Repository for code and generated data for "At-risk marine biodiversity faces extensive, expanding, and intensifying human impacts": O’Hara, C. C., Frazier, M., & Halpern, B. S. (2021), Science, 372(6537), 84–87. https://doi.org/10.1126/science.abe6731
My first pass on this was to simply use the IUCN "impact score" as a weighting factor on stressor layers to determine cumulative impact on a species. However, "impact score" is more of an aggregation of several factors (rate of decline, scope of population affected, timing of stressor) and I think it is better considered more in line with a total impact of that stressor on the population, rather than a sensitivity that can be used to determine the impact of a stressor on a species in any given cell.
CHI 2008 (and presumably all since) consider sensitivity like so:
so if I consider the IUCN impact score as the Ic then I can calculate species sensitivity μ to stressor D as Ic divided by the summed exposure to the stressor. So the cumulative impact of that stressor summed across the total species range is simply the IUCN impact score.
One consequence of this is that, for a given stressor and two species with equal impact Ic but different range sizes, the larger-ranged species will necessarily appear to have a lower sensitivity. That might be OK? But the range of sensitivities is rather large. I considered a range-adjusted version (effectively using mean exposure per cell rather than total exposure) which removed the effect of range size on sensitivity, but those resulting sensitivity numbers looked pretty absurd.
Thoughts on this process?
impacted range
I am also considering different "clips" of the stressors to identify "impacted range" (for doing an analysis similar to Allan et al 2019) - similar to something we're doing for the East Coast EIMA project - where I can identify the smallest area that contains (for example) 95% of the total exposure, a bit different than just clipping to a 99.99%ile. E.g. a highly spatially heterogeneous/highly skewed stressor might have 95% of its cumulative stress acting in a small area, with the remainder of the area exposed to very tiny amounts, potentially severely skewing the impression of how much range is meaningfully impacted. Because many of these stressors are so skewed I think this would I think give a fairer impression of "impacted area" than just counting any non-zero value.
calculating stressor weights
My first pass on this was to simply use the IUCN "impact score" as a weighting factor on stressor layers to determine cumulative impact on a species. However, "impact score" is more of an aggregation of several factors (rate of decline, scope of population affected, timing of stressor) and I think it is better considered more in line with a total impact of that stressor on the population, rather than a sensitivity that can be used to determine the impact of a stressor on a species in any given cell.
CHI 2008 (and presumably all since) consider sensitivity like so:
so if I consider the IUCN impact score as the Ic then I can calculate species sensitivity μ to stressor D as Ic divided by the summed exposure to the stressor. So the cumulative impact of that stressor summed across the total species range is simply the IUCN impact score.
One consequence of this is that, for a given stressor and two species with equal impact Ic but different range sizes, the larger-ranged species will necessarily appear to have a lower sensitivity. That might be OK? But the range of sensitivities is rather large. I considered a range-adjusted version (effectively using mean exposure per cell rather than total exposure) which removed the effect of range size on sensitivity, but those resulting sensitivity numbers looked pretty absurd.
Thoughts on this process?
impacted range
I am also considering different "clips" of the stressors to identify "impacted range" (for doing an analysis similar to Allan et al 2019) - similar to something we're doing for the East Coast EIMA project - where I can identify the smallest area that contains (for example) 95% of the total exposure, a bit different than just clipping to a 99.99%ile. E.g. a highly spatially heterogeneous/highly skewed stressor might have 95% of its cumulative stress acting in a small area, with the remainder of the area exposed to very tiny amounts, potentially severely skewing the impression of how much range is meaningfully impacted. Because many of these stressors are so skewed I think this would I think give a fairer impression of "impacted area" than just counting any non-zero value.