. Martin, C. W. Boal, and B. A. Grisham. 2022. “Structural and Functional Landscape Connectivity for Lesser Prairie‐Chickens in the Sand Shinnery Oak Prairie Ecoregion.” The Journal of Wildlife Management 86: e22146.

Schindler, A. R., D. A. Haukos, C. A. Hagen, and B. E. Ross. 2020. “A Decision‐Support Tool to Prioritize Candidate Landscapes for Lesser Prairie‐Chicken Conservation.” Landscape Ecology 35: 1417–1434.

Segurado, P., and M. B. Araujo. 2004. “An Evaluation of Methods for Modelling Species Distributions.” Journal of Biogeography 31: 1555–1568.

Solomon, M., K. Fricke, C. Kruse, and L. B. McNew. 2025. “morgsol1004/Lesser‐Prairie‐Chicken‐Analysis‐Mixed‐Grass‐Prairie: LesserPrairieChickenAnalysisMGP.” Zenodo. https://doi.org/10.5281/zenodo.15392628.

Spencer, D., D. Haukos, C. Hagen, M. Daniels, and D. Goodin. 2017. “Conservation Reserve Program Mitigates Grassland Loss in the Lesser Prairie‐Chicken Range of Kansas.” Global Ecology and Conservation 9: 21–38.

Sullins, D. S., D. A. Haukos, J. M. Lautenbach, J. D. Lautenbach, S. G. Robinson, M. B. Rice, B. K. Sandercock, et al. 2019. “Strategic Conservation for LEPC among Landscape of Varying Anthropogenic Influence.” Biological Conservation 238: 108213.

Sullins, D. S., J. D. Kraft, D. A. Haukos, S. G. Robinson, J. H. Reitz, R. T. Plumb, J. M. Lautenbach, J. D. Lauthenbach, B. K. Sandercock, and C. A. Hagen. 2018. “Demographic Consequences of Conservation Reserve Program Grasslands for Lesser Prairie‐Chicken.” The Journal of Wildlife Management 82(8): 1617–1632.

Taylor, M. A., and F. S. Guthery. 1980. Status, Ecology, and Management of the Lesser Prairie Chicken. Fort Collins, CO: General Technical Report no. RM‐77. U.S. Department of Agriculture Forest Service, Rocky Mountain Forest and Range Experiment Station.

USFWS [U.S. Fish & Wildlife Service]. 2016. “Endangered and Threatened Wildlife and Plants; Determination of Threatened Status for the Lesser Prairie‐Chicken.” United States Fish & Wildlife Service Federal Register 81: 47047–47048.

USFWS [U.S. Fish & Wildlife Service]. 2022. “Species Status Assessment Report for the Lesser Prairie‐Chicken.” Version 2.3. https://www.fws.gov/media/species-status-assessment-report-lesser-prairie-chicken.

USFWS [U.S. Fish & Wildlife Service]. 2023. “Endangered and Threatened Wildlife and Plants; Lesser Prairie‐Chicken; Threatened Status with Section 4(d) Rule for the Northern Distinct Population Segment and Endangered Status for the Southern Distinct Population Segment.” United States Fish & Wildlife Service Federal Register 87: 72674–72755.

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. Cheyenne, WY: Western Association of Fish and Wildlife Agencies.

WAFWA [Western Association of Fish & Wildlife Agencies]. 2013. Southern Great Plains Crucial Habitat Assessment Tool (CHAT). Cheyenne, WY: Western Association of Fish and Wildlife Agencies. https://www.sgpchat.org/.

Winder, V. L., K. M. Carrlson, A. J. Gregory, C. A. Hagen, D. A. Haukos, D. C. Kesler, L. C. Larsson, et al. 2015. “Factors Affecting Female Space Use in Ten Populations of Prairie Chickens.” Ecosphere 6: 166–175.

Wolfe, D. H., L. C. Larsson, and M. A. Patten. 2016. “The Lesser Prairie‐Chicken in the Mixed‐Grass Prairie Ecoregion of Oklahoma, Kansas, and Texas. Ecology and Conservation of Lesser Prairie‐Chickens.” Studies in Avian Biology 48: 299–314.

Wood, S. 2017. “Generlized Additive Models: An Introduction With R.” Chapman and Hall/CRC 2: 100–160.

Woodward, A. J. W., S. D. Fuhlendorf, D. M. Leslie, and J. Shackford. 2001. “Influence of Landscape Composition and Change on Lesser Prairie‐Chicken (Tympanuchus pallidicinctus) Populations.” The American Midland Naturalist 145: 261–274.

Zeigenfuss, L. C., F. J. Singer, and M. A. Gudorf. 2000. “Test of a Modified Habitat Suitability Model for Bighorn Sheep.” Restoration Ecology 8: 38–46.