Using systematic conservation planning to inform restoration of freshwater habitat and connectivity for salmon

Instream barriers remain ubiquitous threats to freshwater species and their habitats. Decisions regarding barrier removal are often aimed at maximizing habitat area and connectivity for freshwater fish; yet can be challenging due to the sheer number of barriers, uncertainty in species presence, abundance


| INTRODUCTION
Wild Pacific salmon support ecosystems, economies, and cultures. Today, wild Pacific salmon are at record lows and face multiple cumulative pressures (Beamish et al., 2004;Grant et al., 2019). The loss and degradation of freshwater habitat is a key pressure contributing to populations declines (Boyle, 1997;Cohen, 2012;Finn et al., 2021). Despite high levels of investment (Bernhardt et al., 2005), efforts to restore salmon habitat are hindered by a lack of coordination throughout watersheds, small-scale restoration efforts, scarce resources, and uncertainty about what actions to take and where to take them to have the greatest benefit for salmon recovery. In urban areas, the scale of habitat loss and degradation along with intense competition for land amplifies these challenges (Walsh et al., 2005). Salmon habitat restoration decisions are further complicated by the need to consider the potential benefit and cost of alternative restoration actions for multiple species and populations, decisions that are linked upstream and downstream (Albert et al., 2020;Magurran, 2016).
Salmon restoration and rehabilitation efforts are commonly taken at the scale of a single reach or multiple reaches throughout a watershed and have had mixed success due to the challenge of considering the watershed scale processes of ecosystem threats, and historical conditions (Roni et al., 2008). Failure to identify watershed stressors can lead to wasted resources and poor return on investment in the form of ecological recovery (Palmer et al., 2010). Instead of approaches that focus on repairing specific habitat conditions, actions are needed that address the restoration of landscape processes which shape and sustain thriving habitats Roni et al., 2008). Integrated approaches that assess the threats acting on an entire watershed are needed in order to implement holistic actions that address the primary causes of ecosystem degradation . Recognition of the importance of the spatial connectedness and influence of watershed scale processes on the quality of aquatic habitat lend themselves to systematic approaches to planning restoration efforts (Hermoso et al., 2011;Langhans et al., 2016;Salgado-Rojas et al., 2020).
Instream anthropogenic barriers consist of structures that restrict movement of organisms longitudinally through a stream network or laterally through the floodplain (Coleman et al., 2018) and have multiple impacts including limiting movement of nutrients (Tockner et al., 1999), changing flow regimes (Tonkin et al., 2018), changing sediment deposition patterns (Fryirs et al., 2007), and changing thermal characteristics (Gordon et al., 2015). Barriers also block access to habitats required for different life stages of migratory species (Gibson et al., 2005) and can fragment resident fish populations (Esguícero & Arcifa, 2010;Jager et al., 2001). Restoring connectivity of habitat isolated by instream anthropogenic barriers is a priority in watershed restoration due to its often quick biological response and relatively high success rate when suitable upstream habitat exists (Roni et al., 2002). However, deciding which barriers to remove in priority to achieve the greatest benefit for multiple freshwater species remains a challenging question. Site selection is complicated by a variety of factors including the spatial connectedness of barrier removal actions and therefore potential complementarity of multiple barrier removal projects, uncertainty regarding species benefit, limited financial resources, and the goals or values of the entity carrying out restoration. Depending on the species of concern, migration and movement patterns may pose further constraints on the potential benefit of barrier removal (McManamay et al., 2015).
The identification of priority barriers for removal has been approached using a variety of methods (McKay et al., 2017). Broadly, these approaches include mathematical optimization (Lin et al., 2019), scoring and ranking techniques, and approaches that rely on local knowledge, unexpected failures, or reactive responses (McKay et al., 2020). Scoring and ranking methods can be intuitive, and flexible but commonly do not consider spatial relationships of barriers and the complementarity of multiple barrier restoration projects (McKay et al., 2020;O'Hanley & Tomberlin, 2005). Mathematical optimization on the other hand is useful for identifying the complementarity of multiple potential projects while maximizing quantitative criteria, however may be limited by data constraints, and has been criticized for being overly prescriptive, requiring action on a set of barriers in order to achieve benefits (McKay et al., 2020).
To address some of the inflexibility of previous optimization methods we use a systematic conservation planning approach to explore the potential benefits of barrier removal to restore wild Pacific salmon habitat and identify locations that may require additional management action for freshwater restoration to succeed. The systematic conservation planning framework is commonly applied to area-based conservation problems to optimize the spatial representation of conservation action, most commonly the designation of protected area networks (Possingham et al., 2000). Many of the challenges that come with protected area optimization problems are shared with the issue of prioritizing freshwater barrier removal. Often a large number of land parcels may be candidates for protection, each land parcel has a variable cost of acquisition or protection, and they all have varying benefits when it comes to the types and amount of habitat or species that are contained within (Ball et al., 2009;Church et al., 1996). The same can be said when it comes to anthropogenic barriers in freshwater systems, where databases of observed instream barriers are recorded and can be candidates for removal. Each barrier may have varying costs for removal depending on the type of barrier and its location, and will vary in the amount of habitat upstream depending on the species of concern and other barriers in the system. In fact, specific constraints that stress the importance of upstream connectivity in freshwater systems have been incorporated into area-based prioritizations (Hermoso et al., 2011).
In this paper, we use a systematic conservation planning framework to understand opportunity for restoring connectivity for salmonids in the Lower Fraser River, British Columbia (BC), Canada, through the removal of anthropogenic barriers. The Lower Fraser presents an important case in terms of barrier removal as it has some of the highest diversity of salmon populations in the world (Northcote & Larkin, 1989), while at the same time, intense economic development and subsequent landscape change have caused the isolation of an estimated 64% of linear stream length (Finn et al., 2021).
By estimating the cost of restoration and quantifying the amount of habitat upstream of each barrier for 14 populations of wild Pacific salmon including Chinook, coho, chum, and sockeye, and the assessment of four indicators of habitat quality at the watershed scale, we investigate how restoration priorities change across scenarios that emphasize quantity and quality of habitat differently. This approach is then placed into a broader adaptive management context intended to demonstrate how the use of existing systematic conservation planning software can be integrated into watershed planning and implemented in a time and place where decisive action for salmon is urgently needed.

| Study area
The Fraser River watershed covers a quarter of BC with an area of 233,000 km 2 (Northcote & Larkin, 1989). The Lower Fraser is commonly defined as the final 150 km stretch of river downstream of Hope, BC, where the flow diverts westward toward the estuary that empties into the Strait of Georgia. For this study, the Lower Fraser was delineated hydrographically using the watershed groups defined in the BC Freshwater Atlas. All watershed groups that contribute to this final 150 km of the river were used to delineate the Lower Fraser Watershed, including the Fraser Canyon, Chilliwack, Harrison, Lillooet, and Lower Fraser watershed groups ( Figure 1). Although only representing a modest portion of the entire basin, at 20,203 km 2 , the Lower Fraser contains considerable geographic diversity. The valley that surrounds the Fraser mainstem is largely populated by the Vancouver metropolitan area, which is home to 2.5 million people, approximately half the population of BC.
For salmon, the Lower Fraser is disproportionately important, it supports the highest diversity of populations in the entire basin (Nesbitt & Moore, 2016), and acts as a migration route for populations that spawn further up the basin. Pacific salmon in Canada are managed at the level of a conservation unit (CU) (Holtby & Cirunia, 2007), defined as a group of salmon that are isolated enough such that if they were to be lost, re-colonization would not happen within an acceptable timeframe (Fisheries and Oceans Canada, 2005). The watersheds used by each CU have been mapped and these boundary files were overlaid with the study area. CUs were then included if their boundary was completely contained within the study area. Lake-type sockeye salmon were excluded from consideration as their CU boundary is defined by their natal lakes, rather than watersheds. Since the lake must be accessible for the CU to exist, we assume no barriers can be assigned to lake-type sockeye. In total 14 CUs were included (Table 1, Data S1). Many of these CUs are currently at record low abundances, with five CU's designated as threatened or endangered by the Committee On the Status of Endangered Wildlife In Canada (COSEWIC), and a further seven unassessed or data deficient (Grant et al., 2019).

| Define anthropogenic barriers
Anthropogenic barriers were collated from three primary sources including Provincial Stream Crossing Inventory System (PSCIS), Fish Information Summary System (FISS), and Watershed Watch Salmon Society (WWSS) Flood Infrastructure Mapping. Freshwater Atlas obstructions data were also included from the BC Freshwater Atlas (FWA) (Data S1). The PSCIS, FWA, and WWSS data sets were compared against the FISS data set for any redundancy and any overlapping entries within the FISS data set were removed. The barriers were then combined to a single data set. In total 1280 anthropogenic barriers were identified within the study area, and they included hydroelectric dams (hereafter referred to as hydro dams), road culverts, flood infrastructure (floodgates and pump stations), small dams, and weirs. Barriers were removed from the analysis if they were upstream of a natural barrier to salmonids (Data S1), fell on a loop in the system (such as a braided channel), and thus did not create complete barrier to upstream habitat, or if spatial error meant that the barrier could not be placed on a stream. This left 669 barriers to be considered in the optimization.

| Defining habitat features
Several metrics were used to characterize the potential benefit to salmon from barrier removal. Both the amount and quality of habitat that would be restored were considered important. The quantity of habitat upstream of each barrier was estimated by measuring the linear streamline from the 1:20,000 BC Freshwater Atlas stream layer up to the next mapped anthropogenic or natural gradient barrier. This only serves as an indicator for habitat quantity, acknowledging that not every part of this streamline will correspond to useful habitat for each species. We assume that a greater length indicates more potential "habitat" restored and refer to it as such. Species-specific stream gradient thresholds were used to determine the natural accessibility of streams (Washington Department of Fish and Wildlife, 2019). Natural accessibility was then used to estimate the length of salmon-accessible streams upstream of each barrier. Maps of salmon CUs defined whether the habitat upstream of a barrier would benefit a given CU (Data S1). Measurements were then standardized to a proportion of the total alienated habitat length within the study area to improve model performance.
Indicators of habitat quality were quantified so that barriers downstream of poor habitat could be avoided or selectively prioritized (Table 2). For each stream segment, the watershed was determined using the corresponding 1:20,000 watershed layer in the BC Freshwater Atlas. We used landcover data to summarize the impervious surface, riparian habitat disturbance, and watershed disturbance upstream of each stream segment. In addition to landcover indicators of quality, the proportion of upstream habitat that was described as "ditch" in the Freshwater atlas was also quantified (Table 2). Detailed methods for spatial analysis and data used are provided in Data S1. For each indicator of disturbance, a threshold was used to determine whether the threat was present for the stream where the barrier occurred. Thresholds for disturbance levels were adapted from the assessment of wild, threatened, and endangered streams of the Lower Fraser Valley (Precision Identification Biological Consultants, 1998). If a stream was above the threshold, the quantity value of the species with the greatest presence was assigned ( Table 2). The thresholds ensure that each quality feature is treated equally to the quantity value of a single species. The result is that each barrier has an estimate of the amount of habitat upstream, with that value duplicated for each quality indicator that stays over the threshold.

| Estimating the cost of barrier removal and restoration
Cost estimates were used to differentiate the primary types of barriers that inhibit salmon passage in the Lower Fraser River. These barriers are predominantly made up T A B L E 1 Species and Conservation Units (CU) included in the study area.

Species
Conservation unit of hydro dams, flood infrastructure such as floodgates and pump stations, as well as road culverts. Both the Coquitlam and Alouette River dams have had detailed feasibility studies done within the last 20 years that contain cost estimates for the installation of fish passage infrastructure (Gaboury & Bocking, 2004;R2 Resource Consultants, Inc., 2018). These estimates were converted to 2020 Canadian Dollars and applied to their respective dams, and the average of the two was applied to the Ruskin Dam which sits on the Lower Stave River. It should be noted that restoration costs for the two dams do not include ongoing costs of water management to assist fish movement (Gaboury & Bocking, 2004; R2 Resource Consultants, Inc., 2018). Estimates for both culvert restoration and restoration of flood infrastructure were derived from a structured expert elicitation (Hemming et al., 2018) and collated information on past projects. Other barrier types that were less common, such as small dams and weirs were assigned the same cost as culvert replacement which was estimated as $225,000. Cost estimates for flood infrastructure were set at $3,000,000 per site (Table 3). Although these are static estimates for broad "barrier types" a sensitivity analysis was performed to understand the impact of cost variation on results (Data S2).

| Optimization
We used the R package "Prioritizr" (Hanson et al., 2019) with Gurobi Optimization solver (Gurobi Optimization LLC, 2020) to optimize the selection of priority barriers for removal (R Core Team, 2019). We used the "maximum utility" objective function, which seeks to maximize the overall representation across the set of conservation features, subject to a budget. In this case, the features represent the habitat quantity for each species, and the habitat quality indicators. This objective function does not use targets, but rather the representation of features in the output is controlled by feature weights and is expressed as: Where x i is the binary decision variable for whether barrier i is included in the solution, and c i is the cost of barrier i. a j represents the total amount of each feature upstream of selected barriers where r ij represents the amount of feature j upstream of barrier i calculated for all (8) features within the set () of features J. w j is the user given weight of feature j. The scaling factor s is used to shrink the costs so that the problem will return the cheapest solution when there are multiple solutions with the same feature representation. The budget, B, constrains the total possible cost of the solution (Hanson et al., 2019). The "add contiguity constraints" function forces the solution to consist of barriers that are spatially connected and was used to ensure that barriers were selected from the base of the watershed upwards (Data S3).
Individual barrier priority was assessed using two metrics. First was the selection frequency, which T A B L E 3 Estimated costs for the restoration of barrier types and a description of the action that is assumed to be implemented. Average calculated from other dams in the study area represents the number of times the barriers were selected to be part of the optimal solution at a given budget level and scenario. For each budget level and scenario, the 10 solutions closest to optimality were calculated. By generating multiple solutions, a better understanding of the relative importance of each barrier can be understood by counting how frequently sites are selected. The second metric for the importance of specific barriers was the calculation of the irreplaceability score using the replacement cost method, where a replacement cost of zero indicates the site could be swapped in or out of the optimal solution and it would not improve its performance, while values closer to one express rising importance for a site in the optimal solution (Cabeza & Moilanen, 2006, Data S4). Figure 2 shows a conceptual model for how data sets are used to calculate the planning unit attributes that are then entered into the optimization model.

| Scenarios: Habitat quality versus quantity
Potential trade-offs among priorities were examined through the development of three scenarios that optimized different characteristics of habitat upstream of barriers that might motivate restoration (Table 4). The first scenario aimed to maximize the amount of restored habitat across the 14 CUs of salmon that would become accessible if select barriers were removed. This scenario was then contrasted against scenarios that considered the quality of the habitat along with the amount. To do so, within prioritizr, weights can be assigned to different features to emphasize their representation in the final solution. The weights are numeric values that are applied to increase the value of their associated feature value and achieve a change in representation. Thus, to understand F I G U R E 2 Conceptual model for the optimization. Spatial data sets include the barriers that are being prioritized, the streams are used to orient the barriers and determine habitat quantity, and landcover data are used to inform habitat quality indicators. Barrier cost was estimated using structured expert elicitation and published data, and all barrier attributes are then used to spatially optimize barrier removal for a given budget. the impact of habitat quality on the spatial distribution of barrier removal priorities, the final scenario assigned a weight of 20 to each indicator of habitat quality. A weight value of 20 was used for each habitat quality indicator after sensitivity testing indicated that outputs remained similar for weights ranging from 10 to 500. Each scenario was run across a range of budgets covering funding levels between 1 and 200 million dollars with 5 million-dollar increments. A ceiling of 200 million dollars was selected as it aligns with recent government investments focused on salmon habitat restoration in BC (see Data S2). The ceiling can be extended to explore what is possible at higher levels of investment, but as budgets increase, returns diminish as remaining unselected sites become less cost-effective (or "optimal" according to the definition of the objective function).

| Regional connectivity restoration
The optimization identified the "optimal set" of barriers to restore at each budget level. The proportion of alienated habitat that could be restored by removing these barriers was plotted against a range of budgets from 1 to 200 million dollars for all three scenarios and indicated a diminishing marginal returns curve for each scenario (Figure 3). The length-only scenario indicates that the maximum amount of habitat that can be restored with a budget of $200 million is approximately 75% of naturally accessible salmon habitat upstream of mapped barriers. Importantly, the first 60% of this habitat can be restored for approximately $100 million, with dramatically reduced marginal gains past this level of investment. When the quality of the habitat was included in the optimization, the amount of habitat being restored did not change by a large margin. However, as the quality was given more weight, the amount of habitat that was being restored decreased across budgets by about 10% (Figure 3).

| Species representation
Although optimal removal of barriers showed similarities across scenarios in terms of the amount of habitat that was being restored, the species that benefitted from the optimal solutions varied. Fraser River Canyon CUs of Chinook and coho salmon received little attention when it came to identifying priority barrier removals when the quantity of habitat was the priority, however, this changed when the quality of habitat was prioritized ( Figure 4). When the quality of habitat was given weight in the prioritization, barriers impacting the Fraser River Canyon CUs became priorities at a budget of $40 million rather than $100 million. The opposite trend was observed for chum salmon, as the quality of the habitat being restored increased in importance, fewer barriers that would benefit chum salmon were included in optimal solutions. The quality-weighted scenario resulted in about 20% less habitat being restored for chum salmon across budget scenarios (Figure 4).

| Barrier types and priority locations
Priority barriers for removal appeared to be somewhat evenly distributed throughout the study area across all F I G U R E 3 Marginal salmon habitat restoration returns across a range of budgets in millions for three barrier removal scenarios: maximizing habitat length only (solid line), maximizing both habitat length and quality (long dash), and maximizing length and quality with additional weight applied to each habitat quality indicator (short dash). scenarios ( Figure 5). When habitat quality was given more weight, there was a subtle shift towards priority removals in the more northern and eastern (Lillooet, Fraser Canyon, and upper Chilliwack) regions of the study area. This shift was also reflected in the species representation ( Figure 4). Certain barriers remained priorities across all scenarios and budget ranges. For example, the Sumas River and Hatzic Slough floodgates, where the quantity of upstream habitat made these sites priorities regardless of the quality of this habitat. Several culverts were also consistently selected ranging from sites on the coastal Nicomekl and Campbell Rivers in the south to multiple tributaries of the Birkenhead River in the north. Contrasting selection frequencies among scenarios can give insight into what might be driving the priority of specific barriers and if there are additional considerations that needed to be made. Scenarios that emphasized habitat quality generally selected fewer floodgates compared with those that emphasized habitat quantity, the priorities generally shifted towards culverts in more remote areas and investing in restoring passage past the large hydro dams. (Data S4).
The calculation of replacement cost for barriers indicated how important they are to the optimal solution for a given scenario at a given budget level. Figure 6 shows the representation of floodgates steadily increased before leveling out at $$60 million of investment. However, this number drops when habitat quality is included. At the same time, a greater number of hydro dams were included in optimal solutions above $80 million in investment for length only and $30 million when habitat quality was included ( Figure 6). Two of the hydro dams (Alouette and Coquitlam) were consistently selected in optimal solutions. The importance was amplified for both dams in scenarios that also considered habitat quality, with the replacement cost of Alouette Dam increasing to 0.75 at or above investments of $25 million dollars and the Coquitlam Dam at investments levels of > $60 million. The third hydro dam in the area (the Ruskin Dam on the Lower Stave River) was also selected at investment levels larger than $110 million (Data S4). The relative representation of barrier types is the number of each type of barrier normalized to the proportion of that barrier type in the set of all barriers (i.e., if barriers were selected randomly the value should approach 1). It can illustrate broad strategies that might be taken under different management and budget scenarios. The relative representation of hydro dam barriers in Figure 6 could be explained by the relatively low number of these barriers within the study area. Their selection, therefore, is more likely driven by the value of their removal than chance.
F I G U R E 4 Cumulative proportion of habitat that is restored for 14 conservation units of salmon across three scenarios for barrier removal prioritization. Minor random variation has been applied to visualize overlapping lines.

| DISCUSSION
Systematic conservation planning software commonly applied to plan protected area networks was used to prioritize the removal of 669 in-stream anthropogenic barriers to restore salmon habitat connectivity. Expressing barrier removal as a conservation problem in this way provides a flexible framework for the optimization of F I G U R E 5 Spatial distribution of mapped barriers to fish passage and cumulative selection frequency for all restoration scenarios examined. The cumulative selection frequency indicates how many times a barrier was selected in the optimal solution across all budget levels.
F I G U R E 6 Relative representation of barrier types in optimal solutions for each scenario. The relative representation is calculated as the difference between the number of each barrier type in the optimal solution and the expected number based on the proportion of each barriers type in the set of all barriers. different habitat values, across a range of budgetary constraints. Based on this model, a $200 million investment in restoring connectivity for salmonids could restore approximately 75% of the stream length that is currently inaccessible in the Lower Fraser River, while half this investment could restore 60%. With an estimated 64% of naturally accessible stream length currently alienated in the Lower Fraser as a result of barriers (Finn et al., 2021), this represents 1667 and 1333 km of habitat for investments of $200 or $100 million, respectively. Multiple scenarios of optimal restoration were explored to identify where, and for which species, additional restoration, or management actions may need to be taken. Generally, as the quality of habitat was given more weight in the prioritization, the amount of habitat that was being restored decreased, and priority restoration sites shifted out of the Lower Fraser Valley and into the surrounding mountainous areas where there are relatively fewer threats to habitat quality. The resultant shift in priorities across space impacted certain species more than others, suggesting that some CUs may require more than just barrier removal in order to restore productive freshwater habitat. Restoring passage past the Alouette and Coquitlam dams was a consistent priority at high levels of investment. Floodgates were important for restoration, often due to their position in the catchment, but they became less of a priority as habitat quality was given importance in the prioritization, suggesting the quality of some habitats upstream of floodgates may need to be addressed.

| Regional insights
The two dams that were included in optimal solutions have received attention for fish passage remediation in the past. The Alouette Dam was installed in 1928 and has since blocked access for all anadromous species to upstream habitat (Gaboury & Bocking, 2004), while the Coquitlam Dam has blocked anadromous salmonids since 1914 (R2 Resource Consultants, Inc., 2018). Although the dams are included in optimal solutions, they are only included at higher levels of investment for scenarios that prioritize the maximum length of habitat. No dams were selected in optimal solutions with budgets below $80 million. This indicates that at lower levels of investment, more habitat can be restored by addressing a large number of culverts and some floodgates before any hydro dams become priorities. However, the budget at which dams become a priority declines dramatically to $30 million when the quality of the habitat is given weighted priority. The inclusion of these dams in optimal solutions, despite their elevated costs for restoration, supports the findings of previous work that has demonstrated how aligning dam restoration with the restoration of other barriers increases the cost-effectiveness of action (Fitzpatrick & Neeson, 2018). Importantly, this analysis reveals the levels of investment at which those return-on-investment gains can be realized. However, the point at which this happens depends on the restoration goals that are emphasized. An additional consideration for if these dams were restored is that the opportunity costs related to lost energy potential from reservoir releases to facilitate fish movement were not included in the prioritization (Gaboury & Bocking, 2004;R2 Resource Consultants, Inc., 2018).
Recent work has established how floodgates create local upstream environments that may be hotspots for invasive species due to increased water temperatures and lower dissolved oxygen (Scott et al., 2016). While the number of individual floodgates included in optimal solutions decreased when habitat quality was weighted in importance ( Figure 6), there remained several sites that grew in priority as the budget increased, and eventually became required for the feasibility of the solution (according to the replacement cost metric; Data S4). While restoration of habitat quality may be required for many streams with floodgates, the very act of restoring fish passage could improve habitat quality through the associated changes in flow regimes (Gordon et al., 2015;Seifert & Moore, 2018).

| Relevance for species-specific management
Removing barriers to allow access to productive habitat can help raise the carrying capacity of freshwater ecosystems to support salmon, but threats throughout their life cycle including climate change (Crozier et al., 2021), changing marine conditions (Oke et al., 2020;Ruggerone & Irvine, 2018), exploitation (Healey, 2009), and disease (Miller et al., 2014) all play a role in determining how population trends of each species will be impacted by the restoration of freshwater habitat connectivity. From an anadromous species recovery perspective, each species or population will benefit differently from a mix of management actions ranging from high-level fisheries policy reform to direct restoration actions in the freshwater, marine and terrestrial realms Roni et al., 2002;Walsh et al., 2020). For these reasons, it is important that a process like barrier removal optimization be situated into a broader adaptive management framework which is focused on ecosystem process restoration (Linke et al., 2019). A future direction for spatially optimizing connectivity restoration should be to incorporate a more mechanistic understanding of how the removal of barriers contributes to improving the quality of aquatic ecosystems, raises the capacity for spawning and rearing, and contributes to population productivity. Understanding the causal impact of habitat restoration can be exceedingly difficult, especially for actions that require decades to effect habitat change such as riparian planting or complete road removal Tear et al., 2005). However, estimates for the response of coho salmon have shown that restoration actions, including barrier remediation , can produce substantial increases in fish production.
The regional trade-off between the total amount of habitat that is restored, and the quality of habitat varied between CUs. For example, there were fewer streams associated with chum salmon prioritized for connectivity restoration as habitat quality was given more weight in the prioritization because those streams have been heavily impacted by urban and agricultural development. The Fraser River Canyon CUs of Chinook and coho salmon on the other hand, saw more associated habitat being restored with increasing emphasis put on indicators of habitat quality. By running multiple scenarios and emphasizing different components of the prioritization, streams that need additional actions to be restored can be identified as well as those that might provide intact habitat upstream.
Each CU varies in the extent to which they are impacted by stream loss-not just alienation. The relative representation of CUs in the optimal solutions of each scenario ( Figure 5) did not consider the total amount of their historical habitat that was restored, but rather showed the proportion of habitat upstream of barriers that had access restored. The interpretation of how much habitat is being restored should consider the $1700 km of streams that are estimated to be completely lost to avoid shifting baselines in the perception of salmon habitat in the Lower Fraser (Finn et al., 2021).

| Site-specific complexity and implementing action
Our approach integrates a landscape-level understanding of potential benefits from barrier removal, with cost information and complementarity of multiple restoration sites to highlight priorities for restoring connectivity. Our analysis integrated four different data sets on barriers to fish passage which varied in their reliability. While some structures undeniably create complete barriers to movement for all species, variation in swimming and jumping abilities, combined with seasonal variation in water levels mean that some barriers impact some species or life stages more than others. This variation was handled to some extent by assuming different gradient thresholds when estimating natural gradient barriers for each species (Data S1), however, the incorporation of this level of detail is made difficult by the breadth of the problem.
There is a balance to be made between investing in reducing critical uncertainty that may change spatial priorities and investing in taking action (Buxton et al., 2020). Putting off a decision to act, in favor of continued monitoring and information gathering, can result in missed opportunities from funding agencies or result in irreversible population impacts that could have been avoided (Martin et al., 2017). An iterative approach will rely on the involvement of decision-makers and stakeholders to determine the priorities that will influence the site characteristics (e.g., species presence, habitat values) that are included in the prioritization (Er} os et al., 2018;Grizzetti et al., 2016). As priority barrier removals are identified, they will need to be ground-truthed to confirm the impact of the barrier in limiting movement, quality of habitat upstream, and the existence of additional barriers that change the potential complementarity of restoration sites.
Variations in how a given barrier may be restored can be implemented within this framework by adjusting sitespecific cost estimates. Our optimization only considered a single technique for barrier restoration at each site. For culverts and flood infrastructure, we assumed the complete replacement of structures to make it passable for salmonids, while hydro dam connectivity was informed by recommended actions from previous feasibility studies. The complete replacement of infrastructure is not always needed. Improved operation of floodgates (Gordon et al., 2015) and retrofitting existing infrastructure can facilitate passage through culverts (Cabonce et al., 2019;Franklin & Bartels, 2012). Restoration methods that avoid the complete replacement of structures are often much cheaper, meaning our cost estimates for restoring habitat throughout the Lower Fraser River are likely on the high end.
The spatial planning method employed here is embedded into a social-ecological landscape, the characteristics of which vary across space. Outputs from the optimization may be best used to facilitate partnerships between First Nations, other governments and stewardship groups interested in restoring habitat with those who have jurisdiction over the structures that create the barriers. We have not included all potential values or habitat features that could be included and used to inform priorities. Site characteristics that could indicate the feasibility of projects including infrastructure upgrades , social license (Christensen et al., 2009), or ecosystem service improvement (Adame et al., 2015) are project attributes that could be quantified for inclusion in the prioritization and spatial optimization process. Many barrier characteristics often included in spatial optimization describe positive aspects of barrier removal, however, in some locations the removal of barriers can facilitate the spread of invasive species and this threat must be balanced with habitat connectivity restoration (Milt et al., 2018;Pratt et al., 2009). Further efficiencies can be achieved through aligning scheduled infrastructure upgrades with restoration action .
The use of systematic conservation planning optimization software to identify priority barriers for removal while also considering indicators of habitat quality and connectivity with the marine environment provides a significant advance in our ability to identify cost-effective salmon habitat restoration actions. Spatial data on barrier location and the river network are all that are required to define the spatial connectivity needed for the optimization and additional data describing species, habitats, or a range of other values can be incorporated and explored depending on the context. With freshwater barriers being a global issue (Reid et al., 2019), our framework could be carried out anywhere in the world where data on streams and barriers exist.