Invasive species are a primary threat to biodiversity and are challenging to manage once populations become established. But removing them is further complicated when invasions occur in continental, mixed-ownership systems. We demonstrate a rare conservation success: the regional-scale removal of an invasive predator – the barred owl (Strix varia) – to benefit the spotted owl (S. occidentalis) in California, USA. Barred owl site occupancy declined six-fold from 0.19 to 0.03 following one year of removals, and site extinction (0.92) far exceeded colonization (0.02). Spotted owls recolonized 56% of formerly occupied territories within one year, contrasting starkly with removals conducted after barred owls achieved high densities in the Pacific Northwest. Thus, our study averted the otherwise likely extirpation of California spotted owls by barred owl competition. Collectively, leveraging technological advances in population monitoring, early intervention, targeting defensible biogeographic areas, and fostering public-private partnerships will reduce invasive species-driven extinction of native fauna in continental systems.

Barred owls and barred x spotted owl hybrids (hereafter “hybrids”) were located for removals using a combination of (1) intensive passive acoustic surveys on national forests in the northern Sierra Nevada where densities are highest and (2) active broadcast surveys across public and private land across the entire Sierra Nevada. Removal success was monitored using passive acoustic surveys in the high-density northern Sierra Nevada. Thus, while our test of the feasibility of barred owl removals was focused on the northern Sierra Nevada, success in this area is assumed to be indicative of the success and feasibility of removals in the areas of much lower density of the Sierra Nevada.

We conducted passive acoustic surveys at 346 sites in 2018, 983 sites in 2019, and 267 sites in 2020 across 6,000 km² of national forest land in the northern Sierra Nevada (Wood et al. 2019, 2020). Each site was a 4 km² hexagonal grid cell – the approximate size of barred owl territories in the region (Wood et al. 2020; Figure 2). The 346 sites surveyed in 2018 were selected systematically with approximately one of every five of the 1500 available sites receiving a survey; these sites were considered “pre-removal” in our test of removal effectiveness and provided locations of barred owls for removals. Typically, these sites were separated by at least one non-surveyed site to reduce the probability of detecting the same individuals at multiple sites (i.e., minimize false detections). The 983 sites surveyed in 2019 included 330 of those surveyed in 2018 and an additional 653 intervening sites intended to locate as many barred owls as possible – and thus encompassed most (65%) of national forests in the northern Sierra Nevada (Figure 2). Surveys conducted in 2019 were not used to assess the effectiveness of removals (see below). The 267 sites surveyed in 2020 were a subset of those surveyed in 2018 (owing to COVID-19 restrictions) and were treated as “post-removal” sites in our test of removal effectiveness. At each site, we conducted two to five passive acoustic surveys, five to seven nights in duration. We deployed two to three autonomous recording units (ARUs; Swift Recorder, Cornell Lab of Ornithology, Ithaca, New York, USA) per site in areas of high topographic relief with ARUs operating from 2000 to 0600 hours at a sample rate of 32 kHz. We scanned audio data using Program Raven Pro 2.0 (Cornell Lab of Ornithology, Ithaca, New York, USA) and used a previously developed sliding window template detector applied to our audio data to identify barred owl eight-note territorial calls (Wood et al. 2019). This template detector yields >0.98 probability of detecting at least one barred owl call within a bout of calling (Wood et al. 2019). We manually reviewed all possible detections to confirm the identification of barred owls.

Active surveys were conducted across the Sierra Nevada in areas with known, historic barred owl detections using digitally broadcasted barred and spotted owl vocalizations. We typically surveyed multiple points within 1 km of historic detections for 10 minutes each. These detections were supplemented by soliciting information on barred owl detections from management agencies and other researchers that survey for owls in the region. In concert, we expected that our passive acoustic surveys in the northern Sierra Nevada and the extensive active surveys conducted by our and other groups throughout the Sierra Nevada would locate a high fraction of the territorial barred owl population in the region.

We lethally removed barred owls and hybrids from 2018 to 2020, following field protocols established by Diller et al. (2014). Individuals with vertical barring on breast feathers that produced distinct eight-note calls were identified as pure barred owls, while individuals with bars and spots on their breast feathers that produced territorial calls that were neither distinctly barred owl or spotted owl were identified as hybrids (see Fig. S1 in Wood et al. [2021]).

We tested for declines in barred owl occupancy following experimental removals using a multi-season occupancy model (MacKenzie et al. 2003) parameterized with detection/non-detection data from passive acoustic survey data from 2018 and 2020, when removals were conducted before or after the acoustic surveys, respectively. Our primary sampling periods were the two seasons (May to August, 2018 and 2020) and sites were restricted to those 267 sites surveyed in both the pre-removal (2018) and post-removal (2020) years. We excluded data from 2019 because removals and acoustic surveys occurred simultaneously, violating the assumption of closure. We designated two to five secondary sampling periods within each primary sampling period which reflected the five-to-seven night ARU deployments.

We estimated p, the probability of detecting a barred owl at a site given that the site was occupied, ψ, the probability of barred owl site occupancy, and ε, the probability of an occupied site going extinct from 2018 to 2020, and derived an estimate of γ, the probability of site colonization from 2018 to 2020. We used Akaike’s information criterion, corrected for small sample size (AIC_c), to compare models and considered models with ΔAIC_c < 2 to have strong support from our data. We tested for heterogeneity in p between primary sampling periods (years) and among secondary sampling periods (surveys), and for changes in barred owl occupancy from 2018 to 2020 by allowing ψ to vary by year. We considered all combinations of sub-model parameters to find the model structure with the most support (Doherty et al. 2012). We used packages RMark and xlsx in Program R 4.0.2 (R Core Development Team 2020).

Our R script will run the single-species multi-season occupancy model from the spreadsheet (bdo_occupancy.xlsx), but the "core <- read.xlsx" line will need to be changed to the correct pathway on whoever's computer is trying to use this dataset.

On the spreadsheet:

• "Cell_ID" refers to the site ID
• "ch" refers to the detection histories for both primary sampling periods combined (ch_2018 and ch_2020) with 0 = barred owl not detected, 1 = barred owl detected, and . = not surveyed during the particular secondary sampling period.
• "core" refers to whether sites were surveyed in both primary sampling periods (2018 and 2020) with 1 = yes, and 0 = no.
• Dates refer to the initial Julian Date of each secondary sampling period, which we tested for as a covariate, but did not find support for in our modeling.