Accounting for trade among countries

[gclawson1]

Based on what me and @cottrellr talked about last week, I've been taking a look at the food trade repo (I think its private so you won't be able to see) that Jessica Gephart is working on to start brainstorming if we could account for trade in our paper. Hopefully we can get a more accurate picture of what countries contribute to aquafeed pressures/impacts.

"For each crop or animal producing country, we first calculated the proportion of each product that they trade with each consuming country (e.g., Country A trades 10% of their total grain production to Country B). Given that this trade matrix includes self-loops, the proportion of each product that a country produces and then consumes is also included."

• So this makes it seem like they have data that will show us how many tonnes country X imports from country Y, and how many tonnes country X produces and uses locally.

  • Ex: Assume Norway only trades soy with Canada. Norway imports 100 tonnes of soy from Canada, and their local supply (the supply of soy that Norway doesn't export) is 20 tonnes, so their total supply of soy is 120 tonnes.
  • Based on this we could make some assumptions on the amount of that 120 tonnes of soy Norway uses for feed, and attribute that amount of tonnes to Canada as feed pressures.

  "We use FAO Food Balance and Commodity Balance data to estimate the percent of a country's domestic supply quantity that goes to feed for each crop category. The five crop categories we include are grains (barley, maize, sorghum, rice, millet, wheat, rye, oats, and other cereals), palm kernels, soybeans, oil crops (cottonseed, groundnuts, other oil crops, rape and mustard seed, sesame seed, and sunflower seed), and pulses (peas, beans, and other pulses). We gapfilled missing proportion feed estimates with regional values."

  "We adjusted the trade matrix so that the country that consumes the animal receives the related feed consumption pressures for that animal, rather than the country that produced that animal. The end result of this adjusted trade matrix is a trade matrix that connects the feed producing country to the animal consuming country by feed crop and animal system.

  To connect feed pressures to the animals that consume them, we needed to determine how much of each feed crop was going to each animal system in each animal producing country. To determine these values, we used data from Halpern et al. 2022 on the tonnes of feed consumption per crop (aggregated to grains, palm kernels, soybeans, oil crops, and pulses), per animal system, and per country. (Mel I'm not sure if we need to fill this in more or if we can just reference your other paper. Also you calculated tonnes feed consumption for inland aquaculture for this project, which is not in the other paper, so maybe we do need to include some methods on this here.) Then we calculated the proportion of each crop that each animal system consumes as feed in each country (e.g., pigs consume 10% of all grains in country A). Lastly, we multiplied the proportion of each crop that each animal system consumes as feed by the traded tonnes of feed and associated pressures in the adjusted feed trade matrix. This updates the trade matrix so that each row represents the tonnes and pressures traded between each feed producing country, animal producing country, crop, and animal system.

  Since we were not interested in the animal producing country, we summed the traded tonnes and associated pressures by each combination of feed producing country, animal consuming country, feed category, and animal system."

• I think we would actually want the traded tonnes summed as each combination of feed producing country, animal production country, geed category, and animal system.

• Based on all of this, I think they might have basically what we need (at least at the country level, not at a raster level). I'm gonna see if I can find their dataset on Aurora and dig around in it...

[gclawson1]

I've started digging into the trade data that Jessica and co have created, and here is what I've found:

It appears they have trade information (for 2017) for 4 types of salmonid feed from crops:

[1] "grain"   "oilcrop" "pulse"   "soybean"

They have it grouped as feed_producing_country and animal_consuming_country. So if I filter for just Norway as the "animal_consuming_country", I believe that this is all of the salmon fish that Norway produces itself and consumes, and all of the fish that Norway imports, and the associated feed production (tonnage) for that fish.

For example, this data is saying that in 2017, of the salmon that Norway consumed, they imported ~477 tonnes of pulses feed from the UK associated with that salmon?

I think ideally we would actually want information that told us how much feed is associated with the production of salmon in Norway, not the consumption... right? They probably have to have that information in an intermediate step somewhere, so I will look for that.

[cottrellr]

You're right we don't need the feed that is associated with the consumption of salmon but we want the production.

I think using their data could work for the first chapter. But my only concern in the long run is we will need to start thinking about sugar cane for microalgae and fish trimmings and so being able to select the trade categories will be important - rather than just those that are preprocessed. But yes for comparing a fish- to plant-dominant feed I think that could work if they have those intermediate steps.

[gclawson1]

Looked through their script, and it appears they do have that information (they just didn't save it as a csv anywhere), so we could probably get our hands on it. However, I want to think through if this is something we actually need.

Let's say we get this data (and presumably they have something similar for fish oil and meal), how are we planning to use it? So basically we just want to know the amount of a countries crop (soybean, pulses, etc) and FMFO production that actually goes to feed (regardless of where it ends up, since all we really care about in this case is the feed producing country)? Then we would take that amount (probably a percentage) and multiply it by the appropriate crop or fishing rasters? This way we would have rasters for every area that contributes to global aquaculture feed?

Does that make any sense at all??

[gclawson1]

Been thinking through this more. I think this is the workflow we could use to get a more realistic view of which countries are producing feeds.

What our (the raw material origins rasters @cottrellr created) current rasters show:

• Only raw material producing countries (crops) or areas (FMFO catch) are included if they can support the sum of the global demand for salmon feeds.

What we could have:

• Know which countries actually support the salmon industry based on the trade data.
• In this case, we don't really care about the trade pathways (right?)

• What we want:

  • Data frame with the demand sum for each country and feed type:

  Salmon producing country	raw material	animal type	Tonnes of raw material used (that was imported or created locally)
  NOR	soybean	Salmon	50
  CHL	fish oil	Salmon	25
  UK	wheat	Salmon	20
  AUS	Palm oil	Salmon	30

  • Data frame with supply sum for each raw material producing country and raw material type". We know this info because the trade data will have this.

  Feed producing country	raw material	animal type	Tonnes of raw material for feed (that was traded or used locally)	Total raw material produced	prop feed raw material to total
  UKR	soybean	Salmon	100	1000	0.1
  ARG	fish oil	Salmon	200	700	0.29
  USA	wheat	Salmon	150	800	0.19
  BRA	Palm oil	Salmon	300	900	0.33

Now we could take our total raw material production rasters and multiply them by "prop raw material to total" column above:

Global soybean prod:

Multiplied by

Feed producing country	raw material	animal type	prop raw material to total
UKR	soybean	Salmon	0.1
IND	soybean	Salmon	0.29
USA	soybean	Salmon	0.19
BRA	soybean	Salmon	0.33

The resulting raster would be a raster of soybean crop that is used to support global demand for salmon feed. And we could do the same thing for each crop type and fish oil/fishmeal. And probably have some threshold for exclusion if the amount of tonnes is too small or something.

Then we take to pressures (?) and overlay with AOH to get at impacts.

I'm going to take a look at the trade data to see if the numbers of feed produced generally make sense.

[gclawson1]

Wanted to take a look at the top feed producing countries for salmonid aquaculture (inland and marine) from the trade data. Filtered for top 15 and >1000 tonnes for each raw material:

Does this line up with expectations?

Here are tables instead:

feed_producing_country	feed_category	sum_tonnes
NOR	grain	252160.91
GBR	grain	91855.81
CHL	grain	87068.79
RUS	grain	69697.16
TUR	grain	67901.81
ARG	grain	67112.13
CAN	grain	65374.65
USA	grain	63509.80
CHN	grain	53487.20
FRA	grain	50827.35
SWE	grain	48113.20
DEU	grain	39670.78
POL	grain	32006.18
DNK	grain	30340.22
IRN	grain	29253.64

feed_producing_country	feed_category	sum_tonnes
CAN	oilcrop	165413.60
RUS	oilcrop	133855.44
CHL	oilcrop	126680.45
GBR	oilcrop	116391.38
ARG	oilcrop	101457.60
AUS	oilcrop	62988.14
UKR	oilcrop	38906.92
DNK	oilcrop	37396.69
FRA	oilcrop	30166.72
DEU	oilcrop	29162.73
USA	oilcrop	26044.66
LTU	oilcrop	22974.41
ESP	oilcrop	20270.94
LVA	oilcrop	18751.38
ROU	oilcrop	17417.14

feed_producing_country	feed_category	sum_tonnes
GBR	pulse	567074.500
AUS	pulse	100894.067
CAN	pulse	83364.783
IRL	pulse	49557.201
USA	pulse	43000.567
FRA	pulse	38736.080
RUS	pulse	22840.790
SWE	pulse	18309.970
EST	pulse	13399.471
CHL	pulse	8794.677
LTU	pulse	8679.279
ARG	pulse	8308.124
TUR	pulse	7610.836
IND	pulse	5521.726
MMR	pulse	5422.004

feed_producing_country	feed_category	sum_tonnes
BRA	soybean	85079.733
ARG	soybean	75870.490
CAN	soybean	61495.856
PRY	soybean	45294.335
USA	soybean	23550.766
BOL	soybean	14611.776
RUS	soybean	3468.259
UKR	soybean	1620.373

[cottrellr]

• Know which countries actually support the salmon industry based on the trade data.

Yeah I see this as being the importance of the trade data. Norway's dependence will almost certainly dffer to the UK, Canada, Chile etc. So this is why trade will help us spatialise the impacts.

[cottrellr]

• In this case, we don't really care about the trade pathways (right?)

Not sure what you mean by pathways - i.e. how many trade steps there are? Do you mean forage fish could be caught off Peru processed in Thailand and sent to Norway. In this instance we care about the impacts in Peru if that makes sense.

[cottrellr]

• What we want:

  • Data frame with the demand sum for each country and feed type:

Salmon producing country raw material animal type Tonnes of raw material used (that was imported or created locally) NOR soybean Salmon 50 CHL fish oil Salmon 25 UK wheat Salmon 20 AUS Palm oil Salmon 30

Yeah there are some tricky details that we need to pay attention to in trade data. Trade data may have the resolution of "soybeans" (as in the whole raw material) or FM/FO (which is the ingredient level). We need to convert it all to embodied raw material biomass (wet weight of fish or harvested equivalents for crop products).

[gclawson1]

• What we want:

  • Data frame with the demand sum for each country and feed type:

Salmon producing country raw material animal type Tonnes of raw material used (that was imported or created locally) NOR soybean Salmon 50 CHL fish oil Salmon 25 UK wheat Salmon 20 AUS Palm oil Salmon 30

Yeah there are some tricky details that we need to pay attention to in trade data. Trade data may have the resolution of "soybeans" (as in the whole raw material) or FM/FO (which is the ingredient level). We need to convert it all to embodied raw material biomass (wet weight of fish or harvested equivalents for crop products).

Right.. I'm not suuuper familiar with their methods, but I know they do use FAO and mapspam crop names for everything else included, so I think it is safe to assume that the feed data they have made is derived from this. They probably just aggregated it this way for ease of use.

[cottrellr]

• Data frame with supply sum for each raw material producing country and raw material type". We know this info because the trade data will have this.

Feed producing country raw material animal type Tonnes of raw material for feed (that was traded or used locally) Total raw material produced prop raw material to total UKR soybean Salmon 100 1000 0.1 ARG fish oil Salmon 200 700 0.29 USA wheat Salmon 150 800 0.19 BRA Palm oil Salmon 300 900 0.33

Yeah so understanding what proportion of the feed demand in terms of raw material equivalents or ingredients will be likely supplied from where is important. It's up to you when you make this conversion. The trade data could have FMFO as a resolution - but how that maps to forage fish depends on the FMFO yields of the fish harvested (Peruvian anchovy may have a different yield for both FM and FO to Atlantic herring for example). So you could calculate the embedded forage fish after learning what proportion of FMFO comes from where. But if the trade data is also in soybeans (raw material) - you would first how to understand how demand for e.g. soy protein concentrate converts into the soybeans before splitting up among the countries - does that make sense?

[cottrellr]

Now we could take our total raw material production rasters and multiply them by "prop raw material to total" column above:

Global soybean prod:

Yeah once you have the raw material quantities you can populate a raster of feed production for salmon for each raw material as you have. Might also be worth converting different ingredients which depend on the same raw material (e.g. corn gluten meal, corn starch) into one quantity (rather than a quantity per ingredient).

[cottrellr]

Multiplied by

Feed producing country raw material animal type prop raw material to total UKR soybean Salmon 0.1 IND soybean Salmon 0.29 USA soybean Salmon 0.19 BRA soybean Salmon 0.33 The resulting raster would be a raster of soybean crop that is used to support global demand for salmon feed. And we could do the same thing for each crop type and fish oil/fishmeal. And probably have some threshold for exclusion if the amount of tonnes is too small or something.

Then we take to pressures (?) and overlay with AOH to get at impacts.

I'm going to take a look at the trade data to see if the numbers of feed produced generally make sense.

We will also need to think about how we attack AOH once you have all those species habitat layers sorted.

• For two species each with a habitat of 500km2 - 100km2 of soy production might be less of a problem (or could even be a benefit from less competition) for a species that can survive pretty well in farmland. So establishing that sensitivity needs to be done too. This could be by looking through the key words in IUCN threat information e.g. agriculture, fisheries. For fisheries, I think we can assume fisheries are a threat if that species shows up in fisheries catch for instance. So thinking this part through will be important too.

[gclawson1]

I just wanted to have a look at how the trade data on feed producing countries compared to your data on crop demand in the plant based diet using gross economic allocation. Obviously this isn't exactly the same due to the concerns you just raised regarding the resolution of the trade data (e.g. soybeans vs soy protein concentrate), but it actually matches kind of well (aside from pulses, not really sure whats happening there, probably some dry weight conversion or something). Grain difference is probably due to crop loss or something with stalks:

"sum_tonnes" is the producing countries sum of feed (global) "crop_demand" is the mean of your "total_crop_demand" column in this data feed_pressure_mapping/data/tidy_data/demand/sourcing_countries_crops.rds" The groupings and resolutions of the crops probably aren't exactly the same, but this seems promising, at least on a global level

[cottrellr]

Wanted to take a look at the top feed producing countries for salmonid aquaculture (inland and marine) from the trade data. Filtered for top 15 and >1000 tonnes for each raw material:

Does this line up with expectations?

Yeah I think the top players are outlined there. Do you know if this is what the trade data is limited too? "oil crop" seems like a pretty general category and would be very different for canola oil vs palm oil vs linseed oil or soy oil. And just going on this for all oil crops could spit out some eyebrow raising stuff. Like most palm oil coming from Canada or something like that (not that we have palm oil but you know what I mean)

[cottrellr]

I just wanted to have a look at how the trade data on feed producing countries compared to your data on crop demand in the plant based diet using gross economic allocation. Obviously this isn't exactly the same due to the concerns you just raised regarding the resolution of the trade data (e.g. soybeans vs soy protein concentrate), but it actually matches kind of well (aside from pulses, not really sure whats happening there, probably some dry weight conversion or something). Grain difference is probably due to crop loss or something with stalks:

"sum_tonnes" is the producing countries sum of feed (global) "crop_demand" is the mean of your "total_crop_demand" column in this data feed_pressure_mapping/data/tidy_data/demand/sourcing_countries_crops.rds" The groupings and resolutions of the crops probably aren't exactly the same, but this seems promising, at least on a global level

Yeah cool. I think we need to do the conversions to be as accurate as possible but good to know it's in the ball park (no order of magnitude differences). Pulse is weirdly different.

[gclawson1]

Well based on all of this, I think the way forward is to leverage this trade data. I'll keep exploring their repo to see if I can answer the questions regarding what is contained in their feed categories. I'm almost positive that we could recreate their data (or they already have it in an intermediate step) but with better resolution categories for our purposes.

I'll also send out an email to Jessica/Mel/Ben and co to get their thoughts/approval on actually using this data and if they think it will suit our project... I assume it will be ok but I kind of stole it off of Aurora hahaha

[cottrellr]

Perfect yeah - it might be that if they have gotten the data from COMTRADE and then cleaned it that is the best data for what we need. But if they have that, it would be a huge help. Definitely good to ask whether it can be used - but also the source of the data and what was done to it (even if they just provide cleaned raw data). Thinking across the course of your project we will need to start introducing ingredients not covered in what they have so yeah as broad as possible is what we need.

[gclawson1]

The five crop categories we include are grains (barley, maize, sorghum, rice, millet, wheat, rye, oats, and other cereals), palm kernels, soybeans, oil crops (cottonseed, groundnuts, other oil crops, rape and mustard seed, sesame seed, and sunflower seed), and pulses (peas, beans, and other pulses).

[gclawson1]

A little update on where I'm at.

I'm nearly finished working through all of Mel's scripts. I updated the FCRs to match what you did in the raw material origins paper (using Tacon & Metian 2008 values). I also updated the diet composition to match that of Aas et al 2022 (the 2020 diet). As of now I've left out the fish trimmings, microingredients, and "other" category. In Mel's scripts, she accounted for extraction rates from raw mapspam crops to each processed raw material (like going from soybeans to soy protein concentrate), and I've done that too.

I still need to rasterize this information, but here is a bar chart of likely countries of origin for raw material production for crops:

Here is what I have for likely place of origin for forage fish catch going to salmon aquaculture feed (logged):

I think it should be fairly easy to retrofit the code to add other diets, for now I just wanted to get it up and running for diet in Aas et al as a test run.

[cottrellr]

Awesome - this looks great. Digging into the fish trimmings should be interesting, along with the other new additions to that feed. There is probably some information out there about where these trimmings are coming from in terms of species and then we can probably start to look at the trade data to make some assumptions about their source.

Then once we have good raw material conversion factors across all ingredients we can sub them in too.

---

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 21 Mar 2023, at 2:38 pm, Gage Clawson @.**@.>> wrote:

A little update on where I'm at.

I'm nearly finished working through all of Mel's scripts. I updated the FCRs to match what you did in the raw material origins paper (using Tacon & Metian 2008 values). I also updated the diet composition to match that of Aas et al 2022 (the 2020 diet). As of now I've left out the fish trimmings, microingredients, and "other" category. In Mel's scripts, she accounted for extraction rates from raw mapspam crops to each processed raw material (like going from soybeans to soy protein concentrate).

I still need to rasterize this information, but here is a bar chart of likely countries of origin for raw material production for crops:

[image]https://user-images.githubusercontent.com/33332753/226517754-f865da6d-5ba5-4d68-b4f6-3eae67bf5603.png

Here is what I have for likely place of origin for forage fish catch going to salmon aquaculture feed (logged):

[image]https://user-images.githubusercontent.com/33332753/226518095-c513dfbe-d03a-4910-936b-4925af9a03e5.png

I think it should be fairly easy to retrofit the code to add other diets, for now I just wanted to get it up and running for diet in Aas et al as a test run.

— Reply to this email directly, view it on GitHubhttps://github.com/Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping/issues/2#issuecomment-1477270439, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJK3YJEDOQ7Q3JQQ4ZU57MLW5EWETANCNFSM6AAAAAAVQU5ZWY. You are receiving this because you were mentioned.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]

Writing up the workflow for how I've calculated raw material origins. Just want another set of eyes on it to make sure it makes sense, and it'll help when writing up methods eventually

Question: Because I am using trade data to determine raw material origins, am I correct in assuming I don't need to use any allocation approaches like you do, Rich? Since we know (basically) where the materials are coming based on the trade, we don't need to estimate where they originate using allocation?

Workflow showing how to estimate raw material origins based on trade and diet data

Figure out origin of crops using trade data:

Start with production data for all crops

Read in FAO food balance sheets to determine imports, exports, feed, and domestic supply for each country and crop

Join MapSPAM prod data with FBS data This shows us that ~91% of the total supply of barley is local in Norway

FAO provides a trade matrix showing tonnes of each crop imported into each country. Match this to SPAM crop categories

• Join this with SPAM data and make sure that if there is no production according to MapSPAM then there will be no exports for the country/crop.
• do some magical gapfilling

Do a bit more data wrangling and calculate the prop of supply for each country and crop that is imported or grown locally. Ex: This shows that Norway imports ~92% of its total sugarbeet supply from Russia

Grab diet composition data from Aas et al for Atlantic salmon and assume every salmon producing country uses this diet (this will be our "plant diet")

Calculate feed ingredient demand using the same methods as Rich (e.g. eFCR ratios from Tacon & Metian)

Join ingredient demand to MapSPAM production data based on equivalent MapSPAM crop category. E.g. "coconut oil" ingredient is equivalent to MapSPAM "oil of coconuts" category

source_ingredient	product
coconut oil	oil of coconuts
corn gluten	maize gluten
faba beans	faba bean
guar protein	guar protein
linseed oil	oil of linseed
pea flour	pea other
pea protein	pea protein
rapeseed and camelina oil	oil of rapeseed
soy protein concentrate	soy protein concentrate
soybean oil	oil of soya bean
sunflower	cake of sunflower seed
wheat	wheat
wheat gluten	wheat gluten

Determine processing loss. It looks like Mel and team figured out crop loss adjustments based on literature review. So they figured out the amount of loss there is from going from a raw coconut to coconut oil (~79% loss of weight), or maize to maize gluten (16% loss). You can see below that now we know that 377,237 tonnes of soy protein concentrate is equal to ~ 428678 tonnes of raw soy... same process done with fofm.

Now we have the amount of raw materials each country needs to supply their aquaculture production, the import/export stats for total crop/fofm, so we can match those up and determine prop of production in each cell that goes to aquaculture feeds

Now create rasters describing the proportion of each country's forage fish catch going to salmon feed in the location of capture.

• This the step I need to ask Mel/Jessica about. It looks like Jessica created a dataframe that describes the tonnes of FOFM sent between countries for each fed animal system. I'll probably ask her about that and see if I can recreate it.
• Regardless, here we take the Watson fisheries catch data, rasterize it, and multiply it by a raster of the proportion of fofm catch going to feed for salmon aquaculture in each countries' EEZ.

Now create rasters for each crop describing the amount of each crop consumed in each country by salmon aquaculture systems.

This says that ~78% of pulse production in Norway goes to pea protein, regardless of where it ends up.

• This is basically the same method as above, but for crops.
• Rasterize a dataframe that describes the proportion of a crop in a country that contributes to salmon aquaculture feed out of the total crop production (we've figured this out above).
• Multiply that raster for each crop by the MapSPAM crop production rasters, to get the likely tonnes of production in a cell that goes to aquaculture feed for each feed ingredient

Logged tonnes of production of soy protein concentrate for the Aas et al 2022 diet

I glossed over a number of data wrangling things here, but I think this gives the general idea of the workflow.

Now the next step is to add another diet, which should be pretty trivial. I think I'll start with the "fish-dominant" diet Rich uses in his raw material origins paper

[cottrellr]

Hi Gage,

Great that you are figuring out the process. Sorry for banging on about it but I think the conversion from ingredient to raw material needs to be thought about a little more though. The loss is 1 - the extraction rate (yield) I take it? What that yield does not tell you is how much of other products are coming out of the same biomass of raw material as the e.g. wheat gluten. The biomass needed for wheat gluten will also support wheat flour production for example. Soy oil and lecithin both are produced from dehulled soybeans before being processed into defatted flakes which SPC, SBM etc from.

Attributing the whole biomass to one ingredient from the process disproportionately penalises the use of by-products (which typically have low extraction rates e.g. 0.08 so ingredient demand divided by that extraction rate gives A LOT of raw material). Being properly able to distribute the impacts across ingredients needs to account for this. This is why I have used the mass and energetic allocation approaches in my paper. Probably something we can add back in once you have the whole process down….

---

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: Friday, 24 March 2023 at 1:08 pm To: Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping @.> Cc: Richard Cottrell @.>, Mention @.> Subject: Re: [Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping] Accounting for trade among countries (Issue #2)

Writing up the workflow for how I've calculated raw material origins. Just want another set of eyes on it to make sure it makes sense, and it'll help when writing up methods eventually

Question: Because I am using trade data to determine raw material origins, am I correct in assuming I don't need to use any allocation approaches like you do, Rich? Since we know (basically) where the materials are coming based on the trade, we don't need to estimate where they originate using allocation?

Workflow showing how to estimate raw material origins based on trade and diet data

Figure out origin of crops using trade data:

Start with production data for all crops [image]https://user-images.githubusercontent.com/33332753/227387134-17c79dd6-e157-4dfc-82b1-3944f24a2264.png

Read in FAO food balance sheets to determine imports, exports, feed, and domestic supply for each country and crop [image]https://user-images.githubusercontent.com/33332753/227387311-c32a7ae8-2cab-498e-a318-1ae93b33a059.png

Join MapSPAM prod data with FBS data This shows us that ~91% of the total supply of barley is local in Norway

[image]https://user-images.githubusercontent.com/33332753/227388441-5cca6f9a-e2a0-42ea-9613-aae3544f745c.png

FAO provides a trade matrix showing tonnes of each crop imported into each country. Match this to SPAM crop categories

• Join this with SPAM data and make sure that if there is no production according to MapSPAM then there will be no exports for the country/crop.
• do some magical gapfilling

Do a bit more data wrangling and calculate the prop of supply for each country and crop that is imported or grown locally. Ex: This shows that Norway imports ~92% of its total sugarbeet supply from Russia

[image]https://user-images.githubusercontent.com/33332753/227389597-7d17b7d4-0a45-4e51-bc4e-812259bcbaf0.png

Grab diet composition data from Aas et al for Atlantic salmon and assume every salmon producing country uses this diet (this will be our "plant diet")

Calculate feed ingredient demand using the same methods as Rich (e.g. eFCR ratios from Tacon & Metian)

[image]https://user-images.githubusercontent.com/33332753/227394157-848d3e78-97fb-4ae4-b495-28169968039a.png

Join ingredient demand to MapSPAM production data based on equivalent MapSPAM crop category. E.g. "coconut oil" ingredient is equivalent to MapSPAM "oil of coconuts" category source_ingredient product coconut oil oil of coconuts corn gluten maize gluten faba beans faba bean guar protein guar protein linseed oil oil of linseed pea flour pea other pea protein pea protein rapeseed and camelina oil oil of rapeseed soy protein concentrate soy protein concentrate soybean oil oil of soya bean sunflower cake of sunflower seed wheat wheat wheat gluten wheat gluten

Determine processing loss. It looks like Mel and team figured out crop loss adjustments based on literature review. So they figured out the amount of loss there is from going from a raw coconut to coconut oil (~79% loss of weight), or maize to maize gluten (16% loss). You can see below that now we know that 377,237 tonnes of soy protein concentrate is equal to ~ 428678 tonnes of raw soy... same process done with fofm.

• Now we have the amount of raw materials each country needs to supply their aquaculture production, the import/export stats for total crop/fofm, so we can match those up and determine prop of production in each cell that goes to aquaculture feeds

[image]https://user-images.githubusercontent.com/33332753/227396343-e1c5d289-7de1-4d82-9731-59463af97b1a.png

Now create rasters describing the proportion of each country's forage fish catch going to salmon feed in the location of capture.

• This the step I need to ask Mel/Jessica about. It looks like Jessica created a dataframe that describes the tonnes of FOFM sent between countries for each fed animal system. I'll probably ask her about that and see if I can recreate it.
• Regardless, here we take the Watson fisheries catch data, rasterize it, and multiply it by a raster of the proportion of fofm catch going to feed for salmon aquaculture in each countries' EEZ.

[image]https://user-images.githubusercontent.com/33332753/227411685-96066e93-6783-4805-810b-03a0ff4c0360.png

Now create rasters for each crop describing the amount of each crop consumed in each country by salmon aquaculture systems.

• This is basically the same method as above, but for crops.
• Rasterize a dataframe that describes the proportion of a crop in a country that contributes to salmon aquaculture feed out of the total crop production (we've figured this out above).
• Multiply that raster for each crop by the MapSPAM crop production rasters, to get the likely tonnes of production in a cell that goes to aquaculture feed for each feed ingredient

Logged tonnes of production of soy protein concentrate for the Aas et al 2022 diet

[image]https://user-images.githubusercontent.com/33332753/227413458-b2bf57d7-3d74-4493-a500-2e87cd76c967.png

Now the next step is to add another diet, which should be pretty trivial. I think I'll start with the "fish-dominant" diet Rich uses in his raw material origins paper

— Reply to this email directly, view it on GitHubhttps://github.com/Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping/issues/2#issuecomment-1482188016, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJK3YJA6GELILV4P4XOWX6DW5UF2HANCNFSM6AAAAAAVQU5ZWY. You are receiving this because you were mentioned.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]

Unless I have missed something here….soy protein concentrate is a category in MAPSPAM? i.e. this table

source_ingredient product coconut oil oil of coconuts corn gluten maize gluten faba beans faba bean guar protein guar protein linseed oil oil of linseed pea flour pea other pea protein pea protein rapeseed and camelina oil oil of rapeseed soy protein concentrate soy protein concentrate soybean oil oil of soya bean sunflower cake of sunflower seed wheat wheat wheat gluten wheat gluten

---

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: Richard Cottrell @.> Date: Friday, 24 March 2023 at 2:27 pm To: Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping @.>, Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping @.> Cc: Mention @.> Subject: Re: [Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping] Accounting for trade among countries (Issue #2) Hi Gage,

Great that you are figuring out the process. Sorry for banging on about it but I think the conversion from ingredient to raw material needs to be thought about a little more though. The loss is 1 - the extraction rate (yield) I take it? What that yield does not tell you is how much of other products are coming out of the same biomass of raw material as the e.g. wheat gluten. The biomass needed for wheat gluten will also support wheat flour production for example. Soy oil and lecithin both are produced from dehulled soybeans before being processed into defatted flakes which SPC, SBM etc from.

Attributing the whole biomass to one ingredient from the process disproportionately penalises the use of by-products (which typically have low extraction rates e.g. 0.08 so ingredient demand divided by that extraction rate gives A LOT of raw material). Being properly able to distribute the impacts across ingredients needs to account for this. This is why I have used the mass and energetic allocation approaches in my paper. Probably something we can add back in once you have the whole process down….

---

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: Friday, 24 March 2023 at 1:08 pm To: Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping @.> Cc: Richard Cottrell @.>, Mention @.> Subject: Re: [Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping] Accounting for trade among countries (Issue #2)

Writing up the workflow for how I've calculated raw material origins. Just want another set of eyes on it to make sure it makes sense, and it'll help when writing up methods eventually

Question: Because I am using trade data to determine raw material origins, am I correct in assuming I don't need to use any allocation approaches like you do, Rich? Since we know (basically) where the materials are coming based on the trade, we don't need to estimate where they originate using allocation?

Workflow showing how to estimate raw material origins based on trade and diet data

Figure out origin of crops using trade data:

Start with production data for all crops [image]https://user-images.githubusercontent.com/33332753/227387134-17c79dd6-e157-4dfc-82b1-3944f24a2264.png

Read in FAO food balance sheets to determine imports, exports, feed, and domestic supply for each country and crop [image]https://user-images.githubusercontent.com/33332753/227387311-c32a7ae8-2cab-498e-a318-1ae93b33a059.png

Join MapSPAM prod data with FBS data This shows us that ~91% of the total supply of barley is local in Norway

[image]https://user-images.githubusercontent.com/33332753/227388441-5cca6f9a-e2a0-42ea-9613-aae3544f745c.png

FAO provides a trade matrix showing tonnes of each crop imported into each country. Match this to SPAM crop categories

• Join this with SPAM data and make sure that if there is no production according to MapSPAM then there will be no exports for the country/crop.
• do some magical gapfilling

Do a bit more data wrangling and calculate the prop of supply for each country and crop that is imported or grown locally. Ex: This shows that Norway imports ~92% of its total sugarbeet supply from Russia

[image]https://user-images.githubusercontent.com/33332753/227389597-7d17b7d4-0a45-4e51-bc4e-812259bcbaf0.png

Grab diet composition data from Aas et al for Atlantic salmon and assume every salmon producing country uses this diet (this will be our "plant diet")

Calculate feed ingredient demand using the same methods as Rich (e.g. eFCR ratios from Tacon & Metian)

[image]https://user-images.githubusercontent.com/33332753/227394157-848d3e78-97fb-4ae4-b495-28169968039a.png

Join ingredient demand to MapSPAM production data based on equivalent MapSPAM crop category. E.g. "coconut oil" ingredient is equivalent to MapSPAM "oil of coconuts" category source_ingredient product coconut oil oil of coconuts corn gluten maize gluten faba beans faba bean guar protein guar protein linseed oil oil of linseed pea flour pea other pea protein pea protein rapeseed and camelina oil oil of rapeseed soy protein concentrate soy protein concentrate soybean oil oil of soya bean sunflower cake of sunflower seed wheat wheat wheat gluten wheat gluten

Determine processing loss. It looks like Mel and team figured out crop loss adjustments based on literature review. So they figured out the amount of loss there is from going from a raw coconut to coconut oil (~79% loss of weight), or maize to maize gluten (16% loss). You can see below that now we know that 377,237 tonnes of soy protein concentrate is equal to ~ 428678 tonnes of raw soy... same process done with fofm.

• Now we have the amount of raw materials each country needs to supply their aquaculture production, the import/export stats for total crop/fofm, so we can match those up and determine prop of production in each cell that goes to aquaculture feeds

[image]https://user-images.githubusercontent.com/33332753/227396343-e1c5d289-7de1-4d82-9731-59463af97b1a.png

Now create rasters describing the proportion of each country's forage fish catch going to salmon feed in the location of capture.

• This the step I need to ask Mel/Jessica about. It looks like Jessica created a dataframe that describes the tonnes of FOFM sent between countries for each fed animal system. I'll probably ask her about that and see if I can recreate it.
• Regardless, here we take the Watson fisheries catch data, rasterize it, and multiply it by a raster of the proportion of fofm catch going to feed for salmon aquaculture in each countries' EEZ.

[image]https://user-images.githubusercontent.com/33332753/227411685-96066e93-6783-4805-810b-03a0ff4c0360.png

Now create rasters for each crop describing the amount of each crop consumed in each country by salmon aquaculture systems.

• This is basically the same method as above, but for crops.
• Rasterize a dataframe that describes the proportion of a crop in a country that contributes to salmon aquaculture feed out of the total crop production (we've figured this out above).
• Multiply that raster for each crop by the MapSPAM crop production rasters, to get the likely tonnes of production in a cell that goes to aquaculture feed for each feed ingredient

Logged tonnes of production of soy protein concentrate for the Aas et al 2022 diet

[image]https://user-images.githubusercontent.com/33332753/227413458-b2bf57d7-3d74-4493-a500-2e87cd76c967.png

Now the next step is to add another diet, which should be pretty trivial. I think I'll start with the "fish-dominant" diet Rich uses in his raw material origins paper

— Reply to this email directly, view it on GitHubhttps://github.com/Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping/issues/2#issuecomment-1482188016, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJK3YJA6GELILV4P4XOWX6DW5UF2HANCNFSM6AAAAAAVQU5ZWY. You are receiving this because you were mentioned.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]

Oh sorry thats not right... should look like this. Just a lookup table:

And right... the allocation stuff makes more sense now. I wasn't thinking about it that way. We definitely will want to account for the other types of products so not to over allocate impacts that might not be there

[cottrellr]

Yeah sorry, that was me, a function of me copying and pasting in the email rather than through the issue.

What we can do is compare how the estimates of raw material demand are coming out from the way you have it now and from using a conversion factor. If they are approximately the same for both approaches (from a quick look they actually might be) then there is probably no issue. Although for work going forward, we will need a systematic way to think about how ingredients convert to raw materials so we can look into some of thos by-products more as a focus of some of the work.

---

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: Friday, 24 March 2023 at 4:37 pm To: Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping @.> Cc: Richard Cottrell @.>, Mention @.> Subject: Re: [Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping] Accounting for trade among countries (Issue #2)

Oh sorry thats not right... should look like this. Just a lookup table:

[image]https://user-images.githubusercontent.com/33332753/227443945-91e40b20-4035-475c-86f3-9f86f40862be.png

And right... the allocation stuff makes more sense now. I wasn't thinking about it that way. We definitely will want to account for the other types of products so not to over allocate impacts that might not be there

— Reply to this email directly, view it on GitHubhttps://github.com/Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping/issues/2#issuecomment-1482325914, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJK3YJHOVXIK3BVDSIWF67DW5U6LZANCNFSM6AAAAAAVQU5ZWY. You are receiving this because you were mentioned.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]

Retrofitted code to include a fish-dominant diet. Pretty easy to change/add new diets now:

[cottrellr]

Awesome!

[juliablanchard]

Cool!!

---

Professor Julia Blanchard (she/her) ARC Future Fellow Institute for Marine and Antarctic Studies & Centre for Marine Socioecology College of Science and Engineering University of Tasmania

Please note I work Tuesdays-Fridays

E: @.**@.> T: +61 (0)3 62.26.69.32 W: http://www.utas.edu.au/profiles/staff/imas/julia-blanchardhttp://www.utas.edu.au/profiles/staff/imas/julia-blanchard ; www.sizeecology.comhttp://www.sizeecology.com/

I acknowledge the Traditional Owners of the lands and waters on which I work and live, and pay my respects to the Elders, past and present.

[Logo Description automatically generated with low confidence]

---

From: Gage Clawson @.> Sent: Tuesday, 28 March 2023 3:33 PM To: Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping @.> Cc: Subscribed @.***> Subject: Re: [Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping] Accounting for trade among countries (Issue #2)

Retrofitted code to include a fish-dominant diet. Pretty easy to change/add new diets now:

[image]https://user-images.githubusercontent.com/33332753/228128805-281c6ce4-84fa-4bbe-9c5b-cb2c14847066.png

— Reply to this email directly, view it on GitHubhttps://github.com/Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping/issues/2#issuecomment-1486201055, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ABGD7OQO4HVTBYEZYTUC6GLW6JSX7ANCNFSM6AAAAAAVQU5ZWY. 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]

Here is what the raw material origins look like after implementing the mass allocation conversion factors (plant material only; excluding soy oil and coconut oil):

Plant diet:

Fish diet:

[cottrellr]

Nice, what I’ll do is set up an issue for priority ingredients for conversion factors.

---

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 31 Mar 2023, at 2:21 pm, Gage Clawson @.**@.>> wrote:

Here is what the raw material origins look like after implementing the mass allocation conversion factors (plant material only; excluding soy oil and coconut oil):

Plant diet: [image]https://user-images.githubusercontent.com/33332753/229021786-8061525d-b395-4178-8718-a307cdca5e1c.png

Fish diet: [image]https://user-images.githubusercontent.com/33332753/229021869-ac432d46-568e-475c-8e85-3514be0b3198.png

— Reply to this email directly, view it on GitHubhttps://github.com/Sustainable-Aquafeeds-Project/feed_biodiv_impact_mapping/issues/2#issuecomment-1491266574, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AJK3YJEXUOKWJYMN7NPS3BLW6ZLUJANCNFSM6AAAAAAVQU5ZWY. You are receiving this because you were mentioned.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]

Finished rerunning my scripts after adding pea flour, soy oil, and coconut oil, and accounting for all co-products with mass and energetic allocation.

Adjusted source ingredient plots

Fish-dominant energetic allocation:

Plant-dominant energetic allocation:

Fish-dominant mass allocation:

Plant-dominant mass allocation:

Next step is to figure out the fish ingredients.