Barred Owl Tool: Methods

Overview | To obtain the results available in this tool, we followed a process of

downloading and aligning available spatial data
defining current and future threats posed to spotted owls by barred owls
specifying and optimizing spatial prioritizations
summarizing and mapping spatial prioritization results

The sections below provide details about each of these steps.

Data sources | We compiled a suite of spatial data to capture barred and spotted owl landscape use, threatened species ranges, fire risk, fire refugia, federally reserved forest, distance-to-road, and distance-to-trail. Below we describe how these data were aligned for prioritization, but first we provide their sources:

Northern spotted owl pair occupancy: Lesmeister et al. in review (link coming soon).
Northern spotted owl nest/roost habitat: Davis et al. 2022 Northwest Forest Plan—The First 25 Years (1994–2018): Status and Trends of Northern Spotted Owl Habitats.
Barred owl landscape use: Jenkins et al. 2025 Journal of Biogeography (paper; dataset).
At-risk species distributions: We obtained distributional data for 43 at-risk terrestrial vertebrate species from the USGS Gap Analysis Project. The list of species was based on (1) Table 3-29 in the USFWS Strategy, which provides species deemed to be impacted by barred owl competition and predation, and (2) results of Arenas-Viveros et al. (preprint), who used molecular methods to enrich our understanding of barred owl diet in the western US. The full list of at-risk species whose ranges are included in the tool can be viewed here.
Fire risk: Large fire suitability layer from Davis et al. 2017 Forest Ecology and Management (paper).
Fire refugia: Contemporary topoclimatic fire refugia layer from the Eco-Vis portal; see here for details and Krawchuk et al.'s recent Conservation Science and Practice paper for an overview of fire refugia concepts.
Federal reserve: Northwest Forest Plan's Land Use Allocation map.
Distance-to-road/trail: Calculated in house by processing the National Transportation Dataset of The National Map (USGS) using the ‘Euclidean Distance’ tool within the Spatial Analyst toolbox of ArcGIS.

Data compilation | Regardless of native resolution and alignment, all spatial data were rasterized (if necessary), resampled, and aligned to a “master raster” with 30 m resolution. The master raster extent was defined as the area with complete coverage for all input spatial datasets across forested land within the range of the northern spotted owl. We then summarized all spatial data across a grid of 5 km² hexagonal cells (the same grid used for acoustically monitoring northern spotted owls). To establish inclusion of cells in the tool, we summarized coverage of the master raster in each hexagonal cell; cells with less than 25% coverage by the master raster were omitted, yielding a study area consisting of 148,192 hexagonal cells (740,960 km²). Next, within each hexagonal cell, we summarized all spatial data inputs by taking their mean value across all 30 m pixels whose centroid was inside the hexagon. Finally, to improve the selection of weight parameters for prioritization (see below), we rescaled each hex-level variable to have a minimum of 0 and a maximum of 1.

Calculating spotted owl threat | we combined spatial data on barred and spotted owl landscape use to formulate two layers: one that captured the current threat posed to spotted owl populations by barred owls, and one that captured the future threat posed to spotted owl populations by barred owls. Conceptually, the current threat posed to spotted owls by barred owls is greatest where both species co-occur at relatively high densities; in other words, where spotted owls persist behind the barred owl invasion front. On the other hand, the greatest future threat occurs near or beyond the invasion front, in areas where spotted owls have experienced less competitive exclusion, but barred owls will eventually invade (or increase in density) if left unchecked. Thus, on a per-pixel basis, we calculated current threat as the product of spotted and barred owl landscape use probabilities:
Current threat= ψ_{NSO pair} × ψ_BDO
where ψ_{NSO pair} denotes the probability of landscape use (after accounting for imperfect detection and environmental covariates) of a pair of northern spotted owls and ψ_BDO denotes landscape use probability for any barred owl. Conversely, to calculate future threat, we subtracted barred owl landscape use probability from spotted owl landscape use probability:
Future threat= ψ_{NSO pair} - ψ_BDO
These formulations resulted in high current threat values for hexes where both barred and spotted owl landscape use probabilities were high, and high future threat values in hexes where spotted owl landscape use probabilities were high but barred owl landscape use probabilities were low. Formulating future threat in this way made the reasonable assumption that in the absence of management, barred owls will disperse to virtually every portion of the northern spotted owl’s range. In California, we have observed barred owls as far southwest as Marin County and as far southeast as the central Sierra Nevada. Thus, relatively healthy spotted owl populations currently occurring in areas with few or no barred owls will presumably face invasion and increased interspecific competition in the future if no population control of barred owls is carried out.

Spatial prioritization | We used the prioritizr package in R to define prioritization problems and solve them to optimize the spatial allocation of barred owl removal efforts. The prioritizr package provides a platform for optimizing the spatial allocation of conservation actions such as protected area designation, habitat restoration, and reintroductions. Spotted owl layers (current threat, future threat, nest/roost habitat) and at-risk species ranges were included in prioritizations as conservation features, while fire risk, fire refugia, federal reserve, distance-to-road, and distance-to-trail were included as penalties. Weights control the relative importance of features and penalties to the prioritization. Users can "tune" one northern spotted owl feature or one penalty at a time (see How-to guide). When specifying prioritization problems, we used the “maximum utility” objective function, which maximizes coverage of input layers (while accounting for layer-specific weights) given some user-specified budget. We defined cost per planning unit as its land area (i.e., 5 km2) and specified a budgets to be 50% of the land area in the spatial unit being considered. Once fully specified, we used prioritizr to optimize the allocation of barred owl removal effort across hexagonal cells within the study area based on the distribution and weighting of input layers.

Following optimization, we called prioritizr’s “eval_rank_importance” function, which assigns a relative importance score (ranging from zero to one) to each unit (hexagonal cell) that is selected for management. Briefly, the function implements existing methods, in which the prioritization problem is re-solved across a range of evenly spaced, decreasing budgets (we specified 20 iterations). Only those planning units (i.e., hexagonal cells) selected for management in the first optimization are considered viable for inclusion in the second optimization, and so on, such that each subsequent solution contains a subset of previously selected units. Finally, the proportion of iterations in which a given unit is selected is taken to represent that unit’s relative importance, or, in other terms, its cost-effectiveness for meeting the selected objective.

After optimizing each prioritization scenario and calculating importance scores, we joined the optimized solutions with the input spatial data to evaluate coverage, which captures the extent of a given layer’s distribution that is covered by a given solution. To quantify coverage, we reported the amount of each species’ distribution that was captured by each solution across the gradient of relative importance scores. For all hexagonal cells assigned a given relative importance score (distributed between zero and one in 20 evenly spaced increments), we summed each feature’s coverage, then calculated the cumulative sum across relative importance values, moving from high to low (under the assumption that if incomplete solution implementation occurs, it would first incorporate cells with higher relative importance). Finally, to evaluate trends and tradeoffs in penalty layers, we took the mean value of each layer across all hexagonal cells receiving a given relative importance score for a given solution. We then used generalized additive models to fit a smoothed curve to the distribution of mean values across relative importance scores.

See Hobart et al. (under review; preprint) for further methodological details.