Edge thermal niche conservation
We measured the edge thermal niche—the temperatures found at the range
edge—by predicting annual warm and cold temperature extremes at the
range edge position using region-specific GAMs (see Temperature
and distribution data sources ). We then fitted Bayesian linear
regressions to test whether either the warm or cold extreme temperature
at a species’ range edge had changed over time (n = 306, 153
range edges for each of two temperature extremes). Single-species
Bayesian linear regressions were fitted using the “rstanarm package”
(Goodrich et al. 2018) with four chains, 40,000 iterations including
10,000 burn-in draws, and a target average proposal acceptance
probability of 0.99. We selected a normally distributed vague prior with
a mean of 0 and standard deviation of 0.1 °C/year; this standard
deviation was chosen to exceed the largest rates of SST change across
all study regions (see Results ). Models were weighted by
GAM-estimated standard errors (\(\frac{1}{\text{SE}^{2}}\)), so that
estimated temperatures with higher associated uncertainties were less
influential. Models converged for all range edges (Gelman-Rubin
convergence statistic below 1.1). We calculated the mean and 90%
Bayesian credible interval from each single-species posterior
distribution of the year coefficient for either warm or cold temperature
extremes.
If at least one of the two temperature metrics we measured at a range
edge—cold or warm extremes—was constant over time, the range edge
could be tracking that temperature and exhibiting edge thermal niche
conservatism (Hutchins 1947). We categorized range edges according to
whether the range edge maintained a constant warm extreme temperature at
the edge over time, a constant cold extreme temperature, both, or
neither, based on 90% Bayesian credible intervals (Figure 2). In this
method, edge thermal niche conservatism could arise either from the
range edge shifting in space to track temperature, or the range edge
remaining stationary in a location where temperatures have not changed
over time. To disentangle these processes, we compared changes in the
edge thermal niche to changes in the range edge position (Figure 2).
RESULTS
From 1967 to 2018, minimum, mean, and maximum SST in the Northeast all
increased (Figure 1A), translating to more than one degree Celsius of
warming in every SST metric over the fifty years measured (respectively,
0.023 ± 0.007 °C/year, p = 1.4 × 10-3; 0.03 ±
0.004 °C/year, p = 3.7 × 10-9; 0.033 ± 0.006
°C/year, p = 4.4 × 10-7). On the West Coast
(Figure 1B), no significant trends occurred in any temperature metric
from 1976-2018, despite variation of ± 2 °C for individual years
(minimum SST 0.004 ± 0.008 °C/year, p = 0.65; mean SST 0.002 ±
0.007 °C/year, p = 0.77; maximum SST 0.003 ± 0.009 °C/year,p = 0.77). In the Eastern Bering Sea (Figure 1C), warming was
evident in maximum SST change from 1988-2018, which increased 0.038 ±
0.018 °C/year (p = 0.049). Neither minimum nor mean SST increased
significantly in the Eastern Bering Sea, although both had a positive
relationship with year (minimum 0.008 ± 0.01 °C/year, p = 0.48;
0.022 ± 0.013 °C/year, p = 0.10).