Trophic consequences of terrestrial eutrophication for a threatened ungulate

Changes in primary productivity have the potential to substantially alter food webs, with positive outcomes for some species and negative outcomes for others. Understanding the environmental context and species traits that give rise to these divergent outcomes is a major challenge to the generality of both theoretical and applied ecology. In aquatic systems, nutrient-mediated eutrophication has led to major declines in species diversity, motivating us to seek terrestrial analogues using a large-mammal system across 598 000 km2 of the Canadian boreal forest. These forests are undergoing some of the most rapid rates of land-use change on Earth and are home to declining caribou (Rangifer tarandus caribou) populations. Using satellite-derived estimates of primary productivity, coupled with estimates of moose (Alces alces) and wolf (Canis lupus) abundance, we used path analyses to discriminate among hypotheses explaining how habitat alteration can affect caribou population growth. Hypotheses included food limitation, resource dominance by moose over caribou, and apparent competition with predators shared between moose and caribou. Results support apparent competition and yield estimates of wolf densities (1.8 individuals 1000 km−2) above which caribou populations decline. Our multi-trophic analysis provides insight into the cascading effects of habitat alteration from forest cutting that destabilize terrestrial predator–prey dynamics. Finally, the path analysis highlights why conservation actions directed at the proximate cause of caribou decline have been more successful in the near term than those directed further along the trophic chain.


27
We conducted a spatial analysis of factors that predict the vegetation index (ΔEVI) across the 28 598,000-km 2 study area. We used a linear model including all 500-m pixels in the study (n > 29 127,000). The R 2 was 0.44 (F = 9293, df = 127088, p < 0.0001). P-values are not meaningful 30 with such high sample size, but the point was to show the magnitude of multiple factors affecting 31 the vegetation index. This is why we did not link habitat alteration (on its own) directly to 32 vegetation index in the path analysis (even though they are highly correlated, Appendix S1),

38
We obtained moose densities using aerial moose surveys conducted by provincial governments, 39 academic, and industry partners between 2008 and 2018 (Table S3.1). Moose surveys were 40 primarily conducted using either the ver Hoef (2008) geospatial or a stratified random block 41 design (Gasaway 1986) but distance sampling became more frequently used as of 2010 42 (Buckland et al. 2004). Moose density estimates from aerial surveys were not available in the 43 Cold Lake Saskatchewan Wolf Survey Unit (WSU). We therefore estimated the density of 44 moose using remote wildlife cameras, and corrected camera estimates to aerial survey estimates 45 using a correlation analysis. We first evaluated the relationship between moose densities 46 estimated using remote wildlife cameras to densities estimated using aerial surveys across Alberta, and applied this correction factor to estimated moose densities in Saskatchewan from 48 wildlife cameras.  To compare density estimates for moose from cameras deployed, we related the estimated moose 54 density from each of the provincial aerial surveys to estimated moose densities from a wildlife

102
We omitted one outlying datum with aerial density of 0.5 km -2 but camera density of 7.1 km -2 .

103
The extreme camera estimate is from a Wildlife Management Units with only 4 cameras, and is 104 largely due to a single camera with an extended visit from one moose. The 90% confidence 105 intervals for that camera estimate are 1.4 -35.3 km -2 , indicating an extremely uncertain estimate.

106
We included one datum with an outlying aerial estimate of 0.77 km -2 in the analyses.

108
There was a general positive relationship between camera estimates and aerial-survey estimates 109 of moose across Wildlife Management Units, but wide scatter as densities increase ( Figure S3.1).

110
The very wide confidence intervals on the GAM included the linear fit line. The linear models

130
We applied the correction factor to moose densities estimated in Cold Lake Saskatchewan 131 caribou range using remote cameras that overlapped the Cold Lake Saskatchewan WSU.

132
Cameras in the Cold Lake Saskatchewan caribou range were randomly placed within a 12.5 x 4 -133 km area, with a minimum spacing of 1 km between each camera. Cameras collected data from 134 January 2017 to March 2018. We calculated moose density using the approach as described 135 above for each camera, and averaged across the 25 cameras to get one density estimate for that 136 region. We estimated the moose density within the Cold Lake Saskatchewan WSU as 0.0789 137 moose km -2 . We then corrected the estimated density by multiplying by the correction factor, 138 0.478, such that 0.0789 moose km -2 * 0.478 = 0.0377 moose km -2 or 3.77 moose 100 km -2 . were always oriented north-south, positioned randomly in the east-west direction, and spaced 1, transects at least once.

180
We repeated the simulated snow track segment outlined above 100 times for each time series.

181
For each time series, we calculated the proportion of snow track segments that were detected for 182 each combination of transect spacing and segment length. These proportions were presented 183 using box plots. All programming was conducted in R using the following packages: rgdal, 184 lubridate, plyr, reshape, and ggplot2.

186
As expected, detection rates increased when transect spacing was reduced and when the number under recent snowfall and these are also noted and considered when searching for fresh tracks.

195
Finally, because WSUs were large and surveyed in one effort over several days, track 196 detectability increased over time (e.g., 2, 3, 4, 5 etc. nights worth of tracks as survey progressed).

198
After wolf tracks were intercepted along a transect, the tracks were forward-tracked and 199 sometimes back-tracked to count the number of wolves in the group using tracking evidence and