Breeding Productivity—Data Collection

To examine and compare population parameters between the Otay and San Diego Cactus Wren populations, we established two study plots in the Otay population (Salt Creek and Johnson Canyon) and two study plots in the San Diego population (Lake Jennings and Sweetwater; fig. 2). We monitored nesting activity at Cactus Wren territories in these four study plots weekly between February and August in 2015–19. In 2016, we added a third study plot in the Otay population, Wolf Creek, but then abandoned it after that year because site conditions made collecting accurate data impractical.

We initially selected Cactus Wren territories for monitoring in 2015 and included all wren territories at Salt Creek (five territories) and Johnson Canyon (five territories) in the Otay population, all but one wren territory at Lake Jennings (eight territories), and all wren territories at Sweetwater (Sweetwater Summit Regional Park, Sweetwater Reservoir, and San Diego National Wildlife Refuge; six territories) in the San Diego population. In subsequent years, we monitored new territories when previously occupied territories were determined to be vacant.

All Cactus Wren nests were located and a GPS point for each nest was collected. Cactus Wrens build multiple nests, many of which are “roost” or “dummy” nests that are not used for breeding. We checked all nests at each territory weekly until a “brood” nest was identified by the presence of eggs or nestlings. Nests were approached carefully to avoid flushing an incubating female or older nestlings. Brood nests were checked weekly using a fiber optic camera or by reaching a hand into the nest to feel contents to determine the number of eggs, the number and age of nestlings, and whether the nest had successfully fledged young (nestlings reached 19 days of age or were detected alive outside of the nest). Unsuccessful nests were placed into one of three nest fate categories. Nests found empty or destroyed before the estimated fledge date and where adult wrens were not found tending fledgling(s) were considered depredated. Nests failing for reasons such as collapse of the host cactus or the presence of a clutch of infertile eggs were classified as failing because of other causes that were known. Nests that appeared intact and undisturbed but were abandoned with Cactus Wren eggs or nestlings were classified as having failed because of unknown causes. We located and monitored all nesting activity for each focal pair through the entire breeding season.

From these data, we calculated several measures of productivity: egg clutch size, egg hatch rate (number and percent of eggs that hatched), nest hatch rate (number and percent of nests with eggs that hatched), hatchling fledge rate (number and percent of hatchlings that fledged), nest fledge rate (number and percent of nests with hatchlings that fledged), egg fledge rate (number and proportion of eggs that produced fledglings), apparent nest success (number of nests with eggs that produced fledglings), and the number of fledglings produced per pair. We compared the egg hatch rate with the hatchling fledge rate and the nest hatch rate with the nest fledge rate to determine which stage of the nesting cycle was more vulnerable to failure. We also compared egg hatch rate with nest hatch rate and hatchling fledge rate with nest fledge rate to determine if factors that influenced these stages were affecting the entire nest (nest scale) or only affecting some of the nest contents (partial nest failure).

Breeding Productivity—Data Analysis

We used Chi-square and Fisher’s Exact tests to determine if there were differences in proportions of renesting attempts between Cactus Wren populations and years and after successful or failed first nesting attempts and to determine if there were differences between populations in the proportion of failed nests that failed as a result of depredation or other causes. To evaluate the direct influence of precipitation on breeding success and productivity, bio-year precipitation (hereafter, “precipitation”) was calculated from July 1 of the year before breeding through June 30 of the breeding season year. Although this definition of bio-year does not include the entire breeding season, very little precipitation fell between June and the end of the breeding season (September), and therefore, the defined period was functionally inclusive of the entire breeding season. We used logistic regression to determine if precipitation and population (Otay or San Diego) influenced egg hatch rate, hatchling fledge rate, the proportion of pairs that attempted a second nest, and apparent nest success. We used precipitation data from the El Cajon weather station (National Oceanic and Atmospheric Administration, 2020). We used Poisson regression to determine if precipitation and population influenced the number of brood nests produced per pair per year and egg clutch size because Poisson was appropriate for count data. For Poisson and quasi-Poisson regressions, we used incident rate ratios (e^[estimate]) to quantify the amount of change in the response variable for each predictor variable, holding all other variables in the equation constant.

We used the RMark package (Laake, 2013) in Program R (R Core Team, 2020) to model the effects of Cactus Wren population (Otay or San Diego) and precipitation (as described in the preceding paragraph) on daily survival rate (DSR) of Cactus Wren nests (White and Burnham, 1999; Dinsmore and others, 2002). Nest survival was calculated across a 39-day cycle encompassing egg-laying, incubation, hatching, and nestling periods. Age of nests at the time they were discovered was calculated by forward- or backward-dating of nests in relation to known dates of egg-laying, hatching, or hatchling age. We used an information-theoretic approach (Akaike’s Information Criteria for small sample sizes or AIC_c; Burnham and Anderson, 2002) to evaluate support for nest survival models regarding the effects of population and precipitation. We used logistic regression with a logit link to build models. First, we generated a constant survival model to serve as a reference for the effects of Cactus Wren population and precipitation on DSR. We then modeled the covariates and evaluated support for the model in relation to the constant survival model. Well supported models had ∆AIC_c less than 2.

Survivorship and Movement—Data Collection

We banded wrens at nest monitoring sites with a combination of colored leg bands unique to each individual to evaluate survivorship and dispersal of adults and first-year birds. Adults and fledglings were captured opportunistically at monitored territories using mist nets and song playback to attract birds to nets. All hatchlings from accessible nests in monitored territories were banded at 5–15 days old (age target was 6–12 days old; hatchlings outside of this age range were evaluated for banding potential on an individual basis). During banding of nestlings, we recorded age (in days), weight, and collected fecal samples for a diet analysis. During banding of adults, we recorded sex, weight, and reproductive status. We resighted all Cactus Wrens at survey plots and monitoring sites, when possible, to identify individuals based on color-band combinations. When bands were missing or observations were unclear, we returned on non-survey days to obtain photographs using a Canon 7D Mark II digital single-lens reflex camera with the Canon 100–400 millimeter (mm) F/4.5–5.6 zoom lens. Photographs were useful for determining fine color differences (faded bands) or reading numbers on metal bands.

Cactus Wrens do not exhibit obvious sexual dimorphism when observed under normal field conditions. Gender is typically determined by specific behavioral cues (position during copulation, incubation [performed only by the female]) or morphology of birds in the hand (females have brood patches, males have cloacal protuberances). Gender cannot be determined for nestlings without genetic analysis. If none of these cues were observed, for convenience in reporting we assigned an adult as “male” if it sang or called more frequently or was more visually obvious (potentially advertising territory boundaries), although females can also exhibit these behaviors. As a result, we did not have a large sample of confirmed-gender adults, and therefore, we did not attempt gender-related analyses of survivorship or movements except as general summaries.

Survivorship and Movement—Data Analysis

We used the RMark package (Laake, 2013) in Program R (R Core Team, 2020) to model the effects of age (first-year or adult), population (Otay or San Diego), precipitation, and year on Cactus Wren survivorship. We calculated annual survival from 2015 to 2019 by creating an encounter history matrix of all individual Cactus Wrens ever encountered within the study area. Survivorship was assumed to be constant for adults once they survived their first year. For precipitation, we used bio-year in the previous year (July 1–June 30 for the bio-year ending in the previous breeding season (for example, survival for 2017–18 as a function of the precipitation in the bio-year ending in 2017) as a time-varying individual covariate. We used precipitation data from the El Cajon weather station (National Oceanic and Atmospheric Administration, 2020). We used logistic regression with a logit link to build and rank models by AIC_c and present annual real estimates from the top model. Because we expected survival to differ between first-year and adult birds, age was included in all models. If there was support for multiple models (ΔAIC_c less than 2), we averaged all models using AIC_c weights to obtain annual real estimates of survival for adult and first-year Cactus Wrens. We also evaluated the effect of covariates within our top models by examining the odds ratio for each covariate (the odds that the covariate had an effect on survivorship where “no effect” equals 1, negative effect less than 1, positive effect greater than 1). We then calculated the 95-percent confidence interval of the odds ratio to determine the likelihood that the effect was significant. Where the confidence interval was greater than or less than 1, we concluded that we had 95-percent confidence that the covariate had a positive or negative effect on survivorship relative to the reference condition.

We calculated natal dispersal distance for first-year Cactus Wrens by measuring the straight-line distance between the center of an individual’s natal territory and the center of the same individual’s breeding territory the following year. We used the Kruskall-Wallis test to determine whether natal dispersal distance differed by population. Similarly, adult site fidelity was determined by measuring the distance between the center of a bird’s breeding territories in successive years. Wrens exhibited site fidelity if they returned to within 100 meters (m) of their previous year’s territory. We used logistic regression to determine whether site fidelity differed by population or with precipitation in the previous bio-year.

Diet Analysis and Prey Availability—Data Collection

Cactus Wren fecal samples were analyzed to identify arthropod taxa that were considered Cactus Wren prey and we assessed the abundance of these taxa by vegetation type within wren territories. Fecal samples were collected from nestlings when a fecal sac was produced during the banding process. In 2016, 43 fecal samples were collected and analyzed for diet content using deoxyribonucleic acid (DNA) barcoding and screened against the custom database from the San Diego Natural History Museum arthropod collections of our study areas to obtain operational taxonomic units (OTU), which is a quantified representation of each taxon within a fecal sample (A. Vandergast, unpub. data, 2020). The genetic amplification and analysis process introduces some known errors that can skew the volume of an OTU within a sample or introduce contaminants, so a taxon was excluded from further examination if it accounted for less than 1 percent of the OTUs in a given fecal sample. Operational taxonomic units were grouped into arthropod orders and families and identified to species when possible. The sampling unit for analysis was the number of fecal samples that contained each OTU. Any taxa determined to be present in the environment but not likely to come from the fecal sample (for example, parasites such as mites [order Acari] and thrips [order Thysanoptera], or taxa that were likely miscoded such as Decapoda) were removed from further examination. It is worth noting that fecal samples were collected from nestlings only. It is likely that adult Cactus Wrens consumed similar prey to what was fed to nestlings, but there may be some items missing from our analysis that adult Cactus Wrens consumed.

The San Diego Natural History Museum collected arthropods in 2016 at 9 monitored territories in the Otay Cactus Wren population and at 14 monitored territories in the San Diego Cactus Wren population (San Diego Natural History Museum, unpub. data, 2017). Using pitfall traps (for ground and litter-dwelling arthropods) and back-pack vacuums (for canopy-dwelling and aerial arthropods), arthropods were collected in the following vegetation species known to be used by foraging Cactus Wrens: blue elderberry (Sambucus nigra spp. caerulea; hereafter “elderberry”), cactus (cholla or prickly pear), California sagebrush (Artemisia californica; hereafter “sagebrush”), California buckwheat (Eriogonum fasciculatum; hereafter “buckwheat”), lemonadeberry (Rhus integrifolia), native bunch grass, non-native grass (multiple species), and black mustard (Brassica nigra; hereafter “mustard”). Arthropods also were collected from bare ground using pitfall traps (no vacuum samples were collected from bare ground). We refer to the plant species and bare ground collectively as “vegetation types.” Pitfall traps were open for 10 to 15 days and target plants were vacuumed for 30 seconds to collect arthropods from the foliage. Any large arthropods flushed while vacuuming also were collected when possible. Not all vegetation types were adequately represented at every Cactus Wren territory to support a pitfall trap or to provide enough substrate for a vacuum sample in 2016, when prey sampling occurred. Also, pitfall traps were sometimes disturbed, preventing the collection of arthropods at that vegetation type. Plant types that were not sampled at all territories included native bunch grass (sampled at 6 territories), elderberry (sampled at 10 territories), lemonadeberry (sampled at 10 territories), mustard (sampled at 21 territories), buckwheat (sampled at 22 territories), and non-native grass (sampled at 22 territories). Samples were collected during three periods of the Cactus Wren breeding season corresponding to early nesting (courtship, nest building, egg laying, February–early March), peak nesting (feeding hatchlings, May–June), and late nesting (feeding fledglings, independent fledglings, September). Arthropods greater than 5 mm in length were identified, counted, and sorted into groups or taxa with known or presumed significance to Cactus Wren (Hamilton and others, 2020). Arthropod biomass was not calculated.

Diet Analysis and Prey Availability—Data Analysis

To determine which arthropod taxa to include in prey abundance analyses, we used R package adehabitatHS (Calenge, 2006) to calculate resource selection indices (w_i) which examined the relationship between arthropods consumed by Cactus Wrens (found in fecal samples) and available (collected in pitfall traps and by vacuum; Reynolds and others, 2006):

Values of w_i greater than 1 were evidence of selection (used disproportionately more than their availability), values less than 1 were evidence of avoidance (used less than their availability), and values equal to 1 indicated arthropods were used in proportion to their availability. We calculated 95-percent confidence intervals from standard errors (SE) provided for each w_i by adehabitatHS:

Resource selection indices were considered significant if the 95-percent confidence intervals for w_i did not contain 1 (Reynolds and others, 2006). First, we calculated resource selection indices for arthropods by orders in which they were classified. Next, we calculated resource selection indices for arthropod families to identify if there were any families that drove the selection or avoidance at the order level. Families that were avoided were not considered prey and were excluded from further arthropod prey analyses. Arthropod families that were selected or were neither selected nor avoided were considered prey and included in subsequent analyses. We included all unidentified larval arthropods in our Cactus Wren analyses as prey despite the inability to classify them into orders or families.

The sampling units for arthropod prey abundance analyses were the number of arthropod prey collected per sampling event (pitfall trap session or vacuuming session) by vegetation type, Cactus Wren territory (to control for differences in sampling between territories), and sampling period. Because only a small proportion of arthropods were collected by vacuum, we performed further analyses only on arthropods collected by pitfall trap.

We used generalized linear mixed modeling (GLMM; Faraway, 2006) to build a series of models describing the relationship between vegetation type, sampling period, and the number of arthropod prey collected in pitfall traps. We included sampling period as a fixed effect because it did not have enough levels to include as a random effect. Cactus Wren territory was a random effect, and we nested vegetation type (all pitfall trap samples collected at a unique vegetation type location) within Cactus Wren territory also as a random effect and treated the arthropod data as Poisson-distributed. We used AIC_c to evaluate models; the model with the lowest AIC_c and any model within two AIC_c of the lowest model were considered well supported. Odds ratios were examined to evaluate the influence of each level of each variable (fixed effects) on arthropod abundance. If there was support for multiple models (ΔAIC_c less than 2), we averaged all models using AIC_c weights to obtain real estimates of arthropod abundance for each level of each variable.

Vegetation Composition—Data Collection

We collected and analyzed vegetation data at Cactus Wren territories to compare vegetation structure and species composition across territories and relate it to arthropod prey availability and wren productivity. Permanent vegetation sampling plots were established in 2015 at the same 23 monitored wren territories where arthropods were collected. Vegetation plots consisted of 30 points arranged in a 5 x 6 grid with 15 m between points in both directions. Points were revisited annually at the end of the breeding season and the presence or absence of the following species recorded within a 2-m-radius circle surrounding the point: native bunch grasses, non-native annual grasses, cactus, elderberry, lemonadeberry, sagebrush, buckwheat, and mustard. Percent cover of each species was calculated as the percent of the 30 sampling points at which the species was present. The substrate at the center of each point was recorded as plant, bare ground, or litter to calculate the percent of bare ground within each plot. In 2016, we did not estimate cover of shrubs, but instead only recorded presence or absence of herbaceous species.

Vegetation Composition—Data Analysis

Shrub data were not collected in 2016. The lack of data presented several options for our analyses. These included ignoring 2016 altogether, pooling data across years, or interpolating the missing values (estimating 2016 shrub cover based on the known data points in 2015, 2017–19 using linear regression). Ignoring 2016 and pooling data across years entailed a loss of information, particularly in analyses of the relationship between nest success and vegetation cover where inter-annual variation was one of our main interests. Thus, we felt that interpolation of the 2016 cover values, especially if interpretation of patterns for that year was done cautiously, was the more appealing option. We reasoned that woody species are not prone to major shifts in cover between successive years (unless there was a major disturbance such as a fire, landslide, and so forth), autocorrelation between successive years tended to be in the strong to moderate range, 2016 was the fourth consecutive drought year (interannual variation over the last several years was relatively low), the sampling design gave us the capability of estimating cover in each territory, and we would be interpolating and not extrapolating the cover values. These five conditions taken together indicated to us that interpolated cover values for 2016 would be reasonable and provide us greater flexibility and ecological insight than simply removing that year from the analyses or pooling data across years. We used GLMM to estimate the missing cover values and then used those in subsequent analyses. We used a binomial error structure in the model, with cover estimated as the proportion of point intercept hits out of 30 points. Year was specified as the fixed effect and coded as time (for example, 2015=1, 2016=2, and so forth) to avoid problems with convergence because of large numbers. Territory was specified as a random effect, which allowed us to estimate cover in 2016 for each one. We performed a preliminary analysis to evaluate random slope-intercept models, but these did not converge because of singularity (insufficient sample size relative to the number of parameters in the model). Therefore, all estimates of shrub cover in 2016 were derived from random intercept models.

Analysis of vegetation data entailed two separate but related steps. The first consisted of a multivariate analysis (redundancy analysis [RDA]; Buttigieg and Ramette, 2014) focused on differences in vegetation cover among territories. This step allowed us to evaluate: (1) the range in variation of vegetation cover among territories, (2) the degree to which vegetation cover varied by year, (3) the most important variables for differentiating vegetation across the territories, and (4) how distinct the populations (Otay and San Diego) were in vegetation structure. In the second step, we used a GLMM to relate the most important vegetation variables from the RDA to the number of fledglings produced per pair.

We used the RDA to evaluate the degree to which vegetation cover varied among territories and years. Redundancy analysis combines principal components analysis (PCA) and multiple regression to summarize variation in response variables (in our case, percent cover of vegetation types), such as (1) among sampling units (territories within each year), and (2) in relation to a set of explanatory (or predictor, environmental) variables. The explanatory variables can be either continuous (for example, elevation, soil moisture) or categorical (for example, burned or unburned); when explanatory variables are categorical, RDA can be conceptualized as a form of multivariate analysis of variance (ANOVA). Redundancy analysis creates two sets of ordination axes, one based on linear combinations of response variables (vegetation types; the PCA component) and one based on the linear combination of response variables after they are “constrained” by the explanatory variables (territory and year); the multiple regression component. Thus, RDA is among multivariate methods known as constrained ordinations. The PCA step provides the total amount of variation in vegetation species cover and the regression gives the proportion of that variation that is accounted for by the explanatory variables. One of the notable strengths of constrained ordination is it allows simultaneous evaluation of overall variation in a dataset and variation relative to explanatory variables. Other variables can be included as covariates, which further enables evaluation of the relative contribution of explanatory variables, covariates, and residual variation to overall variation in the data. Scores are derived for sampling units, species (vegetation types), and the explanatory variables in ordination space, and significance of the overall ordination and the variables in the model are evaluated with Monte Carlo permutation tests. This makes RDA free of the restrictive assumptions of more traditional parametric multivariate analyses, such as multivariate ANOVA, discriminant analysis, and canonical correlation analysis.

We specified two separate RDA models. Territory was the explanatory variable and year was a covariate in the first model, whereas year was the explanatory variable and territory was the covariate in the second model. This approach allowed us to calculate the relative contributions of territory and year to the overall variation in the data; in other words, the variability among territories within years, and the variability in vegetation across years, which differed in precipitation. Territory and year were each considered a categorical variable in both models. Although uncommon vegetation species have little effect on the results of an RDA, they can distort visualization of the ordination axes and scores. Thus, we performed a preliminary analysis to determine if any of the vegetation variables could be removed from the models. Based on this, we removed elderberry, lemonadeberry, and native bunch grass from further analyses, resulting in a matrix with dimensions of 115 sampling units (23 territories x 5 years) and 6 vegetation variables (bare ground, buckwheat, cactus, mustard, non-native grass, and sagebrush). Monte Carlo tests were based on 999 permutations.

We used Poisson regression with the number of fledglings per pair as the response variable in a GLMM (Christensen, 2019) with territory as a random effect to analyze the effects of vegetation composition and bio-year precipitation on annual productivity. We used the results of the RDA as a guide to determine which vegetation variables to include in the models. We included precipitation in the analysis to compare the relative effect of precipitation to the contribution of the vegetation variables to the model. To compare the magnitude of the effects of each model parameter, we standardized the vegetation and precipitation variables by calculating the z scores for each variable, rescaling the variables to have a mean of zero and a standard deviation of one. Z scores are calculated using the following formula:

Breeding Productivity

Nesting activity was monitored annually at 10–13 territories in the Otay population (fig. 3A, 3B) and 14–18 territories in the San Diego population (fig. 3C, 3D) from 2015 through 2019 (table 5). All territories except three in 2016 (two in Otay and one in San Diego) and two in 2017 (both in San Diego) were considered fully monitored, meaning that all brood nests within the territory were found and documented during the breeding season. One or more brood nests may have been missed at partially monitored territories. All monitored territories were occupied by pairs except two territories that were occupied by single birds in 2015, five in 2016, and two in 2019. We monitored 46–74 Cactus Wren brood nests per year. Of these nests, 46–71 were in fully monitored territories (table 6).

Nesting Attempts

The Cactus Wren breeding season began in January or February and ended in July or August. The earliest first nest initiation (first egg laid in the first nest) was on January 23, 2017, in the San Diego population and February 12, 2019, in the Otay population. The latest first nest initiation was April 12, 2018, in the Otay population and February 26, 2016, in the San Diego population.

Fully monitored Cactus Wren pairs laid eggs in zero to four brood nests each year. The number of brood nests per Cactus Wren pair did not differ between populations; however, there was a significant, relationship between the number of brood nests and the amount of precipitation (for every 1-cm increase in precipitation, there was a 1-percent increase in the number of brood nests per Cactus Wren pair; table 7).

Table 7.    

Results of Poisson regression of the number of Cactus Wren brood nests per pair as a function of the amount of precipitation and by population (Otay or San Diego).

[Incident rate ratio is the multiplicative amount of change in number of nests with every 1-centimeter change in precipitation, holding all other variables constant. Precipitation was compiled for the bio-year (July 1–June 30) ending in the current breeding year. Abbreviations: P, probability that the statistical test result was false; <, less than]

Table 7.    Results of Poisson regression of the number of Cactus Wren brood nests per pair as a function of the amount of precipitation and by population (Otay or San Diego).

Coefficient	Estimate	Standard error	Z-value	P	Incident rate ratio
Intercept	0.31	0.18	1.70	10.09	1.37
Precipitation	0.01	0.004	3.04	1<0.01	1.01
Population (San Diego)	0.03	0.12	0.29	0.77	1.04

¹

Significant result.

The number of pairs that attempted a second nest was significantly and positively related to the amount of precipitation and did not differ between populations (table 8). In 2018, the driest year of the study, Cactus Wren pairs were more likely to renest after a failed first nesting attempt than they were after a successful first nesting attempt (fig. 4; Fisher’s Exact test, P=0.07). A similar, but non-significant, pattern was seen in 2015, another dry year (Fisher’s Exact test, P=0.16). In all other years, Cactus Wren pairs were as likely to renest after a failed first nesting attempt as they were after a successful first nesting attempt (Fisher’s Exact test, 2016, 2017, 2019: P>0.99). Over all years, 75–100 percent of Cactus Wren pairs attempted to renest after a failed first nesting attempt, and 38–100 percent attempted to renest after a successful first nesting attempt.

Table 8.    

Results of logistic regression of the number of attempted Cactus Wren renests after a first nesting attempt as a function of the amount of precipitation and by population (Otay or San Diego).

[Precipitation was compiled for the bio-year (July 1–June 30) ending in the current breeding year. Abbreviations: P, probability that the statistical test result was false; <, less than]

Table 8.    Results of logistic regression of the number of attempted Cactus Wren renests after a first nesting attempt as a function of the amount of precipitation and by population (Otay or San Diego).

Category	Estimate	Standard error	Z-value	P
Intercept	−1.78	0.75	−2.38	10.02
Precipitation	0.10	0.02	4.13	1<0.001
Population (San Diego)	0.26	0.51	0.51	0.61

¹

Significant result.

On average, Cactus Wren pairs fledged the highest number of broods in the wettest years, 2017 (1.9±0.7) and 2019 (1.7±1.1), and the lowest number of broods in the driest years, 2018 (0.9±0.6) and 2015 (1.0±0.7). In 4 of the 5 years of our study, a few fully monitored Cactus Wren pairs failed to fledge any nests (fig. 5). The proportion of pairs that did not fledge any nests was similar between populations from 2015 through 2017 (2015: 25 percent at San Diego versus 20 percent at Otay; 2016: 8 percent at San Diego versus 11 percent at Otay; 2017: 0 percent at both populations) until 2018, when 35 percent of San Diego pairs and 9 percent of Otay pairs failed to fledge any nests. Similarly, in 2019, 33 percent of San Diego pairs and no Otay pairs failed to fledge any nests. Several Cactus Wren pairs fledged two or more broods each year from 2015 to 2019. During 3 of 4 years, 4 (2016) or 5 (2017 and 2019) Cactus Wren pairs fledged 3 broods, and in 2019, 1 Cactus Wren pair fledged 4 broods.

Productivity

We found 1,082 Cactus Wren eggs in 295 monitored nests from 2015 to 2019 in the Otay and San Diego populations (tables 11, 12). Cactus Wren clutch size ranged from two to eight eggs. Although we found nests with single eggs, we could not confirm that these were full clutches (for example, we found nests later in the nest cycle that may have lost eggs or egg laying was interrupted and the nest abandoned). Two nests were discovered with eight eggs at the same territory in 2 different years. Both nests were abandoned shortly after eggs were laid. One other nest was found containing six nestlings, all of which successfully fledged. We do not know if this nest initially contained more than six eggs. Clutch size did not differ between populations but was positively associated with precipitation (for every 1-cm increase in precipitation, there was a 1-percent increase in clutch size; table 13; fig. 6). The effect of precipitation on wren clutch size was most pronounced between the wet years of 2017 and 2019 and the extreme dry year of 2018 (fig. 6).

Table 11.    

Reproductive success and productivity of nesting Cactus Wrens in the Otay population, San Diego County, 2015–19.

[Calculations of average clutch size include only nests that had a known full clutch. Calculations of average young fledged per pair include only fully monitored Cactus Wren pairs. Abbreviation: ±, plus or minus]

Table 11.    Reproductive success and productivity of nesting Cactus Wrens in the Otay population, San Diego County, 2015–19.

Parameter	2015	2016	2017	2018	2019
Nests with eggs	20	27	31	18	31
Eggs laid	78	98	114	56	124
Average clutch size	4.0±0.6	3.8±0.8	3.9±0.8	3.4±0.7	4.5±0.5
Hatchlings	47	63	92	37	102
Nests with hatchlings	15	20	29	13	30
Hatching success
Percent of eggs that hatched	60	64	81	66	82
Percent of nests with eggs in which at least one hatched	75	74	94	72	97
Fledglings	35	57	76	33	92
Nests with fledglings	11	18	24	12	28
Fledging success
Percent of hatchlings that fledged	74	90	83	89	90
Percent of nests with hatchlings from which at least one fledged	73	90	83	92	93
Fledglings per egg	0.4	0.6	0.7	0.6	0.7
Average number of young fledged per pair	3.5±2.3	5.2±3.1	6.3±2.8	2.5±1.9	7.7±2.5
Total number of pairs fully monitored	10	11	12	13	12
Pairs fledging at least one young (percent)	8 (80)	10 (91)	12 (100)	10 (77)	12 (100)

Table 12.    

Reproductive success and productivity of nesting Cactus Wrens in the San Diego population, San Diego County, 2015–19.

[Calculations of average clutch size include only nests that had a known full clutch. Calculations of average young fledged per pair includes only fully monitored Cactus Wren pairs. Abbreviation: ±, plus or minus]

Table 12.    Reproductive success and productivity of nesting Cactus Wrens in the San Diego population, San Diego County, 2015–19.

Parameter	2015	2016	2017	2018	2019
Nests with eggs	26	30	43	30	39
Eggs laid	85	117	170	87	153
Average clutch size	3.1±0.6	3.8±0.7	3.9±0.8	3.0±0.4	4.1±0.7
Hatchlings	46	77	129	61	99
Nests with hatchlings	15	22	38	23	31
Hatching success
Percent of eggs that hatched	54	66	76	70	65
Percent of nests with eggs in which at least one hatched	58	73	88	77	79
Fledglings	34	64	99	35	54
Nests with fledglings	12	18	29	13	18
Fledging success
Percent of hatchlings that fledged	74	83	77	57	55
Percent of nests with hatchlings from which at least one fledged	80	82	76	57	58
Fledglings per egg	0.4	0.5	0.6	0.4	0.4
Average number of young fledged per pair	2.8±2.4	5.3±3.1	6.4±2.7	1.9±2.0	3.6±3.3
Total number of pairs fully monitored	12	13	17	18	15
Pairs fledging at least one young (percent)	9 (75)	11 (85)	15 (88)	11 (61)	10 (67)

Table 13.    

Results of Poisson regression of Cactus Wren clutch size as a function of the amount of precipitation and of population (Otay or San Diego).

[Incident rate ratio is the multiplicative amount of change in clutch size with every 1-centimeter change in precipitation, holding all other variables constant. Precipitation was compiled for the bio-year (July 1–June 30) ending in the current breeding year. The Otay population is the reference and, therefore, is included in the intercept. Abbreviations: P, probability that the statistical test result was false; <, less than]

Table 13.    Results of Poisson regression of Cactus Wren clutch size as a function of the amount of precipitation and of population (Otay or San Diego).

Coefficient	Estimate	Standard error	Z-value	P	Incident rate ratio
Intercept	1.16	0.12	9.97	1<0.001	3.21
Precipitation	0.01	0.003	2.10	10.04	1.01
Population (San Diego)	−0.05	0.07	−0.75	0.45	0.95

¹

Significant result.

The percent of eggs that hatched was higher as precipitation increased and was higher in the Otay population than in the San Diego population (table 14). Disparities within years between the percent of eggs that hatched and the percent of nests with eggs that hatched indicate that the factors affecting hatch rate were operating at the egg level (for example, infertile or inviable eggs, crushed eggs, partial predation) rather than at the nest level. At least one egg was lost from 87 nests before the date the nest failed (26 nests) or was ultimately successful (61 nests), whereas 59 nests failed during the egg stage. In contrast, the percent of hatchlings that fledged and the percent of nests with hatchlings that fledged at least one young were similar within years, indicating that factors affecting fledge rate were operating at the nest level. The percent of hatchlings that fledged was higher in the Otay population than in the San Diego population but was not significantly related to precipitation. In the Otay population, the percent of eggs that hatched was consistently lower than the percent of hatchlings that fledged indicating that factors causing partial nest failure were occurring in late egg or early nestling stages. In the San Diego population, the percent of eggs that hatched was lower than the percent of hatchlings that fledged from 2015 through 2017, but in 2018 and 2019, the relationship was reversed. The number of fledglings produced per egg combined the egg hatch rate and the hatchling fledge rate, and therefore appeared higher in the Otay population but did not track bio-year precipitation.

Table 14.    

Results of logistic regressions of Cactus Wren hatching and fledging success as a function of the amount of precipitation and by population (Otay or San Diego).

[Precipitation was compiled for the bio-year (July 1–June 30) ending in the current breeding year. The Otay population is the reference and, therefore, is included in the intercept. Abbreviation: P, probability that the statistical test result was false; <, less than]

Table 14.    Results of logistic regressions of Cactus Wren hatching and fledging success as a function of the amount of precipitation and by population (Otay or San Diego).

Category	Estimate	Standard error	Z-value	P
Percent of eggs that hatched
Intercept	0.23	0.21	1.08	0.28
Precipitation	0.02	0.004	3.90	1<0.001
Population (San Diego)	−0.23	0.14	−1.91	10.06
Percent of hatchlings that fledged
Intercept	1.61	0.31	5.25	1<0.001
Precipitation	0.01	0.01	0.76	0.45
Population (San Diego)	−0.99	0.19	−5.23	1<0.001

¹

Significant result.

We found 579 fledglings in all fully and partially monitored Cactus Wren territories from 2015 to 2019 in the Otay and San Diego populations (tables 11, 12). The number of fledglings per fully monitored pair ranged from 0 to 12. The average number of fledglings per fully monitored pair was higher with more precipitation and was higher in the Otay population than in the San Diego population (table 15; fig. 7).

Table 15.    

Results of quasi-Poisson regression of the number of Cactus Wren fledglings produced per pair as a function of the amount of precipitation and of population (Otay or San Diego).

[Incident rate ratio is the multiplicative amount of change in number of fledglings per pair with every 1-centimeter change in precipitation, holding all other variables constant. Precipitation was compiled for the bio-year (July 1–June 30) ending in the current breeding year. The Otay population is the reference and, therefore, is included in the intercept. Abbreviations: P, probability that the statistical test result was false; <, less than]

Table 15.    Results of quasi-Poisson regression of the number of Cactus Wren fledglings produced per pair as a function of the amount of precipitation and of population (Otay or San Diego).

Coefficient	Estimate	Standard error	t-value	P	Incident rate ratio
Intercept	0.64	0.19	3.46	1<0.001	1.90
Precipitation	0.03	0.004	6.19	1<0.001	1.03
Population (San Diego)	−0.24	0.11	−2.18	10.03	0.78

¹

Significant result.

Daily Nest Survival

Using Cactus Wren population and precipitation as covariates, we built five models with potential to predict that a nest would survive from one day to the next. The constant model was least supported of all models (table 16). Adding precipitation and population separately moderately improved model fit. The models that included both precipitation and population were the two best supported for predicting Cactus Wren daily nest survival. The top model showed that nests had a higher likelihood of survival in the Otay population than the San Diego population (table 17; fig. 8). The 95-percent confidence interval of the odds ratio did not include 1, indicating that population was a significant contributor to this model. The odds ratio for the San Diego population in this model was 0.59, indicating that when precipitation was held constant, the odds that a nest in the San Diego population would survive 1 day were 41 percent less than the odds that a nest in the Otay population would survive 1 day. In the top model, precipitation had an odds ratio of 1.03 and the lower boundary of the 95-percent confidence interval was slightly above 1, indicating that for every 1-cm increase in precipitation DSR increased by 3 percent. The second-ranked model included an interaction between population and precipitation, specifically that while DSR tracked precipitation in both populations from 2015 to 2018, in 2019, Otay DSR responded to high precipitation while San Diego DSR did not (fig. 8). This interaction, along with the population variable, did not contribute significantly to predicting DSR because the 95-percent confidence intervals of the odds ratios included 1. In the second-ranked model, the odds ratio for precipitation was 1.07, indicating that precipitation had a slightly positive effect on DSR (for every 1-cm increase in precipitation, DSR increased by 7 percent). Population was included in the top three models and in two of these models, DSR for the Otay population was significantly higher than the DSR for the San Diego population (95-percent confidence interval of the odds ratio did not include 1).

Table 16.    

Logistic regression models of daily Cactus Wren nest survival as a function of the amount of precipitation and Cactus Wren population (Otay or San Diego) in southern San Diego County, 2015–19.

[Models are ranked from best to worst based on Akaike’s Information Criterion for small samples (AIC_c), the difference between the model’s AIC_c and the highest-ranked model’s AIC_c (ΔAIC_c), and AIC_c weights. AIC_c is based on −2 x log_e likelihood and the number of parameters in the model. Precipitation was compiled for the bio-year (July 1–June 30) ending in the current breeding year.]

Table 16.    Logistic regression models of daily Cactus Wren nest survival as a function of the amount of precipitation and Cactus Wren population (Otay or San Diego) in southern San Diego County, 2015–19.

Model	Deviance	Number of parameters	AICc	ΔAICc	AICc weight
Population + precipitation	713.2	3	719.2	0	0.45
Population * precipitation	711.8	4	719.8	0.6	0.33
Population	717.3	2	721.3	2.1	0.16
Precipitation	719.6	2	723.6	4.5	0.05
Constant	724.1	1	726.1	6.9	0.01

Table 17.    

Parameter estimate (β), standard error, odds ratios and 95-percent confidence intervals for the top three models explaining daily survival rate of Cactus Wren nests at Otay and San Diego populations, southern San Diego County, 2015–19.

[Models are in order of best-supported to least-supported. Reference represents Cactus Wren nests in the Otay population in 2015. All other values are relative to the reference values. Precipitation was compiled for the bio-year (July 1–June 30) ending in the current breeding year.]

Table 17.    Parameter estimate (β), standard error, odds ratios and 95-percent confidence intervals for the top three models explaining daily survival rate of Cactus Wren nests at Otay and San Diego populations, southern San Diego County, 2015–19.

Effect	β	Standard error	Odds ratio	95-percent confidence interval
Population + precipitation
Reference	4.1	0.3	60.47	132.73–111.75
Precipitation	0.0	0.0	1.03	11.001–1.07
Population (San Diego)	−0.5	0.2	0.59	10.39–0.90
Population * precipitation
Reference	3.7	0.5	39.02	115.56–97.86
Precipitation	0.1	0.0	1.07	11.005–1.14
Population (San Diego)	0.1	0.6	1.12	0.36–3.42
Precipitation * population (San Diego)	0.0	0.0	0.96	0.89–1.03
Population
Reference	4.6	0.2	104.10	173.1–148.32
Population (San Diego)	−0.5	0.2	0.58	10.38–0.89

¹

Values indicate 95-percent confidence intervals that didn’t include 1; those variables were considered significant contributors to the model.

Survivorship and Movement

We banded 629 Cactus Wrens between 2015 and 2019 (table 18). In total, 579 first-year wrens were banded: 260 in the Otay population and 319 in the San Diego population. A total of 50 Cactus Wrens were banded as adults: 9 in the Otay population and 41 in the San Diego population.

During surveys and monitoring between 2015 and 2019, we resighted 301 adult birds with color-bands (153 individuals, some of which were resighted in multiple years; tables 19, 20), 105 in the Otay population (101 individuals) and 196 in the San Diego population (52 individuals), most of which we were able to identify. One additional bird that had originally been banded as a hatchling in the San Diego population was identified in the San Pasqual Valley as a part of a separate study; that bird was excluded from our analyses. We were unable to identify seven birds as a result of missing bands. Adult birds ranged in age from 1 to 8 years old (table 20), with the oldest birds banded initially in 2011 in a previous study (Barr and others, 2015), one of which was present for all 5 years of the project.

Survivorship

The best models explaining annual survival of Cactus Wrens included age (first-year or adult) and either year or precipitation in the previous bio-year (ΔAIC_c=0.3; table 21). Examination of the odds ratios showed that survival of adult Cactus Wrens was significantly higher than that of first-year Cactus Wrens (table 22). Age was an important variable in all of the models. In both models that contained year we found that survival was lower in 2018–19 than in any of the other time intervals. Although there was less support for an effect of precipitation on survival, there was some evidence (odds ratio=1.01, with the 95-percent confidence interval boundary just above 1) that survival may be positively correlated with precipitation (table 22; fig. 9). There was less support for a population effect on survival; examination of the odds ratio in all models that included population suggested that population did not contribute significantly to the model (the 95-percent confidence interval spanned 1).

We averaged models to obtain estimates of annual survival for adult and first-year Cactus Wrens. Annual survival for adults ranged from 60 to 70 percent whereas first-year survival ranged between 20 and 28 percent (table 23; fig. 9). Survival was similar for the San Diego and Otay populations. The highest values occurred in 2017–18 and the lowest in 2018–19 for both age classes and in both populations, coinciding with the years of highest and lowest precipitation (fig. 9). Detection probability was high (0.94) for adult and first-year wrens.

Movement

Diet Analysis and Prey Sampling

A total of 37,111 individual arthropods of 19 orders and at least 128 families were collected in pitfall traps and by vacuum in 2016. Of the 37,111 individual arthropods, 2 percent (902/37,111) of arthropods were collected by vacuum while the remainder were collected in pitfall traps. Of these, 36,663 (99 percent) were classified in orders identified in fecal samples. Of all arthropods that were collected in pitfall traps and by vacuum, 448 belonged to orders not represented in fecal samples (class Arachnida, orders Opiliones: 112, Pseudoscorpiones: 3, and Solfugida: 26) or were not identified to order (class Diplopoda: 48, class Insecta: 13, unidentified larvae: 246). Of all arthropods collected, 15,900 (43 percent) belonged to families that were found in fecal samples (15,823 collected in pitfall traps, 77 collected by vacuum); the remaining 57 percent were either not identified or belonged to families that were not identified in fecal samples.

We found 10 arthropod orders in more than 10 percent of Cactus Wren nestling fecal samples (table 30). We found 16 arthropod families of 9 orders in more than 10 percent of fecal samples. No single arthropod family in the order Neuroptera was represented in more than 10 percent of fecal samples, although collectively, neuropterans were found in 12 percent of fecal samples. All 10 orders and 12 of the 16 families most commonly found in fecal samples were collected in pitfall traps and by vacuum. Four of the arthropod families found commonly in fecal samples were not collected by pitfall traps or by vacuum.

The most abundant arthropod order collected was Hymenoptera (36 percent), followed by Isopoda (23 percent), Diptera (16 percent), Coleoptera (10 percent), and Hemiptera (10 percent). Each of the remaining arthropod orders made up less than 1 percent of all arthropods collected. Cactus Wrens consumed arthropods in the order Hymenoptera significantly less than their availability (table 31). Selection indices for all other arthropod orders were within the 95-percent confidence intervals indicating that Cactus Wrens did not significantly select or avoid them.

We identified 29 arthropod families in fecal samples and that were also collected in pitfall traps or by vacuum. Two of these families (Isopoda Porcellionidae [woodlice] and Hymenoptera Formicidae [ants]) were avoided (consumed significantly less than their availability; table 32). All other families were consumed in proportion to their availability. Because woodlice and ants were avoided, we excluded them from further prey abundance analyses. Additionally, individuals within orders and classes not identified in fecal samples and individuals identified only to class Insecta were excluded from further prey abundance analyses. Unidentified larvae (246) were included in prey analyses.

After excluding the arthropod taxa that were avoided or not represented in fecal samples, the remaining 17,697 individual arthropods were considered Cactus Wren prey. Of the arthropods considered Cactus Wren prey, 95 percent (16,823) were captured in pitfall traps and 5 percent (874) were captured by vacuum. Diptera (32 percent) was the most abundant order within the taxa considered Cactus Wren prey, followed by Coleoptera (20 percent), Hemiptera (20 percent), Hymenoptera (12 percent), and Aranea (9 percent). All other taxa represented 1 percent or less of the number of arthropods that were considered Cactus Wren prey.

We created a series of GLMMs to determine the contribution of vegetation type and sampling period to the abundance of arthropods collected in pitfall traps. The highest ranked model contained the interaction between vegetation type and sample period and the AIC_c weight was 1, indicating that no other models carried enough weight to be considered viable (table 33). In this model, the 95-percent confidence interval for the odds ratio of all vegetation types (when not interacting with sample period) included 1, indicating that arthropod abundance by vegetation type alone did not contribute significantly to the model (table 34). No other model had an AIC_c score within two of the top ranked model. The magnitude of the variance for the random effects (vegetation type within territory=0.50–0.52 for all models; territory=0.03–0.05 for all models) was comparable in scale to the parameter estimates in all models, indicating that the random effects were strong (there was high variability in arthropod abundance between territories and vegetation types within territories).

According to the top model, arthropod prey abundance was 17 percent higher in the peak nesting period than in the early nesting period and was 33 percent lower in the late nesting period than in the early nesting period (table 34; fig. 10). Arthropod abundance was also influenced by interactions between vegetation type and sampling period in the top model. Vegetation types departing from the “peak greater than early greater than late” pattern of arthropod seasonal abundance included elderberry, cactus, buckwheat, native bunch grass, and mustard, all of which exhibited a pattern of “early greater than peak greater than late” arthropod availability (fig. 10). Notably, during the early nesting period, arthropods were most abundant in native bunch grass and least abundant in lemonadeberry. Arthropods were most abundant in elderberry (although not significantly represented in the models) and non-native grass and least abundant in lemonadeberry and mustard in the late nesting period. During peak nesting period, arthropods were most abundant in native bunch grass and in patches of bare ground (although not significantly represented in the models) and least abundant in cactus and elderberry. Differences in arthropod prey abundance between vegetation types alone were minor and no vegetation type stood out as supporting substantially more or fewer arthropods than any other.

Insufficient arthropods were captured during vacuum sampling to incorporate into GLMM analysis. Visual inspection of the capture rate indicates that the highest arthropod abundance was in non-native grass and sagebrush (fig. 11).

Vegetation Composition

Vegetation was sampled at 23 territories, 9 in the Otay population and 14 in the San Diego population, from 2015 to 2019. We measured cover each year for three herbaceous species (native bunch grasses, non-native grasses, and mustard), for five shrub species (cactus, elderberry, lemonadeberry, sagebrush, and buckwheat), and for bare ground. Four of the vegetation types were widespread, occurring at all territories; however, elderberry, lemonadeberry, native bunch grass, and mustard were absent at one or more Cactus Wren territories. Vegetation types that accounted for the most cover in the Otay population were cactus, buckwheat, and non-native grasses, each occurring in at least 60 percent of sampling points across territories in most years (fig. 12). In the San Diego population, cactus, sagebrush, buckwheat, non-native grass, and mustard were present in at least 40 percent of sampling points across territories in most years (fig. 13). One notable difference between the Otay and San Diego populations was the absence of lemonadeberry or native bunch grass at all Lake Jennings territories within the San Diego population. Laurel sumac (Malosma laurina), which is floristically and structurally similar to lemonadeberry, was the prevalent shrub at Lake Jennings. In contrast, lemonadeberry and native bunch grass were present at all but two of the Otay territories. Elderberry was not common in either population and was only sampled at four territories (two in Otay and two in San Diego). The cover of individual shrub species was fairly stable through time, with small annual changes. We found greater annual variation within the herbaceous species and bare ground.

Variation in Vegetation Composition Between Territories and Years

Model 1: Variation Among Territories

The first RDA model characterized the variation in the vegetation composition and cover across territories, with year as a covariate. We found that vegetation cover varied widely among territories. A high proportion (69 percent) of the total variation in the data was explained by the effect of territory (F=19.9, P<0.001); this was approximately four times greater than that explained by the covariate year (table 35). Monte Carlo permutation tests indicated that the first five RDA axes were significant (F>30.6, P<0.001). Only three of the axes had a clear ecological interpretation; thus, our interpretation is restricted to these axes (table 36).

Table 35.    

Partitioning of the variance in vegetation cover from redundancy analysis (RDA) describing the proportion of the variance explained by Cactus Wren territory and year, southern San Diego County, 2015–19.

Table 35.    Partitioning of the variance in vegetation cover from redundancy analysis (RDA) describing the proportion of the variance explained by Cactus Wren territory and year, southern San Diego County, 2015–19.

Category	Proportion of variance
Total	1.00
Year	0.17
Territory	0.69
Unconstrained	0.14

Table 36.    

Vegetation type scores (loadings) from redundancy analysis for RDA axes 1–3 (scores represent the magnitude of each vegetation type on the axis), the proportion of variation explained by each axis and the cumulative variation explained by each successive axis.

Table 36.    Vegetation type scores (loadings) from redundancy analysis for RDA axes 1–3 (scores represent the magnitude of each vegetation type on the axis), the proportion of variation explained by each axis and the cumulative variation explained by each successive axis.

Category	RDA1	RDA2	RDA3
Non-native grass	−0.04	0.27	−0.52
Sagebrush	0.48	0.64	−0.33
Bare ground	0.22	−0.20	0.05
Mustard	0.95	−0.19	−0.04
Cactus	−0.59	0.51	0.33
Buckwheat	−0.67	−0.35	−0.54
Proportion explained	0.43	0.20	0.17
Cumulative proportion	0.43	0.63	0.80

The first three axes accounted for 80 percent of the variation in vegetation composition and cover attributable to territory. The first axis was a gradient from mustard-dominated to buckwheat- and cactus-dominated territories, the second was a gradient from territories with high cover of sagebrush and cactus to those with high cover of buckwheat, and the third was a gradient from territories with high cover of non-native grass and buckwheat to those with high cover of cactus (table 36). There was little indication territories in the San Diego and Otay populations differed in vegetation composition along the first two RDA axes (fig. 14). Centroids of the territories were generally mixed in ordination space (fig. 14). The 95-percent confidence ellipses indicated vegetation composition within territories did not vary greatly among years.

Territories in the San Diego and Otay populations also were intermixed in ordination space along the second and third axes, although not to the degree as along the first and second axes. A higher proportion of the territories in the San Diego population were characterized by sagebrush and cactus whereas a higher proportion of territories in the Otay population had relatively high cover of buckwheat (fig. 15). Consistent with the first two axes, the 95-percent confidence ellipses and graphs of individual year indicated vegetation composition within territories did not vary greatly among years for the second and third axes (fig. 15).

Model 2: Variation Across Years

The second RDA model characterized the variation in vegetation cover across years, with territory as a covariate. Monte Carlo permutation tests indicated that the first three RDA axes were significant (F>6.9, P<0.001). Only two of the axes had a clear ecological interpretation; thus, our interpretation is restricted to these axes (table 37). Because Model 2 simply interchanged the explanatory variable (year) and covariate (territory), the total variation in the data explained by these two sources were mirror images of those in Model 1 (17.3 percent by year, 69 percent by territory, table 35). Although the amount of total variation accounted for by year was notably less than territory, it nonetheless was a substantial effect (F=27.5, P<0.001).

Table 37.    

Vegetation type scores (loadings) from redundancy analysis for RDA axes 1–3 where the scores represent the magnitude of each vegetation type on the axis, the proportion of variation explained by each axis and the cumulative variation explained by each successive axis.

Table 37.    Vegetation type scores (loadings) from redundancy analysis for RDA axes 1–3 where the scores represent the magnitude of each vegetation type on the axis, the proportion of variation explained by each axis and the cumulative variation explained by each successive axis.

Category	RDA1	RDA2	RDA3
Non-native grass	0.15	0.30	−0.12
Sagebrush	−0.14	0.08	−0.04
Bare ground	−0.91	0.04	0.03
Mustard	−0.11	−0.20	−0.22
Cactus	−0.22	0.08	−0.09
Buckwheat	−0.09	0.03	0.03
Proportion explained	0.81	0.12	0.06
Cumulative proportion	0.81	0.93	0.99

The first two axes accounted for 93 percent of the variation attributable to year. The first axis was a strong gradient from territories with high cover of bare ground to those with low cover of bare ground (table 37). The second axis was a clear gradient from territories with high cover of non-native grass to those with high cover of mustard (table 37). Centroids and the 95-percent confidence ellipses of the years indicated a general progression consistent with variation in inter-annual variation in precipitation, with 2017 and 2019 being years with considerably more precipitation than 2015, 2016 and 2018 (fig. 16). This progression was particularly evident from graphs of the individual years (fig. 17).

Relationship Between Vegetation and Cactus Wren Productivity

We performed a GLMM on the effect of vegetation composition and cover on Cactus Wren productivity. We used the vegetation variables identified in the RDA that were found to have a strong effect on the ordination of the territories (non-native grass, sagebrush, bare ground, mustard, cactus, and buckwheat). Three vegetation variables affected the number of fledglings produced per pair. Percent cover of sagebrush (P=0.07; fig. 18A) had a positive effect on productivity whereas non-native grass (P=0.004; fig. 18C) and mustard (P=0.10; fig. 18B) had negative effects (table 38). Precipitation, as previously discussed (see “Breeding Productivity” section) also was significant in this model, with the number of more fledglings produced with increasing precipitation (P<0.001; fig. 18D). Comparisons of the estimates of the slope coefficients reveal that the precipitation effect (0.42) was 2.6 times stronger than the effect of any vegetation cover variables (highest estimate for vegetation cover variables was −0.16, non-native grass; table 38).

Table 38.    

Results from the generalized linear mixed model on the effects of precipitation and vegetation cover on the number of fledglings produced by Cactus Wrens, southern San Diego County, 2015–19.

[Precipitation was compiled for the bio-year (July 1–June 30) ending in the current breeding year. Abbreviations: P, probability that the statistical test result was false; <, less than]

Table 38.    Results from the generalized linear mixed model on the effects of precipitation and vegetation cover on the number of fledglings produced by Cactus Wrens, southern San Diego County, 2015–19.

Category	Estimate	Standard error	Z-value	P
Intercept	1.29	0.08	17.04	<0.001
Precipitation	0.42	0.08	5.47	1<0.001
Non-native grass	−0.16	0.05	−2.90	10.004
Sagebrush	0.15	0.08	1.78	10.07
Bare ground	0.05	0.08	0.59	0.55
Mustard	−0.15	0.09	−1.67	10.10
Cactus	0.02	0.08	0.22	0.83
Buckwheat	0.00	0.08	−0.06	0.95

¹

Significant result.

To determine whether the vegetation effects on breeding productivity were between-year (as a result of annual changes in precipitation or other effects) or within-year differences we also analyzed each year separately. In 3 of the 5 years, we found a negative effect of non-native grass and mustard on the number of fledglings produced per pair, suggesting that these were within-year effects (table 39). The results for other variables were less clear, appearing either in only a single year (sagebrush and cactus) or having different signs in different years (bare ground and buckwheat).

Table 39.    

Summary of the results from five generalized linear models on the effects of vegetation cover on the number of Cactus Wren fledglings per pair in each year, southern San Diego County, 2015–19.

[Significant results are indicated by a + (positive effect; +, P=0.10; ++, P=0.05; +++, P=0.01; ++++, P=0.001), a - (negative effect; -, P=0.10; --, P=0.05; ---, P=0.01), or ND (no effect). Abbreviation: P, probability that the statistical test result was false]

Table 39.    Summary of the results from five generalized linear models on the effects of vegetation cover on the number of Cactus Wren fledglings per pair in each year, southern San Diego County, 2015–19.

Year	Non-native grass	Sagebrush	Bare ground	Mustard	Cactus	Buckwheat
2015	ND	ND	ND	--	ND	ND
2016	---	ND	ND	-	ND	ND
2017	--	ND	---	ND	ND	--
2018	ND	ND	ND	ND	ND	ND
2019	--	++++	++++	-	+	++