Barn owls exert top‐down effects on the abundance and behavior of rodent pests

Abstract

Predators can exert top‐down effects on their prey by direct consumption, by imposing risk of predation, or a combination of both. However, empirical evidence for these effects is mixed and inconsistent. Barn owls in agriculture present a system with attributes that should, according to theory, lead to strong top‐down effects, but research is limited and the distinction between effects on prey abundance and behavior remains unresolved. Understanding the top‐down effect of predators on pests is a cornerstone of integrated pest management and important where barn owls may contribute to the control of rodents on farms. Vegetation can mediate the top‐down effects of predators on prey, but this has received little attention for barn owls and rodent pests, limiting the information farmers can use to manage their fields to harness the effect of barn owls. To fill these gaps, we surveyed aboveground foraging rodents and monitored barn owl nest boxes at six winegrape vineyards in Napa Valley, California, from February to July in 2023. We deployed chew block grids to determine the effect of barn owls on rodent abundance and used giving‐up density (GUD) trays and camera traps to determine the effect of barn owls on rodent perceived predation risk and activity, respectively. We found that, over the range of hunting pressures observed (from nearly no owls to about three occupied boxes with six adults and eight chicks within a 1‐km radius), barn owls reduced rodent (predominantly deer mouse, Peromyscus spp.) abundance by 38%–52% and increased perceived predation risk by 16%–38%. Results also suggest that vegetation cover favors the acoustically attuned barn owls over the visually oriented rodent pests. These results contribute to growing evidence that barn owls can be used as biological control in agricultural systems and that vegetation can be managed to optimize their impact. We advise that growers maintain the vine canopy of leaves, pruned twigs on the ground, and vegetation cover in a cover crop row grown between vine rows to elevate perceived predation risk in mice. We also recommend that, to the extent feasible, farmers time mowing to reduce mouse activity when crops are vulnerable.

INTRODUCTION

Ecological theory predicts that predators may exert top‐down effects on their prey by direct consumption (Bowlby & Roff, 1986; Hairston et al., 1960; Menge & Sutherland, 1976; Terborgh, 1992), by affecting behavior from imposed risk of predation (Abramsky et al., 1996; Brown, 1988; Jacob & Brown, 2000; Wheeler & Hik, 2014), or a combination of both (Orrock & Fletcher, 2014). However, empirical evidence for these effects among vertebrates is mixed and inconsistent, with some studies failing to show direct top‐down control of prey (Sinclair & Norton‐Griffiths, 1979; Wright et al., 1994), while others noting strong top‐down effects (Beschta & Ripple, 2009, 2016; Brown, 1984; Peterson, 1999). Previous reviews suggest the relative strength of top‐down effects of predators on prey abundance varies with several ecological contexts including food chain complexity (Finke & Denno, 2005; Polis, 1999; Polis & Strong, 1996; Strong, 1992), nutritional quality of primary productivity (Cebrián, 1999; Leibold, 1989; Polis, 1999), behavioral avoidance of predation by prey (Persson & Greenberg, 1990; Schmitz et al., 2004), and functional traits of predators and prey (Laundré et al., 2014; Tanaka & Mano, 2012; Wirsling & Ripple, 2011). Likewise, some evidence suggests that predators strongly influence the behavior and spatial distribution of their prey by imposing a heterogeneous “landscape of fear” (Crego et al., 2018; Laundré et al., 2001; Orrock et al., 2004), whereas other work suggests that the effect of predators on prey is negligible (Bramley & Waas, 2001; Carthey & Banks, 2016; Fraser & Huntingford, 1986; Garvey et al., 2017; Irving & Magurran, 1997; Ward et al., 1997). The magnitude of top‐down effects on prey behavior also appears to vary with multiple factors such as landscape heterogeneity (Gilliam & Fraser, 1987; Sih, 1987), the predictability of risk (Schmitz et al., 2004), the degree to which risk is associated with reliable cues (Preisser et al., 2007; Weissburg et al., 2014), and the presence of other predators (Cresswell & Quinn, 2013; Kotler et al., 1992; Murray et al., 1995).

Vegetation and abiotic conditions can play strong roles in mediating top‐down effects of predators on their prey (Embar et al., 2011; Ripple & Beschta, 2004; Tabor & Wurtsbaugh, 1991). Predators may exert stronger effects where vegetation and abiotic conditions enhance their capacity to detect and capture prey, which varies with the sensory faculties and morphology of the predator (Camp et al., 2013). For example, structurally diverse vegetation may obscure prey for visually oriented predators, but auditory‐attuned predators may detect prey more readily in dense vegetation (Konishi, 1973). For some predators, vegetation can physically interfere with and decrease prey capture efficiency (Longland & Price, 1991), whereas it can increase capture efficiency for stealth ambush predators (Preisser et al., 2007; Schmitz & Suttle, 2001). Additionally, visually oriented nocturnal predators may detect more prey under lunar illumination (Huck et al., 2017; Rockhill et al., 2013), and acoustically attuned predators may detect less prey with noises such as wind and anthropogenic sounds (Mason et al., 2016; Rochie et al., 2009; Siemers & Schaub, 2010). Wind can also enhance hunting success for soaring aerial predators (Fritz et al., 2000; Weimerskirch et al., 2012) but interfere with the flight control of predators hunting highly mobile prey (Quinn & Cresswell, 2004).

The effect of predators on prey behavior is influenced by how prey assess risk (Lima & Dill, 1990). Prey organisms perceive risk according to their evolutionary history (Abrams, 2000; Soluk & Collins, 1988), morphology, sensory and cognitive capacities (Jurcak & Moore, 2018), and the reliability of direct and indirect cues (Luttberg & Trussell, 2013). For instance, many prey that evolved with aerial predators have an innate visual response to “looming objects” (Pereira & Moita, 2016) to aid escape from overhead threats. Furthermore, structurally complex vegetation is a reliable cue for risk in prey that evolved with ambush predators (Hopcraft et al., 2005; Lischinsky & Lin, 2019). Generally, vegetation can either provide refugia for prey that typically hide to avoid capture or hinder escape for prey that typically flee to avoid capture (Fležar et al., 2019; Martin, 2009; Riginos & Grace, 2008; White, 2010).

Understanding how vegetation mediates effects of predators on the abundance and behavior of prey is especially important where predators may contribute to the control of agricultural pests. A substantial body of literature has examined the effect of predators on insect pests (Klapwijk et al., 2016; Omkar & Kumar, 2016; Sandhu et al., 2012), but far less work has focused on vertebrate pests (Lal, 2009). Rodents are among the world's most economically damaging agricultural pests, and the ultimate cost of rodent pest damage has been estimated to reach several billion US dollars per year worldwide (Jacob & Buckle, 2018). Rodent abundance—and subsequent crop damage—may be reduced by direct consumption by predators (Browning et al., 2016; Kross et al., 2018; Kross & Baldwin, 2016; Meyrom et al., 2009; Ojwang & Oguge, 2003), though empirical evidence is varied (Heaton et al., 2008; Heywood & Pavey, 2002; Howard, 1967; Hygnstrom et al., 1994; Martin, 2009). The effect of predators on the behavior of rodent pests in agriculture is less studied, despite the potential to reduce crop damage when rodents avoid areas of high perceived risk (Brown et al., 1999; Brown & Kotler, 2004; Suraci et al., 2016; Terborgh & Estes, 2010) or make disadvantageous decisions that limit breeding activity (Arthur et al., 2004; Bonnington et al., 2013; Dudeck et al., 2018; Duffield et al., 2017; Haapakoski et al., 2018).

Earlier work shows promise that barn owls (Tyto spp.), predators commonly attracted to farms with nest boxes, may exert top‐down effects on voles (Microtus arvalis and M. duodecimcostatus; Luna et al., 2020), gophers (Thomomys bottae; Hansen & Johnson, 2022), and rats (Rattus spp.; Mohd Noor et al., 2003, 2024). Taxonomic revisions have recently split the cosmopolitan “barn owl” into several species, including the Western Barn Owl (Tyto alba), the Eastern Barn Owl (Tyto javanica), and the American Barn Owl (Tyto furcata; Aliabadian et al., 2016). Because we draw on studies of all three species, we use the term “barn owl group” or “barn owls” when referring collectively to this clade, and American Barn Owl when referring specifically to the species in our study system (or simply “owls” for brevity, especially in the Methods and Results ). Despite increasing interest in their role as pest predators, most prior work has focused on documenting prey remains or reductions in rodent numbers, leaving unresolved the distinction between effects on prey abundance versus behavior. This distinction is important because predator‐induced changes in behavior can reduce crop damage even without strong demographic suppression, meaning that barn owls may contribute to pest control in ways that extend beyond direct consumption. In addition, it remains unknown how vegetation cover mediates these effects, limiting both ecological understanding and the practical guidance farmers can use to manage habitats in ways that enhance the ecosystem services provided by barn owls.

Here, we investigate whether American Barn Owls exert effects on the abundance and behavior of aboveground foraging rodent pests. In addition, we explore how vegetation cover and abiotic factors mediate barn owl and rodent interactions. We measure the direct effect of American Barn Owls on rodent abundance with chew blocks, and their effect on perceived predation risk and rodent activity with giving‐up density (GUD) trays and game cameras, respectively, in winegrape vineyards in California. Specifically, we test alternative models to examine whether American Barn Owls reduce rodent abundance and activity, and increase perceived risk, and whether environmental contexts (vegetation cover, wind, lunar illumination) mediate these effects.

METHODS

Study system

Study area

Napa Valley is 100 km north of San Francisco, California, between the Vaca and Mayacamas mountains. A Mediterranean climate coupled with diverse soils renders the valley ideal for winegrape production (Napa Valley Vintners, 2014). The region supports a mosaic of almost 500 wineries (about 20,000 ha) surrounded by native oak woodlands, grasslands, forests, and riparian areas (Napa County, 2010; Wendt & Johnson, 2017).

Barn owls

Barn owls provide a compelling case for rodent pest control due to several life history traits. The barn owl group is nearly cosmopolitan, meaning a variety of agricultural systems around the world can benefit (Roulin, 2020). Species in the barn owl group are also generalists that consume rodent species in correspondence to what is locally available, indicating a capacity to prey‐switch and remain in an area even when certain rodent species are at low density (Roulin, 2020; Tores et al., 2005). Furthermore, barn owls can reach high densities (Smith et al., 1974), are readily attracted by nest boxes (Labuschagne et al., 2016), and hunt predominantly near nesting locations (Castañeda et al., 2021; Roulin, 2020).

Rodents

In Napa Valley vineyards, the most common groups of aboveground foraging rodents are deer mice (Peromyscus spp.), house mice (M. musculus), and California voles (M. californicus; Hansen & Johnson, 2022). Voles and house mice are known to prefer dense vegetation (Dickman, 1992; Jensen et al., 2003; Wywialowski, 1987), whereas deer mice are generalists who tend to prefer more open environments (Garman, 1994; Kamler & Pennock, 2004; Kaufman et al., 1988; Wolf et al., 2018). For growers, vole herbivory is particularly harmful to young vines, especially during eruptive years (every 6–8 years; Ross, 2009). Most game camera detections in this study were deer mice (see Results ), consistent with live trapping conducted in 2019, when 98% of captures were Peromyscus spp. (Hansen & Johnson, 2022). Voles were detected in less than 1% of photos in this study, also consistent with 2019 surveys where zero voles were captured (Hansen & Johnson, 2022). American Barn Owl diets in the region vary considerably with local habitat composition (St. George, 2021), ranging from 7% to 53% mice (Mus musculus, Peromyscus spp.; mean = 22.0%), 5%–76% vole (Microtus californicus; mean = 49.6%), and <1%–35% gopher (Thomomys bottae; mean = 17.5%).

Rodent measurements

Index of rodent abundance (chew blocks)

To determine relative rodent abundance, we deployed grids of chew blocks, which are small (~20 g) cubes of wax and seeds that are attractive to aboveground foraging rodents such as mice and voles. Chewing indices effectively monitor population size for numerous rodent species including California voles, house mice, and deer mice (Caughley et al., 1998; Engeman & Whisson, 2006; Engeman & Witmer, 2000; Whisson et al., 2005). Unlike depletable trays of seeds with harvest rates that decline as the food diminishes (see GUD below), the visitation to and evidence of foraging on chew blocks should increase with the density of local aboveground rodents (Engeman, 2016). We deployed four chew block grids on each of six vineyards (n = 24 grids), four vineyards in the southern Carneros region of Napa Valley and two in the mid‐valley near the towns of Yountville and Oakville. On each vineyard, we chose grid locations that spanned a range of anticipated American Barn Owl hunting pressures, based on their distance from existing nest boxes and past occupancy and nest success data (Carlino, 2024). Grids were at least 200 m apart to maintain independence. All grid locations were sampled twice, once in February through March (at the onset of nest box occupancy and courtship when anticipated hunting pressure by owls was low), and again in May–June (when owl nestlings are developing, corresponding to higher hunting pressures; St. George, 2021). Each grid consisted of 36 chew blocks, arranged in a pattern of 6 × 6 blocks spaced approximately 5 m apart (chew blocks were placed in every other vine row). We used NoTox (Liphatech, Inc., Milwaukee, WI) chew blocks, as we found this brand to be more water resistant than DETEX (Bell Laboratories, Inc., Madison, WI). Since the blocks are not truly waterproof, and rain could influence our results, we placed a 16 qt plastic storage container (16.75 × 11.88 × 7 cm) with six rectangular 7 × 7 cm openings in the sides over each chew block to allow rodent access (Appendix S1: Figure S1). We recorded the presence or absence of bite marks after 4 days for each chew block. Eight of the chew blocks in each grid were paired with remote cameras (four Bushnell Core S‐4K No Glow model #119949C and four Browning Strike Force Elite HD model BTC‐5HDE) to provide imagery to confirm the species visiting and chewing on the blocks. We identified species in four groups: deer mice, house mice, voles, and unknown. Voles were distinguished from mice based on their relatively large body size, small tail, and dark coloration. Deer mice were distinguished from house mice based on a bicolored tail, large eyes, and large ears.

Perceived predation risk (GUD)

To determine rodents’ perceived predation risk, we deployed GUD patches in the same locations as chew block grids and again operated the patches twice corresponding to low and high hunting pressure, March and June, respectively. There were at least 2 days between running the chew block and GUD grids at a given location to avoid one method affecting the other. Briefly, GUD patches were trays of sand mixed with a fixed amount of seeds each night, and the weight of seeds left behind the following morning provided an index of the foraging rodents’ perceived predation risk. This follows the marginal value theorem, predicting that foragers will give up sooner, and thus leave behind more seeds, in risky compared to safe environments (Stephens & Krebs, 1986, Stephens et al., 2007; see Appendix S1 for a more thorough conceptual explanation and methodological details). Following the sampling design of Jacob and Brown (2000), each grid had two rows of six trays, for a total of 12 GUD trays per grid, and each tray was run for six nights. We reused the plastic bins from the chew block grids for the GUD trays, flipped right side up (Appendix S1: Figure S1). We put 0.5 × 0.5 m 60 gauge vinyl covers on top of the trays to keep them dry from rain and irrigation. We used a clear cover to reduce the confounding effect of an opaque lid on perceived predation risk. Numerous studies used similar lids in their GUD design and found a significant effect of predation risk on GUD (Denny et al., 2021; Jacob & Brown, 2000; Steinbeiser et al., 2017). To collect nocturnal GUD data, a mixture of 10 g of sunflower seeds and 1 L of sand was deployed in each tray at dusk, and the seeds remaining the following morning were sifted through wire mesh and collected (Johnson & De Leon, 2015). In the laboratory, collected seeds were separated from other particles, dried in an oven (75°C for 2 h), and then weighed (Johnson & De Leon, 2015). Six trays per grid were paired with the same cameras used for the chew blocks set to video to confirm the species visiting and foraging in the trays.

Activity (cameras)

To determine rodent activity, we deployed 12 cameras on each of the six vineyards (n = 72) from February to July 2023. Cameras were spaced at least 75 m apart to ensure independence between sites, and none were within 75 m of the chew block or GUD grids. The cameras were mounted 1 m high in the vine row, pointed straight down at the ground, and baited following Eriksson et al. (2019), using a mixture of oats, peanut butter, strawberry jam, and rabbit food placed inside wire mesh bags fixed to the ground with garden staples. We used the Bushnell Core S‐4K No Glow cameras because they had settings ideal for rodent detection: 0.2 s trigger speed and high sensitivity. Cameras were set to take bursts of three photos, with a 1 s delay on low illumination and high sensitivity. Because vegetation can obscure the camera's ability to detect rodents, we cleared the vegetation immediately surrounding the camera's field of view. Despite this, initial deployment resulted in some SD cards filled with blank diurnal vegetation photos and all rodent detections occurred at night, so we changed the camera settings in mid‐April to only take photos at night. We processed photos in Wildlife Insights (Wildlife Insights 2024) and distinguished mice and voles following the features specified above. As the photos were from a top‐down view, we could not distinguish between house and deer mice. No gophers were detected on the cameras.

Owl presence

To obtain spatially and temporally explicit estimates of American Barn Owl hunting pressure, we used a model of American Barn Owl hunting habitat selection in Napa Valley from Castañeda et al. (2021) and updated by Chavez (2023). The model was a resource selection function (RSF, Northrup et al., 2022) built from 260 used hunting locations obtained from 11 female owls in our study region and briefly summarized here (details in Appendix S1). The RSF had strong negative effects of distance to nest box and distance to oak savanna; owls showed a preference for hunting in riparian and savanna habitats, and they used urban, vineyard, and water land cover classes proportionally less than their availability (Table 1). There was also a strong preference for hunting in areas with less aggregated land cover classes, and weak but positive relationships with forest edge density and landscape edge density (Table 1).

TABLE 1.

The model‐averaged coefficients (± SE) for the top hunting habitat selection models from Chavez (2023), with upper and lower 95% confidence limits (CL).

Coefficient	Estimate ± SE	Lower CL	Upper CL
Intercept	3.69 ± 0.67	2.37	5.01
Grassland	−0.07 ± 0.19	−0.45	0.31
Riparian	0.48 ± 0.22	0.04	0.91
Oak savanna	0.35 ± 0.19	−0.03	0.73
Urban	−1.30 ± 0.27	−1.84	−0.77
Vineyard	−1.12 ± 0.18	−1.48	−0.76
Water	−2.85 ± 0.33	−3.49	−2.20
Distance to nest box	−4.03 ± 0.23	−4.48	−3.58
Distance to oak savanna	−0.52 ± 0.23	−0.97	−0.08
Landscape aggregation index	−2.30 ± 0.66	−3.61	−1.00
Forest edge density	0.49 ± 0.43	−0.35	1.33
Landscape edge density	0.15 ± 0.24	−0.32	0.62

Note: This model was a resource selection function built from 11 GPS‐tagged American Barn Owls tracked in Napa Valley, 2018. All variables were scaled and centered for analyses; the landcover types were binary (yes/no) with forest landcover as the reference category. See text and Appendix S1 for details.

To reflect owl hunting pressure specifically in 2023, we combined the RSF with the observed use of next boxes by breeding owls in 2023 and an index of rodent consumption per box based on the number and age of nestlings in each used box (see Appendix S1 for details). Specifically, we multiplied the RSF output by an index of how many rodents were removed from the vineyard for each occupied box. This index was the number of adults present in the box times adult rodent consumption (10.5 rodents/week, St. George, 2021) plus the number of fledglings multiplied by weekly rodent prey delivery rate (from St. George, 2021, which documented rodent delivery rates for nestlings based on age in three periods [weeks 1–4 = 18.75 rodent/nestling/week, weeks 5–8 = 21.47, weeks 9–10 = 13.15]). The result was a raster layer of weekly predicted owl hunting pressure that recognized spatial effects of the owls' tendencies to favor some habitat over others (e.g., grassland and riparian edges) and to hunt near their nest boxes, while also incorporating the spatial and temporal effects of increased hunting near used boxes with varying rates of food provisioning to nestlings as they age (St. George, 2021; see Appendix S1 for example maps). The values of these rasters then were spatially and temporally aligned with rodent sampling events and used as a predictor in models as explained more fully below.

Vegetation

Vegetation cover

Vegetation cover creates a heterogeneous landscape that can influence the effect of American Barn Owls on the abundance or behavior of rodents. In a vineyard habitat, vegetation cover is provided by the canopy of leaves growing on the winegrape vines, the debris of vine twigs and branches left on the ground after periodic pruning, the grass and forbs grown between the vine rows as a cover crop, and for some vineyards, the grass and weeds growing in line with the vine rows themselves. We measured each of these cover types on a subset of six chew block and two GUD sampling sites within each grid, because vegetation characteristics were very similar within a grid. We also measured the same variables every 2 weeks at each camera deployed to monitor rodent activity. We measured vine canopy cover with a spherical densiometer held in line with the vine row at a height of 18 cm. We measured the maximum grass and forb height with a 1‐m stick at each sampling site directly in line with the vine rows, and in the middle of each adjacent cover crop row. For the cover of vine debris, we visually estimated the percentage of ground covered with dry vine debris within a 1‐m radius hoop centered on each sampling site. After preliminary analysis indicated most vegetation variables showed consistent effects with response variables, and to simplify candidate models and limit the number of parameters, we created a composite score of vegetation cover. We first calculated the median of each of the four variables in each grid (vine canopy cover, cover crop height, grass height in vine row, and percent cover of vine debris), then scaled them between zero and one and summed them to yield a total score for each grid (or camera) at each sampling time that ranged continuously from zero to four. We standardized and scaled the final score for analysis.

Mowing (thatch thickness)

Rodent habitat is abruptly modified when vineyards mow the cover crop growing between vine rows, likely affecting rodent behavior and how rodents and owls interact. Mowing creates an additional form of vegetative cover by producing a layer of grass clippings (hereafter “thatch”). We measured the thickness of the thatch (in centimeters), and we denoted a mowing event by an abrupt increase in thatch between sequential measurements. Since thatch produced from mowing occurs in the cover crop row, and the cameras were mounted in the vine row, thatch likely did not affect the cameras’ ability to detect rodents.

Abiotic variables

Lunar illumination and wind can also influence rodent behavior (Gutman et al., 2011; Orrock & Fletcher, 2014; Zimecki, 2006) and barn owls’ ability to hunt rodents (Longland & Price, 1991). We obtained nightly data on the fraction of the moon illuminated from the US Naval Observatory's sun and moon data portal (US Naval Observatory), and the maximum nightly wind speeds (in meters per second) from the Napa County Airport NOAA weather station. The extent of artificial light was relatively small in our study; the vineyards were all at least 4 km from any urban center, and none of the nest boxes or rodent sampling grids were located near a barn or other farm structure with permanent artificial light. Occasional nocturnal farming operations occurred (e.g., tractors), but tracking this was beyond the scope of this study and likely affected all the vineyards similarly. Thermoregulatory costs may affect foraging behavior (Collier, 2014), so we obtained the maximum nightly temperature from the Napa County Airport NOAA weather station. Precipitation amount was also obtained from the Napa County Airport NOAA weather station.

Analysis

We created generalized linear mixed‐effects models and made separate candidate model sets for each of the three rodent response variables: index of abundance (chew blocks), perceived predation risk (GUD), and activity (cameras) (Appendix S1: Tables S1–S3, respectively). All the models were fit with the glmmTMB package in R (R Core Team, Brooks et al., 2017), and all independent variables were standardized and scaled. Model selection was based on an Information Theoretic approach, where the top model had the lowest AICc score and models within two AICc of the top model were considered competitive (Arnold, 2010). We compared AICc values with the model.sel function from the MuMIn package in R (R Core Team, Bartoń 2023). The weight of evidence for an individual predictor variable was assessed by summing the weights of each model that contained that predictor variable, and it was considered biologically meaningful if its coefficients had 95% CI that did not overlap zero. We assessed the goodness of fit for each model in the candidate model set and for the full models (Appendix S1: Table S4) based on R ² values and by plotting the kernel density estimate of the observed dataset with density estimates of 50 simulated datasets drawn from posterior distributions (Gelman & Hill, 2007). These goodness‐of‐fit statistics were produced with the check_predictions functions from the performance package in R (R Core Team, Lüdecke et al., 2021).

To model the index of rodent abundance, we used the proportion of blocks on a grid that showed evidence of chewing after 4 days of deployment as the response variable with a beta distribution, which outperformed models with a binomial distribution or a Gaussian distribution with a logit link (Appendix S1: Table S4). We also evaluated models with the mass of chew block removed as a continuous response variable, which Engeman (2016) used to correlate chew block indices to live trapping, though model performance was poor and several of the candidate models failed to converge (Appendix S1: Table S4). Since exploratory analysis showed that the relationship between both the binary and continuous response variables and predictor variables was very similar (Appendix S1: Figure S3), we used the proportion of chew blocks with bite marks as our index of rodent abundance. All models also had a fixed effect for precipitation amount as a methodological nuisance variable because ground moisture may affect the integrity of chew blocks. To account for the five replicate grids per vineyard, all candidate models included a random intercept for the grid with an unstructured (general symmetric positive definite) variance–covariance (see Appendix S1: Figures S4–S9 for the evaluation of model assumptions).

To model rodent perceived predation risk, we used the average proportion of seeds remaining (in grams) per GUD grid as the response variable with a beta distribution, which outperformed models with a Gaussian distribution (Appendix S1: Table S4). To account for the well‐known effect of habituation on GUD over multiple nights (reps, Doherty et al., 2015, Steinbeiser et al., 2017), all models included a linear term for each replicate night (1–6) as a fixed effect. All candidate models also had a random intercept for the grid nested in the vineyard with an unstructured (general symmetric positive definite) variance–covariance (See Appendix S1: Figures S3–S9 for evaluation of model assumptions).

To model rodent activity, we used the number of photos (hereafter “camera hits”) per night (defined from 1 h before sunset to 1 h after sunrise), given a 10‐min independence threshold as the response variable. Independence thresholds are commonly used in camera trap studies to group multiple photos generated by a single encounter (Gracanin et al., 2022; Rovero & Marshall, 2009; Villette et al., 2016, 2017). Since we were interested in modeling an index of rodent activity, rather than abundance or presence/absence, we used an independence threshold that optimized model performance. Hypothetically, a single rodent remaining at the bait for the entire night would yield an activity score of 1.0 for that night, whereas a rodent coming into and out of the camera view at 10‐min intervals for 8 h would yield an activity score of 48. Thus, our measure of activity represents an index of the frequency of rodents moving to and from each camera's detection zone, which was 76 × 53 cm. Candidate models were run with a negative binomial distribution, which outperformed models with a Poisson, zero‐inflated Poisson, zero‐altered Poisson, zero‐inflated negative binomial, and zero‐altered negative binomial distribution (Appendix S1: Table S4). The full model had no over‐dispersion (dispersion ratio = 0.68, p = 0.34), which was assessed using the check_overdispersion function from the performance package in R Studio (R Core Team, Lüdecke et al., 2021). Since <1% of camera hits were voles (and the probability of voles using a site was just 6%, Larson unpubl. multi‐season occupancy model), we grouped vole and mouse detections together. We only used camera hits taken two nights before, the night of, and two nights after we measured vegetation variables at the camera sites (every 2 weeks) to limit major changes in vegetation (like mowing and disking) left unaccounted for. We measured vegetation 13 times at each camera site (mid‐February through late July), yielding a total of 13 five‐day “series.” To focus analyses on rodent activity, rather than on a combination of behavior and occupancy, we removed series that never detected a rodent within the 5‐day interval (22% of the series were removed). Therefore, in this dataset, no sites contained zeros over all five nights in a series, but many contained zeros on a given night, reflecting nightly variation in rodent activity owing perhaps to abiotic variables such as wind or temperature. This claim is supported by the high detection probability of rodents (82% from a multi‐season occupancy model; Larson unpubl.), suggesting that absence of detections on some nights was driven mainly by variation in activity rather than site vacancy. Similarly, to model an index of rodent activity rather than abundance, all activity models included group size (the maximum number of rodents in an image during an event) as a fixed effect. Because >95% of detections were mice (primarily Peromyscus spp. based on trapping (Hansen & Johnson, 2022), though Mus musculus was also present), we included the maximum number of individuals detected in an image as a fixed effect to account for occasional simultaneous detections. All models also included a fixed effect for precipitation as a methodological nuisance variable, because precipitation caused some flooding of camera sites and affected bait persistence. To account for the five replicate nights per series and the 12 cameras per vineyard, all candidate models included a random intercept for the camera site nested in the vineyard with an unstructured (general symmetric positive definite) variance–covariance (see Appendix S1: Figures S4–S9 for evaluation of model assumptions). Lastly, bait age was also included as a nuisance variable for all activity models to account for the sampling design.

Illumination and wind provided an environmental context for our study. As there is rich literature on the effects of illumination and wind on rodent ecology (Gutman et al., 2011; Zimecki, 2006), all candidate models for each response variable included illumination and wind as additive effects. Thermoregulatory costs may also affect foraging behavior (Collier, 2014), so all models involving the behavioral responses of rodents (GUD and cameras) included a fixed effect of the maximum nightly temperature along with the previously mentioned abiotic variables.

We tested alternative models for the effect of American Barn Owl hunting pressure and vegetation cover alone, their additive effects, and an interactive effect between the two. We also tested models with lunar illumination and wind mediating the effect of owls on rodents because previous work has suggested that rodent vulnerability to owl hunting can be dampened by dark nights (Falcy & Danielson, 2013) and windy conditions (Carr et al., 1989; Volman, 1994). Because of the fine temporal resolution of the baited camera sites, we also tested models for the effect of thatch cover from mowing and its interaction with owl hunting pressure on rodent activity.

RESULTS

Index of rodent abundance (chew blocks)

The top model of rodent abundance supported the hypothesis that American Barn Owl hunting pressure had a direct negative effect on the proportion of chewed blocks (Figure 1, Table 2), with a confidence interval that did not overlap zero (Table 3). The three top chew block models included this variable (ΔAICc 0.72) and collectively contained 84% of the weight in the candidate model set (Table 2). Over the range of hunting pressures observed (from nearly no owls to about three occupied boxes with six adults and eight chicks within a 1‐km radius), owls reduced rodent abundance by 38%–52% given the upper and lower 95% confidence limits (CL), respectively, for the effect of hunting (Figure 1). The second competitive model included an interaction between hunting pressure and illumination, but the top model did not, and the weight of evidence for this mediating effect of illumination was relatively weak (0.25, Table 2, Appendix S1: Table S1). The third competitive model included an additive effect for vegetation cover; however, the weight of evidence for this effect was also weak (0.15), and the coefficient had a confidence interval that overlapped zero (Appendix S1: Figure S10). All top models also included additive effects of wind, precipitation, and lunar illumination. The proportion of chewed blocks increased with wind and decreased with temperature; the effect of lunar illumination had a confidence interval that overlapped zero (Table 3). Goodness‐of‐fit tests revealed a marginal R ² value of 0.65 for the top model (Table 2).

FIGURE 1.

The effect of American Barn Owl hunting pressure on relative rodent abundance from the top model, as indexed by the proportion of chewed blocks per grid. Line depicts relationship from the top model, shading depicts the 95% confidence limits (CL), and dots depict raw data. Owl hunting pressure was an index between 0 and 1 (see Methods and Appendix S1 for a detailed explanation of how we measured barn owl hunting pressure).

TABLE 2.

Top candidate models for indices of rodent abundance, perceived predation risk, and activity from surveys in Napa Valley, CA, 2023, showing the number of parameters (k), AICc, ΔAICc, model weight (w), and conditional and marginal R ².

Response and ecological effects	k	Covariates	AICc	ΔAICc	w	Conditional R 2	Marginal R 2
Chew block
Owl hunting plus abiotic	7	Precipitation a , Wind b , Hunting d	−1386.65	0	0.35	0.77	0.65
Owl hunting, pressure mediated by illumination, plus abiotic	8	Precipitation, Wind, Hunting × Illumination c	−1385.99	0.66	0.25	0.77	0.65
Owl hunting, cover, plus abiotic	8	Precipitation, Wind, Hunting, Cover e	−1385.94	0.72	0.24	0.77	0.65
GUD
Owl hunting, cover, plus abiotic	8	Rep f , Temperature g , Precipitation, Wind, Illumination, Hunting, Cover	−296.73	0	0.59	0.98	0.25
Owl hunting mediated by cover, plus abiotic	9	Rep, Temperature, Precipitation, Wind, Illumination, Hunting × Cover	−295.99	0.74	0.40	0.98	0.25
Camera
Owl hunting mediated by thatch and cover, plus abiotic	12	BaitAge g , GroupSize h , Temperature, Precipitation, Wind, Illumination, Hunting × Cover, Hunting × Thatch i	12398.33	0	0.55	0.66	0.54
Owl hunting mediated by thatch, plus abiotic	9	BaitAge, GroupSize, Temperature, Precipitation, Wind, Illumination, Hunting × Thatch	12398.73	0.40	0.45	0.66	0.54

^a Sum of nightly precipitation (cm).

^b Maximum nightly wind speeds (m/s).

^c Nightly data on the fraction of the moon illuminated.

^d Owl hunting pressure derived from spatiotemporal model of owl hunting (see Methods and Appendix S1).

^e Index of vegetation cover combined from vegetation height, percent of ground covered with vine debris, and canopy cover of leaves produced from vines in the vinerow and cover crop rows (see Methods ).

^f Replicate night (1–6) of GUD trials.

^g Maximum nightly temperature.

^h Maximum number of rodents in a photo for a given event.

ⁱ Thickness (cm) of recently mowed thatch in cover crop.

TABLE 3.

Coefficients, SEs, and 95% confidence limits (CL) from the top model of relative rodent abundance, as measured with chew blocks, in Napa Valley, CA, 2023.

Covariate a	Estimate ± SE	Lower CL	Upper CL
Fixed effects
Intercept	0.93 ± 0.15	0.63	1.24
Precipitation	−0.46 ± 0.03	−0.52	−0.40
Illumination	0.13 ± 0.03	0.07	0.18
Wind	0.67 ± 0.04	0.59	0.74
Hunting	−1.61 ± 0.10	−1.82	−1.41
Random effects variance
Vineyard	0.13

^a See Table 2 for brief covariate explanations.

Perceived predation risk (GUD)

The top models for perceived predation risk by rodents supported the hypothesis that American Barn Owl hunting pressure and vegetation cover had positive effects on the proportion of seeds remaining in the GUD trays (Table 2), with confidence intervals that did not overlap zero (Table 4). This result is supported by the two top GUD models (ΔAICc 1.46), which collectively contained 99% of the weight in the candidate model set (Table 2). The weight of evidence that owls and vegetation cover had positive effects on the proportion of seeds remaining were 1.00 and 0.99, respectively (Table 2). Over the range of hunting pressures observed, owls increased rodent perceived predation risk by 16%–38% given the lower and upper 95% CL, respectively, for the effect of hunting pressure (Figure 2a). An increase in vegetation cover also increased rodent perceived predation risk (Figure 2b). While the second competitive model included an interaction between hunting pressure and vegetation, the top model did not, and the weight of evidence for this mediating effect of vegetation was relatively weak (0.40; Table 2; Appendix S1: Figure S10). Abiotic variables also strongly affected rodent GUDs. Lunar illumination had a positive effect, while wind had a negative effect on the proportion of seeds remaining in the GUD trays (Figure 2c,d). The GUD also decreased over each six‐night trial as expected, though the relationship varied between GUD trays and grids. Goodness‐of‐fit tests revealed a marginal R ² value of 0.29 for the top model (Table 2).

TABLE 4.

Coefficients, SEs, and 95% confidence limits (CL) from the top model of rodent perceived predation risk, as measured with GUD trays, in Napa Valley, CA, 2023. The model has a random intercept for grid nested in vineyard.

Covariate a	Estimate ± SE	Lower CL	Upper CL
Fixed effects
Intercept	0.97 ± 0.32	0.34	1.60
Rep	−0.12 ± 0.03	−0.17	−0.07
Temperature	−0.16 ± 0.07	−0.30	−0.03
Wind	−0.19 ± 0.05	−0.28	−0.10
Illumination	0.31 ± 0.07	0.18	0.44
Hunting	1.02 ± 0.29	0.46	1.58
Cover	0.35 ± 0.06	0.23	0.47
Random effects variance
Grid: Vineyard	0.54
Vineyard	0.34

^a See Table 2 for brief covariate explanations.

FIGURE 2.

Top ecological predictors of top model for rodent perceived predation risk, as indexed by the proportion of seeds remaining in giving‐up density (GUD) feeding trays, showing the effect of (a) American Barn Owl hunting pressure, (b) vegetation cover, (c) lunar illumination (the fraction of the moon illuminated [US Naval Observatory]), and (d) wind (m/s). Lines depicts partial response curves, ribbons depict 95% confidence limits (CL), and dots depict raw data. All independent variables were standardized and scaled.

Activity (cameras)

The top models for rodent activity supported the hypothesis that the effect of American Barn Owl hunting pressure on the number of camera hits varied with thatch thickness, where barn owls had a strong negative effect on rodent activity in thick thatch, with a confidence interval that did not overlap zero (Figure 3). The weight of evidence for an interaction between thatch thickness and owl hunting pressure on rodent activity was 1.0 (Table 2). The second competitive model also included an interaction between owl hunting pressure and vegetation cover, but the top model did not, and the weight of evidence for this mediating effect of vegetation was relatively weak (0.55) and the confidence interval overlapped zero (Table 2; Appendix S1: Figure S10). Abiotic variables also affected the number of camera hits, where lunar illumination, temperature, and precipitation had negative effects, while wind had a positive effect (Table 5). As expected, activity also increased with rodent group size (Table 5), with confidence intervals that did not overlap zero. Goodness‐of‐fit tests revealed a marginal R ² value of 0.54 for the top model (Table 2).

FIGURE 3.

Marginal effect of American Barn Owl hunting pressure over the range of thatch thickness from the top rodent activity model, as indexed by the number of camera hits per night. Shading depicts the 95% confidence limits (CL). Owl hunting pressure was an index between 0 and 1 and thatch thickness was standardized and scaled; rug marks on x‐axis depict observed data (see Methods and Appendix S1 for a detailed explanation of how we measured barn owl hunting pressure and thatch thickness).

TABLE 5.

Coefficients, SEs, and 95% confidence limits (CL) from the top model of rodent activity, as measured with camera hits in Napa Valley, CA, 2023. The model has a random intercept for camera site nested in vineyard.

Covariate1	Estimate ± SE	Lower CL	Upper CL
Fixed effects
Intercept	−0.24 ± 0.11	−0.44	−0.03
Bait age	−0.02 ± 0.00	−0.03	−0.01
Group size	1.26 ± 0.04	1.18	1.34
Temperature	−0.08 ± 0.02	−0.13	−0.04
Wind	0.06 ± 0.02	0.02	0.09
Precipitation	−0.07 ± 0.02	−0.11	−0.03
Illumination	−0.08 ± 0.02	−0.12	−0.04
Hunting	0.02 ± 0.03	−0.03	0.08
Cover	−0.03 ± 0.03	−0.08	0.02
Thatch	0.09 ± 0.02	0.04	1.14
Cover × Hunting	−0.03 ± 0.02	−0.07	0.01
Thatch × Hunting	−0.13 ± 0.03	−0.18	−0.07
Random effects variance
Site: Vineyard	0.20
Vineyard	0.03

DISCUSSION

Congruent with ecological theory, American Barn Owls in this study exerted top‐down effects on the abundance and behavior of aboveground foraging rodents in winegrape vineyards. Owls reduced rodent abundance, as indicated by their negative effect on the proportion of chewed wax blocks (Figure 1), and induced fear, as indicated by their positive effect on rodent GUD (Figure 2a). This study system has attributes that should, according to theory, lead to strong top‐down effects. Top‐down effects are likely stronger in systems with high production: biomass ratios (the resource quantity and quality hypothesis; Cebrián & Duarte, 1994; Duarte & Cebrián, 1996) and with linear food chains (the community regulation hypothesis; Finke & Denno, 2005; Mohd Noor & Mohd, 2003; Polis & Strong, 1996; Strong, 1992), like those found in irrigated winegrape vineyards with sown cover crops. In farming systems with few natural cavities, nest boxes supplement an otherwise limited resource for members of the barn owl group (exogenous inputs hypothesis; Leroux & Loreau, 2008), which is predicted to also strengthen their top‐down effect. Adding nest boxes is especially likely to enhance top‐down effects for non‐territorial generalists, like American Barn Owls, who hunt available rodents near their nests (Castañeda et al., 2021).

Farmers use nest boxes to attract species in the barn owl group worldwide to manage rodent pests, and this research contributes to a growing body of evidence for their capacity to meaningfully exert top‐down effects in agriculture. In Spain, Western Barn Owls reduced vole abundance in alfalfa fields (Luna et al., 2020). Members of the barn owl group have been shown to reduce the abundance of rats in Kenyan maize fields (Mastomys spp.; Ojwang & Oguge, 2003) and in oil palm plantations in Malaysia (Rattus spp.; Lenton, 1984; Mohd Noor et al., 2003). In California, American Barn Owls reduced gopher activity on winegrape vineyards (Thomomys bottae; Hansen & Johnson, 2022). In Florida sugarcane plantations and elsewhere (Charter et al., 2010; Charter & Rozman, 2022; Martin, 2009; Meyrom et al., 2009; Peleg et al., 2018; Wendt & Johnson, 2017), installing nest boxes inflates the density of owls in the barn owl group, which should enhance their top‐down effect. While these studies all demonstrate strong top‐down effects, our research is the first to quantify both the direct (consumptive) and behavioral effects of a member of the barn owl group on rodents, and to investigate how the strength of these effects may vary with environmental context, especially vegetation.

Our results suggest that vegetation cover favors acoustically attuned American Barn Owls over visually oriented rodent pests. Owls exerted negative behavioral effects on rodent activity only in thick thatch cover after mowing (Table 4; Figure 3). In addition, rodents' perceived predation risk increased with our index of vegetation cover, as indicated by the positive effect on rodent GUD (Figure 2b), although evidence for an interaction between vegetation and owl hunting pressure on GUD was weak. Because members of the barn owl group primarily detect their prey with auditory cues (Bergan et al., 2005; Floyd, 2023; Lisney et al., 2012; Pena & DeBello, 2010; Takahashi, 2010), they may have a comparative advantage over their more visually oriented prey with more vegetative cover. Konishi (1973) found that while barn owls struggle to detect and locate pure tones (like the squeak of a mouse), the rustling of prey under vegetation contains a range of frequencies most suitable for barn owls to locate (Kotler et al., 1991). Furthermore, the barn owls’ strike force is larger than their leap (takeoff) force, a hunting strategy different from most raptors (Schalcher et al., 2024; Usherwood et al., 2014). Because this strike force is still insufficient to kill prey upon impact, it has been hypothesized that this force is an adaptation to help owls penetrate prey through thick litter, augmented by barn owl's distinctly long leg morphology (Usherwood et al., 2014). If American Barn Owls detect and hunt prey most effectively in areas with high vegetation cover, it follows that their prey would be less active and perceive more risk in such areas. Interestingly, the effect of vegetation was restricted to rodent behavior, as we found little evidence for any effect of vegetation on our measure of rodent abundance, perhaps suggesting that rodents adjust their behavior sufficiently to avoid higher predation rates.

Consistent with our results, Schulte‐Hostedde and Brooks (1997) found that, in the presence of auditory predators, deer mice and woodland jumping mice (Napaeozapus insignis) avoided habitats with excess leaf litter and preferred sites that dampened movement‐induced noises. Furthermore, rodents with an innate visual response to aerial predators (Yang et al., 2020; Yilmaz & Meister, 2013) likely perceive even greater risk where vegetation cover obscures their ability to detect aerial predators (Camp et al., 2012, 2013; Potash et al., 2019). This may especially be the case for small, agile deer mice, the primary species detected in this study, who prefer open habitats for fast travel and escape from predators (Morris et al., 2015). However, other studies have found that larger bodied rodents (such as common voles [Microtus arvalis]; Jacob & Brown, 2000) and rodents with larger auditory bullae (like gerbils Gerbillus andersoni allenbyi; Embar et al., 2011) perceive more predation risk in less vegetation cover. Another study found that gerbils (G. allenbyi and G. pyramidum) suffered higher predation rates by barn owls in open sites compared to brush piles in a laboratory experiment (Kotler et al., 1991), so additional work is needed to unravel how vegetation may favor one actor in a particular predator–prey system. Although other predators, such as snakes, have not been studied in this system, we suspect they play a relatively minor role because we rarely encountered them in the field, and they have low ingestion rates relative to owls (Diller & Johnson, 1988).

Abiotic variables further provide an environmental context for American Barn Owls and rodents in this study. Rodents were more active and perceived less risk on windy nights, as indicated by the positive effect of wind on the number of camera hits (Table 4) and by the negative effect of wind on GUD (Figure 2d). There was also evidence for an effect of wind on the proportion of chew blocks chewed (Table 2). A meta‐review on the effect of wind on predator–prey interactions (Cherry & Barton, 2017) found that wind reduces the ability of predators to detect prey, which may hold especially true for acoustic predators like American Barn Owls. Wind can also interfere with the flight control of aerial predators, particularly for predators attempting to target mobile prey (Quinn & Cresswell, 2004). For hawk‐owls (Ninox philippensis spilonota and N. connivens) and masked owls (Tyto novaehollandiae), wind has been shown to negatively affect foraging and roost site selection in the Philippines (Godfrey, 2013) and Tasmania (Bell & Mooney, 2002; Newton, 1994; Olsen, 1999; Taylor et al., 2023). In this study, windy nights likely hindered the owls' ability to hear and hunt rodents, and rodents may have consequently responded with bolder behaviors.

Conversely, rodents were less active and perceived more risk under high lunar illumination, as indicated by the negative effect of illumination on the number of photos (Table 4) and the positive effect of illumination on GUD (Figure 2c). This is consistent with a large body of previous work showing an effect of increased perceived predation risk with lunar illumination (Daly et al., 1992; Farnworth et al., 2019; Johnson & De Leon, 2015; Longland & Price, 1991; Navarro‐Castilla & Barja, 2014; Orrock et al., 2004; Soulsbury, 2021; Taylor et al., 2023; but see Upham & Hafner, 2013). At first glance, this may seem opposite to our vegetation effect: If limited visibility in dense cover increases rodent caution, one might expect brighter conditions to improve visibility and reduce risk. However, we suggest these two factors operate differently. Vegetation likely acts at a local scale, impairing rodents’ ability to detect owls at close range and increasing their acoustic detectability to owls especially, whereas moonlight alters the broader sensory landscape by increasing rodents' conspicuousness to a range of predators at a larger scale. Although brighter conditions may marginally improve rodents' ability to detect predators, the increase in their visibility to predators is much greater, tipping the balance toward heightened perceived risk. This interpretation aligns with the widespread “lunar phobia” response observed across many rodent taxa and ecosystems (Prugh & Golden, 2014), suggesting that rodents use moonlight as a generalized cue of danger.

CONCLUSIONS AND RECOMMENDATIONS

This study provides strong evidence that American Barn Owls exerted top‐down effects on rodent pests, but it is important to acknowledge the spatial context of their pest removal services. Nesting barn owls are central place foragers, concentrating their hunting near their nest site to efficiently provision their young, with up to 53% of their time spent within 500 m of their nest box (Castañeda et al., 2021). Nonetheless, the hunting areas for the barn owl group are large, with recent GPS tracking work showing they will hunt up to 2.86 km away from the nest boxes in California (Castañeda et al., 2021) and 2.76 km in Switzerland (Séchaud et al., 2022). Thus, their pest removal services are spread broadly across the landscape. This is valuable for pest management because the owls can potentially find and respond to local concentrations of rodents, especially where and when they are made more vulnerable by farming practices, such as shortly after mowing events. On the other hand, in heterogeneous landscapes, the owls' wide‐ranging behavior may draw them to hunt in uncultivated habitats and field margins rather than directly within rows of crops where pests inflict the most damage (Castañeda et al., 2021; Séchaud et al., 2022). Future work should use the latest technology to pinpoint where barn owls strike and kill prey (Schalcher et al., 2024) to clarify the spatial distribution of their pest removal services.

We have framed the removal of rodents by barn owls as beneficial. Reducing the abundance and activity of rodent pests could reduce crop damage and/or diminish the need for the application of rodenticides and potential poisoning of nontarget species (Bontzorlos et al. 2024). But there are certain contexts where barn owls could be problematic where they consume potentially agriculturally beneficial insectivores such as shrews (Soricidae; Comay & Dayan, 2018) or rare or endangered rodents living near agricultural plots (Zaitzove‐Raz et al., 2020). As noted in this study and others (Cain et al., 2023), barn owls often hunt away from their nest boxes and show some preference for natural areas, so the use of nest boxes on farms could impact small mammals in nearby conservation areas (Zaitzove‐Raz et al., 2020).

We also advise that growers manage vegetation to elevate perceived predation risk in rodent pests. Based on the results of this study, dominated by deer mice, the vine canopy of leaves, pruned twigs on the ground, and vegetation in a cover crop grown between vine rows will likely increase the top‐down effect of owls on how deer mice perceive predation risk. Evidence also suggests that thick thatch cover shortly after mowing cover crop vegetation enhances the effect of American Barn Owls on deer mouse activity. Thus, to the extent feasible, farmers could time mowing to reduce deer mouse activity when crops are particularly vulnerable. Our data did not allow us to examine the relationship between American Barn Owls, thatch, and the activity of voles specifically, though some studies have found that voles are more abundant with thick thatch (Baldwin et al., 2019; Jacob & Hempel, 2003), so we recommend additional research where and when voles are especially abundant. In particular, future work should examine owl hunting behavior before and after mowing events as a natural experiment.

Given the spatial context of American Barn Owl pest removal, effective integrated pest management should rely on multiple methods to control rodents. Previous videography work was used to estimate that a single family of American Barn Owls can remove 3000–4000 rodents per year (Johnson & St George, 2020), and modeling suggests this is enough to meaningfully reduce rodent pests at scale when rodent density is low to moderate, but that owls alone are insufficient to control pests during periodic population spikes. In addition, as noted above, owls hunt widely and farmers have little influence over where they hunt. Other, more conventional means of killing rodents, such as lethal trapping lines or ecologically safe rodenticides, may be more effective at removing rodents in specific locations and/or when rodent densities are particularly high, though modeling suggests trapping is more labor‐intensive and may be 10 times more expensive over time than deploying owl nest boxes (Larson et al., In press). Thus, we recommend that growers rely on natural predators such as owls to maintain a steady removal of rodents from agricultural ecosystems and be ready to deploy targeted conventional methods when pest populations rise, leveraging the reproductive biology and perceived predation risk of the pest species to optimize efficacy (Baldwin et al., 2019; Krijger et al., 2017; Lloyd & Baldwin, 2021; Singleton et al., 1999).

AUTHOR CONTRIBUTIONS

Katherine Larson contributed to the project idea and development and spearheaded the data analysis and writing. Dr. Matthew Johnson contributed to the project idea and development, provided support for data analysis, and contributed editorial contributions. Dr. Roger Baldwin contributed to the project idea and development, provided field sampling guidance, and contributed editorial contributions.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Appendix S1.

Larson, Katherine C. , Baldwin Roger A., and Johnson Matthew D.. 2026. “Barn Owls Exert Top‐Down Effects on the Abundance and Behavior of Rodent Pests.” Ecological Applications 36(3): e70253. 10.1002/eap.70253

DATA AVAILABILITY STATEMENT

Data (Johnson, 2026a) are available in Figshare at https://doi.org/10.6084/m9.figshare.31828669. Code (Johnson, 2026b) is available in Figshare at https://doi.org/10.6084/m9.figshare.31828681.