Location‐only and use‐availability data: analysis methods converge

Studies of resource selection form the basis for much of our understanding of wildlife habitat requirements, and resource selection functions (RSFs), which predict relative probability of use, have been proposed as a unifying concept for analysis and interpretation of wildlife habitat data. Logistic regression that contrasts used and available or unused resource units is one of the most common analyses for developing RSFs. Recently, resource utilization functions (RUFs) have been developed, which also predict probability of use. Unlike RSFs, however, RUFs are based on a continuous metric of space use summarized by a utilization distribution. Although both RSFs and RUFs predict space use, a direct comparison of these 2 modeling approaches is lacking. We compared performance of RSFs and RUFs by applying both approaches to location data for 75 Rocky Mountain elk ( Cervus elaphus ) and 39 mule deer ( Odocoileus hemionus ) collected at the Starkey Experimental Forest and Range in northeastern Oregon, USA. We evaluated differences in maps of predicted probability of use, relative ranking of habitat variables, and predictive power between the 2 models. For elk, 3 habitat variables were statistically significant ( P < 0.05) in the RSF, whereas 7 variables were significant in the RUF. Maps of predicted probability of use differed substantially between the 2 models for elk, as did the relative ranking of habitat variables. For mule deer, 4 variables were significant in the RSF, whereas 6 were significant in the RUF, and maps of predicted probability of use were similar between models. In addition, distance to water was the top‐ranked variable in both models for mule deer. Although space use by both species was predicted most accurately by the RSF based on cross‐validation, differences in predictive power between models were more substantial for elk than mule deer. To maximize accuracy and utility of predictive wildlife‐habitat models, managers must be aware of the relative strengths and weaknesses of different modeling techniques. We conclude that although RUFs represent a substantial advance in resource selection theory, techniques available for generating RUFs remain underdeveloped and, as a result, RUFs sometimes predict less accurately than models derived using more conventional techniques.

Applications of logistic regression in a used–unused design in wildlife habitat studies often suffer from asymmetry of errors: used resource units (landscape locations) are known with certainty, whereas unused resource units might be observed to be used with greater sampling intensity. More appropriate might be to use logistic regression to estimate a resource selection function (RSF) tied to a use–availability design based on independent samples drawn from used and available resource units. We review the theoretical motivation for RSFs and show that sample “contamination” and the exponential form commonly assumed for the RSF are not concerns, contrary to recent statements by Keating and Cherry (2004; Use and interpretation of logistic regression in habitat-selection studies. Journal of Wildlife Management 68:774–789). To do this, we re-derive the use–availability likelihood and show that it can be maximized by logistic regression software. We then consider 2 case studies that illustrate our findings. For our first case study, we fit both RSFs and resource selection probability functions (RSPF) to point count data for 4 bird species with varying levels of occurrence among sample blocks. Drawing on our new derivation of the likelihood, we sample available resource units with replacement and assume overlapping distributions of used and available resource units. Irrespective of overlap, we observed approximate proportionality between predictions of a RSF and RSPF. For our second case study, we evaluate the classic use-availability design suggested by Manly et al. (2002), where availability is sampled without replacement, and we systematically introduce contamination to a sample of available units applied to RSFs for woodland caribou (Rangifer tarandus caribou). Although contamination appeared to reduce the magnitude of one RSF beta coefficient, change in magnitude exceeded sampling variation only when >20% of the available units were confirmed caribou use locations (i.e., contaminated). These empirically based simulations suggest that previously recommended sampling designs are robust to contamination. We conclude with a new validation method for evaluating predictive performance of a RSF and for assessing if the model deviates from being proportional to the probability of use of a resource unit.

Resource availability and resource use are 2 key concepts in studies of resource selection. Although equal accessibility to resources is one component defining resource availability, we rarely know what restricts access to resources. Consequently for spatially distributed resources, the animal's use of space in association with the occurrence of resources is a frequent basis for inferring resource use and testing for resource selection. For many resources, occurrence can be defined for the population or the individual animal and requires that researchers specify the spatial extent of resources an animal might use during the time interval of study (e.g., the “choice set”). Often the occurrence of resources is defined at multiple scales, which facilitates understanding hierarchical selection patterns. We discuss numerous factors and criteria that should be considered when delineating the area an animal might use during a period of interest. New analytical approaches to resource selection, including resource utilization functions (RUF) and discrete choice modeling, help address some of the issues of defining availability and dealing with the behaviors associated with resource use. A currency of use is a measure of the investment made by an animal in securing resources, avoiding loss of resources, or otherwise optimizing fitness. Common currencies used by researchers include time spent or distance traveled in a cover type, use of event sites (e.g., nest sites, roosting sites, den sites), or amounts of different kinds of foods consumed. Less common, but potentially highly informative, are such currencies as energy expended or predation risk or other risk incurred. Simulation of animal movements interspersed with diel resting periods, through habitat types with activity-dependent energy expenditure and habitat-specific predation risk showed that choice of a currency of use strongly influences inferences about habitat selection. We argue that perhaps the most informative currency of use would be increased risk to fitness accepted by an animal. Although fairly simple conceptually, such application of risk assessment faces formidable empirical challenges and is a worthy goal for the next generation of researchers of animal resource selection.

Is my sdm good enough? insights from a citizen science dataset in a point process modeling framework

Resource selection is often studied by ecologists interested in the environmental drivers of animal space use and movement. These studies commonly produce spatial predictions, which are of considerable utility to resource managers making habitat and population management decisions. It is thus paramount that predictions from resource selection studies are accurate. We evaluated model building and fitting strategies for optimizing resource selection function predictions in a use-availability framework. We did so by simulating low- and high-intensity spatial sampling data that respectively predicted study area and movement-based resource selection. We compared one of the most commonly used forms of statistical regularization, Akaike's Information Criterion (AIC), with the lesser used least absolute shrinkage and selection operator (LASSO). LASSO predictions were less variable and more accurate than AIC and were often best when considering additive and interacting variables. We explicitly demonstrate the predictive equivalence using the logistic and Poisson likelihoods and how it is lost when the available sample is too small. Regardless of modeling approach, interpreting the sign of coefficients as a measure of selection can be misleading when optimizing for prediction.

Presence-only data, where information is available concerning species presence but not species absence, are subject to bias due to observers being more likely to visit and record sightings at some locations than others (hereafter “observer bias”). In this paper, we describe and evaluate a model-based approach to accounting for observer bias directly – by modelling presence locations as a function of known observer bias variables (such as accessibility variables) in addition to environmental variables, then conditioning on a common level of bias to make predictions of species occurrence free of such observer bias. We implement this idea using point process models with a LASSO penalty, a new presence-only method related to maximum entropy modelling, that implicitly addresses the “pseudo-absence problem” of where to locate pseudo-absences (and how many). The proposed method of bias-correction is evaluated using systematically collected presence/absence data for 62 plant species endemic to the Blue Mountains near Sydney, Australia. It is shown that modelling and controlling for observer bias significantly improves the accuracy of predictions made using presence-only data, and usually improves predictions as compared to pseudo-absence or “inventory” methods of bias correction based on absences from non-target species. Future research will consider the potential for improving the proposed bias-correction approach by estimating the observer bias simultaneously across multiple species.

SummaryTypically in resource selection studies, the spatial extent of a home range is defined first and then the available resources within that perimeter are estimated. However, the home ranges (or habitats) of some animals are constrained by linear environmental features (e.g. rivers, shorelines). Traditional home range estimators often overestimate home range extent for such species, which can lead to spurious estimation of resource availability and selection.We used a synoptic model of space use to explicitly account for resource selection of a species constrained by linear features in its environment to compare with traditional home range estimators. We used the endangered Blakiston's fish owlBubo blakistoniin the Russian Far East as our example.Mean annual home range size (± standard error) was more than three times larger when using kernel methods (30·3 ± 15·1 km2) than when using the synoptic model (9·4 ± 2·0 km2,n = 7).Fish owls showed strong selection for areas within valleys, closer to waterways, closer to patches of permanently open water and with greater channel complexity than available sites.Synthesis and applications. The synoptic model solves a long‐standing problem in home range and resource selection studies because it provides an objective way to estimate the space use of a species whose habitat is constrained by linear features in its environment. Improvements in the accuracy of such estimations can lead to identification of important resources across landscapes, the development of more rigorous site‐specific or landscape‐scale management plans, and to scientifically defensible conservation or threat mitigation measures.

Quantifying landscape characteristics that wildlife select is essential for conservation and management action. Models that map wildlife resource selection tend to be informed by telemetry technology which is costly to acquire/maintain and potentially risky to deploy. Therefore, there is value in pursuing alternative data collection protocols, such as citizen scientist approaches to ascertain whether they can reveal results comparable to those derived from telemetry studies. The conservation of African wild dogs (Lycaon pictus) presents an interesting case study to examine this topic. The species is rare and wideranging, hence data collection is both challenging and costly. They are, however, a groupliving species with unique and conspicuous coat markings, making them potentially well-suited to citizen science data collection strategies. Here, we fitted resource selection functions (RSFs) built from Global Position System (GPS) telemetry data, and from citizen scientist data, collected in and around Hwange National Park, Zimbabwe. We assessed comparability of these RSFs by evaluating the relative importance of parameters, parameter coefficients (direction and magnitude of effect), and the spatial predictions of relative probability of use by African wild dogs. The most important predictors in both models were proportion of woodland and bushland, the number of habitat types, and distance to waterhole. Furthermore, spatial predictions from both models displayed a high degree of overlap (r = 0.74), indicating similarities in selected and avoided habitat patches. Our analysis demonstrates that sufficient citizen science data can be a valuable alternative to telemetry data for African wild dogs. We thus encourage the collection and use of citizen science data for similar analyses, particularly when funding is limited. Our work also highlights areas in and around Hwange National Park with the highest probability of being used by African wild dogs, which is where conservation efforts should be intensified.

Resource selection is a multi‐staged process of behavioral responses to various resource cues or stimuli. Previous research suggests some aspects of resource selection may be inherent (i.e., genetic predisposition) or based on early experience and that individuals respond to certain resource cues but not to others. In other words, resource selection may be based on a template that specifies which cues to use in the resource‐selection process and the appropriate response to those cues. We used resource utilization functions (RUFs) to examine the resource‐selection template of translocated three‐toed box turtles (Terrapene carolina triunguis; hereafter turtles) and made comparisons to resident turtles. Translocated turtles, previously residents of a predominantly forested landscape with low edge‐density, used forest openings, forest edges, and southwest‐facing slopes before and after translocation to a fragmented site containing resident turtles. In contrast, resident turtles used forested areas and northeast‐facing slopes within a predominantly open landscape with high edge‐density. Our comparison of resource selection by translocated and resident turtles revealed population‐specific resource selection and consistency in selection following translocation, which reinforces the idea of a resource‐selection template and suggests that in the short‐term box turtles may not adapt their predisposed behavior to local conditions. Thus, translocated animals may evaluate and respond to resource cues as if they were at the original site. Lack of site fidelity may result from individuals seeking additional resources to match their resource‐selection template. Successful translocation of turtles may require an assessment of resource selection prior to translocation and development of management strategies that mitigate turtle response to translocation.

ABSTRACTAimRecent methodological advances for studying how animals move and use space with telemetry data have focused on fine‐scale, more mechanistic inference. However, in many cases, researchers and managers remain interested in larger scale questions regarding species distribution and habitat use across study areas, landscapes, or seasonal ranges. Point processes offer a unified framework for many methods applied in studies of species distribution and resource selection; however, challenges remain in terms of dealing with temporal autocorrelation common in many types of telemetry data collected from animal locations.InnovationSpace–time point processes (STPPs) have a unique property, in that marginalising time offers a connection between individual animal movement and broader point processes, yet this property has seen little attention in both statistical and applied research. In this paper, we first present some of the details of this marginalisation property and methods for applying marginalised STPPs (mSTTPs) to autocorrelated telemetry data and then apply a mSTTP in a case study on the summer space use and habitat selection of female caribou (Rangifer tarandus) in Denali National Park and Preserve, Alaska.Main ConclusionsThe case study demonstrated that an mSTPP approach can improve inference over other commonly used methods in terms of its ability to account for temporal autocorrelation and offers greater precision in parameter estimates and improved predictions of space use. As this method fits conveniently into the existing point process frameworks, it offers a practical solution to dealing with temporal autocorrelation inherent to many types of telemetry data when research questions center around broader scale patterns of animal habitat selection and space use.

Abiotic and biotic factors influence sheep and goat landscape distribution. Resource selection functions allow us to determine which factors influence distribution the most. This study tested distance to water, distance to trees, tree count, wind direction, and vegetation metrics as resources influencing the distribution of sheep and goats on extensive pastures in Queensland, Australia. Vegetation metrics were computed from remotely sensed data, and are measured as green vegetation, nongreen vegetation, bare ground, and total standing dry matter. We found the location of water, trees, and the prevailing wind direction were the most influential factors affecting sheep and goat distribution. Both sheep and goats selected for areas close to water, trees, and in the direction of the prevailing wind. At one site, goats showed a preference for green vegetation mainly in treed areas, while sheep showed a preference for nongreen vegetation which in the drought conditions of this study were found in less treed areas and can be interpreted as a selection for herbaceous material. At the other sites, the influence of water, trees, and wind was too strong to see a direct influence of vegetation on resource selection. Further research in nondrought conditions would help to better explain vegetation influence on sheep and goat landscape utilization.

Conservation planning for species of concern or importance can be aided by resource selection functions (RSFs) that identify important areas or attributes. Models that can be interpreted biologically and provide reasonable predictive capacity may best be based on data from individuals grouped into seasonal selection strategies for particular geographical areas or similarities in topographical and vegetative associations. We used logistic regression, the information–theoretic approach, satellite imagery, and locational data (n = 31 females; 16 803 locations) from global positioning system (GPS) collared woodland caribou ( Rangifer tarandus caribou (Gmelin, 1788)) to model resource selection by animals during calving, summer, fall, breeding, winter, and late-winter seasons. Higher variation in resource use corresponded to times when caribou and their young were most susceptible to predation or when food resources were limited. Even with multiple selection strategies, caribou followed a general progression from higher to lower elevation habitats from calving and summer to late winter. Caribou selected against or completely avoided the burned–disturbed vegetation class in every season except summer. We incorporated RSFs with a raster geographic information system to create selection landscapes. We validated selection landscapes using withheld GPS data (n = 6077), 50 known calving sites, and Spearman’s rank correlation coefficient. Selection models and final selection landscapes performed well in validating use locations of woodland caribou in all seasons (all P < 0.003) and in predicting known calving sites (P < 0.001). When seasonal selection strategies are identified and models are coupled with validation, RSFs are effective tools to assist in conservation planning.

To assess the wolf‐habitat relationship on the home range scale (i.e., third order selection), we developed a resource selection probability function (RSPF) through a multiple logistic regression model based on the winter travel routes of a wolf pack in the northern Apennines, Italy (1991–95). Both travel routes (240 km) and habitat variables were mapped at 1:10 000 scale, digitised as Geographical Information System (GIS) layers, and overlaid with a 100×100 m pixel grid to census all used and unused resource units. Out of 15 covariates, the full model included 10 variables and 4 interaction terms. According to the model, travel routes by wolves were not randomly located within the home range but were clearly associated with selected bio‐physical factors, including human‐related habitat modifications (i.e., roads), which appeared to affect the wolves' resource selection and, ultimately, habitat quality. Using a jackknife procedure, the model correctly classified 73.1% of used resource units and 63.2% of unused resource units. A Monte Carlo test showed a non‐significant effect on the model coefficients of 4 increasing sub‐sampling levels of used resource units, suggesting that autocorrelation of snow‐tracking data exerted little influence on the point estimates of the coefficients. However, given the increased standard error at higher sub‐sampling levels, autocorrelation might have caused an underestimation of the theoretical variance. Although wolves are generally considered habitat generalists, this study shows that patterns of habitat selection are disclosed at finer scales of analysis. In this perspective, resource selection probability functions at finer scales offer different and complementary insight with respect to regional landscape applications, and provide a useful management tool for assessment of habitat quality at the local scale.

Bryological research carried out from 2008 in Tajikistan and Kyrgyzstan brought interesting data on the occurrence of epiphytic bryophytes which have not been recorded yet there. One of the species was recently described as a new (<em>Orthotrichum pamiricum</em>) and some of the other newly recorded species are considered as rare or endangered in the region of Middle Asia. To make detailed field monitoring of the species with the aim of mapping their distribution in a wild and complex mountainous terrain, it was necessary in the first instance to identify the area with suitable conditions for the occurrence of these species. We present an innovative modeling program MaxEnt (maximum entropy modeling), which have not previously been used for modeling either epiphytic bryophytes or in the Middle Asia region. Using 205 samples (presence-only data), percent tree cover, and seven uncorrelated bioclimatic variables, regions suitable for the occurrence of the studied species were identified. Distribution models for eight most interesting species of <em>Orthotrichum</em> are presented here (<em>O. affine</em>, <em>O. anomalum</em>, <em>O. crenulatum</em>, <em>O. cupulatum</em>, <em>O. pallens</em>, <em>O. pamiricum</em>, <em>O. pumilum</em>, and <em>O. speciosum</em>). They indicated appropriate areas for the most probable occurrence of the species in western Tajikistan, and southwestern and northeastern Kyrgyzstan. These results could serve as guides for future survey expeditions, and aid in the conservation of target species and our understanding of their ecology. Different environmental variables for various species were selected as the most important for modeling. However, for most species higher minimum temperatures and higher precipitation in the wettest month and mean diurnal range were the variables with the greatest contribution to the models.

Investigating the distribution of archaeological sites: Multiparametric vs probability models and potentials for remote sensing data