Avian biodiversity in central California vineyards

Point counts

To quantify the avian community among Edna Valley Vineyards, we visited each point count location three times between April 21 and June 10, 2024. On each visit, we conducted two back-to-back 5-min point counts, thus totaling 6 point counts per location. All surveys were performed between 6:00 am and 10:15 am to capture high and stable avian activity (Ralph, Droege & Sauer, 1995), and surveys were done by the same person to eliminate differences in inter-observer bias. During these point counts, we kept track of individuals to only count individual birds once per survey. We recorded the distance, bearing, behavior, activity, and detection method for the first detection of each individual as either aural or visual. For subsequent analyses, observations were limited to those within 100 or 150 m to ensure that we had a reasonable probability of observing any individuals present. Detection probabilities were very similar for both a 100 m and 150 m radius (Table S1). Most species observations were limited to within 100 m, but we used a larger 150 m radius for California scrub-jay (Aphelocoma californica), acorn woodpecker (Melanerpes formicivorus), Nuttall’s woodpecker (Dryobates nuttallii), red-tailed hawk (Buteo jamaicensis), red-shouldered hawk (Buteo lineatus), mourning dove (Zenaida macroura), Eurasian collared-dove (Streptopelia decaocto), California quail (Callipepla californica), and American crow (Corvus brachyrhynchos) observations because they are larger, more conspicuous species that are easy to detect from farther away.

Noise levels

Because human and natural sources of background noise have been shown to influence the distribution of individual bird species and the structure of avian communities (Gomes et al., 2021; Bayne, Habib & Boutin, 2008; Francis, Ortega & Cruz, 2009) and also influence detection probabilities during point counts (Ortega & Francis, 2012; Pacifici, Simons & Pollock, 2008), we quantified background sound levels in two ways. First, to characterize sound levels typical of each point count location, an AudioMoth (1.10.0) was placed on fences, utility poles, trees, and other structures within 4 m of the point count location and at a height of approximately 1 m to record sounds for 24 h. We completed each recording on a weekday to ensure differences in traffic volume were not due to differences in weekday and weekend traffic. The AudioMoths were placed in a plastic bag and a small pouch made of faux fur that functioned as a custom windscreen. AudioMoths were set to record on medium gain with a 48 kHz sample rate (16 bit). We used end-to-end calibration to calibrate each AudioMoth with a LarsonDavis 824 Sound Pressure Meter. From the 24-h recordings we calculated time-averaged A-weighted decibels (L_eq, re 20 µPa) using Kaleidoscope Pro 5.6.8. In preliminary analyses we explored whether morning sound levels (5 am-11 am) were better than the full 24-h recordings in describing diversity patterns, but Akaike Information Criterion corrected for small sample size (AIC_c) comparisons suggested that sound levels based on this truncated window were equivalent or less useful than sound levels derived from full recordings. Second, to account for the possible influence of background sound levels on detections during point counts, we measured the time-averaged A-weighted decibels (L_eq, re 20 µPa) of average background noise with a MicW i436 omni directional microphone paired with the SPLnFFT Sound Meter iPhone application (v7.1) during each survey. Although audio recorders could potentially impact the privacy of people by inadvertently recording them, pedestrian traffic is exceedingly uncommon at all recording points. Pedestrians were never encountered during any of the surveys or other field work. Furthermore, the audio recordings were only used to quantify sound levels at each site by audio processing software, therefore any comments made by a person that was captured by recorder would never be heard by any person.

Structural and land cover variables

Structural and land cover variables were calculated for the area within 150 m of each site. Land cover was hand digitized based on satellite imagery provided by Esri et al. (2024) and from in person observations in order to ensure that any recent land cover alterations were included. Land was categorized as grassland, shrubland, woodland, vineyard, orchard, row crop (e.g., oats, strawberries, or vegetables), or developed. The 31 survey sites did not have adequate variation in woodland cover surrounding them, so this variable was excluded from subsequent analyses. Surrounding vineyard cover varied from natural reference sites with 0% vineyard cover to sites with about 80% vineyard cover.

We used Light Detection and Ranging (LiDAR) data from USGS 3D Elevation Program (U.S. Geological Survey, 2021) to calculate percent canopy cover, the percent of an area covered by trees over 4 m tall, and variation in canopy height with the ‘lidR’ 4.1.2 package in R (Roussel et al., 2020; Roussel & Auty, 2024). We used the rasterize_canopy function to create a raster of the canopy height within a 150 m radius of each survey point and calculated the percent canopy cover as the percent of canopy over 4 m tall within each site. We limited canopy to that over 4 m to exclude grape vines and fruit trees in orchards as they are accounted for in the vineyard and orchard percent land cover variables; however, this variable captures canopy cover from naturally occurring trees and ornamental trees. Variation in canopy height was calculated as the standard deviation of the canopy raster for each site.

Distance to surface water was calculated as the distance from the survey point to the nearest stream or other surface water based on a USGS national hydrography dataset with manual corrections for human-made ponds that are not included in the dataset, but observed during site visits or on satellite imagery (U.S. Geological Survey, 2019).

Data analysis

All analyses were conducted in R version 4.4.1 (R Core Team, 2024). To determine which, if any, environmental variables were correlated with each community metric that we describe below, we fit three models based on our original hypotheses that (1) structural complexity, (2) anthropic environmental features, and (3) natural land cover influenced bird communities at vineyards. Structural complexity models included percent canopy cover, canopy height variation, and distance to surface water. Anthropic environmental feature models included vineyard cover, row crop cover, orchard cover, developed land cover, and noise level. Natural land cover models included grassland and shrubland cover. Model fit was assessed by checking normality of residuals, normality of random effects, multicollinearity, and homogeneity of variance using the check_model function in the “performance” R package (Lüdecke et al., 2021). We ranked each model from the three hypothesis categories based on AIC_c. We chose this modeling technique as opposed to a global model to avoid overfitting the model with too many variables and a relatively small sample size. We then assessed the strength and precision of estimated effects in the top model of each hypothesis category using effect sizes and error metrics, plus p- and F-values. Predictor variables from each hypothesis category that were considered to have an apparent effect on the response were then incorporated into a post hoc model that combined influential variables from each hypothesis category. Interpretation of the influence of predictor variables on responses were based on this final post hoc model. Following recent advice from statisticians (Halsey, 2019; Hurlbert, Levine & Utts, 2019; Wasserstein, Schirm & Lazar, 2019), in this study we sought to use a more nuanced approach to reporting responses among the avian community to landcover variation than the dichotomous approach of significance testing. As such, for linear models we report estimated effect sizes and standard errors (SE) alongside t/z-values and p-values to help convey the magnitude and precision of estimated effects. For analyses involving beta-diversity we complement the presentation of F- and p-values with R²-values to provide a more rounded understanding of which predictor variables explain the most variation in community structure. Thus, we do not consider any effects “significant,” but instead present all apparent trends and attempt to contextualize the magnitude and precision of estimated effects in the text.

Species occupancy

We found notable associations in occupancy and the various environmental variables for six different species (Fig. 2; Table S2). Sound level was most commonly influential for species occupancy. It was positively related to California quail and cliff swallow (Petrochelidon pyrrhonota) occupancy and negatively related to acorn woodpecker occupancy (California quail β = 0.808, SE = 0.403, z = 2.000, p = 0.045; cliff swallow β = 0.366, SE = 0.219, z = 1.670, p = 0.095; acorn woodpecker β = −0.403, SE = 0.199, z = −2.020, p = 0.043). California quail and acorn woodpecker were also positively associated with canopy cover with California quail occupancy reaching 0.50 at about 10% canopy cover and acorn woodpecker occupancy reaching 0.60 at about 10% canopy cover (California quail β = 0.469, SE = 0.246, z = 1.910, p = 0.057; acorn woodpecker β = 0.130, SE = 0.064, z = 2.020, p = 0.044). The only species with a notable association with another agriculture type was the European starling which demonstrated a slight decline in occupancy with orchard cover (European starling β = −0.1226, SE = 0.067, z = −1.838, p = 0.066).

Wrentit occupancy was associated with shrubland cover, and American robin (Turdus migratorius) occupancy was negatively associated with canopy height variation (Wrentit β = 0.215, SE = 0.118, z = 1.830, p = 0.068; American robin β = −0.985, SE = 0.549, z = −1.795, p = 0.073). None of the species’ occupancy was strongly influenced by the amount of vineyard cover. However, vineyard cover was included in the final model for several species (Table S2).

Taxonomic community diversity

Results for coverage-standardized Shannon index and species richness were very similar, with canopy cover, vineyard cover, and shrubland cover included in the final models for both metrics (Table S3). For simplicity, we refer to both Shannon index and species richness as community diversity. Vineyard cover initially appeared informative for community diversity based on the anthropic cover model. However, when added to the post hoc model with canopy and shrubland cover, it was not a strong predictor for community diversity (Tables S4, S5). Canopy cover was strongly correlated with community diversity, and diversity increased with canopy cover up to about 35%, after which there were too few point count locations with higher canopy cover to produce precise estimates (Tables S4, S5) (Fig. 3).

As in the community diversity models, vineyard cover was selected for the beta diversity post hoc model (Table S6); however, it had no explanatory power for community composition in the post hoc model (Table S7). Shrubland cover was marginally useful in explaining community turnover (Shrubland R² = 0.042, F = 1.442, p = 0.171), but canopy cover and average sound level were much more informative (Canopy cover R² = 0.065, F = 2.202, p = 0.016; sound level R² = 0.050, F = 1.697, p = 0.062) (Fig. 4). Overall, the final model explained about 24% of turnover among communities.

Functional community diversity

Based on linear models, we found that community functional evenness increased with canopy cover up to about 35% (β = −38.132, SE = 20.515, t = −1.859, p = 0.045; see Table S10 for other polynomial estimates) (Table S9; Fig. 5). Communities also had lower functional evenness when they were closer to surface water; however, the magnitude of this effect was small (surface water β = 0.063, SE = 0.031, t = 2.049, p = 0.050).

Functional dispersion of the bird communities was strongly and positively correlated with both row crop and canopy cover (row crop β = 0.689, SE = 0.312, t = 2.209, p = 0.036; canopy cover β = 0.713, SE = 0.302, t = 2.360, p = 0.026) (Tables S11, S12; Fig. 5). However, only nine of the 31 sites were surrounded by row crops, thus functional dispersion relationships with this variable should be interpreted with caution.

Canopy cover was again included in the final model for functional richness and had a strong positive influence (β = 1.078, SE = 0.385, t = 2.798, p = 0.009) (Tables S13, S14; Fig. 5). Distance to surface water was weakly correlated with functional richness; however, the effect was very minimal and potentially not biologically relevant (β = −0.057, SE = 0.033, t = −1.728, p = 0.095).

Species occupancy

Of the 22 species that were analyzed, only six species appeared to have occupancy patterns that were related to the environmental variables we considered (Fig. 2; Table S2). Both California quail and cliff swallows were positively associated with sound level, and acorn woodpeckers were negatively associated with sound. These different responses may reflect underlying differences in sensitivity to anthropogenic conditions. Acorn woodpeckers and California quail were also positively associated with increasing canopy cover which aligns with their woodland and scrub habitat preferences. Similarly, wrentits were more likely to occupy vineyards with more shrubland cover corresponding to their known habitat preference for shrubland. American robins were negatively associated with canopy height variation, which likely reflects their preference for environments with tall trees.

Surprisingly, the amount of vineyard cover was not strongly correlated with the occupancy of any species. This could indicate that vineyards are tolerable for the species we analyzed, and that certain management interventions such as increasing canopy cover could make vineyards of varying sizes habitable for birds.

Taxonomic community diversity

Our taxonomic diversity analyses showed clearly that canopy cover is the most important predictor for avian community diversity in and around Edna Valley vineyards. Community diversity increases sharply as canopy cover increases up to about 35%, indicating that even small increases in canopy within vineyards could attract a diverse community of birds. This could be caused by the increase in microhabitats available to birds, or the presence of these trees could be a valuable source of food, shelter, and nesting substrates or materials for the birds. Other research in agricultural and urban areas have also found that tree cover can increase avian diversity (Assandri et al., 2016; Barth, FitzGibbon & Wilson, 2015; Wood & Esaian, 2020). Taxonomic diversity appears to decrease above 35% canopy cover, although the precision of taxonomic diversity estimates with more canopy was low because we only sampled three communities in areas with high canopy cover. This pattern reflects that most current vineyards have relatively low canopy cover. However, it is possible that taxonomic diversity does actually decrease with higher percent canopy cover because the habitat becomes less attractive to grassland and shrubland birds. To fully evaluate whether high levels of canopy cover truly is associated with declines in taxonomic diversity, future research could target vineyards that have or are surrounded by a high proportion of canopy cover. Also needed are studies that explicitly test whether the benefits of tree cover for avian diversity might depend on whether trees are native, naturally occurring trees or if they are non-native ornamental trees.

We also found that community turnover was influenced by anthropic and structural characteristics of the sites. Canopy cover appears to influence the community composition of these bird communities as well as the overall diversity, and sound is only associated with a shift in the community composition. Quiet sites with high canopy cover had bird communities that were very different from loud sites with low canopy cover. This could reflect the differences in urban tolerant and urban intolerant bird communities that have been noted in other studies, as some birds are less susceptible to interference from loud noise (Francis, 2015), such as the cliff swallow and California quail in this study (Table S2). The community turnover associated with variation in canopy cover could be a reflection of a transition from grassland associated birds that prefer lower canopy cover to woodland associated birds that prefer higher canopy cover. Previous research supports this finding as canopy cover can increase the abundance of woodland birds in human-modified environments (Bennett et al., 2014). However, as discussed previously, canopy cover was also associated with higher overall taxonomic diversity compared to communities with low canopy cover.

Functional community diversity

The percent canopy cover at each site was overwhelmingly informative for predicting the functional diversity of the avian community. Canopy cover was positively correlated with functional evenness, dispersion, and richness. In other words, increases in canopy cover is not only associated with more ecological function, it also promotes a more even distribution of birds across functional space and is associated with more dissimilar functions. This finding suggests that even small quantities of trees serve as a keystone structure for birds and helps support a diverse and resilient avian community in vineyards and even a small number of trees appear to have a similar benefit to avian diversity in other human-modified landscapes. For example, scattered trees are associated with high bird diversity and abundance in livestock grazing fields, urban parks, and conifer plantations (Fischer, Stott & Law, 2010; Stagoll et al., 2012; Yamaura, Unno & Royle, 2023). This can in turn increase the services that the avian community can provide for vineyards including insect and rodent management and tourism opportunities (Barbaro et al., 2014; Mols & Visser, 2002), which become more important as agricultural lands expand.

Our results indicate that variation in crop type cover generally increases the diversity and evenness of avian communities in vineyards. Row crop cover was positively associated with functional dispersion indicating that row crops attract different birds than vineyards do. These birds include red-winged blackbirds (Agelaius phoeniceus) and Brewer’s blackbirds (Euphagus cyanocephalus), which were almost exclusively found at vineyards with oats growing nearby. Incorporating other crops such as oats or strawberries into vineyards could benefit vineyard owners by allowing them to maintain or even increase crop productivity while sharing the land with a diverse avian community. Understanding whether the row crop type contributes to taxonomic and functional diversity is necessary to confirm whether all row crops positively influence avian communities.

Community functional diversity is also influenced by proximity to surface water and the proportion of surrounding land covered by vineyards. Nearby natural and human made streams and ponds increase the trait space that avian communities in vineyards occupy. Based on observations during the surveys, the increase in functional richness was likely driven by mallards (Anas platyrhynchos) and American coots (Fulica americana), which were commonly detected in ponds. Woodland birds could also be contributing to the increased richness near streams as riparian vegetation has proven to be very beneficial for woodland birds in agricultural areas (Jedlicka, Greenberg & Raimondi, 2014). Our results also reveal that communities at sites with the highest proportion of vineyard cover had lower functional evenness indicating that large concentrations of vineyards may create an unbalanced distribution of functions within the community. That is, communities embedded within high vineyard cover are dominated by common species with few rare species. This can result in a vulnerable community where the loss of one of these rare species could cause a large difference in ecosystem functions that the bird community provides.

Overall, sites with high taxonomic diversity also had high functional richness (Fig. 6). However functional richness does not take into account the spread of species in trait space, and functional dispersion of these communities is not strongly correlated with taxonomic diversity. While high canopy cover is associated with high functional dispersion and taxonomic diversity, many sites with higher species richness do not have particularly high functional dispersion. This is perhaps because many of the species have similar traits, so they do not increase the functional dispersion of the community, highlighting the importance of using both functional and taxonomic diversity to assess avian communities (Mazel et al. 2018). Vineyard managers are likely most interested in functional diversity of the avian community on their land, as it can be closely tied to ecosystem functions that the birds can provide (Barbaro et al., 2014). Specifically, increasing the presence of functionally distinct birds such as raptors, and insectivorous songbirds could help manage a host of agricultural pests (Kross, Bourbour & Martinico, 2016; García, Miñarro & Martínez-Sastre, 2018).