Closed gclawson1 closed 5 months ago
Also, some of the blue in Canada and the US could be due to wheat impacts. Some of the blue in Brazil and USA could be corn.
good sleuthing, and I was wondering about these results. I need to process more the question about if/whether to change the methods. Do you have an intuition of which path makes sense? I do worry that the results as they are will be a red flag for reviewer critique...
I think we have some options:
Keep it the way I currently have it; the "average proportion of habitat impacted" method
The "average total proportion of habitat impacted" method
No averaging at all, just use sums and division; the "total proportion of habitat impacted" method
I would be partial to options 1 or 2, as there is precedent for taking mean percentage change in habitat area within each cell. Williams et al., 2020 reports their results as a mean per taxon. They also present a "change in total habitat (mean habitat loss in a cell multiplied by the number of species present)."
The only concern I have about the averaging options is that I don't think changing to option 2 will solve the problem I've described in previous comments. Because for the plant-dominant scenario, cells where there are only impacts from SPC and soybean oil, everything is equal (nspp exposed and impacted, where pressures are located) aside from the amount of pressure, meaning it would just be a regular average. I think we'd still see that the soybean oil is bringing down the values in those cells, making impacts less than the fish-dominant scenario (which only has SBM impacts).
My feeling is because of averaging across ingredients for the same raw material - as you point out in your bullet above this is maybe what is causing the distortion. We are implicitly saying that SPC and soy oil can come from the same unit of production in a given map so averaging impacts across these ingredients doesn't quite make sense. Because we are looking at the production of soybeans needed that creates the impact. And given we are not dealing with processing disturbance pressures at all (with good reason as our other paper shows), maybe raw material is the focus we need. Soybeans from Brazil or US or Australia is what we are saying creates the impacts not SPC from Brazil, US or Australia. So yeah, I think summing all soybeans or wheat km2 for a given diet in a given cell is what matters and then the species weighting can happen. With the species held the same perhaps this would rectify the issue? Unless I'm missing something.
Yeah that makes sense, and I think that would work to fix the issue. I think I just need to add/change a step in where we aggregate impacts to the raw material, rather than average. Something like this:
the "average proportion of habitat impacted" method
Cool. If you retained the proportion of raw material that each ingredient represent in terms of km2 (i.e., have a column that has raw_material_prop e.g., based on your plant dominant area estimates SPC might get at prop of ~0.97 and soy oil ~0.03) you could still keep impacts later on to the ingredient level if you wished….
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=X1a9t90AAAAJ&hl=en&authuser=1 | ORCIDhttps://orcid.org/my-orcid?orcid=0000-0002-6499-7503 | @RichCottrell22https://twitter.com/RichCottrell22
From: Gage Clawson @.> Date: Thursday, 18 April 2024 at 10:54 AM To: Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping @.> Cc: Richard Cottrell @.>, Comment @.> Subject: Re: [Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping] Exploring fish-dominant higher impact in soybean dominated areas (Issue #18)
Yeah that makes sense, and I think that would work to fix the issue. I think I just need to add/change a step in where we aggregate impacts to the raw material, rather than average. Something like this:
the "average proportion of habitat impacted" method
— Reply to this email directly, view it on GitHubhttps://github.com/Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping/issues/18#issuecomment-2062795403, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJK3YJG75HCVQB5EHT7EGTDY54KTJAVCNFSM6AAAAABGKPMTDKVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDANRSG44TKNBQGM. 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.
@cottrellr based on our conversation a few mins ago, is this what you are suggesting to show for that global map?
If so, where does the species weighting come in?
So I think for each cell summing across all materials but per species. For each species divide by total km2 in cell. Then a mean for each cell. Sorry not species weighted then, that would just be a mean.
Hmmm ok.. I need to think about how to accomplish this.
So I would end up with a raster describing this for example?:
Then from there I could take a nspp weighted mean for each cell to get across all taxa?
I'm not sure that I can do it (computationally) without chunking it into taxa first because there are so many species
Oh I was thinking not by taxa but all species. Yeah but I see that each cell and each species gets very big very quickly. Is there a way to chunk the analysis but save into lists or folders? That way all layers can be pulled back in and the mean taken by multicore terra function like lapp or app (never remember which one it is).
I'm honestly not sure how I could make that work... I already chunk it so much (by taxa and per every 500k cells).
I'm just looking at amphibians right now for example, and for soybeans alone if I tried to save it per species it would be a list of 14 lists of dataframes, and the max number of rows it could be in each of those data frames is ~1840 species * 27000 cells impacted = ~49 million rows
Then from there extrapolate out to the 14 dataframes with 49 million rows* 14 data frames = ~700 million rows I would need to save
And extrapolate that out from soybeans and amphibians to all of the other raw materials (even if we just sum the raw materials to a total pressure map) and species, there would be a lot more cells to add to that... especially considering how dispersed the fishing pressures are, so the marine side of things would be an entirely different beast.
^^ That math isn't exactly correct/not sure if that even makes sense... but just trying to illustrate the mess
Does what you propose above address the issue? i.e.,:
Maybe weighted by taxa could work nicely?
Yeah I think it would; there's just not averaging involved there. That would be showing a map of total km2 impact / total AOH in each cell
For those following along - Rich and I decided on a plan of attack and I've nearly finished updating the analysis with this new methodology.
Before, I was taking averages by taxa and then averaging again across taxa (and other combinations of things, depending on what I wanted to present). I was basing my methodology on a lot of stuff done by Casey, but in our case we don't think it is appropriate to take averages of averages since we aren't rescaling in the same way as Casey (I.e., mean cumulative impact vs mean proportion of habitat impacted).
We figured out how to save species level km2 of impact. From there I am able to calculate proportion of habitat impacted (and extinction risk, just for fun) on a species level, and then average across taxa and raw material (or other combinations like by taxa across material, by material across taxa, etc). This methodology makes more sense and removes some uncertainty associated with taking averages of averages. I've also coded up to save standard deviations and number of species impacted maps.
I have all of these impacts calculated across taxa and ingredient, so hopefully I can update with a new global map soon and we'll see some of the problems discussed above fixed. Once those are done, I'll do other combinations of averaging for SI materials (across taxa by material, across material by taxa, etc).
Updated plot with new methodology - and the problem described above is fixed!!
I'll post an update with the other part of this map and some explanation of what the methodology changed in the results in a separate issue (spoiler alert - maximum average impacts are much higher. A cell in plant-dominant has a average prop impacted of 0.22!)
Why does it appear that the fish-dominant diet has higher impacts in places like Brazil, Argentina, and the USA, where there is a lot of soybean production? See the below delta plot, you'll notice there are many dark blue areas in these areas:
Presumably the plant-dominant diet would have higher impacts given that 1) SPC and soy oil account for a higher percentage of the plant-dominant diet than SBM does in the fish-dominant (thus there should be more demand and production of SPC), 2) SPC has a higher allocation factor (so there should be higher pressures, and presumably higher impacts).
SBM (fish-dominant):
SPC (plant-dominant:
I isolated the mean proportion of habitat impacted maps for the soy protein concetrate compared to the soybean meal, and it appears that the plant-dominant diet (SPC) always has larger impacts than the fish dominant diet (SBM)... which would make sense!:
What this indicates to me is that this is an artifact of the species weighted averaging across taxon and raw materials. Particularly when I average the impacts between SPC and soybean oil for the plant-dominant scenario (since the plant-dominant scenario has both)... If we look at the same map, but with SPC and soy oil averaged for the plant-dominant, then the fish-dominant ends up having higher average impacts for soybeans
Currently, I calculate for each cell, for each species within each taxanomic grouping and raw material/ingredient, the proportion of habitat impacted. Then I take the average of that, and save that, ending up with a raster that describes for a particular taxon and ingredient/raw material, the average proportion of habitat impacted in each cell. For example, I have a raster that is the average proportion of habitat impacted for Birds impacted by soy protein concentrate in each cell (see below):
Then from there, I calculate a species weighted average of proportion of habitat impacted in each cell across all of the taxanomic and raw material combinations, which is where we get the weird SBM vs SPC problem. I think that we see artifacts of this for soybeans, wheat, pulses, and "other crops", as they have multiple ingredients associated with their raw materials (and thus, lots of overlap in cells with "large" and "small" impacts). But obviously soybeans are the big problem considering there is such a large difference of values between SPC and soybean oil in the plant-dominant diet, without much overlap with other crops.
So, the question arises, should I change my methodology?
The other way I could conceive of accomplishing calculating impacts would be to: