Territorial species are often predicted to adhere to an ideal despotic distribution and under-match local food resources, meaning that individuals in high-quality habitat achieve higher fitness than those in low-quality habitat. However, conditions such as high density, territory compression, and frequent territorial disputes in high-quality habitat are expected to cause habitat quality to decline as population density increases and, instead, promote resource matching. We studied a highly human-subsidized and under-matched population of Steller’s jays (Cyanocitta stelleri) to determine how under-matching is maintained despite high densities, compressed territories, and frequent agonistic behaviors, which should promote resource matching. We examined the distribution of fitness among individuals in high-quality, subsidized habitat, by categorizing jays into dominance classes and characterizing individual consumption of human food, body condition, fecundity, and core area size and spatial distribution. Individuals of all dominance classes consumed similar amounts of human food and had similar body condition and fecundity. However, the most dominant individuals maintained smaller core areas that had greater overlap with subsidized habitat than those of subordinates. Thus, we found that 1) jays attain high densities in subsidized areas because dominant individuals do not exclude subordinates from human food subsidies and 2) jay densities do not reach the level necessary to facilitate resource matching because dominant individuals monopolize space in subsidized areas. Our results suggest that human-modified landscapes may decouple dominance from fitness and that incomplete exclusion of subordinates may be a common mechanism underpinning high densities and creating source populations of synanthropic species in subsidized environments.

Study system and sampling. We studied populations of Steller’s jays in two campgrounds within Big Basin Redwoods State Park, Santa Cruz County, California (hereafter Big Basin; Figure 1), to test our hypotheses about resource under-matching. Because of the availability of human food subsidies and previously established high fitness of jays utilizing campground areas, we considered campgrounds to be high-quality habitat and surrounding forest areas to be of lower quality (West and Peery 2017; West et al. 2019). We collected data during three breeding seasons, mid-May – mid-August, in 2017, 2018, and 2019. We captured and banded jays with unique color combinations for individual recognition and to assess individual fitness (see next sections). We used call playback and a combination of mist nets (Avinet Research Supply) and live traps (Havahart and homemade) to capture jays during all three years of the study. To characterize jay space use, we also deployed radio transmitters (Model A1070, Advanced Telemetry Systems) on jays using backpack-style harnesses made of 0.1” natural tubular spectra tape (Bally Ribbon Mills) in each year of the study. Handling time was kept to a minimum (usually <5 minutes, slightly longer for birds receiving radio tags), and samples taken are discussed in the subsequent sections. All appropriate guidelines for humane and ethical use of animals in research were followed, and research was conducted under IACUC protocol A005411-R01-A01 and scientific collection permit SC-13714. To our knowledge, no Steller’s jays were injured or died as a result of our activities.

To examine jay space use, we tracked each radio tagged jay to determine their precise location (± 10 m) 25-35 times per season. We allowed at least two hours between relocations of the same individual to ensure independence between relocations (Swihart and Slade 1985), and we varied the time of day during which we tracked individuals. We also collected roost locations (between 2200 and 0300) to ensure that we fully characterized jay home ranges. We tracked birds by searching on foot with telemetry equipment and marked jay locations using a handheld GPS unit. Observations of jays and jay behavior throughout the breeding season allowed us to assess the breeding status of jays and determine the identities of socially monogamous jay pairs. We only used male jays for these analyses because female jays are generally subordinate to males (Brown 1963), and our sample size of uniquely identified females did not allow dominance assessment.

Defining dominance classes. We classified jays into dominance classes by conducting controlled feeding trials (hereafter ‘behavior trials’) at picnic tables. Because jays have site-based dominance where territorial defense typically weakens as distance from the nest site increases (Brown 1963), we determined a dominance ranking of jays at individual picnic tables dispersed throughout the entire campground to ensure we fully captured spatial variation in dominance for each individual. During each trial, we placed approximately 10 peanuts at the center of a picnic table and then observed jays as they interacted with conspecifics to exploit the food source (Brown 1963; West and Peery 2017). We recorded every banded individual present at each trial, the winner and loser of each interaction, and the aggression level of each interaction on a 0-5 scale. An aggression level of zero indicated that individuals did not interact when feeding at the same time on a table, and so no winner was recorded. Aggression levels were defined as follows: 1: one jay wing-flapped and vocalized with an ‘aap’ or ‘wek’ call at another; 2: one jay displaced another; 3: one jay chased another; 4: jays aggressively sidled with one another but did not make contact; 5: jays physically fought with one another (West and Peery 2017). We evaluated the results of behavior trials to determine the most dominant bird at each table. To be considered dominant at a table, an individual had to win at least 3 interactions at that table. At each table, an individual was considered dominant if it won the most interactions at that table or if it always won interactions against the bird that won the most interactions. In cases where there was not a consistent winner between two individuals that consistently won against all other individuals, or where consistent winners did not interact with one another, we classified the bird with the higher average aggression score in contests that they won as dominant. In cases where individuals’ wins and aggression scores tied, both were considered dominant at a given table. An individual was also considered dominant at a given table if it was the only individual (with the exception of its mate) to appear for two or more 15-minute trials at a specific table on different days. There were occasionally tables at which not enough interactions occurred to determine a dominant bird. We conducted between one and six trials at each of 49 picnic tables in Bloom’s Creek Campground and 65 picnic tables in Huckleberry Campground each year.

We determined dominance for each year separately because dominance and core areas could shift from year to year. Within each year, we overlaid core areas (see below for core area delineation methods) in ArcMap (version 10.7; ESRI 2019) with the results of the behavior trials at each picnic table, and then classified jays into three social classes. ‘High’ dominance included individuals that were dominant at tables within and outside their core area, ‘medium’ dominance included individuals that were dominant only within their core area, and ‘low’ dominance included individuals that were not dominant anywhere within the campground (Figure 2). We used this method, rather than traditional Elo-ratings or other established methods because jay dominance hierarchies shift spatially throughout a given area (Brown 1963) and because we were simply interested in identifying the most dominant individual rather than revealing the entire dominance hierarchy.  

Human food subsidy consumption. We evaluated individual consumption of human food subsidies using stable isotope analysis of δ¹³C in primary feathers. δ¹³C is a useful indicator of human food consumption because human foods are often made up of corn (a C₄ plant) and corn byproducts, making them enriched in the heavy isotope of carbon. This makes them isotopically distinguishable from natural prey items in western North America because primary production in this area is driven by native C₃ plants (Newsome et al. 2010; West et al. 2016). We clipped approximately 50 mm of the most recently grown new primary flight feather from each captured jay at the end of the breeding season (early-mid August) at least 40 days after the conclusion of behavior trials. Because feathers incorporate the isotopic signature of the diet during periods of feather growth (Hobson and Clark 1992) and a primary feather takes approximately 30 days to grow, these feather samples represented breeding season diet but were not contaminated by any peanut consumption that occurred during the behavior trials. We rinsed feather samples thrice in 2:1 Chloroform:Methanol solution to remove surface contaminants and then homogenized them using scissors. Homogenized feathers were dried for approximately 72 hours at 55˚C. Analysis of δ¹³C was conducted at the University of New Mexico Center for Stable Isotopes using a Thermo Scientific Delta V mass spectrometer connected to a high-temperature conversion elemental analyzer and a Costech 4010 elemental analyzer. We report δ¹³C results as parts per mil (‰) ratio relative to the international standard, Vienna-Pee Dee Belemnite limestone. We examined the relationship between dominance class and human food subsidy consumption using a linear mixed model with individual as a random effect because we captured some of the same individuals in multiple years of the study. We used δ¹³C as the continuous response variable and categorical dominance class (i.e., low, medium, high) as the fixed effect. We also included year as a fixed effect to correct for some heteroscedasticity in the residuals. Results are presented as the estimated marginal mean averaged over the three years of the study and a 95% confidence interval.

Fitness metrics. To understand how fitness was distributed among birds in different dominance classes, we collected data on body condition (i.e., body mass and growth bar width) and annual fecundity (i.e., number of fledged young). We conducted analyses using the ‘lme4’ package (version 1.1-21; Bates et al. 2015) and compared means between dominance classes when necessary using the ‘emmeans’ package (version 1.6.1; Lenth 2021) in the R Statistical Environment (R Core Team 2020). 

We measured two indices of body condition: body mass and growth bar width. We measured body mass using a Pesola scale when birds were recaptured at the end of the breeding season (early-mid August). We standardized body mass for body size using tarsus length cubed, an index of body volume (West and Peery 2017). We measured tarsus length using calipers during the same capture event at which body mass was measured. We then multiplied these values by 10,000 for ease in reporting results. We first evaluated body mass among the dominance classes using a linear mixed model with individual as a random effect. However, the variance of the random effect was estimated as zero, and thus we removed it and simply used a two-way ANOVA. Standardized body mass was the continuous response variable, and we used dominance class and year as categorical factors. We did not include an interaction effect between dominance class and year because there was no reason to expect a different relationship between body mass and dominance among years.

 We used growth bar width as another indicator of individual body condition. A feather growth bar consists of one dark band produced during the day, and one light band, produced at night (Wood 1950). Together, one set of bands constitutes feather growth in a 24-hour period (Wood 1950; Grubb 2006). Feather growth is energetically costly, and therefore the ability to grow feathers faster (i.e., wider growth bars) is positively correlated with nutritional status (Grubb 1991). We collected the newest newly grown rectrix, determined by molt pattern and presence of feather sheaths, from each jay recaptured at the end of the breeding season, at least 40 days after the conclusion of behavior trials. Growth bars reflect nutrition during the time of feather growth and thus, because we took a newly grown feather, growth bar width was not contaminated by any peanuts that may have been consumed during the behavior trials. We standardized growth bar width by body size using tarsus length cubed, as we did for body mass, and multiplied it by 100,000 for ease in reporting. We then used a linear mixed model with individual as a random effect to assess the relationship between dominance class and growth bar width. We used standardized growth bar width as the continuous response variable and we used dominance class and year as additive fixed effects. Results are presented as the estimated marginal mean averaged over the three years of the study and a 95% confidence interval.

We estimated annual fecundity (i.e., number of fledglings produced) for individual jays by either locating and monitoring nests until the young fledged (n = 4) or by following radio-tagged birds and observing how many fledglings they interacted with (e.g. begging or feeding behavior; n = 47). Fledgling Steller’s jays follow and receive food from their parents for 30 days or more after fledging (Walker et al. 2020; EHW and KB pers. observations), so it was possible to determine nest success and number of fledglings by closely observing both members of the pair after nesting was completed. When possible, we also banded juveniles so that we could discern identities when we observed family groups on multiple occasions. While locating nests and measuring components of reproduction like lay date, clutch size, hatching success, and nestling condition would have been ideal, we did not have the resources to conduct thorough nest searches. We instead used fledging success as our measure of annual fecundity (Weatherhead and Dufour 2000). 

We treated annual fecundity two different ways for analysis. We first considered annual fecundity as continuous and compared the mean number of fledglings per male per year among dominance classes using a linear mixed model with individual as a random effect. The variance of the random effect was estimated to be zero, so we removed it from the model and assumed independence among all annual fecundity estimates. Due to small sample sizes and non-normality in the distribution of fecundity data, we used a Kruskal-Wallis test for the final comparison of mean annual fecundity between dominance classes. In our second approach, we treated fecundity as a binary indicator of nest success by grouping birds according to whether they successfully fledged offspring or not. We then conducted a chi-squared test to determine if there was any relationship between ordinal dominance class and nest success.

Space use. To characterize jay space use, we delineated home ranges and core areas for each radio-tagged individual. We defined a home range as the 95% Utilization Distribution for each individual jay (West et al. 2016), and we analyzed relocation data using the adehabitatHR package (Calenge 2006) in the R Statistical Environment (R Core Team 2020). Next, we delineated core areas following methods from Vander Wal and Rodgers (2012). Specifically, we plotted each jay’s utilization distribution area against the isopleth volume (Figure S1) and identified the isopleth at which the slope was closest to one (Vander Wal and Rodgers 2012). This point represents the threshold at which proportional home range area begins to increase at a greater rate than the probability of use, and we used that isopleth to define the borders of each jay’s core area (Vander Wal and Rodgers 2012). We chose this method, rather than using an arbitrary 50% utilization distribution, in an effort to ensure that our core areas accurately represented the areas that received the greatest use (see Appendix S1 for a comparison of methods). Finally, we removed seven individuals from all further analyses because their core areas did not overlap campground areas, so we could not accurately assess their dominance class.

We originally planned to examine both core area size and home range size in relation to dominance class, but because these two measurements were highly positively correlated (r = 0.97), we only considered core area size. We log-transformed core area size to meet the assumption of normal distribution and then used a linear mixed model with individual as a random effect to examine the relationship between dominance class and core area size. The variance of the random effect was estimated as zero, so we removed it from the model and used a two-way ANOVA with dominance class and year as categorical factors. Results are presented as the estimated marginal mean averaged over the three years of the study and a 95% confidence interval.

To compare the spatial distribution of core areas in relation to high-quality habitat, we calculated the proportion of overlap between each individual’s core area and the campground area. To delineate the campground boundary, we created a 10-m buffer around each campsite and used the minimum bounding geometry tool in ArcMap (version 10.7.1; ESRI 2019) to create minimum convex hull polygons around the two campgrounds. We calculated the area of overlap between individual core areas and campgrounds and then divided by the total area of individual core areas to determine the proportion of each individual’s core area that overlapped with the campground. Steller’s jays in their first breeding season rarely breed (Brown 1963, West and Peery 2017), and we observed that some of these individuals utilized a strategy of being subordinate ‘floaters’ in which they maintained a high degree of overlap with campground areas. Our hypotheses were restricted to territorial, breeding jays, and we therefore included only individuals known to be in at least their second breeding season in this analysis. To test whether highly dominant individuals had more overlap with campgrounds than subordinates, we conducted a two-way ANOVA, with the proportion of core area overlap with campgrounds as the response, and dominance class and year as explanatory variables.

Data fle is in csv format, can be opened using Microsoft Excel and statistical programs (all analyses were completed in the R Statistical Environment).