Predator-prey interactions can be profoundly influenced by vegetation conditions, particularly when predator and prey prefer different habitats. Although such interactions have proven challenging to study for small and cryptic predators, recent methodological advances substantially improve opportunities for understanding how vegetation influences prey acquisition and strengthen conservation planning for this group. The California Spotted Owl (Strix occidentalis occidentalis) is well-known as an old-forest species of conservation concern, but whose primary prey in many regions – woodrats (Neotoma spp.) – occurs in a broad range of vegetation conditions. Here, we used high-resolution GPS tracking coupled with nest video monitoring to test the hypothesis that prey capture rates vary as a function of vegetation structure and heterogeneity, with emergent, reproductive consequences for Spotted Owls in Southern California. Foraging owls were more successful capturing prey, including woodrats, in taller multilayered forests, in areas with higher heterogeneity in vegetation types, and near forest-chaparral edges. Consistent with these findings, Spotted Owls delivered prey items more frequently to nests in territories with greater heterogeneity in vegetation types and delivered prey biomass at a higher rate in territories with more forest-chaparral edge. Spotted Owls had higher reproductive success in territories with higher mean canopy cover, taller trees, and more shrubby vegetation. Collectively, our results provide additional and compelling evidence that a mosaic of large tree forests with complex canopy and shrubby vegetation increases access to prey with potential reproductive benefits to Spotted Owls in landscapes where woodrats are a primary prey item. We suggest that forest management activities that enhance forest structure and vegetation heterogeneity could help curb declining Spotted Owl populations while promoting resilient ecosystems in some regions.

See README DOCUMENT

Naming conventions

*RSF or prey refers to prey capture analysis

*delivery in a file name refers to delivery rate analysis

*repro in a filename means that file is for the delivery rate analysis

 

Setup

*files with vegetation data should work with minimal alteration(will need to specify working directory) with associated R code for each analysis

*Shapefiles were made in ArcGIS pro but they can be opened with any GIS software such as QGIS.

 

_____________________

Locational data files

_____________________

NOTE LOCATIONAL DATA IS SHIFTED AND ROTATED FROM THE ORIGINAL -due to the sensitive nature of this species. The locational_data includes:

*          All_2021_owls_shifted

            *          Point file showing all GPS tag locations for prey capture analysis

            *          Attributes include:

                        *          TERRITORY ID: Numerical identifier for each bird

                        *          Year: year GPS tag was recorded

                        *          Month: month GPS tag was recorded

                        *          Day: Day GPS tag was recorded

                        *          Hour: Hour GPS tag was recorded

                        *          Minute: minute GPS tag was recorded

*          All_linked_polygons_shifted

            *          Polygon file showing capture polygons for prey capture analysis

            *          Attributes include

                        *          Territory ID: numerical identifier for each bird

                        *          Polygon id: numerical identifier for each capture polygon for each bird

                        *          Shape area: area of each polygon

*          SBNF_camera_nests_shifted

            *          Point file showing spotted owl nests for prey capture analysis

            *          Attributes include

                        *          Territory id: numerical identifier for each bird

*          C95_KDE_2021_socal_shifted

            *          Polygon file of owls 95% kernel density estimate for prey delivery rate analysis

            *          Attributes include

                        *          Id: numerical identifier for each territory(bird)

                        *          Area: area of each polygon

*          San_bernardino_territory_centers

            *          Point file showing Territory centers for historical SBNF territories – shifted for repro success analysis

            *          Attributes include

                        *          Repro Territory id: unique identifier for each territory in broader set of territories

 

Besides the sifted locational data we have included - For the Resource selection function vegetation data, for the delivery analysis we have included an overview of prey deliveries by territory and vegetation data used, and for the reproductive analysis we have again included vegetation data as well as an overview of reproductive success. these are labled as follows:

___________________________________

Files for the prey capture analysis

___________________________________

 

Socal_RSF_data.txt

------------------------------

*description: Text file with vegetation data paired with capture locations both buffered polygons used in prey capture analysis and the unbuffered ones which were not used.(Pair with Socal_rsf_code R script)

*format: .txt

*Dimensions: 2641 X 35

 

*Variables:

            *ORIG_fid: completely unique identifier for each row

            *unique_id: unique identifier for each capture polygon(shared between a buffered capture location and its unbuffered pair)

            *territory_id: unique numerical idenifier of territory

            *Polygon_id: within territory unique prey capture polygon id

            *buff: bianary buffered or unbuffered (1=buffered, 0=unbuffered)

            *used: bianary used=1 available=0

            *prey_type: prey species associated with polygon unkn:unknown, flsq:flying squirel, wora:woodrat, umou:mouse, pogo:pocketgopher, grsq: grey squirel, ubrd: unknown bird, umol:unknown mole, uvol, unknown vole.

            *area_sqm: area of polygon in square meters

            *CanCov_2020_buff: average canopy cover in polygon

            *CanHeight_2020_buff: average canopy height in polygon

            *Canlayer_2020_buff: average number of canopy layers in polygon

            *Understory_density_2020_buff: average brushy vegetation density in polygon

            *pix_COUNT: count of pixels in polygon (not needed for analysis)

            *p_chaparral: percent of polygon comprised of chaparral habitat    

            *p_conifer: percent of polygon comprised of conifer habitat

            *p_hardwood: percent of polygon comprised of hardwood habitat

            *p_other: percent of polygon comprised of other habitat types

            *Calveg_cap_CHt_gt10_CC_30to70_intersect_buff: percent of polygon comprised of trees taller than 10m with 30-70percent canopy cover (used to check data)

            *Calveg_cap_CHt_gt10_CCgt70_intersect_buff: percent of polygon comprised of trees taller than 10m with greater than 70percent canopy cover (used to check data)

            *Calveg_cap_CHt_lt10_intersect_buff:percent of polygon comprised of trees less than 10m (used to check data)     

            *p_sm_conifer: percent of polygon comprised of conifer trees less than 10m (used to calculate diversity)        

            *p_lrg_conifer_sc: percent of polygon comprised of conifer forests >10m tall with sparse canopy(used to calculate diversity)

            *p_large_conifer_dc: percent of polygon comprised of conifer forests greater than 10m tall with dense canopy (used to calculate diversity)

            *p_sm_hard: percent of polygon comprised of hardwood trees less than 10m (used to calculate diversity)

            *p_lrg_hard_sc: percent of polygon comprised of hardwood forests greater than 10m with sparse canopy(used to calculate diversity)   

            *p_lrg_hard_dc: percent of polygon comprised of hardwood forests greater than 10m dense canopy (used to calculate diversity)

            *p_forests_gt10_verysparse_CC: percent of polygon comprised of trees less than 10m with very sparse canopies (used to calculate diversity) 

            *primary_edge: total distance in meters of primary edge in a polygon        

            *normalized_by_area_primary_edge: total distance in m of primary edge in a polygon divided by the area of the polygon    

            *secondary_edge: total distance in meters of secondary edge in a polygon 

            *normalized_by_area_secondary_edge:total distance in m of secondary edge in a polygon divided by the area of the polygon

            *coarse_diversity: shannon diversity in each polygon (see methods below)

            *fine_diversity: shannon diversity in each polygon (see methods below)

            *nest_distance: distance from polygon center to nest for each polygon in meters

_________________________

For the Delivery analysis

__________________________

 

note: For information on determining average prey biomass see methods as well as zulla et al 2022 for flying squirels and woodrat masses

Zulla CJ, Jones GM, Kramer HA, Keane JJ, Roberts KN, Dotters BP, Sawyer SC, Whitmore SA, Berigan WJ, Kelly KG, Gutiérrez RJ, Peery MZ. Forest heterogeneity outweighs movement costs by enhancing hunting success and fitness in spotted owls.  doi:10.21203/rs.3.rs-1370884/v1. PPR:PPR470028.

 

prey_deliveries_byterritory.csv

*Description: overview file of prey delivered to each nest 

*format: .csv

*dimensions:332 x 8

 

*Variables:

            *SITE: Unique numerical identifier for each territory

            *DATE: date prey was delivered (in UTC)

            *CAMERA TIME: time in UTC prey was delivered

            *VIDEO TIME: time on video prey was delivered - unrelated to real time just original file          

            *PREY ITEM: prey species delivered to nest unkn:unknown, uncr: unknown if delivery(removed from eventual analysis due to small sample size), flsq:flying squirel, wora:woodrat, umou:mouse, pogo:pocketgopher, grsq: grey squirel, ubrd: unknown bird, umol:unknown mole, uvol, unknown vole.

            *DESCRIPTION: related description of delivery as needed M=male F=female if blank nothing of note           

            *DATE (FILENAME):          Original video file name (with terrtory changed to unique identifier)

            *SIZE for unknowns: relative size estimate for unknown prey (x-small, small, medium, large)     

            *Prey_mass: estimated mass of prey item in grams

 

 

territory_covariates_with_biomass-delivery.csv

*Description: file with vegetation data paired with hourly delivery rates (this is the data file to be paired with prey_biomass_analysis R script) 

*format: .csv

*dimensions: 11 x 16

 

*Variables:

            *site: Unique numerical identifier for each territory

            *y1:average delivery rate in individual prey delivered per hour for each territory with uncertain deliveries included (not used in final analysis)

            *y2:average delivery rate in grams of prey delivered per hour for each territory with uncertain deliveres included (not used in the final analysis)

            *y1a:average delivery rate in individual prey delivered per hour for each territory with uncertain deliveries removed

            *y2a:average delivery rate in grams of prey delivered per hour for each territory with uncertain deliveres removed

            *area_m: area of each territories KDE in meters squared

            *cht: average canopy height in each territory

            *clc: average number of canopy layers in each territory

            *lfd: average density of brushy vegetation in each territory

            *p_hardwood: percent composition of hardwood dominate forest in each territory

            *coarse_div: shannon diversity in each territory (see methods)

            *fine_div: shannon diversity in each territory (see methods)

            *prim_edge: primary edge measured in meters within each territory divided by the area of the territory

            *sec_edge: secondary edge in meters within each territory divided by the area of the territory

            *p_chaparral: percent composition of chaparral dominate habitat in each territory

            *p_conifer: percent composition of conifer dominated forests in each territory

 

_____________________________________

For the Reproductive success analysis

_____________________________________

* Repro_success_00to20_notcleaned (overview of reproductive success for owl territories)

*Description: uncleaned data for reproduction in paired owls used to examine reproductive success

*format: .csv

*dimensions: 631 x 8

 

*Variables:

            *CDFG #: unique identifier for each territory

            *Year: year of sample

            *Occupied by: data is all for pairs of owls (see methods)

            *Nesting?: did owls nest yes or y = yes and no or n = no

            *Fledged?: did owl pair fledge young y=yes n=no ?=unknown

            *# Fledged: count of number of young fledged 0-3 and n/a where owls were not nesting

            *Occupancy: number of adult owls occupying site (was used as data checking)

            *Protocol Surveys done?: were a protocol number of surveys done yes or Y =yes ?=no or unlikely

 

 

* joined_repro_covariate (file with vegetation data paired with reproductive data)

 

*Description: reproductive sucess (number of young fledged per year) paired with vegetation covariates (this is data file to pair with repro_success R script)

*format: .csv

*dimensions: 269 x 16

 

*Variables:

            *uid: unique identifier for each sample

            *site_id: unique identifier for each territory

            *year: year of sample in actual date  

            *year_1_15: year of sample numbered 1-15 1=2006 15 = 2020

            *fledged: number of young fledged 0-3

            *can_ht: average canopy height in a territory

            *can_cov: average canopy cover in a territory

            *can_lay: average number of canopy layers in a territory

            *lfd: average brushy vegetation density in each territory

            *p_chaparral: percent of polygon comprised of chaparral habitat    

            *p_conifer: percent of polygon comprised of conifer habitat

            *p_hardwood: percent of territory comprised of hardwood habitat

            *coarse_div: shannon diversity in each territory (see methods below)

            *fine_div: shannon diversity in each territory (see methods below)

            *prim_edge:total distance in m of primary edge in a polygon divided by the area of the territory

            *sec_edge: total distance in m of secondary edge in a polygon divided by the area of the territory

 

____________________________________________________________________

METHODS for gathering data below copied directly from the manuscript

____________________________________________________________________

GPS Tagging and Video Monitoring

We used standard call-based survey methods to locate breeding pairs of Spotted Owls in 2021 (Franklin et al. 1996), from which we selected 10 owl pairs for GPS tagging and video camera monitoring of prey deliveries to nest sites between late April and early June. Sites with nesting owls were selected to include a wide range of elevations (~1,200-2,300m), contain the three dominant vegetation cover types (conifer, hardwood, and chaparral), and not have been exposed to widespread (>10% of the territory) high severity wildfire. Additionally, while first ensuring a range of elevations and cover type abundances, we prioritized sites closer to roads and trails, with males that could be captured, and accessible nests. At each site, we captured male Spotted Owls using noose poles and “hand capture” techniques (Wood et al. 2021) and attached GPS tags (Alle-300, Ecotone, Poland) via tail mounts (Kramer et al. 2021b). Males are responsible for the vast majority of foraging by Spotted Owls during incubation and brooding, such that focusing on males should provide reasonable insights into foraging patterns during nesting for this species (Delaney et al. 1999). We programmed GPS tags to record a location every two minutes, for 10 hours each night (2000–0600 PDT) over the life of the tag (x̄ = 6 nights). Median positional error was 45 m (individual tag median range = 15–74m; S.A. Whitmore and H.A. Kramer, unpublished data) and GPS filtering to improve accuracy was not possible with the details associated with each GPS location (i.e., dilution of precision or other similar metrics were not reported by the tags).

We installed infrared cameras (AXIS Q1786 – LE, 4 megapixel) in a tree adjacent to each nest tree to monitor nests during the same 10-hour period that the GPS tags were collecting locations, plus an additional 30 minutes each morning (2000–0630 PDT). Owing to the effort required to install and remove cameras, we frequently acquired extra nights of video data (x̄ = 8.9 nights, x̄ = 93.5 hours). Cameras were placed at least 10 m from the nest tree to minimize disturbance to nesting activities. To install video cameras, we climbed trees using single rope technique, which avoids climbing spurs and thus does much less damage to trees and increases the safety and efficiency of climbers (Anderson et al. 2015).

Identifying Successful Hunting Locations

To identify successful hunting locations, we first reviewed all video data and identified the time of prey delivery and the species delivered (when possible). We then examined GPS data for clusters of GPS points recorded just before a relatively straight movement path back to the nest (Marsh et al. 2014a, Wood et al. 2021). Following Zulla (2022) we defined clusters as containing up to 10 GPS locations prior to the male’s return to the nest, where all points in the cluster occurred over a span of 20 minutes or less and with no more than 250 meters between the farthest points (Figure 2). After we linked prey deliveries observed in video recordings to a GPS point cluster, we calculated a minimum convex polygon around those points, then enlarged the perimeter of (“buffered”) that polygon by 50 m (hereafter “capture polygon”) to better account for GPS positional error, the potential effects of surrounding habitat on prey acquisition, and to maintain consistency with a similar study completed in the Sierra Nevada (Zulla et al. 2022). To compare vegetation conditions in these polygons to the overall composition of the owl’s territory (see statistical analysis for more details), we generated 10 available polygons that matched the shape and size of each capture polygon within Spotted Owl home ranges. We placed each of these available polygons at a random location within the owl’s home range and rotated each to a random orientation. We defined approximate home ranges using a 95% kernel density estimate (KDE) of each tagged owl, estimated using fixed kernel width using the adehabitat package in R (x̄ = 142.57 ha, range = 86.50–238.55 ha; Seaman and Powell 1996, Calenge 2006).

Estimating Prey Delivery Rates

To estimate prey delivery rate, we recorded the number of prey deliveries captured on video per hour. To further investigate the relationship between vegetative conditions and prey delivery rate to nest, we analyzed how prey biomass (i.e., the estimated mass in grams of prey items delivered to nests) was related to vegetative conditions in a home range. For prey identified to the species level, we assigned body mass (grams). For woodrats and flying squirrels (Glaucomys sabrinus), we estimated an average body mass of individuals captured by owls by relating measurements of bones collected from regurgitated pellets to bones of museum specimens of known body mass following Zulla (2021). Briefly, this approach involved developing allometric relationships between body mass and skeletal measurements using museum specimens and then using these relationships to predict the average body mass of woodrats and flying squirrels in owl pellets based on skeletal measurements in those pellets (Zulla 2021). We then assumed that the average body mass of these two prey species was equivalent between pellets and the prey we observed delivered to nests in this study. For other identified species, we obtained an average mass for delivered individuals of each prey species by averaging masses of museum specimens from the southern California region (latitude: 32.3 to 34.6˚N, longitude: -117.9 to -115.9˚ W) using the ARCTOS collections database (https://arctos.database.museum; specimens from collections MVZ, UWBM, MSB and DMNS). In cases that we could not identify to the species level (“unknown”), we assigned prey to one of four size classes (tiny, small, medium, and large) based on the prey’s size relative to the owl delivering it and other, known prey items delivered to the nest. To calibrate the size ranges in each category, we assigned identified prey species into size classes (e.g., deer mice [Peromyscus spp.] were small, whereas woodrats were large). We then used the average mass of all identified species within each size class to assign a mass to that size class, and thus assigned that mass to each unknown prey in that size class.

Characterizing Vegetation Conditions

We characterized vegetation conditions within both used and available polygons to test for selection of vegetation that was associated with capture sites (Table 1). We also characterized vegetation conditions within home ranges to understand how vegetation conditions are related to prey delivery rates and reproductive success (Table 1). We estimated vegetation structure using remotely sensed data from the California Forest Observatory (CFO) which uses deep learning models to link available airborne LiDAR with satellite imagery using pattern recognition to predict vegetation structure across the state of California at a spatial scale of 100 m2 on an annual basis (CFO 2020). We used CFO data from 2020 to estimate average canopy cover (%), canopy height (m), canopy layer count (0 to 3), and ladder fuel density (%). For clarity, we renamed ladder fuel density “shrubby vegetation density,” henceforth, since the term “ladder fuel” is often associated with vegetation under overstory vegetation, yet this metric is simply an estimate of the density of vegetation between 1 and 4 m, agnostic to overstory. We used vegetation type data from the Classification and Assessment with Landsat of Visible Ecological Groupings (CALVEG) vegetation classification system, specifically the Anderson land use/land cover level two classification system (Anderson et al. 1976), a mid-level vegetation mapping product that classifies existing dominant vegetation with a rough minimum polygon size of 2 hectares (Brewer et al. 2015). We used the CALVEG dataset, last updated in 2018, to estimate percentage of conifer, hardwood, and chaparral-dominated vegetation types within polygons. To minimize the temporal mismatch between the timing of vegetation metric and foraging data collection, we avoided collecting GPS and camera data in areas with large impacts from recent fires or other vegetation-altering events.

We then used a combination of the CFO and CALVEG data to estimate two measures of relative edge and two measures of vegetation heterogeneity within capture polygons, available polygons, and home ranges (Table 1, Figure S1). We defined “primary edge” as any boundary between chaparral and a forest type (conifer or hardwood). We defined “Secondary edge” as any boundary between vegetation types and between short (<10m canopy height) or tall (>10m canopy height) vegetation of the same or different type. We normalized primary and secondary edge measures in polygons and home ranges by dividing the sum of edge length by the area of each polygon so that they represented a measure of the “density” of edge. We estimated coarse heterogeneity within capture polygons using the Shannon Diversity Index of the proportions of each of the three main vegetation classes (conifer, hardwood, chaparral) and a class representing any other (“other”) vegetation. We estimated “fine heterogeneity” similarly to coarse heterogeneity, but we split conifer and hardwood forest classes into three classes each, short forests (canopy height <10m), tall forests sparse canopy (canopy height >10m, canopy cover <70%), and tall forests dense canopy (canopy height >10m, canopy cover >70%).

Assessing Reproductive Success

To examine the relationship between vegetation conditions and reproductive success, we used a broader set of reproductive outcomes for Spotted Owl territories in both the San Bernardino and San Jacinto Mountains collected between 2006 and 2020 (Tempel et al. 2022). These historical reproductive outcomes were estimated by assuming that a pair of owls was not nesting if (1) when offered mice by researchers, one member ate or cached four or more mice, or (2) a female was observed roosting for more than 45 minutes before 15 May (Tempel et al. 2022). For nesting owls, the number of young fledged was counted during follow-up visits following fledging young. Reproductive data collected prior to 2006 were not considered because southern California experienced widespread drought related tree mortality between 2002 and 2005 (Preisler et al. 2017) and the remotely sensed vegetation data we used (see characterizing vegetation conditions above) were collected after this period. Hence, the vegetation conditions during the study were different prior to 2006.

Shapefiles were made in ArcGIS pro but they can be opened with any GIS software such as QGIS.