https://doi.org/10.5061/dryad.w6m905qzc

Description of the data and file structure

Study system

The species that comprise the owl assemblage in the Sierra Nevada co-occur at the landscape scales but occupy distinct ecological niches (Wood et al. 2019a). While all forest owl species rely on trees for nesting, great horned (GHOW; Bubo virginianus) and western screech (WESO; Megascops kennicottii) owls occupy a wide variety of habitats and often forage where canopies are relatively open (Davis & Weir, 2010; Johnson, 1992). Flammulated owls (FLOW; Psiloscops flammeolus) breed in mid-successional mixed conifer forests often dominated by yellow pine and Douglas fir (Linkhart et al. 1998). California spotted owls (SPOW; Strix occidentalis occidentalis) rely on closed-canopy forest for roosting and nesting, while benefiting from forest heterogeneity—specifically edges between younger and more mature forest—for access to prey (Zulla et al. 2022). Northern pygmy (NOPO; Glaucidium gnoma) and northern saw-whet (NSWO; Aegolius acadicus) owls are more general in their habitat associations but often nest in mature forests (Hayward and Garton 1988, Hinam and Clair 2008, Groce and Morrison 2010). Finally, western screech, flammulated, northern pygmy and northern saw-whet owls are secondary cavity nesters, nesting in cavities excavated by other species (Scott et al. 1977, Bull et al. 1997) that are more prevalent following fires (Tarbill et al. 2015) but are also present in unburned forest.

Acoustic monitoring in the Sierra Nevada

We conducted ecosystem-scale passive acoustic monitoring (PAM) surveys in 1648 sites across the Sierra Nevada in 2021. Our surveys spanned the western slope of the Sierra Nevada, including coverage in all seven National Forests, three of the four National Parks, and some private land (Kelly et al. 2023). We divided this area into 6236 4 km2 hexagonal grid cells, which are comparable in size to spotted owl and great horned owl territories in this region (Bennett and Bloom 2005, Kelly et al. 2023) and likely encompass smaller owl territories (Peery 2000), to obtain a total sampling area of 24,494 km2. In 2021, we surveyed 845 non-adjacent grid cells to reduce the possibility of double-counting potential spotted owl and great horned owl territories (Wood et al. 2019b). Cells were excluded if they intersected highways, were over 50% water, or lacked road/trail access.

We deployed 1-3, but generally 2, autonomous recording units (ARUs; SwiftOne recorder, K. Lisa Yang Center for Conservation Bioacoustics) in each surveyed grid cell. When possible, no ARUs in this project were closer than 500 m to one another and ARUs were placed at least 250 m from the edges of cells. ARUs had a single omni-directional microphone with -25 dB sensitivity, 62 signal to noise ratio, and recorded 20:00 – 08:00 PDT at a sample range of 32 kHz, 16-bit resolution, and gain of + 33 dB. We began deployments in early-May, and surveys lasted through mid-July. Most locations were surveyed for approximately five weeks continuously.

Forest owl detections

To identify forest owl vocalizations, we used the BirdNET algorithm, a deep convolutional neural network designed to identify 984 North American and European bird species by sound (Kahl et al. 2021). We developed a customized version of BirdNET (Kahl et al. 2021) that was overfit to the vocalizations of species of interest in this region, including the six forest owls in this study. BirdNET outputs a unitless numeric prediction score, ranging from 0-1, for each species in every 3-second interval of audio data. This prediction score indicates confidence in the identification, with larger numbers indicating greater confidence.

Acoustic validation

For all species except spotted owls, we designed species-specific probability-based thresholds in the prediction score to minimize false positives in our acoustic identifications. For each species, we manipulated thresholds for both the minimum BirdNET prediction score for an observation, as well as the minimum number of observations within an hour, such that an hour-long sample was marked as a true positive only if the number of BirdNET observations above a selected prediction score was above a selected number of calls per hour. For each of these species, we manually validated a random subset of at least 200 hour-long acoustic data files that each contained at least one BirdNET identification with a prediction score of at least 0.1. For each hour-long sample, we used RavenPro 2.0 (Cornell Lab or Ornithology, Ithaca, NY) to manually scan potential observations to either 1) confirm the presence of at least one true call or 2) identify false positives where no true calls were present. In each hour-long sample, we counted the number of BirdNET identifications over a series of prediction score thresholds (0.1, 0.2, …, 0.9, 0.91, …, 0.99). We then estimated the probability of a random hour of acoustic data representing a false positive as a function of the number of BirdNET observations over each prediction score. We fit logistic regressions in which the true positive/false positive status of an hour-long acoustic data file was the binary response and the number of BirdNET observations above a prediction threshold was the predictor (lme4; Bates et al., 2015). We did this for multiple prediction score thresholds for each species. We extrapolated false positive rates to a seven-day sampling period using the following equation: 1-(1-FP)n, where FP is the false positive rate per hour and n is the number of hours within the sampling period in which ARUs were recording (84 hours total).

For FLOW and GHOW, we identified a call rate and score threshold at which the false positive rate for the secondary sampling period was about 0.01 (Appendix S1: Figure S1). We used these thresholds to filter detections used in encounter histories. Maximizing precision for three species (NSWO, WESO, NOPO) with BirdNET thresholds produced recall too low to be usable for occupancy models, so we employed an alternative strategy for these species. First, we used a more liberal call rate and prediction score threshold that resulted in higher false positive rates (Appendix S1: Figure S1). We manually validated all remaining BirdNET observations for these species, which were then included in encounter histories. All SPOW vocalizations above a threshold of 0.989 were validated separately from the other forest owl species as part of a separate, species-specific monitoring program (Kelly et al. 2023) and were included in a final encounter history for the species.

To account for imperfect detection, we divided the continuous sampling in 2021 into 2 4-week-long secondary sampling periods starting on Julian day 130 and ending on 193. Each week of acoustic sampling was separated by one day. Specifically, the first week of sampling occurred on Julian days 130–136, the second week of sample occurred on Julian days 137–144, and so on. We determined the presence of either a manually validated or threshold validated detection in each secondary sampling period. If an ARU was not recording at any point during a particular secondary sampling period, we would consider that period null. For all smaller forest owls (FLOW, WESO, NSWO, NOPO), we evaluated detections at the scale of single ARUs. We made this decision because these species have smaller home ranges, and their calls are quieter and propagate over shorter distances than the larger species. For the larger species (GHOW and SPOW), we created encounter histories at the scale of sampling hexagons because these species have larger home ranges and there is a greater chance that multiple ARUs in a sampling hexagons are recording calls from the same individual (Reid et al. 2022).

Files and variables

Files: WESO_DH.csv, FLOW_DH.csv, NOPO_DH.csv, NSWO_DH.csv

Description: Detection histories for western screech owls, flammulated owls, northern pygmy owls, and northern saw-whet owls

Variables

• Unit: Code for each site (250m buffer around autonomous recording unit)
• V1: Visit #1 (Julian days 130–136); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• V2: Visit #2 (Julian days 138–144); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• V3: Visit #3 (Julian days 146–152); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• V4: Visit #4 (Julian days 154–160); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• V5: Visit #5 (Julian days 162–168); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• V6: Visit #6 (Julian days 170–176); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• V7: Visit #7 (Julian days 178–185); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• V8: Visit #8 (Julian days 186–192); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• UTM_N: latitude at the location of the autonomous recording unit
• CC_70: proportion of a site with canopy cover greater than 70%. Canopy cover was obtained from California Forest Observatory Database (CFO; Salo Sciences, 2020) https://forestobservatory.com/
• Elevation: elevation of the autonomous recording unit
• Rugged: standard deviation of elevation in a site
• LM_1719: proportion of low- to moderate-severity fire between 2017–2019 (2–4 years prior to the 2021 sampling). For all fire variables, NAs indicate areas that did not overlap areas with the associated burn severity; NA = no fire occurred (converted to 0 for occupancy modeling). All fire data were derived from https://mtbs.gov/
• LM_1116: proportion of low- to moderate-severity fire between 2011–2016 (5–10 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_0110: proportion of low- to moderate-severity fire between 2001–2010 (11–20 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_8600: proportion of low- to moderate-severity fire between 1986–2000 (21–35 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_20: proportion of low- to moderate-severity fire in 2020 (1 year prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_1719: proportion of high-severity fire between 2017–2019 (2–4 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_1116: proportion of high-severity fire between 2011–2016 (5–10 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_0110: proportion of high-severity fire between 2001–2010 (11–20 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_8600: proportion of high-severity fire between 1986–2000 (21–35 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_20: proportion of high-severity fire in 2020 (1 year prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_1719_pd: patch density of low- to moderate-severity fire between 2017–2019 (2–4 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_1116_pd: patch density of low- to moderate-severity fire between 2011–2016 (5–10 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_8600_pd: patch density of low- to moderate-severity fire between 1986–2000 (21–35 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_20_pd: patch density of low- to moderate-severity fire in 2020 (1 year prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_0110_pd: patch density of low- to moderate-severity fire between 2001–2010 (11–20 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_1719_pd: patch density of high-severity fire between 2017–2019 (2–4 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_1116_pd: patch density of high-severity fire between 2011–2016 (5–10 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_0110_pd: patch density of high-severity fire between 2001–2010 (11–20 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_8600_pd: patch density of high-severity fire between 1986–2000 (21–35 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_20_pd: patch density of high-severity fire in 2020 (1 year prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• Hours: number of hours ARU recorded across the primary sampling period
• p.Linear1: a continuous covariate related to the first sampling period
• p.Linear2: a continuous covariate related to the secondary second period
• p.Linear3: a continuous covariate related to the secondary third period
• p.Linear4: a continuous covariate related to the secondary fourth period
• p.Linear5: a continuous covariate related to the secondary fifth period
• p.Linear6: a continuous covariate related to the secondary sixth period
• p.Linear7: a continuous covariate related to the secondary seventh period
• p.Linear8: a continuous covariate related to the secondary eight period

####

File: GHOW_DH.csv, SPOW_DH.csv

Description: Detection histories for great horned owls California spotted owls. Note that detections for spotted owls were all manually validated for a separate project.

Variables

• Cell: Code for each site (400 ha hexagonal sampling cell with 1-3, usually 2, autonomous recording units deployed within). All data at the ARU scale (including detections) are aggregated across cells
• V1: Visit #1 (Julian days 130–136); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• V2: Visit #2 (Julian days 138–144); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• V3: Visit #3 (Julian days 146–152); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• V4: Visit #4 (Julian days 154–160); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• V5: Visit #5 (Julian days 162–168); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• V6: Visit #6 (Julian days 170–176); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• V7: Visit #7 (Julian days 178–185); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• V8: Visit #8 (Julian days 186–192); NA = survey did not occur; 0 = survey occurred with no detection; 1 = at least one hour of acoustic data in sampling period contained a species-specific number of BirdNET observations with prediction scores over species-specific thresholds; 2 = at least one hour of acoustic data in sampling period contained a manually validated observation
• Elevation: Mean elevation of ARUs in cell
• Rugged: standard deviation of elevation in a site
• UTM_N: Mean latitude of ARUs in call
• CC_70: proportion of a site with canopy cover greater than 70% obtained from California Forest Observatory Database (CFO; Salo Sciences, 2020)
• LM_1719: proportion of low- to moderate-severity fire between 2017–2019 (2–4 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_1116: proportion of low- to moderate-severity fire between 2011–2016 (5–10 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_0110: proportion of low- to moderate-severity fire between 2001–2010 (11–20 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_8600: proportion of low- to moderate-severity fire between 1986–2000 (21–35 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_20: proportion of low- to moderate-severity fire in 2020 (1 year prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_1719: proportion of high-severity fire between 2017–2019 (2–4 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_1116: proportion of high-severity fire between 2011–2016 (5–10 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_0110: proportion of high-severity fire between 2001–2010 (11–20 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_8600: proportion of high-severity fire between 1986–2000 (21–35 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_20: proportion of high-severity fire in 2020 (1 year prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_1719_pd: patch density of low- to moderate-severity fire between 2017–2019 (2–4 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_1116_pd: patch density of low- to moderate-severity fire between 2011–2016 (5–10 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_8600_pd: patch density of low- to moderate-severity fire between 1986–2000 (21–35 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_20_pd: patch density of low- to moderate-severity fire in 2020 (1 year prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• LM_0110_pd: patch density of low- to moderate-severity fire between 2001–2010 (11–20 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_1719_pd: patch density of high-severity fire between 2017–2019 (2–4 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_1116_pd: patch density of high-severity fire between 2011–2016 (5–10 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_0110_pd: patch density of high-severity fire between 2001–2010 (11–20 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_8600_pd: patch density of high-severity fire between 1986–2000 (21–35 years prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• H_20_pd: patch density of high-severity fire in 2020 (1 year prior to the 2021 sampling); NA = no fire occurred (converted to 0 for occupancy modeling)
• Hours: Average number of hours ARUs in cell recorded 
• Count: Number of ARUs deployed in cell
• p.Linear1: a continuous covariate related to the first sampling period
• p.Linear2: a continuous covariate related to the secondary second period
• p.Linear3: a continuous covariate related to the secondary third period
• p.Linear4: a continuous covariate related to the secondary fourth period
• p.Linear5: a continuous covariate related to the secondary fifth period
• p.Linear6: a continuous covariate related to the secondary sixth period
• p.Linear7: a continuous covariate related to the secondary seventh period
• p.Linear8: a continuous covariate related to the secondary eight period

Code/software

Please reach out to the corresponding author for related code for any step of organizing data.