Deer abundance estimation at landscape-scales in heterogeneous forests

Abstract

Reliable estimates of deer abundance support effective management of source-sink population dynamics in complex landscapes and improve understanding of the relation between deer density and biodiversity impacts. Performance of distance sampling using thermal imaging of Reeves’ muntjac Muntiacus reevesi and roe deer Capreolus capreolus was examined across 123km2 of conifer forest in Eastern England, sampling 1567km in total. For muntjac distance sampling was compared to estimates from drive counts in 2007. For each of three subsequent winters (2008–2010), we compared the magnitude and precision of forest-wide abundance estimated from analytical designs that: (i) ignored potential habitat-specific detectability, either with uneven or balanced sampling effort; (ii) controlled for sampling effort and/or density among seven forest blocks (mean=18.8km2±11.1 SD); (iii) accounted for potential movement prior detection; (iv) accounted for varying detectability among habitat classes (as a covariate), while controlling for differing densities among blocks. Detectability was further examined in models that stratified to estimate habitat-specific Effective Strip Width (ESW).Estimated muntjac densities from distance sampling were of similar magnitude to estimates from drive counts. Over 2008–2010, we observed 1926 muntjac and 921 roe groups; allowing robust abundance estimation and habitat-specific analysis. ESWs in open habitat were 31% and 27% greater than in mature and 45% and 46% greater than in dense habitat, for roe and muntjac respectively. Although differences in densities among model designs were not large, ignoring block or habitat effects gave higher estimates, while models that accounted for habitat-specific detectability gave lower (−8%) and more precise (38% reduction in CV) estimates (n=3, muntjac: 5.3–7.5% CV; roe deer: 8.8–12.6% CV). The similarity of density estimates between ungrouped and grouped data and analysis of behaviour of detected deer support the conclusion that distance estimates were not biased by avoidance.We conclude that distance sampling using thermal imaging is a robust and powerful method for estimating deer density. In heterogeneous forest density estimates will be improved by accounting for varying detectability among growth stages or habitats.