Skip to main content
Dryad logo

Data from: Inter-annual climate variation influences nest initiation date and nest productivity of the Red-cockaded Woodpecker at the northwestern edge of its range

Citation

Fullerton, Matthew; Walters, Jeffrey; Will, Rodney; Loss, Scott (2021), Data from: Inter-annual climate variation influences nest initiation date and nest productivity of the Red-cockaded Woodpecker at the northwestern edge of its range, Dryad, Dataset, https://doi.org/10.5061/dryad.fn2z34tss

Abstract

Climate change, including directional shifts in weather averages and extremes and increased inter-annual weather variation, is influencing demograhy and distributions for many bird species. We examined how temperature and precipitation coinciding with multiple nesting seasons affected overall nesting success and productivity for two red-cockaded woodpecker (Dryobates borealis, RCW) populations at the species' northwestern range periphery. We used 26 years of nesting data (1991-2016) from the two RCW populations to determine if inter-annual weather variation has affected nesting pehnology and productivity. We conducted analyses at both broad nesting periods (30 and 60 days before nesting; 40 days overalpping the nesting period up to fledging) and short windows to capture the effects of temperature and precipitation extremes on individual nests. For both RCW populations, warmer early spring temperatures generally advanced nesting and increased clutch size and fledgling number. However, effects of average precipitation varied depending on amount and duration of precipitation in different time periods. At the nest level, temperature and precipitation extremes were unrelated to nest success and loss of nestlings (brood loss). Our results indicate that RCWs are responding to the effects of climate change in various ways, with warming trends having a positive benefit on the species at its northwestern range periphery.

Methods

Nesting data (and associated weather variables) for each population of Red-cockaded Woodpeckers were compiled and formatted into Excel .CSV tables from a combination of hand-written field notes and annual summary reports housed within the Oklahoma Department of Wildlife Conservation and the U.S. Forest Service. The statistical code contained in the text document was written by the lead author (M. Fullerton) using the open-source statistical program R and R Studio (https://www.R-project.org/). We combined data for nesting variables across all nests for each year and treated years as replicates (n=25 years for Oklahoma; n=23 years for Arkansas). Nesting variables assessed at this population level included, for first nests (i.e., excluding renests following nest failure), median nest initiation date (median date first egg was laid across all nests for a given year; treated as Julian date for analyses), average clutch size, and average estimated number of nestlings fledged. We used the median for nest initiation date. We defined 3 time windows for which all weather variables—including average daily maximum and daily minimum temperatures, as well as total precipitation—were separately calculated. We defined 30 and 60 day windows preceding the median first initiation date across all years for each population. Further, we defined a 40 day window with a start date based on the same median first nest initiation date as above. This 40 day period was selected to capture the average egg-laying period, incubation period, and period from hatching to fledging.For population-level analyses, we used generalized linear models with weather variables as fixed effects and a Gamma distribution with a log link. For nest initiation and clutch size analyses, we tested weather variables for the two windows (30 and 60 days) prior to nest initiation. For the fledgling analysis, we tested weather variables for the 40 day window from nest initiation to fledging. We conducted pairwise correlation tests among all predictor variables for each population and analysis. Before creating a global model, each predictor variable from a correlated pair was tested for correlation against the response variable of interest. Further, we tested independent effects of predictor variables. We used a backwards stepwise model selection approach, beginning with the global model containing all uncorrelated predictor variables and removing one variable at a time. We also conducted analyses at the individual nest level (i.e., individual nesting attempts 222 treated as replicates; n=232 nests for Oklahoma; n=612 nests for Arkansas) to assess effects of discrete weather events on nesting success and partial brood loss. For these analyses, we used generalized linear mixed models in the lme4 package, with year and cluster as random effects, and binomial and Poisson distributions for nest success and brood loss analyses, respectively. We treated cluster as a random effect. For nest success analyses, all active nests used in analyses were found within three days of the first egg being laid. We used initiation dates for individual nests to define unique, nest-specific time windows. We also defined 2 windows prior to nest initiation for each nest (7 and 14 days). Three windows were also defined for after nest initiation, including the 11 day incubation period, the 7 day nestling period before nestlings were banded and brood reduction events were identified, and the 19 day nestling period between banding age and fledging at day 26. For all nest-level analyses, we used the above-described GLMM model structure, and for fixed effects, we included nest initiation date and weather variables (absolute maximum and minimum temperature, and absolute maximum daily precipitation value for each time window). For the partial brood loss analysis, we included variables for the 11 day incubation window and 7 day window between hatching and nestling banding age. However, we used all 5 time windows for this analysis. For all response variables assessed at the nest level, we used a similar approach to identify and exclude correlated predictor variables as described for population-level analyses. We used AIC to compare an a priori set of candidate models. We first constructed single variable models; based on rankings for these models, we then systematically built models containing more than one variable. We inferred that models were strongly supported when ∆AIC values were between 0 and 2 and at least two (2) less than the null (i.e., intercept-only) model, and when they did not include uninformative parameters.

Funding

Oklahoma Department of Wildlife Conservation, Award: F17AP00203