Effects of Spatial Scale and Heterogeneity on Avian Occupancy Dynamics and Population Trends in Forested Mountain Landscapes

Publications Type: 
Publication Number: 

Frey, Sarah J. K. 2014. Effects of Spatial Scale and Heterogeneity on Avian Occupancy Dynamics and Population Trends in Forested Mountain Landscapes. Corvallis, OR: Oregon State University. 158 p. Ph.D. dissertation.


Population trends and patterns in species distributions are the major currencies used to
examine responses by biodiversity to changing environments. Effective conservation
recommendations require that models of both distribution dynamics and population trends
accurately reflect reality. However, identification of the appropriate temporal and spatial scales
of animal response, and then obtaining data at these scales present two major challenges to
developing predictive models. In heterogeneous forested mountain landscapes I examined: A)
the relative drivers of climatic variability at fine spatial scales under the forest canopy
(‘microclimate’), B) the influence of microclimate on local-scale occupancy dynamics of bird
communities, and C) the effects of spatial scale and imperfect bird detection on long-term avian
population trends.
Climate change has been predicted to cause widespread biodiversity declines. However,
the capacity of climate envelope models for predicting the future of biodiversity has been
questioned due to the mismatch between the scale of available data (i.e., global climate models)
and the scales at which organisms experience their environment. Local-scale variation in
microclimate is hypothesized to provide potential ‘microrefugia’ for biodiversity, but the relative
role of elevation, microtopography, and vegetation structure in driving microclimate is not well
known. If the microrefugia hypothesis is true, I expected to see areas on the landscape that
remained relatively cooler (i.e., buffered sites). To test this, I collected temperature data at 183
sites across elevation and forest structure gradients in complex terrain of the H. J. Andrews
Experimental Forest in the Cascade Mountains of Oregon, USA (Chapter 2). I used boosted
regression trees, a novel machine learning approach, to determine the relative influence of
vegetation structure, microtopography, and elevation as drivers of microclimate and mapped
fine-scale distributions of temperature across the landscape. Models performed extremely well
on independent data – cross-validation correlations between testing and training data were 0.69 –
0.98 for ten selected climate variables. Elevation was the dominant driver in fine-scale
microclimate patterns, although vegetation and microtopography also showed substantial relative
influences. For instance, during the spring-summer transition, maximum monthly temperatures
observed in old-growth sites were 2.6°C (95% CI: 1.8 – 3.3°C) cooler than plantation sites and
minimum temperatures during winter months were 0.6°C (95% CI: 0.4 – 0.8°C) warmer. This
suggests that older forest stands mediate changes in temperature by buffering against warming
during summer months and moderating cold temperatures during the winter.
Climate is generally considered most influential on species distributions at large spatial
scales; however much microclimate variability exists within regional patterns. I tested whether
this high degree of microclimate variability has relevance for predicting species distributions and
occupancy dynamics of the Andrews Forest bird community. I collected bird occurrence data in
2012 and 2013 at all 183 sites with fine-scale temperature measurements. I used dynamic
occupancy models to test the effects of temperature on occupancy and apparent within-season
bird movement while statistically accounting for vegetation effects and imperfect bird detection
(Chapter 3). Most species (87%) exhibited within-season shifts in response to local-scale
temperature metrics. Effects of temperature on within-season occupancy dynamics were as large
or larger (1 to 1.7 times) than vegetation. However, individual species were almost as likely to
shift toward warmer sites as toward cooler sites, suggesting that microclimate preferences are
species-specific. My results emphasize that high-resolution temperature data provide valuable
insight into avian distribution dynamics in montane forest environments and that microclimate is
an important variable in breeding season habitat selection by forest birds. I hypothesize that
microclimate-associated distribution shifts may reflect species’ potential for behavioral buffering
from climate change in complex terrain.
Factors influencing population trends often differ depending on the spatial scale under
consideration. Further, accurate estimation of trends requires accounting for biases caused by
imperfect detection. To test the degree to which population trends are consistent across scales, I
estimated landscape-scale bird population trends from 1999-2012 for 38 species at the Hubbard
Brook Experimental Forest (HBEF) in the White Mountains of New Hampshire, USA and
compared them to regional and local trends (Chapter 4). I used a new method – open-population
binomial mixture models – to test the hypothesis that imperfect detection in bird sampling has
the potential to bias trend estimates. I also tested for generalities in species responses by
predicting population trends as a function of life history and ecological traits. Landscape-scale
trends were correlated with regional and local trends, but generally these correlations were weak
(r = 0.12 – 0.4). Further, more species were declining at the regional scale compared to within
the relatively undisturbed HBEF. Life history and ecological traits did not explain any of the
variability in the HBEF trends. However, at the regional scale, species that occurred at higher
elevations were more likely to be declining and species associated with older forests have
increased. I hypothesize that these differences could be attributed to both elevated rates of landuse
change in the broader region and the fact that the structure of regional data did not permit
modeling of imperfect detection. Indeed, accounting for imperfect detection resulted in more
accurate population trend estimates at the landscape scale; without accounting for detection we
would have both missed trends and falsely identified trends where none existed. These results
highlight two important cautions for trend analysis: 1) population trends estimated at fine spatial
scales may not be extrapolated to broader scales and 2) accurate trends require accounting for
imperfect detection.