AOH overlap with feed production

gclawson1 commented 1 year ago

Coded up a script to overlap terrestrial area of habitat (AOH) with crop production for salmon feed.

Here is how production of soy protein concentrate (SPC) in our plant dominant diet using mass allocation overlaps with AOH for giant anteaters. So what this shows is the cells that have SPC production and are habitat areas for the giant anteater.

For reference, here is the total AOH for giant anteaters (so most of south america):

And here is what all of the SPC production (km2) looks like in South America:

Did a quick check, and the SPC crop area that overlaps giant anteater AOH represents ~35% of SPC production area globally, and ~54% of SPC production area in South America. Only ~0.02% of the giant anteaters AOH is affected though.

gclawson1 commented 1 year ago

Just realized the above is hectares, not km2. Fixing that now

gclawson1 commented 1 year ago

Just wanted to check another example, with the American black bear

Again, ~0.02% of the habitat is affected by SPC soybean production.

gclawson1 commented 1 year ago

Did some testing on how long this might take for all 16k species and I'm getting inconsistent results because all of the AOH rasters are different sizes. I'm seeing that it could take anywhere from 8 to 27 days to run, provided that I am parallelizing everything correctly using 14 cores... However, it doesn't take too much memory.

I'm gonna keep exploring and see if there is a better way of coding this/if I can speed it up at all.

cottrellr commented 1 year ago

Awesome - is that using app? Or some other apply feature?

Richard S. Cottrell Research Fellow in Aquaculture Sustainability Institute for Marine and Antarctic Studies College of Sciences and Engineering University of Tasmania

Theme Co-Lead, Sustainable Futures and Planetary Health Centre for Marine Socioecology University of Tasmania

Size Ecology Labhttps://www.sizeecology.org/ | Centre for Marine Socioecologyhttps://marinesocioecology.org/themes/sustainable-futures-and-planetary-health/ Google Scholarhttps://scholar.google.com/citations?user=1pLCMKIAAAAJ&hl=en | ORCIDhttps://orcid.org/my-orcid?orcid=0000-0002-6499-7503 | @RichCottrell22https://twitter.com/RichCottrell22

On 7 Jul 2023, at 7:06 am, Gage Clawson @.**@.>> wrote:

Did some testing on how long this might take for all 16k species and I'm getting inconsistent results because all of the AOH rasters are different sizes. I'm seeing that it could take anywhere from 8 to 27 days to run, provided that I am parallelizing everything correctly using 14 cores... However, it doesn't take too much memory.

I'm gonna keep exploring and see if there is a better way of coding this/if I can speed it up at all.

— Reply to this email directly, view it on GitHubhttps://github.com/Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping/issues/9#issuecomment-1624309762, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJK3YJEEDC6WSJ7OUQILW4DXO4SF5ANCNFSM6AAAAAAZ42EHVY. You are receiving this because you are subscribed to this thread.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.

gclawson1 commented 1 year ago

Currently using mclappy, but I'm gonna try out future_lapply based off of a chatgpt recommendation. Also think I might be able to speed it up a bit more by reordering my operations some

cottrellr commented 1 year ago

Cool - yeah future lapply works great for me but doing stuff inside the function with spatrasters is what gives the annoying error re ‘external pointer not valid'. But you might find a way round that. If not the app or lapp options with the "cores =“ could be worth a go...

Richard S. Cottrell Research Fellow in Aquaculture Sustainability Institute for Marine and Antarctic Studies College of Sciences and Engineering University of Tasmania

Theme Co-Lead, Sustainable Futures and Planetary Health Centre for Marine Socioecology University of Tasmania

Size Ecology Labhttps://www.sizeecology.org/ | Centre for Marine Socioecologyhttps://marinesocioecology.org/themes/sustainable-futures-and-planetary-health/ Google Scholarhttps://scholar.google.com/citations?user=1pLCMKIAAAAJ&hl=en | ORCIDhttps://orcid.org/my-orcid?orcid=0000-0002-6499-7503 | @RichCottrell22https://twitter.com/RichCottrell22

On 7 Jul 2023, at 1:12 pm, Gage Clawson @.**@.>> wrote:

Currently using mclappy, but I'm gonna try out future_lapply based off of a chatgpt recommendation. Also think I might be able to speed it up a bit more by reordering my operations some

— Reply to this email directly, view it on GitHubhttps://github.com/Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping/issues/9#issuecomment-1624588374, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJK3YJDEI4KUSFKHDZMACYDXO55DRANCNFSM6AAAAAAZ42EHVY. 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.

gclawson1 commented 1 year ago

Wasn't able to get future_lapply working for me, but I don't really need it. I reordered my code a bit and provided all else is equal, I estimate it'll take ~24 hours to complete! Way better than 8 days hahaha

cottrellr commented 1 year ago

Ha! Yeah 8 days would be killer. Great that it has come down - I’m interested to see your approach.

Richard S. Cottrell Research Fellow in Aquaculture Sustainability Institute for Marine and Antarctic Studies College of Sciences and Engineering University of Tasmania

Theme Co-Lead, Sustainable Futures and Planetary Health Centre for Marine Socioecology University of Tasmania

Size Ecology Labhttps://www.sizeecology.org/ | Centre for Marine Socioecologyhttps://marinesocioecology.org/themes/sustainable-futures-and-planetary-health/ Google Scholarhttps://scholar.google.com/citations?user=1pLCMKIAAAAJ&hl=en | ORCIDhttps://orcid.org/my-orcid?orcid=0000-0002-6499-7503 | @RichCottrell22https://twitter.com/RichCottrell22

On 8 Jul 2023, at 4:33 am, Gage Clawson @.**@.>> wrote:

Wasn't able to get future_lapply working for me, but I don't really need it. I reordered my code a bit and provided all else is equal, I estimate it'll take ~24 hours to complete! Way better than 8 days hahaha

— Reply to this email directly, view it on GitHubhttps://github.com/Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping/issues/9#issuecomment-1625847343, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJK3YJEEIY6U2DJQBU23IVLXPBJATANCNFSM6AAAAAAZ42EHVY. 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.

gclawson1 commented 1 year ago

Managed to revamp my code and get the overlap down to ~6 hours!

I'm gonna put this in some slides for Wednesday, but here are some results from a couple of species I chose because they have a lot of overlap:

this is just clipped to North and South America

This shows the overlap of SPC crop area and AOH for these species

This shows that of the total amount of SPC production area globally, ~65% overlaps with American Black Vulture Habitat, ~55% for Grassland Sparrow, etc.
Important to note that these species all have really large AOHs, so while a large percentage of the SPC area overlaps with their habitat, only a small percentage of their total AOH overlaps with SPC.

I've also implemented species-specific assessments of whether each species can survive and reproduce in agricultural land to calculate total area of habitat (AOH) affected for each species. To do this, I used IUCN data on if cropland is suitable, marginal, or not suitable habitat for each species, and if it was suitable, then 0 AOH affected, if marginal then multiply overlap AOH by 0.5, and if not suitable, then it is fully affected.

gclawson1 commented 1 year ago

A couple of more ways to visualize:

This shows all SPC production area with AOH underneath (black areas)

This shows areas where there is overlap between production and AOH (yellow and red areas) and AOH (black areas)

This is just AOH

gclawson1 commented 1 year ago

cottrellr commented 1 year ago

Managed to revamp my code and get the overlap down to ~6 hours!

I'm gonna put this in some slides for Wednesday, but here are some results from a couple of species I chose because they have a lot of overlap:

this is just clipped to North and South America

This shows the overlap of SPC crop area and AOH for these species

This shows that of the total amount of SPC production area globally, ~65% overlaps with American Black Vulture Habitat, ~55% for Grassland Sparrow, etc.

Important to note that these species all have really large AOHs, so while a large percentage of the SPC area overlaps with their habitat, only a small percentage of their total AOH overlaps with SPC.

I've also implemented species-specific assessments of whether each species can survive and reproduce in agricultural land to calculate total area of habitat (AOH) affected for each species. To do this, I used IUCN data on if cropland is suitable, marginal, or not suitable habitat for each species, and if it was suitable, then 0 AOH affected, if marginal then multiply overlap AOH by 0.5, and if not suitable, then it is fully affected.

This is great. I like your approach to sensitivity - is the 0, 0.5, 1 taken from somewhere else or something arbitrarily assigned for now?

cottrellr commented 1 year ago

Which species is this area of habitat for?

gclawson1 commented 1 year ago

The 0, 0.5, and 1 are from Dave Williams paper https://www.researchgate.net/publication/347537953_Proactive_conservation_to_prevent_habitat_losses_to_agricultural_expansion

we can adjust as we see fit though.

The bar chart is all ~16000 species combined

cottrellr commented 1 year ago

That's great. I think Dave's work is ideal to draw from. Great that you have this already - is this 16000 species globally? I think it would be awesome now to step into average AOH across the species with error bands, including disaggregating by taxa so knowing whether birds, mammals, reptiles etc are more affected on average. I'm sure this is already in your plans of course and not sure how much of a lift this would be for the meeting. But either way would be interesting to start disaggregating some of this total AOH data. Fantastic work, Gage.

gclawson1 commented 1 year ago

Yeah absolutely, I think that should be pretty easy. I have every thing laid out per species, ingredient, and allocation, so the information is all there!

cottrellr commented 1 year ago

Cooking with gas..awesome.

Richard S. Cottrell Research Fellow in Aquaculture Sustainability Institute for Marine and Antarctic Studies College of Sciences and Engineering University of Tasmania

Theme Co-Lead, Sustainable Futures and Planetary Health Centre for Marine Socioecology University of Tasmania

Size Ecology Labhttps://www.sizeecology.org/ | Centre for Marine Socioecologyhttps://marinesocioecology.org/themes/sustainable-futures-and-planetary-health/ Google Scholarhttps://scholar.google.com/citations?user=1pLCMKIAAAAJ&hl=en | ORCIDhttps://orcid.org/my-orcid?orcid=0000-0002-6499-7503 | @RichCottrell22https://twitter.com/RichCottrell22

On 8 Aug 2023, at 11:37 am, Gage Clawson @.**@.>> wrote:

Yeah absolutely, I think that should be pretty easy. I have every thing laid out per species, ingredient, and allocation, so the information is all there!

— Reply to this email directly, view it on GitHubhttps://github.com/Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping/issues/9#issuecomment-1668782977, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJK3YJHFAOMK7QTXVIEII4TXUGJ4XANCNFSM6AAAAAAZ42EHVY. 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.

gclawson1 commented 1 year ago

Recently realized that the habitat data is only for mammals and birds, so no amphibians/reptiles. I have emails out requesting data from these two papers for the amphibians and reptiles, so hopefully that will come through:

gclawson1 commented 1 year ago

Here is the total amount of habitat affected, per birds versus mammals

gclawson1 commented 1 year ago

Average AOH with error bars:

So in general we see birds are more heavily affected, but they also have much wider ranges and habitats. And also really big standard deviations overall..

And without the error bars:

gclawson1 commented 1 year ago

Fisheries overlap update

I've just finished an initial run of overlapping the FMFO fisheries disturbance with the aquamaps "suitable habitat" maps. So far, I've kept it pretty simple and done this:

Filtered the suitable habitat maps for each species for "Predicted (native) species distributions (HSPEC) " > 0.5 (per the aquamaps team suggestion). We could consider changing the probability cut off to be more conservative (i.e., a larger number than 0.5).
- "Includes predicted relative probabilities of environmental suitability based on species-specific computed envelope settings restricted to the native range of a species defined by available information (e.g. lat / lon bounding boxes). Probabilities < 0.5 often correspond to potential rather than realized niche and probabilities 0.5 tend to correspond to the core habitat of a given species. Excluding species with minimal available occurrence records (< 10 presence cells)." Example of great white shark >0.5 map:
- So here, we assume that anywhere >0.5 probability is the "species area of habitat".
Pulled species depth information from fishbase, and clipped to a bathymetric to constrain neritic and shallow-water species to areas no deeper than 200 m. This is what Casey O'Hara has done in the past and suggested doing here.
- ~half of the species in aquamaps data had depth information. For those that didn't have depth information, I just clipped the the bathymetry layer so as to cut out and land area.
  1. Overlapped the depth clipped species suitable habitat maps (>0.5) with each of our fisheries disturbance layers for FMFO forage and trimmings. This is essentially the same as we did for the crops. I guess this means I have just assumed that when a habitat cell overlaps with disturbance cell, the whole cell (and all species present) will be impacted.

So what do we see for prelim results?

This example is for the plant-dominant diet, mass allocation, and just fish meal from forage fish.

A total of 3783065 km2 of impacted habitat area. This is more than all of the crop overlap for birds and mammals combined for the same diet and allocation scenario (878014.0 km2). There are currently more fish species included (~23k for fish vs ~18k for birds/mammals; and we still don't have reptiles/amphibians included). Some species, like humpback whales and great white sharks basically span the entire world. However, this does seem like a very large number in comparison, especially considering this is just for fish meal from forage fish.
The average area of habitat affected for fish is ~159 km2, while for mammals it is ~3km2 and for birds it is 3.87km2.

Possible next steps

Further refine these methods so that we don't assume an entire cell and all species within it are affected.
- Possibly incorporate depth or gear information to determine what fish species are actually affected by the harvest. A couple of considerations if we do this:
  - Not all of the species in aquamaps have depth data. Only about half of them do. What do we do about this? Assign an average of a larger taxonomic class? Ignore it and chalk it up to uncertainty?
  - If we can match the FMFO disturbance to a gear type, do we then assign a depth parameter to each gear type? Then we assume that if a species AOH is within a depth that overlaps with a specific gear type and disturbance, then that species AOH is affected?
  - Could we incorporate some sort of proportion of AOH affected based on how much of the depth overlaps with the gear type depth?
Consider using a probability that is larger than 0.5. The aquamaps team recommends using at least 0.5, and through some conversations with Casey, he seems to think choosing and even higher limit would be better.
Incorporate economic allocation?

gclawson1 commented 1 year ago

And to clarify, once I filter for probability of suitable habitat > 0.5, I assume all the areas over 0.5 are where species are present (I.e. they get a value of 1).

Instead of assuming all areas >0.5 are indicators of species area of habitat (and giving them a value of 1), we could consider just using the probability values as a multiplier (I.e., assuming that if a cell has a probability value of 0.5, then 50% of that cell is area of habitat). I'm not sure if this is what the aquamaps team would recommend/is an appropriate method, but I could ask if that seems like a good idea. Doing this would likely decrease the AOH affected by quite a bit.

bshalpern commented 1 year ago

I think it makes most sense to convert all values above the threshold to 1, but agree it could be worth exploring a different threshold. I think aquamaps has often used 0.8 as the threshold. can't remember for sure...

cottrellr commented 1 year ago

This is all great stuff. A few thoughts because I agree, I think this is skewing to larger AOH estimates because of the diffuse nature of fisheries.

Did Gabriel say we should consider using 0.9 instead of 0.5 as a threshold? If so I think I lean that way because of the uncertainty about whether habitat is now "unsuitable" or not if fishing occurs in it. I think not having a high probability of occurrence in a cell amplifies this uncertainty even more. I actually really like your idea of scaling disturbance by the probability of occurrence. The problem I have with that though is how this approach differs to land. There will absolutely be uncertainty about habitat on land that we largely ignore. So perhaps just high confidence for occurrence in the ocean is the best approach for habitat distribution.
Something I changed for my paper, which I think is appropriate, is restricting gears to purse-seine, bottom trawl, and midwater trawl for the dedicated forage fisheries - and all gears for trimmings sourcing. To my mind, if a species is caught with a gear anywhere it is vulnerable. But is there something similar to unsuitable, marginal, and suitable habitat we could apply here as we have on land? I would have thought we could give it 0 if it is not caught in commercial fisheries at all, 0.5 if it is a bycatch species (we could classify as caught in our gears of interest by the country you are sourcing it from) or 1 if it is a target (target being within spp we filter for forage fisheries or those fisheries for trimmings )? Would IUCN have bycatch as a threat per species? I know marine spp data is poor.
I now have complete economic allocation data for the ingredients (marine and terrestrial) in the old allocation folder on Teams. I can transfer it across to the new folder and we can export from there. I think this is the way forward for attributing low impact to byproducts when it's appropriate (and to adjust for when they are the materials driving the demand).

gclawson1 commented 1 year ago

Chatted with Ben and Casey yesterday, in the past Casey has used Watson data to determine for each gear type, the species that are caught as bycatch by that particular gear. We could possibly create some weighting metric (like the unsuitable, marginal, and suitable weights for terrestrial) based on the amount of bycatch per species and gear. I'm going to start exploring this today to see if there is some sort of appropriate method based on Ben & Casey's work.

Casey once told me he chatted with Gabriel and Gabriel had indicated that 0.9 might actually be more appropriate. I'll look into other work done by/with the aquamaps team to see if they have used anything other than 0.5.

gclawson1 commented 1 year ago

Exploring using a different relative probability of occurrence

So far, I've mostly found examples of papers using 0.5 or 0.6 as the threshold value:

https://www.nature.com/articles/s41598-017-08758-5; Uses 0.5 threshold based on threshold used by other articles.
Paper by Aquamaps team: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0019653; States that 0.6 is the "best" probability to use based on species richness data.
- "For display purposes we used the RES threshold >0.6, associated with the second lowest AIC for all surveys combined at rarefaction basis n = 50 (i.e. expected species per 50 sightings). This threshold was associated with the lowest possible ES basis to ensure the widest possible geographic coverage, while at the same time consistently producing significant relationships and good model fits at the level of individual surveys (Table S1). We note, that the species richness maps based on the RES>0.6 threshold correspond to areas of overlap in highly suitable habitat across many species, but species may also occur in habitat predicted to be less suitable than the selected threshold."
Casey's paper: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0175739; Uses 0.5 threshold for comparability across assessments done by other researchers
https://www.nature.com/articles/srep17539: Uses 0.5
https://doi.org/10.1016/j.biocon.2018.02.021; Uses 0.6, see justification below
- "To take a more conservative approach with respect to assumptions about species occurrence, we produced risk maps based on binary (presence/absence) range maps using the core habitat, defined as species present in any cell with a species-specific predicted probability threshold of ≥0.60. AquaMaps maximum range extents for marine mammals show a very good correspondence with IUCN distribution both in terms of spatial alignment and overall size (O'Hara et al., 2017), but in some cases predicted maximum ranges correspond more closely to the potential or historical niche of a species. However, validation analyses showed the best fit between observed number of species and predicted species richness (indicating an increase of regular occurrence of species in areas of predicted suitability above the probability threshold of 0.60 (Kaschner et al., 2011). Therefore, species-specific risk maps based on predictions of core habitat correspond to locations where probability of exposure to the risk is likely highest."
https://doi.org/10.1016/j.biocon.2020.108536; Used > 0.00 to define species prescence
https://onlinelibrary.wiley.com/doi/full/10.1111/gcb.14616?saml_referrer; Used >0.5, see below for justification
- "We used a probability threshold for AquaMaps range maps of ≥0.5 to define species presence. A sensitivity analysis that was carried out with different thresholds showed that this threshold allowed us being conservative in regard to the total species range that overlaps hydrocarbon activities (Methods S1). This threshold value is also consistent with other global studies that assessed marine biodiversity using AquaMaps (Davies, Maxwell, Kaschner, Garilao, & Ban, 2017; Jones et al., 2018; Klein et al., 2015)."
https://onlinelibrary.wiley.com/doi/10.1111/gcb.16612; Uses 0.5
- "Estimated probabilities of occurrence were converted to presence/absence (1/0) information using a cut-off value of 0.5. This value is arbitrarily chosen but has only a minor impact on global biodiversity patterns, which affects ranges of cosmopolitan or generalist species rather than distribution patterns of specialist or endemic species (Jones & Cheung, 2015). Hereinafter, we use the term core-habitat distribution to refer to the species distribution comprising cells with >0.5 probability of occurrence values."

Based on all of this, 0.5 or 0.6 are the values the scientific community have used most often. If we want to be conservative, and have things to cite for justification, 0.6 is the way to go... unless we can find other instances using something larger.

gclawson1 commented 1 year ago

Another option to consider would be to use the bycatch vulnerability score for each species from the trait-based assessment paper: https://esajournals.onlinelibrary.wiley.com/doi/10.1002/ecs2.3919

We would just have the vulnerability score (0-1) for bycatch for every species. Then multiply the overlapping area by the vulnerability score?

cottrellr commented 1 year ago

Nice - 0.6 works then if that is a widely used threshold. Looks like there's plenty of justification for it.

I think the vulnerability score works really well as a sensitivity metric - are these gear specific?

Also should we consider trying to treat this the same as on land? Applying values of 0,0.5, and 1 by rounding or making some kind of decision about what we are going to classify as "marginal" for marine species using these vulnerability scores. The only reason I suggest this is to try and standardise the sensitivity metric. I know how these small differences scale and then start to create curious contrasts in the data. Of course this loses detail - @bshalpern perhaps you have some thoughts about that?

gclawson1 commented 1 year ago

The vulnerability scores are not split by gear type. I'll start coding this up for now, and make sure all of the species align. I agree, standardizing to 0, 0.5, and 1, so as to match the terrestrial methods seems appropriate.

To clarify for Ben, for terrestrial species we pulled habitat information from the IUCN, where they report if species can coexist in agricultural land, and they classify as unsuitable, marginal, and suitable. So we applied weights of 0, 0.5, and 1 (same methods as Dave Williams' biodiversity paper)

cottrellr commented 1 year ago

Ok yeah. I mean if they are targeted then they are vulnerable. If they are not target species we use the vulnerability for bycatch right? Is that what we’re saying?

Sent from my iPhone

On 23 Aug 2023, at 10:28 pm, Gage Clawson @.***> wrote:

The vulnerability scores are not split by gear type. I'll start coding this up for now, and make sure all of the species align. I agree, standardizing to 0, 0.5, and 1, so as to match the terrestrial methods seems appropriate.

To clarify for Ben, for terrestrial species we pulled habitat information from the IUCN, where they report if species can coexist in agricultural land, and they classify as unsuitable, marginal, and suitable. So we applied weights of 0, 0.5, and 1 (same methods as Dave Williams' biodiversity paper)

— Reply to this email directly, view it on GitHubhttps://github.com/Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping/issues/9#issuecomment-1689873113, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJK3YJFHB6KJE7LUYS4CXZ3XWXZMJANCNFSM6AAAAAAZ42EHVY. 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.

gclawson1 commented 1 year ago

Yeah exactly. So if Sardinella habitat is what we are intersecting our disturbance raster with, since they are a targeted FMFO species, they would get a vulnerability of 1 (most severe)

cottrellr commented 1 year ago

Ok cool so then perhaps we can assign 0.5 or 1 to bycatch species depending on their vulnerability. We could do it based on quartiles or percentiles? Like anything that is in the interquartile range gets a 0.5 or something like that? Under the premise that if a species is not targeted and is in the lowest quartile of vulnerability, it is basically suitable habitat because fisheries are dynamic anyway and it’s not like all biomass for a species is ever excluded really (Hilborn’s gripe).

Sent from my iPhone

On 23 Aug 2023, at 10:49 pm, Gage Clawson @.***> wrote:

Yeah exactly. So if Sardinella habitat is what we are intersecting our disturbance raster with, since they are a targeted FMFO species, they would get a vulnerability of 1 (most severe)

— Reply to this email directly, view it on GitHubhttps://github.com/Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping/issues/9#issuecomment-1689904476, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJK3YJFCNNPBEZLQQ6VLIMLXWX36JANCNFSM6AAAAAAZ42EHVY. 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.

bshalpern commented 1 year ago

this mostly makes a lot of sense. The one thing to also consider is whether to include life history traits in the vulnerability score too. We have that info in Butt et al. The idea being that fast reproducing (e.g., sardines) species that are targeted would be much less vulnerable that slow reproducing (e.g., bluefiin tuna) species that are targeted...

gclawson1 commented 1 year ago

I've recalculated using probability of aquamaps >0.6, and present a couple of different ways.

Regular AOH overlap (no vulnerability multiplier).
Multiplying the AOH overlap by the bycatch vulnerability multipliers presented in Butt et al. 2022.
Multiplying the AOH overlay by the bycatch vulnerability multipliers presented in Butt et al. 2022, where any vulnerability < 0.25 receives vulnerability of 0 ("suitable habitat"), any vulnerability >=0.25 & <= 0.75 gets vulnerability of 0.5 ("marginal habitat") and any vulnerability >0.75 gets a vulnerability of 1 ("unsuitable habitat).

Results for forage fish-fish meal plant dominant diet under mass allocation:

test <- overlay_vuln %>%
  filter(diet == "plant-dominant",
         allocation == "mass", 
         ingredient == "forage fish_fish meal")

### 1 
sum(test$aoh_overlap, na.rm = TRUE) # 3342859 km2... that's a lot compared to crops! But there are more species, with larger AOH ranges, and more dispersed harvesting area
mean(test$aoh_overlap, na.rm = TRUE) # 141.359 km2 mean

### 2 
sum(test$aoh_overlap_vuln, na.rm = TRUE) # 1297965 km2... A lot less now! 
mean(test$aoh_overlap_vuln, na.rm = TRUE) # 54.88688 km2 mean

### 3 
sum(test$aoh_overlap_vuln_quartile, na.rm = TRUE) # 1611213... a bit more than using regular vulnerability values. 
mean(test$aoh_overlap_vuln_quartile, na.rm = TRUE) # 68.13314 km2 mean

gclawson1 commented 1 year ago

Going to gapfill the depth values since we are missing about half of species in this order: genus, family, order, phylum, kingdom

gclawson1 commented 1 year ago

Exploratory plots with marine ingredients and species

Using the method described above, here are what the new plots look like. These are really prelim and rough, but a good way to start exploring the data. The marine species and raw materials/ingredients really wash out all of the terrestrial stuff, particularly marine mammals because their ranges are so large...

Some notes:

For the reptiles and amphibians, these are the just the species included in aquamaps, so they are marine.
For marine microorganisms, I am assuming that they have no habitat affected, as there are not values for their vulnerability to bycatch, and I'm not sure how I could interpolate for those species.
Invertebrates are all marine

Total AOH Affected across all species + ingredients/raw materials

Total AOH Affected across all species + ingredients/raw materials split by species type

Average AOH Affected across all species + ingredients/raw materials split by species type

Average AOH across all species split by species type

gclawson1 commented 1 year ago

Can also split/aggregate these species groups however we see fit (e.g., combining fish and invertebrates, leaving out microorganisms, etc)

cottrellr commented 1 year ago

I think perhaps we need to dig into and discuss a little more how we come up with the sensitivities for marine species again. Perhaps we could do this in our meeting with us all. It is almost counterintuitive that species with the largest ranges have the greatest impact and part of that is how we are assigning sensitivity to fishing activities.

Just thinking - we assign 1 on land to mean a species is displaced from that area, 0.5 if it can survive there but it's not ideal. And 0 if it can survive there no problem. For many harvested marine species, if they are targeted specifically by fishing then they are never really entirely displaced. Fisheries management depends on this - so I guess assigning 1 to targeted marine species is probably not reflective. We should think of a way that address the fact that harvesting (at least as a static assessment) rarely fully displaces species but depletes them (kind of comparable to marginal).

cottrellr commented 1 year ago

It's almost like the equivalent should be 0, 0.25, and 0.5 for example but we need to think of a robust way of establishing these thresholds

cottrellr commented 1 year ago

Exploratory plots with marine ingredients and species

Using the method described above, here are what the new plots look like. These are really prelim and rough, but a good way to start exploring the data. The marine species and raw materials/ingredients really wash out all of the terrestrial stuff, particularly marine mammals because their ranges are so large...

Some notes:

For the reptiles and amphibians, these are the just the species included in aquamaps, so they are marine.

For marine microorganisms, I am assuming that they have no habitat affected, as there are not values for their vulnerability to bycatch, and I'm not sure how I could interpolate for those species.

Invertebrates are all marine

Total AOH Affected across all species + ingredients/raw materials

Total AOH Affected across all species + ingredients/raw materials split by species type

Average AOH Affected across all species + ingredients/raw materials split by species type

Average AOH across all species split by species type

I did also skip over the fact that this is awesome work Gage! My suggestions are important details , but it's so great to see draft results coming in!

gclawson1 commented 1 year ago

Thanks!

Yeah, I agree with the sentiment that we are definitely over-estimating these sensitivities and how species are affected by fishing. I think that targeted species in particular it makes sense to have a high value (given that they literally die). Other species, like marine mammals, are receiving high values (1, e.g. unsuitable) because their sensitivity values to bycatch are so high. I don't think it makes sense for a targeted species (like sardines for FOFM) to have the same sensitivity value as an un-targeted species (like a whale), which is currently what is happening.

And yeah, based on your logic about a species never fully being displaced by fishing, using the 0, 0.5, and 1 values doesn't entirely match up. I'll think on this some and look into other similar assessments and how the IUCN comes up with their "unsuitable, marginal, and suitable" habitat values for terrestrial species, and we can discuss in our coming meeting(s).

I also think that the aquamaps data just shows reallllly large ranges for things and is way more dispersed than the terrestrial AOH maps we have. I know I mentioned above that using 0.6 as the probability threshold was mostly what other researchers have done, but I wonder if we could justify using an even larger value, so as to be more comparable to the terrestrial data?

Sustainable-Aquafeeds-Project / feed_biodiv_impact_mapping