Estimating and forecasting spatial population dynamics of apex predators using transnational genetic monitoring

Abstract

The ongoing recovery of terrestrial large carnivores in North America and Europe is accompanied by intense controversy. On the one hand, reestablishment of large carnivores entails a recovery of their most important ecological role, predation. On the other hand, societies are struggling to relearn how to live with apex predators that kill livestock, compete for game species, and occasionally injure or kill people. Those responsible for managing these species and mitigating conflict often lack fundamental information due to a long-standing challenge in ecology: How do we draw robust population-level inferences for elusive animals spread over immense areas? Here we showcase the application of an effective tool for spatially explicit tracking and forecasting of wildlife population dynamics at scales that are relevant to management and conservation. We analyzed the world's largest dataset on carnivores comprising more than 35,000 noninvasively obtained DNA samples from over 6,000 individual brown bears (Ursus arctos), gray wolves (Canis lupus), and wolverines (Gulo gulo). Our analyses took into account that not all individuals are detected and, even if detected, their fates are not always known. We show unequivocal quantitative evidence of large carnivore recovery in northern Europe, juxtaposed with the finding that humans are the single-most important factor driving the dynamics of these apex predators. We present maps and forecasts of the spatiotemporal dynamics of large carnivore populations, transcending national boundaries and management regimes.

Keywords: density surface; imperfect detection; noninvasive monitoring of large carnivores; spatial capture–recapture; vital rates.

Fig. 1.

The process of turning long-term monitoring data, such as noninvasive genetic samples and dead recoveries, into population density maps and vital rate estimates. The OPSCR model combines submodels for the ecological and observation processes. The model produces estimates of parameters describing individual space use (home range size and movements), abundance and density, and population dynamics. The model can be used further to generate spatially explicit projections of abundance for user-defined scenarios (see also Fig. 4).

Fig. 2.

Annual density maps for brown bears (U. arctos), gray wolves (C. lupus), and wolverines (G. gulo) in Norway and Sweden between 2012 and 2019, with corresponding population estimates (black violins, total; blue violins, Sweden; red violins, Norway). Gray bars in the violin plots indicate the total number of individual carnivores detected during noninvasive genetic sampling. Time series are labeled with the season or year during which the majority of samples were collected for each species (SI Appendix). Bear image credit: Staffan Widstrand Photography. Wolf image credit: Shutterstock/Kjetil Kolbjørnsrud. Wolverine image credit: Shutterstock/Karel Bartik.

Fig. 3.

The OPSCR model generates population-level density (color scale in Fig. 2) and abundance estimates (with 95% credible intervals), as shown here for wolverines on the Scandinavian Peninsula in 2019. These can be broken down into estimates for the national or local jurisdictions at the scale at which carnivore management takes place (e.g., large carnivore management regions in Norway and counties in Sweden). This top–down approach is preferable to methods that piece together estimates derived locally because they risk biased estimates by double counting animals that are shared by multiple jurisdictions (14). Abundance estimates are not shown for all regions on the map; for a complete tally across all jurisdictions and scales see SI Appendix, Table S3. Equivalent maps for wolves and bears can be found in SI Appendix, Figs. S7 and S8.

Fig. 4.

Estimated (white background) and projected (hatched background) size of the female wolverine population in Scandinavia. Abundance estimates were obtained through OPSCR analysis of 7 y of noninvasive genetic sampling and dead recoveries. Forecasting was accomplished through posterior sampling from the same model. Forecasts are shown for five different levels of culling pressure, compared to the average annual adult culling mortality between 2013 and 2019. The approach also captures the increase in uncertainty (longer violins) as predictions are made farther into the future.