Integrating broad-scale data to assess demographic and climatic contributions to population change in a declining songbird

Abstract

Climate variation and trends affect species distribution and abundance across large spatial extents. However, most studies that predict species response to climate are implemented at small spatial scales or are based on occurrence-environment relationships that lack mechanistic detail. Here, we develop an integrated population model (IPM) for multi-site count and capture-recapture data for a declining migratory songbird, Wilson's warbler (Cardellina pusilla), in three genetically distinct breeding populations in western North America. We include climate covariates of vital rates, including spring temperatures on the breeding grounds, drought on the wintering range in northwest Mexico, and wind conditions during spring migration. Spring temperatures were positively related to productivity in Sierra Nevada and Pacific Northwest genetic groups, and annual changes in productivity were important predictors of changes in growth rate in these populations. Drought condition on the wintering grounds was a strong predictor of adult survival for coastal California and Sierra Nevada populations; however, adult survival played a relatively minor role in explaining annual variation in population change. A latent parameter representing a mixture of first-year survival and immigration was the largest contributor to variation in population change; however, this parameter was estimated imprecisely, and its importance likely reflects, in part, differences in spatio-temporal distribution of samples between count and capture-recapture data sets. Our modeling approach represents a novel and flexible framework for linking broad-scale multi-site monitoring data sets. Our results highlight both the potential of the approach for extension to additional species and systems, as well as needs for additional data and/or model development.

Keywords: Avian demography; Cardellina pusilla; Monitoring Avian Productivity and Survivorship; North American Breeding Bird Survey; climate variation; integrated population model; transient life table response experiment.

Figure 1

Adult male Wilson's warbler in California's Sierra Nevada. Photography credit: Gabriel Gonzalez

Figure 2

Breeding Bird Survey (BBS) routes (squares) and MAPS stations (circles) sampled between 1992 and 2008 where Wilson's warbler was detected or captured. The three genetically distinct breeding regions (Ruegg et al., 2014) included the Pacific Northwest (pnw; purple), coastal California (cca; blue), and the Sierra Nevada (sne; orange). Birds of all three breeding regions winter in northwest Mexico (blue), although migratory connectivity data suggest that only the cca breeding region includes the southern Baja California portion of the wintering range. Points for which spring migration wind data were used are shown for each breeding region (purple circles for pne; green ×'s for cca; orange +'s for sne)

Figure 3

Graphical representation of the hierarchical model that integrates North American Breeding Bird Survey (BBS) and Monitoring Avian Productivity and Survivorship (MAPS) data and includes climate covariates of vital rates. The overall model can be characterized by three sub‐models: (1) a state‐space model for the BBS count data (solid blue); (2) a state‐space Cormack‐Jolly‐Seber (SS CJS) model for MAPS capture‐recapture data of adult birds (left; dashed red); and (3) a binomial model of productivity based on age‐specific MAPS capture data (right; dotted yellow). Additional parameters and hyperparameters accounting for spatial and temporal variation in model components are not shown. Data inputs are represented by open rectangles (y = BBS counts; r = MAPS observed residency; c = MAPS adult capture histories; HY = number of young [hatching year] captures; Nind = number of captures; cmd = winter drought index; tw = tailwind; temp = spring temperature). State variables are represented by shaded rectangles (R = residency state; z = alive state; s = number of survivors; g = number of recruits; n = s + g). Parameters associated with observation processes of state‐space models (i.e., “nuisance” parameters) are represented by open circles (p = recapture probability; ρ = observed residency probability. Estimated population parameters are represented by shaded circles. Residency probability, π, and the productivity index, RI, are shaded to match sub‐models informing them. Adult survival probability, ϕ and recruitment, $\gamma$ are shaded intermediate colors to highlight their dependence on information shared between sub‐models. Stochastic relationships (i.e., model likelihoods) are represented by solid arrows. Climate covariate (cmd, tw, and temp) relationships are shown as dashed arrows. First‐year survival/immigration, ι, is a latent parameter (black/gray) not directly informed by the monitoring data. See Section 2.3 for detail

Figure 4

Annual estimates of abundance and demographic rates (means ± 95% credible intervals) for each Wilson's warbler population

Figure 5

Estimated relationships between demographic parameters and climate covariates (means ± 95% credible intervals). (a) adult apparent survival probability declined with increasing drought anomaly for the Sierra Nevada (sne) and coastal California (cca) populations; and (b) productivity increased as a function of spring temperature anomaly for the sne and Pacific Northwest (pnw) populations

Figure 6

Demographic contributions to variation in population growth rate (means ± 95% credible intervals). The pnw population (left) is represented by blue squares, the sna population (middle) by orange circles, and the cca population (right) by green triangles. Recruitment parameters ($\mathit{RI}$ and $\iota$; gray region) contributed substantially more to explaining annual variation than did adult survival ($\varphi$; white region)

Figure 7

Annual population growth rates v. covariates that influenced vital rates (a, c, e) and contributions of demographic parameters to changes in population growth rates v. annual changes in climate covariate values (b, d, f). Points represent means and error bars delineate 95% credible intervals