Variable spatial management implications based on different seabird distribution and diversity metrics

Abstract

As resource managers strive to balance marine ecosystem health and services with diverse ocean uses, the spatial and temporal dynamics of species distributions—and their overlap with both static and dynamic human activities—should be carefully considered. Anthropogenic activities and management decisions rarely affect only one species, underscoring the need for spatially and seasonally resolved, multispecies approaches to inform conservation prioritization for effective marine spatial planning. We conducted at‐sea visual surveys for seabirds using strip transect methods during spring, summer, and fall in the Northern California Current over 2 years (2022–2024). Using this dataset, we developed density surface models to describe and predict the seasonal distribution and density of the 22 most abundant seabird species. We then applied spatial clustering to identify seasonal hotspots and coldspots of three distinct metrics: (1) seabird density, (2) species diversity, and (3) relative abundance of conservation‐priority species. To demonstrate management implications associated with using each of these metrics, we evaluated how hotspots and coldspots overlapped with proposed offshore wind energy areas and seasonal commercial fishing activity. Hotspots and coldspots differed both spatially and seasonally, depending on the metric used to construct them. For example, some locations simultaneously supported a density coldspot and a diversity hotspot, and others were a coldspot for conservation‐priority species in spring but a hotspot in summer. Notably, there were no locations where hotspots defined by all three metrics overlapped. Though this study is not an impact assessment, our results have implications for marine spatial planning. In particular, multispecies metrics associated with abundance and diversity may have contrasting implications for spatial management strategies, necessitating careful a priori identification of management objectives. Our study illustrates how hotspots and coldspots can be used to evaluate potential conflicts between marine species and static (offshore wind development areas) and dynamic (seasonal fishing activity) human uses of marine regions. The choice of metrics to define hotspots and coldspots can shape conclusions about conflict or compatibility between species conservation and human use of marine resources.

INTRODUCTION

Managing multiple human uses of the marine environment while protecting species with varied ecologies and population statuses is a complex challenge. Management strategies such as marine spatial planning have emerged as a result, with the goal of informing the spatial distribution of activities in the ocean so that existing and emerging uses can be maintained, conflicts reduced, and ecosystem health and services can be sustained for future generations (Douvere, 2008; Foley et al., 2010). Marine resources are notoriously patchy and ephemeral, rendering unique challenges for spatial conservation and leading to the emergence of dynamic ocean management approaches (Hyrenbach et al., 2000; Maxwell et al., 2015; Oestreich et al., 2020). Although some human uses and protection of marine spaces may be constrained to static locations (e.g., offshore energy infrastructure, marine protected areas), many species move dynamically in space and time, resulting in seasonal differences in their vulnerability to impacts (Southall et al., 2023). Other human uses such as fisheries are likewise dynamic, creating potential space‐use conflicts as resources shift in space and time (Hazen et al., 2018; Orben et al., 2021; Santora et al., 2020). It is therefore critical to consider both the spatial and temporal (e.g., seasonal) dynamics of overlap between marine species and potentially conflicting human uses of the marine environment.

Because mobile marine predators aggregate around resource patches in the ocean (Benoit‐Bird, 2024; Santora et al., 2017), anthropogenic activities typically simultaneously impact multiple species with diverse ecologies, life histories, and population statuses. A holistic approach to marine spatial planning should consider hotspots (areas of consistently high habitat suitability) and coldspots (areas of avoidance or low use) for multiple species when determining how to minimize overlap with threats. However, how hotspots and coldspots are defined can have a dramatic impact on the resulting management recommendations, and the same location may be simultaneously a hotspot and a coldspot, depending on the metrics used (Baylis et al., 2021; Price, 2002). Areas with the greatest number of animals (highest density), greatest number of species (highest diversity), or greatest importance to species of conservation need (highest relative abundance of threatened species) are all informative to calculate (Buckland et al., 2005) and arguably critical to prioritize, but may occur in different locations. Therefore, considering seasonal hotspots and coldspots of species distribution and diversity using multiple metrics can inform management prioritization.

Seabirds are a speciose group with diverse life histories, foraging habitats (e.g., nearshore vs. offshore), foraging strategies (e.g., surface feeders vs. divers), and life stages (e.g., colony‐constrained breeders vs. wide‐ranging nonbreeders) (Gaston, 2004). They are also vulnerable to a variety of static and dynamic threats, with some seabird populations showing dramatic and concerning population declines (Croxall et al., 2012; Paleczny et al., 2015; Parsons et al., 2008; Piatt et al., 2007; Renner et al., 2024). Additionally, seabirds are visible from a distance at sea and easy to identify relative to most marine species, and are distributed worldwide, thereby making them excellent candidates for ecological monitoring programs (Ballance, 2007; Piatt et al., 2007; Silva et al., 2022). Thus, we utilize seabirds to examine the impact of selecting different summary metrics in the context of spatial management.

The Northern California Current (NCC) is a productive coastal upwelling system, with enhanced productivity that supports a high diversity of marine megafauna (Block et al., 2011) including numerous seabird species that differ in their life history, phenology, and conservation status. The NCC also supports economically and culturally important fisheries and may in the future support renewable energy generation through the construction of floating offshore wind energy infrastructure (Feist et al., 2025). Notably, the NCC is highly seasonal, with intra‐annual variability in the physical processes such as upwelling (Bograd et al., 2009; Santora et al., 2020) and biological processes such as primary production (Kahru et al., 2009). Such variability in these biophysical processes drives the seasonal availability of key forage species at the base of the food web (Shaw & Fisher, 2026; Thompson et al., 2014), which in turn impacts the distribution and habitat use of mobile predators such as seabirds and dynamic human uses of the marine environment such as fisheries.

In this study, we use data from comprehensive at‐sea surveys to describe and predict the seasonal distribution of 22 seabird species in the continental shelf‐slope region of the NCC using density surface models (DSMs), expanding our knowledge of seabird ecology and habitat use in the region. We then use spatial clustering analysis to define seasonal hotspots and coldspots for three metrics: (1) seabird density, (2) species diversity, and (3) relative abundance of species of conservation concern. To demonstrate how these approaches could influence marine spatial planning, we present an exploratory case study examining the overlap between derived hotspots and coldspots with proposed wind energy areas and seasonal commercial fishing activity. As prioritizing all aspects of species distribution and diversity simultaneously while enabling sustainable human use of marine resources is challenging, if not impossible, careful forethought is warranted to inform spatial management based on available data and clear conservation objectives. We therefore examine how using different metrics to define hotspots and coldspots can inform management approaches and outcomes.

MATERIALS AND METHODS

At‐sea surveys

Our study area included waters from the coast to the continental slope edge (~3000 m depth), from the Columbia River to Cape Mendocino in the NCC (Figure 1). Visual surveys were conducted aboard R/V Pacific Storm and NOAA Ship Bell M. Shimada between August 2022 and April 2024, totaling 10 research cruises and 71 days of survey effort. Observer height was approximately 8.2 m (Pacific Storm) and 14.5 m (Bell M. Shimada) above the water and both vessels traveled at 8–10 knots during surveys. Survey methods followed (Ballance, 2018). Briefly, on all surveys, a single trained observer used strip transect methods, counting all birds observed within a 90° quadrant on one side of the vessel's bow (with best viewing conditions), and extending out to 300 m in good visibility conditions. In addition to number of individuals and species (or lowest possible taxonomic) identification, we recorded the distance from the vessel (<100, 100–200, 200–300 m), behavior (sitting, indirect/directed flight, ship‐attracted), and flight height (<10, 10–20, or >20 m). Visibility conditions were classed relative to bird size, allowing for observation strip width to vary considering the fact that in poor conditions, larger birds could be observed farther from the vessel than smaller birds (Appendix S1: Tables S1 and S2). Seabird observations were recorded by the observer using a laptop computer running SeaLog version 2.2.8 (Swingley et al., 2023), and a GPS trackline was recorded at 30 s intervals. The vessel trackline was annotated when visibility conditions changed, observers went on and off effort, and when the vessel made a turn >10° (Derville et al., 2022). These periods of consistent observation effort were then partitioned into approximately 5 km segments (mean = 5.02 ± 0.48 km) and the observation width was used to calculate the area surveyed.

FIGURE 1.

Study area in the Northern California Current (NCC). Orange lines show survey effort across all seasons. Lines that appear darker orange have overlapping effort across multiple surveys. Continental shelf is outlined by the gray line (200 m isobath). Geographic locations discussed are labeled.

Environmental data

Static environmental variables included distance from shore, depth, seafloor slope, and distance to nearest canyon. The coastline for the distance from shore calculation was obtained from the OpenStreetMap dataset (OpenStreetMap contributors, 2023). Depth and slope were obtained from the General Bathymetric Chart of the Oceans (GEBCO; 15 arcsecond resolution) (GEBCO Compilation Group, 2022). Distance to the closest submarine canyon was calculated from a worldwide geomorphological map (Harris et al., 2014). All static environmental variables were computed at a 5 km² resolution.

Dynamic environmental variables were obtained from daily fields of the near‐real‐time configuration of the Regional Ocean Modeling System (ROMS; Neveu et al., 2016) for the California Current system, with a resolution of 0.1° (~10 km²). Metrics were derived from ROMS to describe physical forcing and vertical and horizontal structures in the physical oceanography that can influence biological productivity and ecosystem state (Becker et al., 2016; Cimino et al., 2020; Derville et al., 2022). Surface measurements included sea surface temperature (SST in degree Celsius), sea surface height (SSH in meters), and wind stress curl (CURL in Newtons per cubic meter). Vertical water column structure was described by the isothermal layer depth (ILD in meters), water column stratification calculated as the bulk buoyancy frequency (BBV averaged over the upper 200 m, also known as Brunt‐Väisälä frequency, in inverse seconds). Horizontal structure was measured via eddy kinetic energy calculated from surface current velocities (EKE, in kilogram meter squared per second squared), and the spatial variability in SST (SSTSD) and SSH (SSHSD) calculated as the standard deviation (SD) over 0.3° squares (Appendix S1: Table S3).

Fishing vessel activity was considered as an additional dynamic environmental variable to account for the ship‐attraction behavioral tendencies of some seabirds (Orben et al., 2021; Wahl & Heinemann, 1979). Global Fishing Watch uses automatic identification systems (AIS) to identify and track fishing vessels, including identifying periods of fishing activity from vessel movements (https://globalfishingwatch.org/) (Kroodsma et al., 2018). Global Fishing Watch data were obtained at a daily resolution across the study region using the “gfwr” package in R (Clavelle et al., 2025), and the total number of commercial fishing vessels (across all gear and vessel types) was summed per grid cell at the resolution of the ROMS data.

All static and dynamic environmental variables were extracted at the midpoint location of each survey segment for seabird density model construction. Additionally, daily layers were acquired for all dynamic environmental variables in spring (April–June), summer (July–September), and fall (October–December) between summer 2022 and spring 2024 for spatial predictions.

Seabird density modeling

Seventy‐six bird species were recorded during at‐sea survey effort (Appendix S1: Table S4). However, many species were observed too infrequently to warrant model construction. Therefore, criteria for model development were established based on (1) species‐level taxonomic resolution of observations, (2) ≥25 survey segments with observations, (3) model predicted abundance estimates within the range of reported population sizes, and (4) when Criteria 3 was not met and only one observation was ever made for a season, that single seasonal observation was removed and the model and predictions re‐run; then, the species was either included or excluded based on Criteria 1–3. For species for which Criteria 3 and 4 were not met, the species–environment relationships likely could not be captured through model fitting with the data at hand. This iterative selection process yielded 22 seabird species for which density models could be constructed, evaluated, and interpreted (Table 1). One exception was made to Criterion 1, which was to combine Sooty Shearwaters Ardenna grisea and Short‐tailed Shearwaters Ardenna tenuirostris into a single taxonomic category, due to the difficulty of distinguishing them at sea and ecological similarities (Gillson, 2008; Raymond et al., 2010).

TABLE 1.

Focal seabird species for which density surface models were constructed, the number of survey segments with observations for each species, and density surface model performance evaluation metrics: Percent deviance explained averaged over the training folds, cross‐validated root mean squared error (RMSE), and Spearman's rho and corresponding p‐value for correlation between observed and predicted densities from cross‐validation.

Species model	Segments w/observations	Deviance explained	RMSE	Spearman's rho	Spearman's p‐value
Sooty/Short‐tailed Shearwater Ardenna grisea/Ardenna tenuirostris	1000	58.68	116.35	0.36	0.001
Black‐footed Albatross Phoebastria nigripes	647	53.35	5.04	0.36	0.025
Common Murre Uria aalge	555	71.16	22.71	0.68	<0.001
Pink‐footed Shearwater Ardenna creatopus	508	49.66	7.9	0.35	0.057
Western Gull Larus occidentalis	499	31.55	1.49	0.43	<0.001
California Gull Larus californicus	370	30.27	1.68	0.25	0.065
Cassin's Auklet Ptychoramphus aleuticus	365	41.03	10.79	0.24	0.026
Northern Fulmar Fulmarus glacialis	333	69.38	6.65	0.24	0.095
Red‐necked Phalarope Phalaropus lobatus	253	46.64	12.52	0.08	0.281
Fork‐tailed Storm‐Petrel Hydrobates furcatus	249	83.79	30.59	0.26	0.060
Rhinoceros Auklet Cerorhinca monocerata	238	20.35	1.03	0.09	0.267
Buller's Shearwater Ardenna bulleri	201	56.85	9.02	0.25	0.071
Leach's Storm‐Petrel Hydrobates leucorhous	141	66.01	1.62	0.29	0.039
Red Phalarope Phalaropus fulicarius	134	48.71	2.89	0.13	0.197
Sabine's Gull Xema sabini	121	55.10	3.06	0.14	0.160
Arctic Tern Sterna paradisaea	79	66.91	1.63	0.27	0.012
American Herring Gull Larus smithsonianus	66	35.27	0.25	0.15	0.192
Pomarine Jaeger Stercorarius pomarinus	39	49.09	1.65	0.12	0.207
Parasitic Jaeger Stercorarius parasiticus	37	36.72	13.77	0	0.535
Long‐tailed Jaeger Stercorarius longicaudus	36	54.34	0.7	0.15	0.093
Brandt's Cormorant Urile penicillatus	32	80.89	>1000	0.17	0.091
Tufted Puffin Fratercula cirrhata	26	55.02	0.56	0.06	0.343

Density surface models (DSMs) were fit for the 22 species that met all the above criteria. DSMs use species observations aggregated by survey segment and implement a spatial model that accounts for the varying strip width along the transect, thereby obtaining species density estimates. The spatial component of the DSMs were fit using generalized additive models (GAM), which are semi‐parametric regression models capable of accounting for nonlinear relationships between predictor and response variables using smoothing functions (Wood, 2017). The strip width, as defined by the observation conditions and seabird size category (Appendix S1: Tables S1 and S2), was included as an offset to account for variable detection distance per segment.

For many seabird species, distribution at large scales is a function of phenology (i.e., breeding vs. nonbreeding), as well as habitat at smaller scales; therefore, an interaction term between distance from shore and season was included in the density models to capture these seasonal patterns. Both static (distance from shore × season, depth, slope, distance to canyon) and dynamic (SST, SSH, CURL, ILD, BBV, EKE, SSTSD, SSHSD, vessels) predictor variables were included in all models. All smoothing terms were fitted with a restricted maximum likelihood (REML) smoothing term. Smooth terms were constrained by setting the number of knots used for basis construction to k = 5, to reduce model overfitting. Variable selection was conducted with a shrinkage approach implemented, which adds an extra penalty to each smoother and penalizes nonsignificant variables to zero (Marra & Wood, 2011). All DSMs were implemented in the “dsm” package in R (Miller et al., 2022).

Models were evaluated using 10‐fold cross‐validation, grouped over survey days to account for the hierarchical structure of the data (i.e., segments from the same survey day were not split over multiple folds). The percentage of deviance explained by each of the 10‐fold runs was calculated over the training fold, and each of the 10 folds was used as a test dataset to compute external evaluation of the density models. Model accuracy and discrimination power were then assessed by calculating the root of mean square error (RMSE) and Spearman's correlation coefficient (rho) between observed and predicted densities in the test fold. Functional response curves for each model were produced for all significant predictors to evaluate their influence on species abundance.

DSMs were used to make daily predictions for all days in spring (April–June), summer (July–September), and fall (October–December) between summer 2022 and spring 2024. Extrapolation in the dynamic predictor variables was dealt with via “clamping,” whereby values outside the bounds of what was observed in the training data are replaced with the 99th (if greater than the maximum observed) or first (if below the minimum observed) percentile value for that variable in the training data. Outlier cells were identified as grid cells with predicted abundance >2SD above the mean and removed. Finally, the predicted abundance per grid cell and associated uncertainty (CV) were averaged across all daily predictions for spring, summer, and fall to obtain seasonal maps. Predicted abundance was then converted to density by dividing the predicted abundance by the area of each grid cell.

Hotspot and coldspot analysis

We computed three metrics to summarize patterns across all predicted seasonal grids: density, species diversity, and mean relative abundance. Density is the predicted number of individuals per unit area (birds per square kilometer) calculated from the predicted abundance per grid cell, summed across all species for each grid cell. Species diversity was calculated as the Shannon index (H′), which measures species diversity by the uncertainty in predicting the species of a randomly selected individual from a given location (Shannon & Weaver, 1998). In our case, it effectively measures how evenly the predicted abundance per grid cell is distributed among all species in each grid cell and is more sensitive to the inclusion of species with lower abundances. We computed Shannon index using the “vegan” package in R (Oksanen et al., 2024). Relative abundance was calculated by scaling species‐specific predicted abundance per grid cell relative to the maximum value for that species across all seasons; then, relative abundances are averaged across species to obtain mean relative abundance. In this case, the geometric mean (whereby values are logged, then averaged, then the result is exponentiated) is better suited for relative abundance data, as they represent values that are multiplicative in nature (i.e., proportional changes). The geometric mean therefore minimizes the influence of some species with very high relative abundance. This approach is particularly important when considering the relative abundance of species of conservation concern, which may be comparatively rare (Buckland et al., 2005).

Density and species diversity were computed for each season across the full study area using all 22 species. The mean relative abundance was computed for species of conservation concern, defined as species with international, federal, or state listings as endangered, threatened, vulnerable, or sensitive (Appendix S1: Table S5). These included Sooty Shearwater, Black‐footed Albatross Phoebastria nigripes, Pink‐footed Shearwater Ardenna creatopus, Cassin's Auklet Ptychoramphus aleuticus, Fork‐tailed Storm‐Petrel Hydrobates furcatus, Rhinoceros Auklet Cerorhinca monocerata, Buller's Shearwater Ardenna bulleri, Leach's Storm‐Petrel Hydrobates leucorhous, and Tufted Puffin Fratercula cirrhata.

We examined spatial clustering patterns of the three metrics for each season using the Getis–Ord (Gi*) statistic, which indicates spatial association by identifying areas in which clusters are distinct from patterns in the surrounding area (Getis & Ord, 1992). This method of examining spatial dependence by highlighting areas with unusually high (hotspots) or low (coldspots) values has been successfully used for defining important seabird habitat areas (Cleasby et al., 2020). Gi* was computed for each of our three metrics using the “spdep” package in R (Bivand & Wong, 2018). To assess statistical significance, we used a Monte Carlo randomization approach with 1000 permutations, which generates a null distribution by repeatedly shuffling observed values across the spatial grid; observed Gi* values were compared against this null distribution to obtain pseudo p‐values. We defined hotspots or coldspots using a p < 0.01 threshold and extracted the corresponding bounding polygons. Hotspot and coldspot analyses were conducted separately within each season (e.g., a hotspot in spring does not necessarily represent higher absolute values than in summer); because the method is based on relative clustering, some seasons may contain no significant hot or coldspots if no clustering is detected.

An aggregated score was calculated per grid cell by overlaying all hotspot and coldspot polygons across the study area. Each grid cell was assigned a value of +1 for each instance of a hotspot occurring for any metric and −1 for each instance of a coldspot occurring, or 0 if a grid cell did not fall within any polygon (i.e., the range of possible values for each seasonal grid is between −3–3).

Overlap with anthropogenic activities

As a case study of marine spatial planning, we examined the spatiotemporal overlap of the hotspot and coldspot polygons relative to an example static anthropogenic use (proposed wind energy development areas) and an example dynamic anthropogenic activity (areas of seasonal commercial fishing intensity). Wind energy areas included the proposed locations off Northern California (BOEM, 2024a) and Oregon (BOEM, 2024b). We note that the proposed areas in Oregon have since been retracted, but we present them here for illustrative purposes. Commercial fishing intensity during our study period was summarized by aggregating the daily number of apparent fishing hours per grid cell from Global Fishing Watch and averaging by season. Kernel density estimates were calculated for fishing activity each season using the “spatstat” R package (Baddeley et al., 2015), weighted by fishing hours, with a Gaussian kernel and fixed bandwidth of 0.2°. The resulting density surfaces were normalized by the annual maximum to preserve inter‐seasonal patterns and converted to contour polygons representing the 50% utilization areas, which correspond to the smallest area containing 50% of the total probability density and are commonly used to define the “core area” of use (Laver & Kelly, 2008). We do not intend for this to be a comprehensive analysis of fishing activity in the region; rather, it is an illustrative example of a dynamic (seasonal) human use of the study system that overlaps with the distribution of our seabird species of interest. The percentage of the example anthropogenic activity polygons that overlapped with hotspots and coldspots was calculated for each combination of metric, season, and anthropogenic activity. This aggregated grid was then overlaid with the example anthropogenic activities, and the median score was calculated within the wind energy and fishing polygons.

Finally, we used Monte Carlo resampling to propagate uncertainty from the species‐specific DSM predictions through to all subsequent analyses of hotspots and coldspots. Species‐specific predicted abundance was drawn for every grid cell from a lognormal distribution parameterized by the mean and CV for that species and cell, for 1000 iterations. For species with very high CVs (notably Brandt's Cormorant and Tufted Puffin; Table 1), CVs were capped at 2.0 prior to resampling to reduce the influence of a few species with greater uncertainty due to sparse observations and poorer model fit. Calculation of the three distribution and diversity metrics, Gi* hotspot definition, aggregated score calculation, and overlap with anthropogenic activity polygons was repeated for each iteration. The median value across iterations was reported for the percentage overlap between hot and coldspots and anthropogenic threat polygons and for the aggregate scores within the threat polygons, with 95% CIs from the percentile values across all iterations. All maps illustrating the distribution and diversity metrics and their associated hotspots and coldspots were generated using mean model predictions without resampling, representing the most likely single spatial realization. The stability of hotspots and coldspots under uncertainty was assessed by examining the proportion of iterations for which each grid cell fell within a hotspot or coldspot, for each season and metric.

RESULTS

Species‐specific DSMs

At‐sea data collection yielded 1864 segments representing 9361 km of survey effort for seabird density model construction. Model performance varied across the 22 focal species, and the evaluation metrics highlighted different strengths and weaknesses of different models (Table 1). Deviance explained ranged between 20% and 83%, with generally good cross‐validated correlation when predicting to withheld folds (Spearman's rho and associated p value) and variable RMSE.

The functional relationships between density and environmental covariates varied by species, as did the predicted seasonal distribution patterns. Individual species model performance, functional response curves, and predicted distribution maps can be found in the (Appendix S1: Figures S1–S88). Here, we highlight three species with different ecological characteristics for comparison: Black‐footed Albatross showed a dispersed, offshore distribution, particularly in spring and fall; Sooty/Short‐tailed Shearwaters were distributed on the shelf in spring and summer, with high densities over Heceta Bank and off the Columbia River; Cassin's Auklets had a nearshore distribution with high densities around Cape Blanco and Cape Mendocino in summer but a dispersed, offshore distribution at other times (Figures 2 and 3).

FIGURE 2.

Example partial dependence plots for the Sooty/Short‐tailed Shearwater (A), Black‐footed Albatross (B), and Cassin's Auklet (C) models. Partial dependence plots display the functional response curves representing the effect of the smoothed function for the significant predictor variables on the trend in bird density. Different axis ranges between panels. BBV, bulk buoyancy frequency; CURL, wind stress curl; DEPTH, bathymetric depth; DIS_CANYON, distance to nearest submarine canyon; DIS_SHORE, distance to shore; EKE, eddy kinetic energy; ILD, isothermal layer depth; SSH, sea surface height; SSHSD, standard deviation in SSH; SST, sea surface temperature; SSTSD, standard deviation in SST.

FIGURE 3.

Seasonal density maps for the Sooty/Short‐tailed Shearwater (A), Black‐footed Albatross (B), and Cassin's Auklet (C) models. The seasonal distribution maps show the seasonal mean of the daily predicted density per grid cell. Color scale ranges differ between panels.

Hotspots and coldspots

The spatial clustering analysis using the Gi* statistic identified significant spatial clusters of both high and low values for all metrics in all seasons (Appendix S1: Figures S89–S94), which captured the observed spatial and seasonal patterns (Appendix S1: Figures S4–S6). Hotspots and coldspots were generally stable across Monte Carlo iterations (Appendix S1: Figure S95). Seabird density hotspots were observed on the shelf in the spring and summer with a particularly high‐density area just south of the Columbia River in summer, and coldspots in offshore waters (Figure 4). Species diversity hotspots occurred predominantly in continental slope waters, with coldspots near shore (Figure 5). Hotspots for relative abundance for species of conservation concern spanned shelf and slope waters particularly in the northern part of the study region, with the highest relative abundance values occurring off the Columbia River in summer, and coldspots in very nearshore and offshore waters (Figure 6).

FIGURE 4.

Total density of seabirds by season across the study region, computed from the seasonal predicted densities of all 22 seabird species. Red and blue polygons represent hot and coldspots, respectively, defined by significant (p < 0.01) clusters of high and low values based on the Getis–Ord statistic (Gi*). Solid orange polygons represent the proposed wind energy areas as an example of a static anthropogenic activity, and dashed orange polygons represent the 50% kernel density areas of apparent fishing activity as an example dynamic anthropogenic activity.

FIGURE 5.

Seabird species diversity (evenness) by season across the study region, computed from the seasonal predicted densities of all 22 seabird species. Red and blue polygons represent hot and coldspots, respectively, defined by significant (p < 0.01) clusters of high and low values based on the Getis–Ord statistic (Gi*). Solid orange polygons represent the proposed wind energy areas as an example of a static anthropogenic activity, and dashed orange polygons represent the 50% kernel density areas of apparent fishing activity as an example dynamic anthropogenic activity.

FIGURE 6.

Mean relative abundance of seabird species of conservation concern by season across the study region, computed from the seasonal predicted densities of the nine species with international, federal, or state listings as endangered, threatened, vulnerable, or sensitive. Red and blue polygons represent hot and coldspots, respectively, defined by significant (p < 0.01) clusters of high and low values based on the Getis–Ord statistic (Gi*). Solid orange polygons represent the proposed wind energy areas as an example of a static anthropogenic activity, and dashed orange polygons represent the 50% kernel density areas of apparent fishing activity as an example dynamic anthropogenic activity.

The total scores summarizing the hot and coldspots measured by all three metrics ranged from −3 to 2, indicating that there were some instances of three coldspots overlapping each other and no overlapping hotspots, and there were no instances of three hotspots overlapping with no overlapping coldspots (Figure 7).

FIGURE 7.

Score representing the aggregated hot and coldspot polygons generated from the three different metrics (total density, species diversity, relative abundance of species of conservation concern), where each grid cell was assigned +1 for each overlapping hotspot and −1 for each overlapping coldspot. Overlaid purple polygons represent static (solid) and dynamic (dashed) anthropogenic activity.

The area covered by the example anthropogenic activity (wind energy and fishing) polygons that overlapped seabird hotspots or coldspots varied seasonally and by metric used to define hot and coldspots (Table 2). For example, the wind energy polygons (2299 km²) did not overlap with any seabird density hotspots and 7%–29% of their area overlapped with coldspots across different seasons. However, across seasons, 18%–49% of the wind energy polygon area overlapped with species diversity hotspots and had no overlap with species diversity coldspots. The apparent fishing activity during our study period revealed seasonally concentrated areas in fishing effort in the northern part of the study region in spring, the northern and central study region in summer, and more dispersed and less intense effort in the fall (Appendix S1: Figure S96). While we do not focus here on specific fishing fleet dynamics, we note that the majority of fishing hours (60%) in the study region across all seasons were attributable to trawl fishing (Appendix S1: Table S6). The kernel density analysis of apparent fishing activity resulted in 50% utilization polygons in spring (3311 km²) and summer (5191 km²). For the fishing density polygons, 0%–51% of the area overlapped with seabird density hotspots across seasons, 0%–29% overlapped with species diversity hotspots, and 0%–61% overlapped with relative abundance of species of conservation concern hotspots.

TABLE 2.

The percentage of the total area covered by each group of example anthropogenic activity polygons that overlaps with either coldspots or hotspots, defined by the three different metrics of seabird distribution and diversity, within each season.

			Percentage overlap
Metric	Anthropogenic activity	Classification	Spring	Summer	Fall
Seabird density	Fishing	Cold	0.00 (0.00–0.00)	0.31 (0.00–0.45)	0.00 (0.00–0.00)
		Hot	5.42 (0.04–16.7)	51.26 (44.43–57.47)	0.00 (0.00–0.00)
	Wind energy	Cold	29.25 (16.71–40.59)	7.80 (5.12–12.58)	12.26 (4.31–15.56)
		Hot	0.00 (0.00–0.00)	0.00 (0.00–0.00)	0.00 (0.00–0.00)
Species diversity	Fishing	Cold	0.00 (0.00–0.00)	36.79 (30.09–43.57)	0.00 (0.00–0.00)
		Hot	29.2 (21.04–37.03)	3.97 (1.3–5.65)	0.00 (0.00–0.00)
	Wind energy	Cold	0.00 (0.00–0.00)	0.00 (0.00–0.00)	0.00 (0.00–0.00)
		Hot	18.15 (7.41–28.34)	49.11 (36.68–55.03)	22.91 (8.28–36.40)
Species of conservation concern	Fishing	Cold	23.24 (17.45–27.92)	0.5 (0.5–0.5)	0.00 (0.00–0.00)
		Hot	26.9 (19.17–35.18)	61.78 (55.26–67.02)	0.00 (0.00–0.00)
	Wind energy	Cold	1.23 (0.06–2.47)	2.21 (0.00–3.22)	3.22 (3.22–3.22)
		Hot	6.61 (6.61–8.56)	1.96 (0.00–12.67)	13.68 (5.10–21.15)

Note: Percentages represent the median value across 1000 Monte Carlo iterations propagating uncertainty in the species‐specific density surface model predictions; values in the parentheses indicate the 95% CI. For combinations with 0% overlap across all classifications, 95% CIs were (0.00–0.00), indicating this lack of overlap was consistent across all iterations; note that no fishing activity polygon was delineated for fall, so overlap is therefore necessarily 0% across all iterations.

Median aggregated score within the wind energy polygons ranged between −0.07 and 0.64, and median aggregated score within the fishing polygons ranged between 0.44 and 0.81 across seasons (noting that there is no fall score due to no defined fall fishing polygons). For comparison, the median score for the entire study area ranged between −0.27 and −0.17 across seasons (Table 3).

TABLE 3.

Median score within all wind energy polygons and fishing activity polygons, and median score across the full study area, by season.

Polygon	Median score
	Spring	Summer	Fall
Fishing	0.44 (0.31–0.62)	0.81 (0.71–0.92)	Not applicable
Wind energy	−0.07 (−0.29 to 0.21)	0.64 (0.43–0.86)	0.36 (0.07–0.64)
Total study area	−0.17 (−0.26 to −0.13)	−0.27 (−0.31 to −0.23)	−0.25 (−0.3 to −0.21)

Note: Score per grid cell is calculated by summing the number of hotspot polygons (+1 per metric) and coldspot polygons (−1 per metric) that overlap each grid cell. Scores represent the median value across 1000 Monte Carlo iterations propagating uncertainty in the species‐specific density surface model predictions; values in the parentheses indicate the 95% CI. This represents our case study of example usage of the hotspot and coldspot calculations and should not necessarily be interpreted as a threat assessment for the region.

DISCUSSION

Spatial management implications differ between metrics

We produced DSMs for 22 seabird species, revealing seasonal distribution patterns and providing new insights into seabird habitat use across a range of species with diverse life histories, foraging strategies, and conservation statuses (Figures 2 and 3; Appendix S1: Figures S1–S88). In turn, this species‐specific distribution information enabled us to focus further on the overall ecological patterns that arise across all 22 species (Figures 4, 5, 6, 7). Our spatial assessment and comparison of three seabird metrics in the NCC identified different areas as hotspots and coldspots by season. Each metric evaluated—seabird density, species diversity, and relative abundance of species of conservation concern—highlighted different areas that, if applied in a management context aimed at seabird conservation, could result in distinct regulatory strategies. For example, the divergent spatial and seasonal patterns across these metrics resulted in variable overlap with the proposed wind energy areas and seasonal fishing effort polygons examined in our case study (Tables 1 and 2). Our findings therefore highlight the need for forethought in marine spatial planning, as the choice to prioritize conservation of the highest densities, the greatest species diversity, or areas of importance for species of particular concern would support different spatial management strategies. Importantly, while the trade‐offs are between different conservation objectives, the underlying ecological patterns are themselves not necessarily contradictory. Rather, they reflect the complex reality of a dynamic ecosystem that hosts a multitude of species, and the numerous ways to measure spatiotemporal patterns of species occurrence.

We present an approach to spatially summarize how multiple hotspots and coldspots overlap using the aggregated scores (Figure 7) as a way to synthesize multiple ways of measuring the distribution and diversity of species. While not comprehensive of all interacting social, economic, and ecological components of the ecosystem, a strength of this method is its simplicity. Alternatively, spatially explicit trade‐off analysis is a more complex but promising approach within marine spatial planning to optimally minimize losses and maximize benefits across multiple sectors, including renewable energy development, fisheries, wildlife conservation, and tourism (Feist et al., 2025; Lester et al., 2013; White et al., 2012, 2024). Perhaps the trade‐off analysis framework can be extended to support consideration of multiple metrics of species distribution and diversity or even incorporate the aggregated score approach to guide management decisions. While there is unlikely to be a single consistent pathway to attaining successful marine spatial planning, we highlight the need for stakeholder value‐driven policy guidance regarding which aspects of species distribution and diversity should be prioritized in conservation management efforts. This process moves important conversations beyond spatial ecology, as presented here, and into the realm of conservation policy.

Seabird hotspots and coldspots in the NCC

The DSMs constructed from at‐sea survey data yielded key information on the seasonal distribution of 22 seabird species, representing underlying species–environment relationships that enable prediction in space and time. The resulting seasonal distribution maps, both individually (Appendix S1: Figures S1–S88) and in combination as we have synthesized here (Figures 4, 5, 6, 7), contribute to a greater understanding of seabird ecology in the NCC region, and can inform a broad range of research and management applications. The predictions from the DSMs developed in this study of the NCC generally align with similar analyses at the larger scale of the California Current (Leirness et al., 2021; Nur et al., 2011) but provide regionally and seasonally resolved predicted densities for the NCC species.

Our analyses of the combined distribution across species revealed both stark and nuanced differences in the hotspots and coldspots identified for different metrics and seasons. The results demonstrate that multiple ecological realities can exist simultaneously, reflecting the dynamic ecosystem that supports a rich diversity of species with variable population abundances and spatiotemporal distribution patterns. For example, the southernmost part of our study area simultaneously supported a density coldspot and a diversity hotspot, with the spatial extent of each varying by season (Figures 4 and 5). In the northern part of the study area, there was a seasonal inshore‐offshore shift for mean relative abundance of species of conservation concern, whereby the same locations fell within a coldspot in spring and a hotspot in summer (Figure 6). While there was never complete concordance in hotspots across metrics and seasons anywhere in our study region, two particular areas were identified as hotspots more frequently than others: Heceta Bank and the Columbia River plume (Figure 7). Heceta Bank is a bathymetric feature that generates recirculation and retention of nutrients (Barth et al., 2005; Castelao & Barth, 2005), supports recruitment and productivity at the base of the food web, and which has previously been identified as an important area for seabirds in the NCC (Ainley et al., 2005; Nur et al., 2011). The Columbia River plume, while dynamic in space and time, also represents an area that consistently aggregates numerous seabird species (Bliss et al., 2024; Russell et al., 2023; Zamon et al., 2014). Although there was never full alignment of multiple hotspots, these two locations represent places with higher agreement among metrics, lending further support for the value of these areas to seabirds in the NCC.

Overlap between seabirds and example anthropogenic activities

We examined wind energy areas and seasonal fishing activity as case study examples of how identifying seasonal hotspots and coldspots using different metrics can facilitate informed marine spatial planning, but with variable results. However, we emphasize that this study is not a rigorous impact analysis and results are more illustrative than actionable. Notably, initial plans to develop these wind energy areas have recently been revoked, and seasonal kernel density analysis of active fishing hours is only one simplified way to quantify activities of a diverse fishing fleet. When assessing the potential impacts of offshore wind energy infrastructure on seabirds, it is important to consider how spatial overlap may render certain species more or less vulnerable to collision risk or displacement impacts based on their behavior, life history, or population status (Croll et al., 2022; Kelsey et al., 2025). Similarly, spatial overlap between seabirds and fishing vessels does not necessarily equate to interactions (Torres et al., 2013), and seabird bycatch risk is more likely driven by a combination of environmental factors, seabird behavior, and specific fishing practices and fleet dynamics (Orben et al., 2021). Furthermore, static and dynamic threats to seabird populations act in concert with the impacts of large‐scale environmental variability, and seabirds are already contending with the impacts of climate change via distributional, phenological, and demographic changes (Gasbarro et al., 2025; Sydeman et al., 2015). Nevertheless, our results demonstrate how summarizing the spatial distributions of multiple species to define hotspots and coldspots can facilitate overlap analyses designed to minimize the impact of anthropogenic activities, and the approach can be further tuned to address specific management objectives.

CONCLUSIONS

Our assessment of seasonal seabird hotspots and coldspots defined using three different metrics is illustrative of the complexity and contradictions in species conservation. Despite the lack of convergence among them or the emergence of a single guiding metric, we highlight the importance of evaluating the question of “what to conserve” from multiple perspectives. While we applied these methods to seabirds in the NCC, they are adaptable and transferable across species, regions, and anthropogenic activities. The consideration of different hotspot and coldspot definitions can add depth of information to marine spatial planning efforts beyond species presence.

AUTHOR CONTRIBUTIONS

Lisa T. Ballance, Leigh G. Torres, Rachael A. Orben, and Dawn R. Barlow conceived the ideas and designed methodology. Lisa T. Ballance and William L. Kennerley collected the data. Dawn R. Barlow, William L. Kennerley, and Rachael A. Orben analyzed the data with guidance from Leigh G. Torres and Lisa T. Ballance. Dawn R. Barlow led the manuscript writing with substantial input from all coauthors. All authors contributed critically to the drafts and gave final approval for publication.

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.

Supporting information

Appendix S1.

Barlow, Dawn R. , Ballance Lisa T., Kennerley William L., Orben Rachael A., and Torres Leigh G.. 2026. “Variable Spatial Management Implications Based on Different Seabird Distribution and Diversity Metrics.” Ecological Applications 36(7): e70327. 10.1002/eap.70327

DATA AVAILABILITY STATEMENT

Data and code (Barlow, 2026) are available in Figshare at https://doi.org/10.6084/m9.figshare.31083766.v2. External datasets used in this study are referenced in the Materials and Methods section and in Appendix S1: Table S3.