Mechanistic home range capture–recapture models for the estimation of population density and landscape connectivity

Spatial capture–recapture models (SCRs) provide an integrative statistical tool for analyzing animal movement and population patterns. Although incorporating home range formation with a theoretical basis of animal movement into SCRs can improve the prediction of animal space use in a heterogeneous landscape, this approach is challenging owing to the sparseness of recapture events.

In this study, we developed an advection–diffusion capture–recapture model (ADCR), which is an extension of SCRs incorporating home range formation with advection–diffusion formalism, providing a new framework to estimate population density and landscape permeability. we tested the unbiasedness of the estimator using simulated capture–recapture data generated by a step selection function. We also compared accuracy of population density estimates and home range shapes with those from an SCR incorporating the least-cost path. In addition, ADCR was applied to real dataset of Asiatic black bear in Japan to demonstrate the capacity of the ADCR to detect geographical barriers that constrain animal movements.

Population density, permeability, and home range estimates of ADCR were unbiased over randomly determined sets of true parameters. Although the accuracy of density estimates by ADCR was nearly identical to those of existing models, the home range shape could be predicted more accurately by ADCR than by an SCR incorporating the least-cost path. For the application to bear dataset, ADCR could detect the effect of water body as a barrier of movement which is consistent with previous population genetic studies.

ADCR provides unique opportunities to elucidate both individual- and population-level ecological processes from capture–recapture data. By offering a formal link with step selection functions to estimate animal movement, it is suitable for simultaneously modeling with capture–recapture data and animal movement data. This study provides a basis for studies of the interplay between animal movement processes and population patterns.

Methods
Study site
              Our survey was conducted in the eastern Toyama prefecture, Japan. Our study site locates at the western foot of Tateyama mountains and partly overlapped to the Chubusangaku National Park. It contains a wide range of topography from lowland, hill to mountains. In the hilly area, agricultural lands along the rivers divide the forest landscape. The deciduous coniferous trees (Fagus crenata, Quercus crispula and Q. serrata) which offer food for bears in autumn are dominant species of the forest (Arimoto et al. 2011). As in other parts of Japan, a hard crop of acorns causes behavioral changes in black bears that increase conflicts with human (Ohnishi et al. 2011).

Survey design
              From 2013 to 2015, we conducted a camera trap capture-recapture survey at 86 locations in the forest (Fig. S1). The survey were conducted from May to October, which is active season for bears. In each location, we set a camera trap (Trophycam ; Bushnell Outdoor Products, Overland Park) with video-recording mode. The duration of video was 30 seconds, and lag time after a trigger was set to minimal value. For efficient photographing of a chest mark as a key to individual recognition, we used an odor stimulant (mixture of honey and red wine) to encourage bears to stand up in front of camera by the protocol shown by Higashide et al. (2013). The odor stimulant was filled in a plastic bottle covered by a robust polyvinyl chloride tubing and fixed to the surrounding trees for protection from bear attachs. We visited each location every one to two months to replace batteries and SD cards and to refill the odor stimulant. The records of the same individual at a location within 60  minutes were grouped into a detection event. An image library of chest marks was developed from the video footage taken, and identical individuals were matched manually (Higashide et al. 2012). For fitting the capture recapture models, we aggregated the numbers of detections for each camera trap, individual and year.