Pacific North coast case study

The Pacific North coast of British Columbia, Canada is host to a wealth of resources important for ecological, economic and cultural reasons, many of which are unique to the region; for example glass sponge reefs, globally significant seabird populations, salmon, eulachon, and resident orca. A broad range of human activities occur in this region, as reviewed in [24]. Sea-based activities include fishing, aquaculture, tourism, utility and transportation. Coastal activities also influence the marine and estuarine resources in this region, including human settlement, ports and marinas, and log storage and handling. Land-based activities occurring in the watersheds are connected to coastal marine systems through freshwater runoff and include forestry, agriculture, mining and pulp and paper mills. The region is also subject to impacts from long-range and global stressors such as climate change, pollutants and debris (Table 1). Activities that include vessel use additionally include the stressors associated with either small or large vessel use in their cumulative risk.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Sector, human activities and associated stressors included in the risk assessment.

https://doi.org/10.1371/journal.pone.0162932.t001

Assessing Risk

We utilized a risk assessment framework for evaluating the risk to species and species groups (hereafter called ecosystem components, Table 2) from stressors associated with various human activities [11]. Risk was estimated as the product of the Exposure of a population to a specific human stressor and the Consequence or the sensitivity of that population to that same stressor. Risk scoring of each ecosystem component was done considering the species or habitat’s entire life history (rather than examining different life history stages separately). More details of the risk framework can be found in [11] and S1 File.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Ecosystem components assessed in the qualitative risk assessment.

Trophic groups with * are also included in the food web as prey species.

https://doi.org/10.1371/journal.pone.0162932.t002

Qualitative Risk Assessment

The relative Risk_ij to an ecosystem component was calculated by multiplying its exposure to a stressor by the consequence of that exposure, or Equation 1 where Risk_ij to ecosystem component j by stressor i is the product of the Exposure_i of stressor i and the Consequence_ij to ecosystem component j at being exposed to stressor i; where ecosystem component j is one of the pilot ecosystem components selected for this analysis (Table 2) and stressor i is a stressor produced by one of the sea or land-based activities (Table 1). Qualitative scoring of the risk variables used the same scoring methodology defined by [11] and scored by [12] (Table A in S1 File). Exposure_ij of ecosystem component j to stressor i was the product of three variables (scored between 0–3; score maximum = 36): Temporal Scale_i (TS), Spatial Scale_i (SS), and Load_i (L): Equation 2

Consequence was scored from 1 to 6 for the ecosystem component at the scale of individual stressors. Consequence_ij was squared to make the scale of the score comparable to Exposure_i (such that both have a maximum score of 36). This way, exposure and consequence effectively vary along the same scale even though we broke exposure into three subcomponents that we could score separately and had only one overall score for consequence. An example of the risk calculation (Eq 1) is presented in S1 File. The original risk scores for individual ecosystem component-stressor combinations were assigned after review of the available literature and consultation with experts. The scores are available as S3 File and scoring justifications in ([12] Appendix 4).

Incorporating Uncertainty into Risk Scores

For each of the four risk variables an uncertainty term was assigned (Table B in S1 File), and Monte Carlo simulation was used to incorporate this uncertainty explicitly into the calculation of risk. Each risk variable score (Temporal Scale, Spatial Scale, Load and Consequence) was assigned as the mean of a normal distribution with standard deviation as a function of the level of uncertainty assigned; higher uncertainty had higher standard deviation. The distribution was bounded by the minimum and maximum scores for each risk variable so that the scores estimated to be higher or lower than the variable’s scale were assigned as the minimum and/or maximum (e.g., a Load score higher than 3 would be scored as 3). The score of each risk variable was then randomly sampled from this distribution with 1000 replicates so that each risk variable was therefore an array with 1000 entries. The final risk score for each ecosystem component-stressor relationship was a product of the four arrays representing the risk variables (Risk = SS x TS x L x C²), where the first score generated from each variable array is multiplied across all four risk variables, followed by the second, and so on for all 1000 replicates and resulting in a final risk array of 1000 scores. The median and 10% and 90% quantiles from this final array of the overall risk to each ecosystem component-stressor relationship was reported. Quantiles were used instead of standard deviation or standard error because the resulting distribution of risk scores was non-normal.

Cumulative Risk

Cumulative risk to each ecosystem component was calculated by summing the total risk score produced for each ecosystem component across all its stressors, within each iteration of the Monte Carlo simulation. The statistical platform R was used to generate and run the code for the uncertainty scoring (R Development Core Team 2008); code available in S2 File. Direct cumulative risk CRisk to the ecosystem component is calculated by summing risk across all stressors so that Equation 3

Direct and Indirect Risk

In order to incorporate indirect risk through food web and habitat relationships we developed a series of linkage pathways between the ecosystem components and the prey and habitat ecosystem components on which they each depend (Fig 1). The linkage pathways were developed specifically for the study region using literature review (e.g. [25]) and expert consultation. The modeled relationships between species are known to vary based on location, season, year, etc and therefore the pathways represent generalized prey and habitat linkages. Including indirect risks, the comprehensive cumulative risk to each ecosystem component j is calculated as Equation 4 which includes 100% of its direct cumulative risk (CRisk_j), plus a proportion (p) of the risk to each of its supporting species (s), including prey species or biogenic habitat (if applicable). We applied a conservative, well-accepted energy transfer relationship of 10% [26] to reflect relationships between species in the network. Because calculating substitutability of prey items and incomplete spatial overlap between predator and prey is extremely data intensive and uncertain, even for a single species, we have made the assumption that, in the absence of complete dependency, the risk is diminished an order of magnitude in the transfer, following the 10% energy transfer principle. If the prey or habitat relationship is obligate (the species is entirely dependent on the prey or habitat species), 100% of the prey or habitat risk is added. For example, we applied this obligate relationship in the case of resident fish-eating orca and their obligate prey species salmon. In case the risk to a known prey species was not scored during the case study, the risk to the trophic groups was estimated by adding 10% of the risk to the ecosystem component representative (e.g. for any species consuming a pelagic forage fish other than Pacific herring, we added to its cumulative risk 10% of the cumulative risk to Pacific herring; all representatives for trophic groups are described in Table 2). As a full example, comprehensive cumulative risk to the ecosystem component Steller sea lion is estimated by the direct risk to Steller sea lion plus 10% of the risk to its prey items (relationships demonstrated in Fig 1): anadromous fish (salmon), mobile pelagic invertebrates (prawn), forage fish (Pacific herring), benthic fish (lingcod). In turn, the comprehensive cumulative risk to salmon was calculated using its linkage framework, including the biogenic habitats kelp and eelgrass.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Risk pathway (food web and biogenic habitat) for all species considered.

Dashed line = Obligate relationship (100% risk transfer).

https://doi.org/10.1371/journal.pone.0162932.g001

Analyses

Median and error (10% and 90% percentiles) of direct, indirect and cumulative risk were calculated and compared across ecosystem components. We tested for a relationship between the median direct and indirect risk scores for each ecosystem component using the nonparametric Spearman’s rank correlation test. Because the indirect risk framework depends on the structure and number of linkages in the food web, we used the Spearman rank correlation test again to compare the relationship between number of prey and habitat linkages and indirect risk. Because results of this test were strongly driven by an outlier, resident orca, which also differs from the other ecosystem components as being the only one with an obligate feeding relationship (with salmon), the test was repeated without this species.

Limitations and Uncertainties

Cumulative risk scores are highly dependent on the quality and detail of data regarding the linkages between predator and their choice of prey species, the predator’s ability to switch prey species (generalist vs. specialist), and species’ needs for specific habitats. Our study used a 10% risk-transfer relationship, where 10% of risk from a prey species is included as indirect risk to the predatory component, based on simple energy-transfer theory [26]. This 10% risk-transfer assumption may not appropriately represent the incomplete dependency of one species on another, due to substitutability with other prey sources, incomplete spatial overlap between the predator and prey, and context dependent relationships that vary across space and over time. For example, predator species that consume a variety of prey may nevertheless be largely dependent upon one or more species at particular times or life history stages. Calculating substitutability of prey species and spatial overlap is very challenging, so here we used a simple assumption, that risk is diminished an order of magnitude in the transfer. In future cumulative risk assessments, other transfer relationships could be used when there is specific evidence of relative prey dependence and/or habitat utilization. For example, gut content or stable isotope analysis has been used to estimate relative prey importance in marine mammals [34], [36]. Whereas uncertainty is explicitly incorporated in the risk scoring process, we have not incorporated uncertainty regarding prey choice or habitat use. Incorporating multiple potential food web linkages and conducting sensitivity analyses on the structure of the food web for risk scoring could overcome this limitation.

The incorporation of uncertainty scoring directly into the qualitative risk scoring is an advance over qualitative risk scoring methodology as it allows a more complete characterization of risk. Providing the end users and managers with error bars representing uncertainty in the scoring process allows them to visualize the data gaps and assign resources to address them when possible. Uncertainty in the exposure variables can be more easily addressed with additional research or data sharing and allows refinement of resource allocation where overlapping human activities and stressors are a concern. This method of uncertainty incorporation also allows risk assessment in data poor situations and produces risk estimates that can undergo sensitivity analysis to test various scenarios. Use of these methods therefore supports the precautionary approach in ecosystem-based management.

In these types of semi-quantitative risk assessment methods, there is uncertainty and bias introduced by assigning scores to exposure and consequence factors. For example, by multiplying exposure and consequence, the overall risk score is influenced by the factor with higher overall scores. Furthermore, we squared the consequence term in order to balance the influence of these two factors. While this reduces one kind of bias, it inflates the non-linearity of the consequence term, biasing the influence of higher consequence scores on overall risk, a precautionary decision. However, the initial scores themselves are already non-linear, in the sense that—from the ecosystem component’s perspective, as measured by biomass lost, or other such metrics—a score of 3 is more than 50% worse than a score of 2, a score of 5 is more than 25% worse than a score of 4, etc. Further, having an unequal number of variables on either side of the equation affects uncertainty. There are three exposure variables with associated uncertainty and a single variable for consequence, which means that uncertainty can be inflated for exposure. Thus, careful interpretation of the differences in the risk values and the associated uncertainty is warranted.

The risk scores shown here should not be considered predictions of impact. Our risk scores do not account for system thresholds, by which non-linear dynamics can yield long-term impacts from short-term exposures (as but one example). Similarly, because sensitivity of the ecosystem component is evaluated at the scale of individual stressors, the method does not capture contexts in which a slow-maturing, low fecundity species might be vulnerable to population crashes in response to a combination of stressors [37]. Further advances in risk assessment could also account for sensitivity at the scale of each ecosystem component, not just at the scale of individual stressors facing each ecosystem component.

As in previous cumulative effects analyses, our method for estimating cumulative effects currently assumes risk to be additive, though the interaction of various stressors is commonly synergistic, antagonistic or multiplicative [38–40]. There is little known about the mechanisms behind interaction of stressors, and additional study is required to investigate the nature of these relationships using both ecological experimentation and modeling. Even with additional research, the results are likely to be specific to the study area, scale and species and may not be easily generalizable.

Incorporating bottom-up effects of food webs in our comprehensive cumulative risk framework is an advance on commonly used, direct risk frameworks. We did not consider top-down effects [41] or competitive interactions [17], [42–43], which also have the potential to alter risk to these species. Top-down effects are common in marine food webs and have been demonstrated as controlling the structure of marine communities. For example, the presence of sea otters, a higher level marine predator, alters the structure of kelp forest and seagrass ecosystems through their impact on prey populations [44–46]. Competitive interactions can mediate or alter the effects of stressors, either by direct impact on a competitive interdependent species or by altering the nature of the interaction between species [17]. Where information is available about the nature of inter-specific interactions and the vulnerability of each ecosystem component to these system-wide changes, this information could be used to modify the risk assessment framework to better account for the complexity of marine food webs. The complexity of marine ecosystems and their inherent variability across spatial and temporal scales makes estimating the cumulative risk from multiple stressors a difficult and challenging task. Risk assessments must make use of appropriate spatial and temporal scales in order to produce meaningful results and add additional layers of ecological complexity where data exists.

Conclusions

Risk assessments are a commonly used tool for prioritization of limited management resources and may be used to prioritize species for conservation action or human activities for mitigation of impacts. We showed that including indirect risk in comprehensive risk assessments may improve the evaluation of species at-risk, by incorporating risk mediated through trophic and habitat linkages. The comprehensive cumulative risk framework we propose is easily scalable and can be modified to include new information about top-down or bottom-up ecosystem effects. As this comprehensive risk method is a relatively simple modification of a commonly used risk framework, managers and decision-makers can easily incorporate it into ecosystem-based management, improving our understanding of risk from complex cumulative effects of human activities on marine resources and the ecosystem services they provide.