Exploring marine species vulnerabilities

[gclawson1]

As I start exploring options for accounting for marine species vulnerabilities, and how to better reflect the realities of habitat loss due to fishing, here are some definitions and initial thoughts.

For terrestrial species, we are using the same method as Williams et al. to apply species tolerances. So when a species has a tolerance of "suitable" for a habitat, then there is no area of habitat affected, when a species has a tolerance of "marginal", we apply a weight of 0.5 to the overlap of AOH and aquafeed production, and when the habitat is "unsuitable", the AOH is fully affected:

From Williams et al.

• "We used species-specific assessments of whether each species can survive and reproduce in agricultural land to calculate changes in total area of habitat (AOH) for each species”
• "We used the IUCN classification to define species as able to survive in agricultural land or not. We defined IUCN habitat classes 14.1 and 14.3 (“Artificial – Terrestrial – Arable land”; “Artificial – Terrestrial – Plantations”) as “cropland” and 14.2 (“Artificial/Terrestrial - Pastureland”) as pastureland. For each species, we noted if each habitat was “suitable” or “marginal” and took the maximum value of all habitats for both cropland and pastureland. i.e. if a species has 14.1 as “marginal” and 14.3 as “suitable” then we defined cropland as “suitable” for the species."

Additionally, from the IUCN:

• Suitable - the species occurs in the habitat regularly or frequently.
• Marginal - The species occurs in the habitat only irregularly or infrequently, or only a small proportion of individuals are found in the habitat.
  • Unknown - The habitat is of unknown importance to the species.

We want to create a similar metric for marine species, however, this is proving challenging, as marine species respond completely different to fishing than terrestrial species do to agricultural land. For example, if a terrestrial species habitat is unsuitable for agriculture, then it is reasonable to assume that their species is fully displaced, losing that habitat. However, with marine species, their suitable habitat can be fished, and the fish species will likely return, as the habitat isn't necessarily "unsuitable" for them, it is just stressed.

Our initial thought was to use the vulnerability scores from Butt et al for bycatch, and rescale the vulnerabilities to be 0, 0.5, and 1, so as to match our methods for the terrestrial side of things. However, doing this treats many species as fully displaced by fishing for aquafeeds, which isn't necessarily correct, as noted above. Another thought was to scale the vulnerabilities down to 0, 0.25, and 0.5, however, I believe this still vastly overestimates the impact of fishing on marine mammal species. We see really large values for AOH affected because many fish and marine mammal species have much larger ranges than terrestrial species (perhaps an artifact of there being more ocean area). Because of this, we are likely vastly overestimating the impact of aquafeed fishing on marine species.

To clarify, vulnerability in Butt et al. is defined as: "We define a species' vulnerability to a stressor as a function of its sensitivity (the degree to which it is affected by a stressor), and adaptive capacity (ability to adapt to or recover from a stressor)"

Upon further review, there are other options for vulnerability scores we could explore, as well as refine the bycatch to better reflect AOH displacement from aquafeed fishing. Other vulnerability values of interest include biomass removal, entanglement, and bycatch. Perhaps we could do something like this:

• Combination of entanglement and bycatch vulnerability values for marine mammals and birds?
• Biomass removal vulnerability value for targeted species (forage fish and species used for trimmings)
• Bycatch vulnerability value for all other species, with a caveat (See below)

For the bycatch vulnerability, I think we could account for gear type in some way. We could look at the gear types used to catch forage and trimmings fish species, and see the species that are caught as bycatch for those gear types. If a species is caught as bycatch from the gear types used, then they would receive the vulnerability value for bycatch, otherwise, they are not vulnerable (E.g., no habitat loss, as it isn’t a targeted species, and it isn’t associated as bycatch with any of the gear types used). We could also pare this down even more using depth values for species. For example, if a species cannot be found in any of the spatial or depth zones typically associated with the gear types used for forage fish and trimmings fish used in FMFO, the vulnerability will be categorized as 0.

We can also further rescale to 0, 0.25, and 0.5 as before, I will just need to come up with a scientific justification for this.

[gclawson1]

Had a very productive chat with Ben + team today regarding this. Here are some notes from that:

Problem: Marine species “AOH affected” is way higher than that of terrestrial, because we are treating it the same (e.g., full displacement based on vulnerability values). How do we make this more realistic?

• Because forage fish gear types are highly selective, there probably isn’t much bycatch (meaning that most of the aquamaps species probably are not actually harmed). We can look at the Watson data to see for each gear type of targeted forage species, how much (%) of the catch is bycatch. It is probably very low. If it is less than some %, then we can assume that other species are not affected by harvest, or weight by that %.

• Watson data includes discard estimates - "Discard catch rates that were estimated were not expected to be the same taxon as the reported landed taxon for each record, but rather would be comprised of a variety of taxa, targeted and non-targeted, many of which would not be represented in the original landings source data." Through this, discards in any cell can be summed up across all targeted species, countries, and gear type.

  • Can do the same thing for trimmings species, although given that we are using economic allocation now, and that there is much less disturbance, this probably won’t be much of an issue anyways.

• Before all of that, I need to determine if species are exposed. To do this incorporate depth. Casey has code to split pressure and species maps into depth ranges.

  • Separate into pelagic, benthic, benthopelagic, and reef, for disturbance metric and for species AOH maps. If they overlap, then they are exposed, if not, they are not (e.g., no AOH overlap).

• Possible source for rescaling vulnerabilities to 0.5: Halpern et al. 2022 (food paper) has all biomass removal scaled to 0.5. Can reference this and response to Hilborn.

I thought about all of this, and I think a path forward could be to do something along these lines:

• ~19% of forage catch and associated gear types have bycatch proportions > 0.1 (range from 0.1 to 0.38), the rest have values <0.03 of catch within the targeted forage fish species are bycatch. Generally, bycatch is very very low for the targeted forage fish species, so it is unlikely that un-targeted species will be affected.
  • For example, of the Atlantic horse mackerel catch that is caught by bottom trawls, ~37% of it is bycatch. And of all the forage fish catch, atlantic horse mackerel from bottom trawling only represents ~1%.

• create disturbance rasters for each forage fish species and depth range based on gear types (binned into ranges pelagic, benthic, benthopelagic, and reef)
• calculate AOH exposed based on if aquamaps species are exposed in the depth ranges (if there is no depth overlap, then they are not exposed, and no habitat is affected).
  • note to self: Will need to save columns for each gear type and targeted forage species
• calculate % of each targeted forage fish species and gear type catch that is bycatch (discards).
• Multiply these percentages by the exposed aquamaps species, gear type, and targeted forage species combinations disturbances (e.g., those that have depths that overlap with gear types used for forage fish catch). This will effectively reduce the AOH exposed by anywhere from 97% to 62% for a species and gear type.
• sum across each disturbance value (km2 exposed) to get one value for each aquamaps species
• Then multiply those new AOH overlap values by the bycatch vulnerability values, rescaled to 0, 0.25, and .5 (further reducing the AOH affected value; justified by Halpern et al. methods)

Ideally we could tell if each aquamaps species were caught as bycatch by each gear type, however, we don't have that information. Watson and sea around us only tell you how much of a targeted species catch is bycatch (without info on what species the bycatch is).

Regardless, I need to incorporate the depth exclusion for the exposure, so I'll get chugging away on that before I move forward with anything else!

[gclawson1]

Exposure methods:

• Species AOH maps binned into depth ranges:
  • Benthic, bentho-pelagic, pelagic, and reef
• Disturbance rasters binned into:
  • Depth ranges; Benthic, bentho-pelagic, pelagic, and reef
  • Allocation type; mass, economic, energetic
  • Diet type; plant-dominant, fish-dominant
  • Ingredient type; fish oil or fish meal
  • Targeted fish type; forage or trimmings species

Overlap those two rasters to get an exposure raster of km2 of each species per each disturbance type (disturbance type being the combination of all the categories listed above).

Vulnerability methods:

Per each depth range (benthic, bentho-pelagic, pelagic, and reef), calculate, based on gear types (which gear types affect which depth range is determined from O'Hara et al. 2023 (in prep)), the % of the catch of these species that is bycatch, globally. Do this for both forage species and trimmings species as a whole. For example, we will have:

• % bycatch for forage species in the pelagic range
• % bycatch for forage species in the benthic range
• % bycatch for forage species in reef
• % bycatch for forage species in bentho-pelagic
• % bycatch for trimmings species in the pelagic range
• % bycatch for trimmings species in the benthic range
• % bycatch for trimmings species in reef
• % bycatch for trimmings species in bentho-pelagic

For each species exposure metric (calculated above), multiply by the appropriate depth range bycatch % to see the amount (km2) of habitat that is exposed AND sensitive to FMFO harvest for both forage and trimmings species. This will decrease AOH exposed anwhere from ~70% to ~99.9% (for forage fish, haven't looked at trimmings fish yet), given that the majority of forage fisheries have low-bycatch gears (highly targeted). I suppose we could make this even more refined by using percentages based on specific targeted species (anchovies, sardines, etc) and gear types (bottom trawl, mid water trawl), etc, but I'm not sure that the numbers will actually change much, given this is a global scale and it is all proportional. Additionally, Casey is taking the same approach, and his paper using this method is in revision now (soon to be published).

I think that is likely sufficient, but if we wanted to pare it down even more, we could then multiply by the bycatch vulnerability values provided in Butt et al. 2022.

[gclawson1]

All examples below are based on forage fish, fish meal, mass allocation, and plant-dominant diet

After incorporation of the depth exposure constraints (whether a species is exposed depends on if a species falls into a depth range or not), the AOH overlap is cut by ~70%!

• This is likely because previously we were counting all benthic organisms (molluscs, etc), as being exposed to pelagic fishing, and vice versa for other fish types. Now that is fixed.
• Previous AOH overlap was ~3.3 million km2, now it is down to ~900000 km2.

I have calculated the AOH affected in a number of ways, which we can discuss in our meeting tomorrow. A quick overview:

Method 1:

• Multiply the AOH overlap by the vulnerability values from Butt et al. 2022 rescaled to 0, 0.5, and 1. I think this is assuming full displacement, which probably isn't what we want.
• Total AOH affected = 450772.2 km2
• Mean AOH affected = 9.5 km2

Method 2:

• Multiply the AOH overlap by the vulnerability values from Butt et al. 2022 rescaled to 0, 0.25, and 0.5, based on quartiles. This does not assume full displacement. Ben indicated we could cite the food footprint paper fisheries disturbance metric for rescaling to 0.5, as in that paper we assumed that biomass removal accounted for 0.5 of the disturbance?
• Total AOH affected = 225386.1 km2
• Mean AOH affected = 4.769975 km2

Method 3:

• Multiply the AOH overlap by depth associated vulnerabilities calculated as the proportion of discards (bycatch) catch for forage fish or trimmings fish species.
  • For example, gear types that affect the pelagic zone have ~2.6% bycatch (according to Watson data). We multiply 0.026 by the AOH overlap value for a pelagic species affected by forage fish disturbance.
• Total AOH affected = 64466.13 km2
• Mean AOH affected = 1.364334 km2

Method 4:

• Combination of methods 1 and 3. Multiply AOH overlap by vulnerability values (0, 0.5, 1) and the depth vulnerability values.
• Total AOH affected = 27780.4 km2
• Mean AOH affected = 0.5879324 km2

Method 5:

• Combination of methods 2 and 3. Multiply AOH overlap by vulnerability values (0, 0.25, 0.5) and the depth vulnerability values.
  • Total AOH affected = 13890.2 km2
  • Mean AOH affected = 0.2939662 km2

Method 5 is the most conservative option, and I think probably makes the most sense. However, I could see an argument for Method 2 if we don't feel comfortable with the "depth vulnerability" values I have calculated, as they do decrease the AOH affected values by either ~2% or 30%, and it is citeable (e.g., using established vulnerability values from Butt et al.).

I'll work on putting together exploratory plots using method 5, so as to compare to terrestrial ingredients now!

[gclawson1]

Updated plots using Method 5 for vulnerabilities; using both Butt et al. 2022 rescaled to 0, 0.25, and 0.5, and proportion of catch within a depth (pelagic or benthic) that is bycatch of our targeted species:

With this approach, even in the fish dominant scenarios, the terrestrial impacts are much higher than marine impacts, when looking at total km2. We see more interesting variations when looking at average impacts.

• There are birds in the Aquamaps data, however, their AOH affected is so low that it is hard to see on these plots.

For comparisons sake, here are the same plots, using Method 2 outlined above (just using the vulnerability values from Butt et al. 2022 rescaled to 0, 0.25, and 0.5 for marine spp):

• Based on this, the total amounts even out more when comparing marine and terrestrial ingredients using this method. However, when looking at average amounts, we still see a problem with marine mammals impacts being too large, so perhaps Method 2 is not the way to go.

[gclawson1]

The same plots, using method 5 for marine spp vulnerabilities, and only economic allocation (this is just easier to parse out differences)

[cottrellr]

Rather than using the discards data, I think I lean towards:

• Alter exposure with depth as you have done
• Apply vulnerability data rescaled to 0.5 (or some other maximum that isn't 1) based on some justification from growth curves or the relative biomass status at BMSY averaged across taxa or something like this. Specifically targeted species getting the max and non-target species getting their rescaled vulnerability score. This would assume that 1) specifically targeted spp have the greatest "displacement" compared to those incidentally caught 2) even when the displacement is at it's maximum - this leads to a drop of X% in the biomass of the species. We know from the impacts of overfishing that this a highly conservative figure and with some need for discussion but I think one that is necessary to make and fine if we can robustly justify.
• Along with the total, calculate the proportional habitat loss for each species, and then the average proportion can be taken. Given the huge variation in the size of the ranges this is probably more meaningful. A humpback whale can lose much more habitat than many frogs in absolute terms before it becomes a persistence problem for example, although the same proportional loss could be worse (insert caveats that not all habitat is equal for each species with certain portions being more important for hunting or reproducing than others).

[gclawson1]

Thought about this more this morning, and chatted with Ben and co, and realized that I believe I need to change my order of operations/methods a bit.

Currently, I am overlapping each targeted and non-targeted species by the FMFO catch for forage or trimmings (reported + IUU catch). Then to get at the amount of that which is affects non-targeted species (e.g., the bycatch amount) we multiply by the prop discards, which in reality is more of an exposure metric, rather than a vulnerability metric. We want to know how much km2 exposure there is for non-targeted species, and they are theoretically only exposed if they are bycatch. So, we need bycatch rasters.

It occurred to me that we actually have the amount of bycatch tonnage associated with FMFO catch in the Watson data. The Watson data has data for reported, IUU, and discards. We use the reported + IUU for the catch. I could calculate a proportion per species per cell that is discards, based on that information. Then I could create catch rasters (what I already have; reported + IUU) and bycatch rasters (discards) associated with forage species and trimmings species. Then, for the exposure (km2 overlap) we intersect the targeted species with the full catch rasters (reported + IUU), and the non-targeted species with the bycatch rasters (discards).

Then, once we have that overlap, we multiply by a vulnerability value (biomass removal for targeted species, bycatch vuln for non-targeted), rescaled to 0, 0.25, and 0.5, via some justification TBD (like above).

This essentially gets at what I wanted to do before, but it much more resolute. Now, we will be able to tell how much of FMFO catch in a cell is bycatch (non-species specific, just a lump sum) vs how much is actual catch used for FMFO. Does that make sense?

For Casey's latest paper, he is using bycatch rasters derived from the discards data in the Watson data, and overlapping with species ranges.

Additionally, had a really productive conservation with Ben and co today, and they gave some great feedback and ideas on how to better present the data!

[cottrellr]

This all sounds really good regarding these methods if you can use it to get at how bycatch is distributed across species. Quick thought is that I would leave out IUU. Th main reason is that for the main feed suppliers we are interested in commercial operations. It’s plausible there maybe some IUU in there that they are unaware of in reality but that won’t be something that passes the filter with Biomar when it comes to publication. They are pretty certain about the fisheries they get ingredients from. The second reason is chatting to Reg multiple times about this a few years back, he used to caution about putting too much stock in the IUU estimates. Reported is sufficient for your framework I think.

Sent from my iPhone

On 19 Sep 2023, at 4:24 am, Gage Clawson @.***> wrote:



Thought about this more this morning, and chatted with Ben and co, and realized that I believe I need to change my order of operations/methods a bit.

Currently, I am overlapping each targeted and non-targeted species by the FMFO catch for forage or trimmings (reported + IUU catch). Then to get at the amount of that which is affects non-targeted species (e.g., the bycatch amount) we multiply by the prop discards, which in reality is more of an exposure metric, rather than a vulnerability metric.

It occurred to me that we actually have the amount of bycatch tonnage associated with FMFO catch in the Watson data. I could create catch rasters (what I already have; reported + IUU) and bycatch rasters (discards) associated with each targeted species. Then, for the exposure (km2 overlap) we intersect the targeted species with the full catch rasters (reported + IUU), and the non-targeted species with the bycatch rasters (discards).

Then, once we have that overlap, we multiply by a vulnerability value (biomass removal for targeted species, bycatch vuln for non-targeted), rescaled to 0, 0.25, and 0.5, via some justification TBD (like above).

This essentially gets at what I wanted to do before, but it much more resolute. Now, we will be able to tell how much of FMFO catch in a cell is bycatch (non-species specific, just a lump sum) vs how much is actual catch used for FMFO.

For Casey's latest paper, he is using bycatch rasters derived from the discards data in the Watson data, and overlapping with species ranges.

Additionally, had a really productive conservation with Ben and co today, and they gave some great feedback and ideas on how to better present the data!

— Reply to this email directly, view it on GitHubhttps://github.com/Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping/issues/12#issuecomment-1724148817, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJK3YJCXMD5CZMOXOHHCGV3X3CGWTANCNFSM6AAAAAA4H32K6Q. You are receiving this because you commented.Message ID: @.***>

This email is confidential, and is for the intended recipient only. Access, disclosure, copying, distribution, or reliance on any of it by anyone outside the intended recipient organisation is prohibited and may be a criminal offence. Please delete if obtained in error and email confirmation to the sender. The views expressed in this email are not necessarily the views of the University of Tasmania, unless clearly intended otherwise.