Who prioritizes what? A cross‐jurisdictional comparative analysis of salmon fish passage strategies in Western Washington

Conservation planners often rely on heuristic indices when challenged with prioritizing potential projects under a constrained budget. This paper presents a comparative analysis of several prioritization indices (PIs) of culvert fish passage barriers, which can contribute to declines in anadromous fish populations. A federal injunction requires Washington state to restore 90% of habitat blocked by state‐owned culverts by 2030, prompting the development of numerous PIs, by various entities (i.e., counties, cities) within the injunction area. Our comparative analysis of PIs within the injunction Case Area investigates their ability to distinguish between barriers, their transferability in terms of scoring metrics, how scoring weights differ, and the preferences implied thereby. We document the use of six distinct PI methods by 10 entities and find that some PIs used many shared metrics, whereas others used a high percentage of unique metrics that would be difficult to replicate outside the entity's jurisdiction. Although habitat potential, habitat quantity, and connectivity were considered across all PIs, we found a high level of variation in terms of the metric weights. Our methods can be employed in other geographies or for other restoration PI planning efforts, and our results may facilitate the development and refinement of future PIs.


| INTRODUCTION
The conservation planning literature has long focused on developing frameworks for selecting efficient portfolios of actions.These frameworks are designed to achieve a conservation objective at minimal cost or maximize a conservation metric subject to a budget constraint.Such methods often rely on numerical optimization and have been applied in diverse settings including reserve site selection (e.g., Ando et al., 1998), the spatial configuration of conservation agriculture practices (e.g., Rabotyagov et al., 2010), invasive species eradication (e.g., Epanchin-Niell & Wilen, 2012;Lampert et al., 2014), and the restoration of fish passage barriers (e.g., King et al., 2017;Neeson et al., 2015).Despite the availability of such methods in scientific literature, uptake of these methods has been sparse in many contexts as planners seek to make timely decisions with limited information and resources.These real-world limitations often preclude the development of such frameworks in applied settings (Cook et al., 2013).Planners instead often rely on heuristic algorithms or rankings to inform prioritization.In this paper, we present a comparative analysis of several of these heuristic ranking methods developed in parallel to address a single conservation objective (migratory fish habitat connectivity) in Western Washington state, USA.
Around the world, culverts at road-stream crossings allow for safe transport across waterways and provide flood management, including millions of such structures across the United States (Martin, 2019;McKay et al., 2020).However, road culverts often create barriers for fish passage and can fragment habitat (Januchowski-Hartley et al., 2013), which is an issue of particular concern for migratory species such as salmonids (Nathan et al., 2018).For example, habitat fragmentation has been found to limit salmon access to critical spawning and rearing habitat and reduce genetic exchange over time (Sheer & Steel, 2006;Tudorache et al., 2008).
To address stream habitat fragmentation created by barrier culverts, natural resource managers often develop and apply prioritization strategies to guide barrier culvert removal and replacement (McKay et al., 2020).To develop a plan, managers must have a reliable inventory of culverts within their planning area, which requires advanced mapping capabilities and extensive ground truthing efforts (Kemp & O'Hanley, 2010;Mckay et al., 2017).Barriers in an inventory are then commonly assessed using quantitative and qualitative metrics, which typically consider habitat quantity and potential, connectivity and passability, cost, and species-specific metrics (i.e., life history metrics and relative species importance) (Martin, 2019) to identify high-priority barriers for restoration (Kemp & O'Hanley, 2010;Mckay et al., 2017).Specific metrics used to quantify these criteria vary, and entities can have distinct priorities, which are reflected in their chosen methods.
There are many different potential prioritization approaches including mathematical optimization, scoring and ranking, selecting projects based on expert knowledge, and responding to opportunities presented through either auxiliary construction or funding (McKay et al., 2020).Prioritization indices (PI), the most commonly employed method (Kemp & O'Hanley, 2010), use a weighted ranking of multiple metrics to score individual culverts which are candidates for restoration (McKay et al., 2020).The PI approach is advantageous because it is flexible to multiple objectives, does not require mathematical expertise, is not a "black box," and facilitates stakeholder buy-in (McKay et al., 2020).
Noted limitations of PIs include their inefficiency if not considering connectivity between barriers and the interdependency of barrier removal efforts (McKay et al., 2020).For example, restoring a high scoring upstream barrier will not lead to habitat gains unless all barriers downstream are also restored.In addition, metric weighting is highly subjective and requires a lengthy recursive development process (McKay et al., 2020).However, for many entities, the advantages of PIs outweigh their limitations and as such are commonplace.Thus, it is important to understand what metrics are used, how they are weighted, and the implicit tradeoffs in different PI equations, particularly in Western Washington where barrier restoration efforts are rapidly accelerating in response to the injunction.
Here, we summarize and analyze barrier culvert PIs developed within a region of Washington state, USA, subject to a 2013 Federal injunction requiring the state government to replace many state-owned culverts that block passage for migratory salmon.The injunction has inspired a flurry of activity across levels of government, and in collaboration with nongovernmental organizations, as other barrier culvert owners work to build on court-mandated state investments in fish passage.This study capitalizes on the rare opportunity to study multiple PIs created within a well-defined timeframe and geographic region and addressing a single conservation issue.Comparing PI methods across entities improves transparency and understanding of the implications of various PI methods while potentially facilitating the sharing of data and methods.This builds upon existing work to facilitate coordination and communication within barrier removal efforts (McKay et al., 2020).This research is especially timely in light of the injunction and recent federal infrastructure funding, which will accelerate ongoing barrier restoration efforts (Cordan & Bradley, 2021).
Our cross-jurisdictional comparative analysis addresses the following questions: (1) How clearly do PI scores distinguish high scoring barriers from low scoring barriers?(2) How transferable are the PI scores in terms of scoring metrics utilized?(3) How do PI metric scoring weights differ?(4) What are the tradeoffs between objectives implied by the various PI equations?We identified six entities who developed unique PIs within the Case Area and four additional entities that share a PI method developed by a state agency.
We found that PI scoring methods are publicly available, although often buried within lengthy technical reports.This article addresses this issue by synthesizing across entities to provide detailed information on PI equation metric inputs and weighting.Among the PIs studied, we found a high level of variation in terms of goals, geographic area, culvert ownership type, and metrics employed.We found that all PIs, except one, have non-normal positively skewed distributions which clearly distinguish high-priority barriers.Some PIs used more transferable methods with many shared metrics, whereas others used a high percentage of unique metrics that would be difficult to replicate outside the PI geography.One common theme is that habitat potential, habitat quantity, and connectivity are considered across all linear PIs.Among the PIs studied, we found a high level of variation in terms of the relative weight given to each metric.The methods we present are applicable to other geographies or for other restoration PI planning efforts and may facilitate the development and refinement of future PIs in broader restoration planning and management.

| Study area and systematic review
In Western Washington state, fish passage barrier restoration projects are accelerating as state and local governments respond to a unique political and policy landscape (Blumm, 2017;Roni et al., 2002).In 2007, a group of 21 Indigenous Tribes obtained a federal court ruling that fish-blocking culverts on state property infringe upon their treaty-protected rights to take fish in their usual and accustomed areas (Blumm, 2017).After that ruling, the Tribes and State attempted unsuccessfully to negotiate an appropriate remedy.After a trial on the remedy, the federal court issued a permanent injunction mandating that state agencies, including Washington Department of Fish and Wildlife (WDFW) and Washington Department of State Transportation (WSDOT), restore 90% of fish habitat blocked by outdated noncompliant culverts on their lands or under their roads by 2030 (United States v. Washington 2017).The injunction Case Area includes the lands and waters ceded by the Tribes party to the lawsuit to the state of Washington in exchange for protection of their reserved fishing rights in the Stevens Treaties (United States v. Washington 2017) (Figure 1).While the injunction only applies to state-owned culverts, 16,185 additional barrier culverts owned by a variety of different entities (i.e., cities, counties, private landowners) have been identified within the Case Area, and many of these entities have accelerated barrier restoration with support from both local and state funding.This ruling created a unique political environment that prompted the rapid development of culvert inventories and prioritization strategies within the Case Area.
We defined our study area as the Washington Culverts Injunction Case Area (Figure 1), hereafter referred to as the Case Area.Within the Case Area, we conducted a directed internet search for documentation of culvert PIs published since the 2013 injunction.Specifically, a three-phase search-and-screen procedure, adapted from Prokopy et al. (2019), was employed to identify all uses of PI methods to score-and-rank barrier culverts.
The first screen was a multilayered internet search for specific keywords conducted through a general Google search on January 10, 2022, for the keywords "Washington culvert prioritization," and selecting the first 20 search results to explore for documentation of an existing PI program.For each search result, we used the Find function to search for the inclusion of the word "prioritization."Using this method, we identified nine entities (two state agencies, two nonprofits, one basin-wide effort, three counties, and one city) that matched the search criteria.
A second search was conducted of county webpages, completed for each of the 14 counties within the Case Area on January 17, 2022.We searched for the keyword "culvert prioritization" on each county's website and selected the first 10 search results to further explore, which uncovered one additional county entity that matched our search criteria.
The second phase involved reviewing webpages of the 10 entities engaged in developing PI scores for barrier culverts to identify supplementary reports, data sources, and professional points of contact.Our team reached out via email to lead authors and referenced contact persons identified via jurisdictional reports or websites to acquire additional data, reports, and presentations for review.
The third phase involved review of materials, where two of our team members read the screened reports and presentations, followed by quality checks for the barrier data.Missing or incomplete information was noted.

| Data assembly
We synthesized publicly available information and datasets provided by personal communications to produce a dataset containing variables including PI developer (entity), ownership extent, geographic extent, whether or not the PI developer used a unique PI formula or adopted an existing formula, the number of barrier culverts, owned by the entity with PI scores, use of interactive tools, and additional variables characterizing the PI formula and its inputs.Table 2 summarizes the dataset, and the Table S1 contains a more detailed table defining each metric.

| All PIs: Distribution analysis
We began with a distribution analysis for each PI to better understand the spread of scores generated by each method and how clearly high scoring barriers are distinguished from low scoring barriers.There are multiple possible interpretations to PI distributions, which relate to the relative utility of different distribution features.A non-normal distribution with a positive skew distinguishes a clear hierarchy of high-priority barriers where restoration efforts can begin.However, if the jurisdictional entity (entity) intends to correct all barriers, a uniform distribution would reduce the likelihood of two barriers sharing the same score.We also recognize that some entities may have chosen to produce a coarse hierarchy of barrier scores that are then further screened with methods outside of the PI score, for example, through opportunistic or expert decision-making (Pressey & Bottrill, 2008).
We then calculated summary statistics on the PI scores of barrier culverts from each entity and conducted Shapiro-Wilk tests for normality.To facilitate comparison across entities, PI scores for each entity were normalized to a scale of 0 to 100, with 100 assigned to the maximum possible score, or the largest observed score in the case of WDFW method, which does not have an upper bound.

| All PIs: Metric categorization
We categorized all metrics contributing to PI scores for each entity to measure how implied priorities vary across the Case Area in terms of metrics considered.To compare metrics across entities, we categorized them according to stated objectives (Table 1).These objectives were determined through the description of the metrics in prioritization reports and manuals, where possible, and through identifying qualitative similarities where documentation was missing.

| All PIs: Shared metric analysis
To assess transferability of scoring methods outside of each entity, we used a qualitative comparison matrix to identify which metrics were shared between two or more PIs.Metrics that were labeled shared measured the same type of attribute, such as number of downstream barriers or distance of upstream habitat but may have used different methods or units for measuring that attribute.For example, one PI could have measured downstream barriers as a binary (yes/no) metric, whereas another PI measured the total quantity of downstream barriers.Metrics that were shared by at least one other PI were coded "shared," and metrics that were not shared by any PIs within the Case Area were coded as "unique" metrics.We report the percentage of shared metrics for each PI in total and individually by each metric category.

| All PIs: Formula categorization
We categorized the PI methods according to whether the scores combine metrics through linear (i.e., weighted sum) or nonlinear (i.e., the weight of one metric depends on the value of another) combination.The distinction is important because metric weighting and implied tradeoffs cannot be directly compared across formula types.Our review identified multiple linear PI formulas and one nonlinear PI formula.For the remainder of the analysis, we compare weights of the linear PIs before providing a separate analysis of the behavior of the nonlinear PI method under different representative scenarios.

| Linear PIs: Category weights
By sorting metrics for all PIs into categories, we can directly compare the weight of different categories across all methods to understand how implied priorities vary.Metric weighting was identified for each PI through personal correspondence or technical reports.Linear PI scores are of the general form: where the α i are weights on individual metrics X c i , and superscript c denotes the category of metric i (see Table 1).However, many of the linear PI formulas include multiple metrics that contribute to the same objective.Thus, to quantify the weights that entities assign to objectives, we calculate the scoring weight of each category according to the following: In other words, the category weight is defined as the maximum fraction of the highest PI score that can come from a given category.

Species metrics
Metrics that factor in relative species importance or life history characteristics Note: Specific metrics assigned to each category are described in Table S1.

| Nonlinear PIs: Weights
We identified one nonlinear PI equation in the Case Area, developed by WDFW and used by four entities: where the individual PI sums over each species j the geometric means of four individual metrics X ij .
T A B L E 2 General information extracted from search-and-screen process.Unlike linear formulas, the impact of increasing a metric by one-unit changes based on the product of values of the metric and all other metrics in the formula, which makes it challenging to assess the weight that managers place on a single objective.Thus, to understand the implicit tradeoffs of the nonlinear PI formula, we examine how the PI scores vary based on changes in two key metrics holding all else constant: (1) cost, which can take on values of High, Medium, and Low; and (2) Endangered Species Act (ESA) listing, which takes on values of Listed, Species of Concern, and Not Listed.
We analyze the impact of key metrics on PI scores for each species separately, using typical values for the X ij 's held constant, ascertained from our review of the WDFW manual (WDFW, 2020).Production potential (P) is one such metric that is a species-specific estimate of annual adult salmonid production potential for every square meter of the habitat.
Among prioritizing entities, there was a high level of variation in terms of geographic and jurisdictional scope, and the metrics included the PI scores (Table 2).Each PI entity considered different culvert owners, some choosing to only score culverts owned by the PI entity, and some choosing to score all culverts regardless of ownership.In our sample, the number of scored barriers ranged from 32 to 4259 and the geographic extent ranged from 30.51 to 71,300 mi 2 .Four of these entities used the same nonlinear formula, developed by the WDFW, and five entities developed their own linear PI equation.The remainder of our results focuses on the six unique PI methods.

| All PIs: Distribution results
The Shapiro-Wilk tests for normality identified Bellingham as the only normal ( p < .19)and negatively skewed distribution (Figure 2).The remainder of the PIs had a right skew, and WDFW had the greatest skew.Thurston County had the largest interquartile range (32.6), and Chehalis had the smallest (9.06).A visual comparison shows that WDFW, Chehalis, and CWCC have evenly tapered tails, whereas Thurston and King have longer and more irregularly shaped tails.

| All PIs: Shared metric analysis
As a first step to comparing metrics across PI methods, we analyzed all metric definitions to determine whether the metrics used in the PI formulas were unique or shared with other entities.Overall, the number of unique metrics is greater than the number of shared metrics at 18 to 13 respectively.Next, using a qualitative metric comparison matrix (provided in Table S2), for each PI, we calculated the percent of total metrics that were shared with at least one other entity (Figure 3A).We found that Thurston was the most unique PI, with the fewest shared metrics.Chehalis and WDFW were the least unique as all of their metrics were shared by at least one other PI studied.
Finally, we pooled the PIs and analyzed the percent of shared metrics within each category (Figure 3B).We found that coordination was the most unique among categories of metrics and habitat quantity was the least.Table 3 provides further detailed information about the individual metrics coded as "shared" or "unique." The most highly shared metrics were upstream habitat gain (100%), barrier passability (83%), number of downstream barriers (83%), and species-specific presence at barrier (83%).

| Linear PIs: Category weights
Five of the PIs used a linear equation, with each metric as a weighted input.The similarity between linear equations across entities allowed for direct comparison of the scoring weight of metric categories (Figure 3C).A visual comparison shows that PIs assign weights differently.For example, King assigned a relatively high weight to habitat quantity, whereas Bellingham assigned a relatively low weight to this category.In addition, some entities included many categories, such as Bellingham, which results in overall reduced weights for individual categories.

| Nonlinear PI (WDFW method): Weights
Although a nonlinear PI formula does not contain explicit weights, and it is difficult to analyze tradeoffs across metric categories, the implicit tradeoffs within metrics can be measured.Here, we examine how the PI scores vary based on changes in cost and ESA listing status.

| Scenario 1: Varying costs
The WDFW PI method, used by four entities, uses road class size as a proxy for cost, with larger roads considered higher cost, and smaller roads considered low cost.We model the change in the WDFW PI score in relation to changes in barrier restoration cost using the median PI score (13) as a point of reference for comparison (Figure 4).We found that the scores exhibited a fixed ratio, in terms of upstream lineal habitat gain (ULHG) needed to achieve an equivalent PI score across cost scenarios (see the Supplementary Material S3 for a derivation).All else equal, a low-cost culvert would need a third of the ULHG as a high-cost culvert to achieve the same PI score.Likewise, a medium-cost culvert would need half the ULHG as a high-cost culvert to achieve the same PI score.We further investigated the WDFW's assumption on costs to see if the road class cost categories, that is, a score of 1-3 for low-, medium-, and high-cost projects, respectively, fit data on restoration costs.Specifically, we examine cost ratios for medium-and low-cost barrier restoration projects as reported by the Pacific Northwest Salmon Habitat Restoration Project Tracking Database (PNSHP) data (Pacific Northwest Salmon Habitat Project Database, 2021;Van Deynze et al., 2022).The PNSHP ratio of median costs for barrier restoration projects on the middle-cost and lower-cost roads was 1.46, closely matching the ratio of 1.5 used in the WDFW PI (Table 4).

| Scenario 2: Varying species ESA status
The WDFW formula accounts for ESA-listed species through a metric that takes on three values (ESA Listed = 3, Species of Concern = 2, and Non-listed = 1).Thus, the ESA metric also generates a constant ratio of habitat required to equalize scores across ESA status (1/ 3:1/2:1) or (2:3:6), as seen in the previous cost scenario.
Although the constant ratio property holds for ESA status, the total PI score for a culvert is the sum of species-specific scores, with species that have a higher production potential (P) such as sockeye salmon (P = 3) having higher values for metrics in the habitat quality category than species with a low P such as bull trout/ dolly varden (P = 0.0007).Therefore, the impact of ESA listing while holding habitat constant varies across species (Figure 5).Specifically, we find sockeye and chum salmon realize the largest PI score gains from a listing  status of Endangered/Threatened (Figure 5).Production potential is also high for pink salmon, although there are no ESA-listed pink salmon populations in the Case Area (Figure 5).

| DISCUSSION AND CONCLUSION
The need to replace barrier culverts to increase iconic salmon populations and uphold Tribal treaty rights is gaining traction across Western Washington, even for those entities not bound to do so by the federal injunction.As a result, in the last decade multiple prioritization frameworks have emerged in the region.Here, we conduct a comparative analysis of six prioritization frameworks adopted by 10 entities in Western Washington, together covering an estimated 9786 prioritized barrier culverts.Through our analysis, we increase transparency of restoration priorities in the Case Area and evaluate how attributes of a high-priority barrier are either similar or different across prioritization methods.
We found that all entities in the Washington Case Area with documented prioritization methods have The impact of habitat gain of different Endangered Species Act (ESA) listing categories on prioritization index (PI) scores for culverts, calculated with the Washington Department of Fish and Wildlife method.Species with ESA-listed runs within the Case Area were modeled separately for ESA status with a solid line.The dashed line indicates possible scores for ESA listings that do not exist within the Case Area.Species were organized from the highest PI to lowest to demonstrate overall prioritization weight by species.adopted PI scores to decide which projects to fund in the near term, for example, in contrast to optimization methods or relying on expert opinion alone.Half of these entities used a common nonlinear PI equation, developed by WDFW, and the remaining five entities developed a customized linear PI equation specific for their region or jurisdiction.We found that there is variability in both the functional form of the PI formulas used as well as inputs to the function, including variables and the weights on those variables.
Review of PI method documentation and discussions with practitioners suggest that this customization is often driven by the feasibility for data collection (i.e., whether data for variables is constantly available across candidate barriers); the number of culverts under consideration; the specific habitat needs of salmonid species targeted by the plan; and the desire to incorporate feedback from local experts, stakeholders, and Tribal comanagers.The widespread use of custom and jurisdictionally specific PI methods means that, even within a region, identical barrier culverts may be prioritized very differently, depending on barrier ownership.However, 4 out of 10 entities chose to use the same nonlinear method, which suggests a strong interest in a coordinated and consistent statewide methodology or a lack of local capacity to develop customized methods.
Some commonalities between the PIs in the Case Area exist in terms of included factors.For example, Connectivity, Habitat Potential, and Habitat Quantity are included in all of the PI formulas.These commonalities suggest core themes and consistent values across entities, which could form the basis for a shared consensus prioritization method.
In addition, a Species metric, or a metric that rewards barriers where the presence of specific species is documented or expected, is included in four of the five linear PI formulas and in the nonlinear WDFW framework.In linear formulas, Species metrics were generally given low weight, whereas in the nonlinear PI, the importance of ESA listing status (a type of Species metric) depends on the productivity of the species, with sockeye and chum receiving the largest increase in PI from an ESA listing.That is, although there is near consensus that barriers where species of concern are likely to be found should be prioritized, the degree to which this contributes to overall scores is limited.Surprisingly, Cost is not included in five of the six linear PI frameworks, although the widely adopted, nonlinear WDFW framework gives considerable weight to this metric, requiring a high-cost barrier contribute approximately three times the upstream habitat than a low-cost barrier for a similar score.In Bellingham, the only entity with a linear PI framework including Cost, the scoring weight on Cost is 12.9%, a lower influence on the overall score compared with the WDFW framework.Although costs are treated inconsistently across prioritization frameworks, they are often considered outside the formal scoring process.The ubiquity of barrier restoration projects has allowed analysis of patterns in costs across the study region, presenting an opportunity for the comparison of expected barrier restoration costs on a consistent basis, which may inform consistent representation of costs in future prioritizations (Van Deynze et al., 2022).
Variables representing metrics in PI frameworks are often represented by unique metric inputs, with the number of unique metrics outnumbering the number of metrics shared across methods.Differences in metric inputs further demonstrate the ways in which entities customize their prioritization processes.For example, coordination was the most unique category, which included metrics for a range of collaborative examples including educational opportunities, stakeholder engagement, or construction feasibility.
Customization offers various advantages, such as empowering local entities to establish their own priorities and allowing them to leverage all available data and information deemed valuable for informed decisionmaking.However, customization does introduce challenges in terms of transparency if methods are not sufficiently documented and published; coordination across entities (e.g., none of our PI scores give increased weights to barriers hydrologically connected to planned projects of other entities) and comparing candidate projects across entities as the range of scores varies considerably across entities (e.g., some frameworks have maximum scores while the WDFW score is unbounded).Cross-entity comparisons are particularly important for external funding agencies seeking to allocate funds across the region to the projects with the greatest benefits.Alternatively, funding agencies can establish prioritization criteria that reflect their own or consensus objectives.Establishing clear and shared prioritization criteria can allow for improved coordination across jurisdictional boundaries.
Although PIs are a valuable source of information, entities seldom rely on them as the sole deciding factor in removing barriers.Instead, in practice, PIs are typically considered alongside other factors that determine which barriers should be corrected first.Opportunistic restoration projects are common, for example, when other routine road maintenance is scheduled.These processes are often informal and undocumented, and we are unable to systematically assess what factors are included in the prioritization process outside of the PI scores.Thus, it is unclear whether these unobserved factors drive increased similarity in PI frameworks across the Case Area or divergences.For example, if cost information is as a part of the prioritization process for those entities that do not include cost formally in their PI scores, differences may not be as great as they seem from an evaluation of the PI scores alone.However, the ability of the prioritization process to include local knowledge and local preferences may exacerbate differences.Future research on the larger prioritization process is needed to more completely understand how entities are deciding which barrier culverts to fund in the near term with limited budgets.

AUTHOR CONTRIBUTIONS
CAB drafted the original manuscript, conducted data analysis, and produced figures and tables.CLS and CAB conducted review and gray literature search to identify candidate prioritization methods.SLJ and BVD conceptualized research ideas and provided supervision of research process.All authors contributed to editing and review of the final manuscript.

F
I G U R E 1 Map of study area showing the Case Area within the state of Washington.Each prioritization index geography is shown within the Case Area.Shading is used to denote PIs that overlap geographically.Regions where no PI was located are shown as an unlabeled aggregated land mass.Washington Department of Fish and Wildlife and Washington State Department of Transportation are not shown because their geographies span the Case Area.
that include upstream habitat gain from barrier removal

F
I G U R E 2 Histograms of the prioritization index (PI) scores for barrier culverts.The mean and median are displayed as a solid and dashed vertical line, respectively.The interquartile range is shown as a shaded region.The histograms are ordered by sample size (n), with 100 bins for each histogram.All PIs were normalized to 100 as the maximum PI score.The scale of the y axis is determined by the maximum individual score frequency for each entity.CWCC, Cold Water Connection Campaign; WDFW, Washington Department of Fish and Wildlife.

F
I G U R E 4 The impact of habitat gain for culverts of different cost categories on prioritization index (PI) scores calculated with the Washington Department of Fish and Wildlife (WDFW) method.The solid horizontal line shows the WDFW median PI score.Dashed lines show where the different cost scenarios intersect with the median PI score.
Geographic extent includes watersheds, known as Water Resource Inventory Areas (WRIA).
Road class 3 matches Washington Department of Fish and Wildlife's (WDFW's) proxy for low cost, road class 2 is WDFW medium-cost proxy, and road class 1 is WDFW high-cost proxy.Note that there are no barriers in the Case Area in the Pacific Northwest Salmon Habitat Restoration Project Tracking Database (PNSHP) sample that fall within the WDFW Class 3 definition.N/A is used to denote not applicable because data were not available for road class size 1. Note: