Daily abundance of Dall's sheep peaks during late summer in a seasonal habitat of high-management interest

Informing conservation and management decisions for habitats frequented by
species of high management interest often face the challenge of limited
resources for conducting wildlife surveys. When surveys are focused on
local areas or sparsely distributed species, it may also be difficult to
obtain counts sufficient for implementing abundance models that account
for imperfect detection. With replicated aerial surveys collected within a
70.25 km2 portion of the Eastern Alaska Range, Alaska, USA during the
summers of 2013–2015, we estimated daily abundance of Dall’s sheep using
two different estimation methods: Bayesian N-mixture models and Poisson
regression models. We then compared estimates of relative abundance from
both model types while paying special attention to the assumption of
closure within individual survey units. With abundance estimates obtained
from individual survey days, we then estimated the average number of
Dall’s sheep within the survey area for the period 1 July–1 October. Daily
ewe abundance followed a quadratic pattern, with 10–20 ewes being within
our survey area in early July and late September, and approximately 90
ewes within the survey area in mid-August. Lamb to ewe ratios averaged 0.2
from July–September, while ram to ewe ratios averaged 0.4 from July until
mid-August before increasing to about 1.0 by the end of September. These
results indicate that our survey area is an important habitat to local
Dall’s sheep populations when lambs are vulnerable to predators.
Accordingly, human recreation and military training within the survey area
should be minimized 1.5–3.0 months after parturition to minimize
disturbance. We also found that N-mixture models displayed a pattern of
abundance estimates that increased in magnitude as model complexity
increased. We thus recommend an a priori approach to N-mixture model
construction that balances the risk of overfitting models to modest data
against the risk of fitting models that do not explain heterogeneity in
abundance and detection probability. Lastly, we suggest simple
improvements to replicated, aerial surveys for species like Dall’s sheep
focused on reducing violations of the closure assumption within individual
survey units, which can reduce bias of density estimates obtained with
N-mixture models.