Impacts of the global food system on terrestrial biodiversity from land use and climate change

The global food system is a key driver of land-use and climate change which in turn drive biodiversity change. Developing sustainable food systems is therefore critical to reversing biodiversity loss. We use the multi-regional input-output model EXIOBASE to estimate the biodiversity impacts embedded within the global food system in 2011. Using models that capture regional variation in the sensitivity of biodiversity both to land use and climate change, we calculate the land-driven and greenhouse gas-driven footprints of food using two metrics of biodiversity: local species richness and rarity-weighted species richness. We show that the footprint of land area underestimates biodiversity impact in more species-rich regions and that our metric of rarity-weighted richness places a greater emphasis on biodiversity costs in Central and South America. We find that methane emissions are responsible for 70% of the overall greenhouse gas-driven biodiversity footprint and that, in several regions, emissions from a single year’s food production are associated with global biodiversity loss equivalent to 2% or more of that region’s total land-driven biodiversity loss. The measures we present are relatively simple to calculate and could be incorporated into decision-making and environmental impact assessments by governments and businesses.

I have some questions regarding the methods used that need to be clarified (see Specific comments).
In the policy implications section it would be interesting to have a more targeted and in depth discussion on the application of the work to the current policy frameworks.This would add novelty to the work, a lot of developments in policy arena have happened since the publication of other biodiversity footprint related work.

Specific comments:
Line 145 -147 -not clear which impacts refer to food waste.Can you please clarify?Line 147 -149 -you mention that "we regard land use as a one-off cost that results in a change in biodiversity and we assume that once land has been converted, it can be used repeatedly without the biodiversity cost increasing".I don't understand how can the impact be a one-off but then calculated for an yearly economic flow.Does this mean that if you would calculate the impacts for 2012, there will be almost no impacts since they were attributed to 2011?I don't understand this sentence, could you please clarify?Line 174 -nice figures, it would be more clear if you add the units in all squares so one can follow.It would make it easier to understand the final unit and how it is used.
Line 189 -190 -Where all of these data points/studies used?I suppose you filtered the PREDICTS database to select only the metrics relevant to you, and a selection of studies with primary vegetation and another land use type.Please specify how many data points were used in the end.In the SI you should also identify this per model.It would be also nice to add a map with the geographical distribution of the studies used.
Line 197 -do you follow an approach presented in any other biodiversity footprint paper?If yes please refer to it in this section.If not, can you explain what are the novelties and differences in your approach?Line 216 -not clear why and how you determine the amount of area needed to produce €1M of product.
Line 221 -I don't understand the units.Normally, the characterization factor represents the impact per unit of pressure (species lost/km2 of land use).If the unit of the CF is species x km2 when you multiply it by km2 (from EXIOBASE) a different unit is achieved.This needs to be clarified.
Line 228 -Why final demand?Shouldn't you compute the emissions associated with production activities so that it is compatible with the environmental extension.
Line 238 -You choose to use Global Temperature Change Potential instead of the most commonly used approach using Global Warming Potential.Why is that the case?Please justify why you do not use GWP.I would also recommend calculating the impacts using GWP for two main reasons, first is that it is the most commonly used metric, which would increase the comparability of your results and second as a sensitivity analysis.
Line 241 -The choice of 20 years period should be better justified.If methane accounts for only a small percentage of the total gases I would argue that you should choose a different time horizon.Line 251 -is the use of RCPs consistent with the use of GTP? Line 255 -In the SI file you mention that "Species' projected responses to climate change were derived from10".What was the exact relationship?Can you add this information in SI? Line 265 -266 -you mentioned before emissions from final demand, but here you mention emissions from food production of each product p in each region k.It is not clear what emission were used, please clarify.
Line 269 -please provide more information on the relationship used.
Line 272 -please clarify the unit (see comment referring to line 221) Line 285 -I would say that considering land use as a one-off impact is not compatible with the analysis of the economic flows of one year.
Line 343 -Does this mean that the countries in the top 10 positions are the same for all the different metrics analyzed?Line 370 -372 -Why does a ratio of 100 indicates that reducing GHG biodiversity footprint is a priority?Line 390 -I wonder if this is a real result of somewhat related with uncertainties of the MRIO table.Please check this papers to better understand if this is the case for this result: Line 519 -%21 -I don't think you can state this without a proper analysis.Also check the literature on the differences of the results obtained with physical and monetary approaches.
For example: https://www.sciencedirect.com/science/article/abs/pii/S0921800913003583https://www.sciencedirect.com/science/article/abs/pii/S0921800915000932Line 530 -536 -See my previous comments on the methods.I suggest you follow the same approach (at least as sensitivity) so that the comparison is possible.
Line 566 -I think you cannot really make this statement since you focus only on food products.

Reviewer #2 (Remarks to the Author):
I reviewed the analysis title "Impacts of global food supply on biodiversity via land use and climate change".
The submitted manuscript investigates new methods by combining different datasets and using an EEMRIO to calculate land-and GHG-driven biodiversity footprints globally.Overall, the analysis is interesting, novel, and well-structured, however, there are a couple of issues and open questions that need to be addressed.Lines 86-88: For clarification, it would be helpful for the reader if you could state that your data are building on spatially explicit information but is aggregated and averaged to reflect heterogeneity but, in the end, do not represent spatially explicit information any more.
Lines 95-97: Do you refer to both production and consumption footprints here?Line 117: When using Y from EXIBOASE the matrix is 7 x i, right?Because there are 7 different final demand categories.However, not sure if you need to mention that when you aggregate into one final demand category for each region.You could say "associated with the final demand of households, governments, capital formation, … of each region."Line 130: "spare vector populated in the entries for …" --Sounds a bit odd.It shows which sectors/production activities require direct land-use.The same allocation principle applies to GHGs, right?Lines 136-137: I understand, but I still have to flag this, it is quite old data and I'm unsure if 12-year-old data represent a proper fit to present data, also since you derive policy implications and not only methodological proof of concept.There are no land-use or spatial crop data available after 2011?Lines 147-149: How do you do that?A share of the upstream land-use flows based on spatial explicit information of land-use?Figure 1a): So, for each biome you calculated a mean species richness/rarity richness and multiplied by the total land-use area reported in EXIOBASE?Lines 321: Could you provide the total values for the GHG footprints of all food-related products, globally and/or nationally?This number could be compared to the numbers mentioned in the introduction in lines 31-32.
Lines 333-334: So, with their food system production they harm their domestic biodiversity as well as global biodiversity (GHG-driven).This could be mentioned to illustrate and differentiate.
Lines 342-343: I still don't know the difference between production and consumption footprints, but they appear remarkably close to one another.That is not production-vs consumption-based approaches?! Also, I realize these are total values, but I would recommend to at least mention why there are no European countries shown nor discussed.
Lines 359-361: I don't see this in e) or f).Isn't it Africa and C&S America that stand out?Or where can I find that information?Lines 440-442: If correct, you could state here, again for illustration, that these countries are relatively biodiversity poor but highly industrialized in their land use system.
It would be useful to also mention how net-trade is calculated and what negative and positive values represent.
Lines 485-486: However, land-use intensity is still not reflected (probably early on in the manuscript).You should at least mention this and that other indicators, like HANPP (and embodied HANPP) can account for that.
Lines: 502-503: Actually, it would be interesting to also see the aggregated effects of both in a figure and to see it discussed in the text.
Line 514: You should not that there is an EXIOBASE version with higher national detail (n=214) but only captures land use as an environmental extension so far, not GHGs (see https://zenodo.org/record/2654460).
Lines 550-551: And method, or resolution respectively.Lines 625-626: I'm not sure yet how you calculated the GHG footprints of food-related sectors/products.But since it's a supply-chain perspective, also machinery that is required by food producing sectors would fall into that, right?Lines 626-628: It would be interesting to see (from a comparison) if it was useful from a biodiversity footprint perspective to onshore production to N America and W Europe from other regions to reduce land-driven biodiversity footprints and increase GHG-driven biodiv footprints but still decrease the total biodiv footprint by regionalizing production for consumption.

REVIEWER COMMENTS
We thank the reviewers very much for their comments and have addressed them all, as detailed below.We have paid particular attention to clarifying the methodology and we have also improved the policy implications section (see details below).Following our revisions, we believe the manuscript is much improved.In addressing the revision regarding the total carbon footprint we came across a bug in our code relating to the calculation of the GHG-driven footprint which had led us to underestimate this footprint.The main message of the results has not changed although the ratio of land-driven:GHG-driven footprints has decreased and we have updated the results, figures and discussion accordingly.In making the updates we realised that actually it would be interesting to split out the GHG-driven footprint into the separate contributions made by carbon dioxide, methane and nitrous oxide which we were able to show using stacked bar charts in place of the previous bar charts (Figures 2d,3d,4d,5d).No extra figures have been added.
In addition to the revised document, we have uploaded a revised document with changes tracked.Line numbers of both documents are included in our responses, the numbers of the tracked changes document are in brackets.We have also included a tracked changes version of the revised Supplementary Information.

Reviewer #1 (Remarks to the Author):
This manuscript presents the analysis of food-related biodiversity footprints, it considers both land-driven footprints and GHG-driven footprints and two biodiversity metrics.The manuscript is very well-written, clearly structured and with high quality figures.However, I think it does not add sufficient novelty in its methods and findings.Therefore I cannot recommend this manuscript for publication in Nature Communications.I hope the authors find my comments below useful.

General comments
My main issue with the manuscript in its current format is that it does not add sufficient novelty to the current state of the art.Marquardt et al. 2019 compared different biodiversity footprint metrics between them and also with land footprint.However the authors of this manuscript do not cite this work (https://www.sciencedirect.com/science/article/abs/pii/S1470160X19302687?via=ihub).Wilting et al. 2017 quantified land-driven and GHG-driven biodiversity footprints.However their study is more complete since they cover more sectors in their GHG-driven footprints (https://pubs.acs.org/doi/10.1021/acs.est.6b05296).
Thank you for pointing out this oversight.We agree that we should have referred to Marquardt et al 2019.We have now remedied this both in the introduction (lines 74-76 (84-85)) and in the discussion (lines 596-599 (636-640)) as discussed later in our responses to the Specific Comments.
However, although our analysis builds on that by Marquardt et al and by Wilting et al, we respectfully disagree with reviewer 1 that our study is not sufficiently novel.The differences between species richness and rarity-weighted richness footprints are a key part of our analysis -this adds a valuable addition to the biodiversity footprint comparisons made by Marquardt et al 2019 since their paper does not analyse footprints based on rarity-weighted richness.Moreover, our analysis has greater spatial sensitivity than Wilting et al's since it has the novel combination of (i) allowing sensitivity to land use to vary by land-use type and biome; (ii) allowing for the natural variation of species richness across political regions; and (iii) allowing for spatial variation in future temperature change and in species' responses to that temperature change.Our in-depth focus on the production and consumption of food-related products complements Wilting et al's broader sector approach, addressing a different suite of questions.We use a different approach to Wilting et al and believe that it is important that land-driven and GHG-driven biodiversity impacts are further compared using different trade models and biodiversity metrics.Indeed, our discussion section covers these important points of comparison (section 4.1.).Furthermore, we have now split our GHG-driven footprint into the impacts driven by carbon dioxide, methane and nitrous oxide and show methane to contribute to 70% of the total food-related production footprint, thus providing further novelty to the manuscript.

I have some questions regarding the methods used that need to be clarified (see Specific comments).
Thank you for these comments, they were very constructive -we have now clarified the methods (see responses below in the Specific Comments).
In the policy implications section it would be interesting to have a more targeted and in depth discussion on the application of the work to the current policy frameworks.This would add novelty to the work, a lot of developments in policy arena have happened since the publication of other biodiversity footprint related work.
We agree that the policy implications of the work are interesting and we have now added to this section, discussing how the work could be applied to sustainability regulations in trade deals and highlighting that our results showing the high footprint due to methane emissions provides further support for policies regarding dietary shifts and the transition of farmers away from livestock (lines 619-714 (663-770).

Specific comments: Line 145 -147 -not clear which impacts refer to food waste. Can you please clarify?
We have now made it clear that the 'food waste' impacts refer to the treatment of food waste (lines 160-162 (173-175)).
'…impacts of food waste refer to the waste treatment and decomposition processes, and do not include the production impacts of the wasted food.' Line 147 -149 -you mention that "we regard land use as a one-off cost that results in a change in biodiversity and we assume that once land has been converted, it can be used repeatedly without the biodiversity cost increasing".I don't understand how can the impact be a one-off but then calculated for an yearly economic flow.Does this mean that if you would calculate the impacts for 2012, there will be almost no impacts since they were attributed to 2011?I don't understand this sentence, could you please clarify?
Ensuring the reader understands this is really key so thank you for bringing this up.EXIOBASE provides the land area used in each year.So in order to calculate the change in land use costs between 2011 and 2012 using EXIOBASE, you would subtract the 2011 land use from the 2012 land use.If you wanted the 2012 land use you would use the figures for 2012.What we were trying to explain is that the biodiversity impact of land use arises due to conversion from natural habitat.The land used in 2011 may have been converted one year ago or 1000 years ago and we are assuming that whenever that conversion occurred, the cost to biodiversity is the same.This is what we meant by the term 'one-off cost'.In contrast, emissions accrue year on year.Using EXIOBASE to compare biodiversity loss from land use to GHGs in 2011 means that you are comparing biodiversity loss that has occurred over centuries (land use) to biodiversity loss that will be caused by a single year of emissions.It's critical that the reader understands this hence why we provide the explanation in lines 147-149.However, we may perhaps have over-complicated our explanation.We have rewritten it as follows (lines 162-171 (178-188)): 'We calculate the biodiversity change associated with all of the land area used in food production in 2011 and assume that the biodiversity change associated with land conversion is immediate.GHG emissions that are released during land conversion are not considered, because, without detailed land-history knowledge, we cannot estimate the proportion of emissions that have dissipated since conversion, nor apportion food-production emissions across years.To put this gap in our coverage of GHGs into context, direct emissions from agriculture contribute 5.1-6.1 Pg CO2eq,/yr while the clearing of native land for agriculture contributes around 5.9 (SD 2.9) Pg CO2eq/yr 34 .Consequently, our ratio of land-driven to GHG-driven biodiversity change compares the impacts of the centuries-long process of global agricultural land conversion to the impacts associated with just a single year of GHG emissions.'Line 174 -nice figures, it would be more clear if you add the units in all squares so one can follow.It would make it easier to understand the final unit and how it is used.
Units have now been added to all squares (please note we were not able to track changes for this.)Line 189 -190 -Where all of these data points/studies used?I suppose you filtered the PREDICTS database to select only the metrics relevant to you, and a selection of studies with primary vegetation and another land use type.Please specify how many data points were used in the end.In the SI you should also identify this per model.It would be also nice to add a map with the geographical distribution of the studies used.
This information along with the map has been added to the Supplementary Information (pages 4-6).

Line 197 -do you follow an approach presented in any other biodiversity footprint paper? If yes please refer to it in this section. If not, can you explain what are the novelties and differences in your approach?
Thank you -reading back we realise we did not make the novelty of our study sufficiently clear.We have now updated the introduction to emphasise the novelty (lines 77-92 (86-102)): 'We build on these prior analyses, introducing three novel aspects.(i) We calculate the biodiversity impacts of agricultural land use and GHG-emission footprints using models that directly output metrics of terrestrial biodiversity change in the same units, allowing the drivers' impacts to be compared and splitting emissions into carbon dioxide (CO2), methane (CH4) and nitrous dioxide (N2O).(ii) We consider change in local rarity-weighted species richness relative to an unimpacted baseline in addition to local species richness.Species richness, although easy to measure, captures only one of the many dimensions of biodiversity, and does not always decline with global biodiversity loss 33 .Rarity-weighted richness gives greater weight to species with small geographic range size (range size correlates with species extinction risks 34 ) and so declines if rare species are replaced by more common ones.(iii) We use biodiversity models that allow us for the first time to capture regional variation in the sensitivity of biodiversity both to land-use differences and to climate change 31 .We base our biodiversity metrics on local measures of biodiversity averaged across the relevant agricultural areas as opposed to a value averaged across an entire exporting region, meaning that we better account for the wide variation in species richness that occurs within regions.
Nevertheless, there will still likely be substantial variation in biodiversity responses within our agricultural aggregations.'Line 216 -not clear why and how you determine the amount of area needed to produce €1M of product.
We take the area directly from EXIOBASE and have now made this clear in the text (lines 233-234 (251-252)): 'where  , is the area of agricultural land used to produce €1M of product in land-use type i in region k, as given by EXIOBASE' €1M is the unit of production, as explained in lines 127-129 (141-143): 'f is the (1  ) direct intensity vector, representing the environmental pressures (e.g.area of land, mass of CO2 emissions) associated with one unit (€1M) of production for each product sector in each region.' Line 221 -I don't understand the units.Normally, the characterization factor represents the impact per unit of pressure (species lost/km2 of land use).If the unit of the CF is species x km2 when you multiply it by km2 (from EXIOBASE) a different unit is achieved.This needs to be clarified.
We made our biodiversity footprint in equivalent units to those in the satellite table so it is the biodiversity cost of producing 1 Million Euros of product and already incorporates the land area cost.We have now made this clearer in the text (lines 230-242 (248-260)): "The characterisation factor CFi,k was then calculated as:  , = ∆ , ×  , (Equation 4) where  , is the area of agricultural land used to produce €1M of product in land-use type i in region k, as given by EXIOBASE).….
The characterisation factors for species richness have units of number of species × km 2 and can be thought of as the count of the species lost, with this loss extending over the area of land required to produce €1M of product, as given in EXIOBASE" Line 228 -Why final demand?Shouldn't you compute the emissions associated with production activities so that it is compatible with the environmental extension.
You are absolutely right -by 'final demand' we meant consumption but actually the first step is indeed to calculate the production (which is what we did).We have corrected the text, changing 'final demand' to 'production' (line 247 (265)).
Line 238 -You choose to use Global Temperature Change Potential instead of the most commonly used approach using Global Warming Potential.Why is that the case?Please justify why you do not use GWP.I would also recommend calculating the impacts using GWP for two main reasons, first is that it is the most commonly used metric, which would increase the comparability of your results and second as a sensitivity analysis.
We have added an explanation for our choice of GTP (lines 257-264 (276-283)): 'GHG-induced warming is often described in terms of the Global Warming Potential (GWP), a metric which compares the radiative forcing integrated over a time period caused by the emission of 1 kg of an agent relative to the integrated forcing caused by the emissions of 1 kg of CO2 50 .However, GWP has been criticised as it does not translate into a climate response that is intuitively understood 51 .
We therefore calculated the increase in global surface temperature due to GHG emissions using the Global Temperature Change Potential (GTP), a metric designed to be an intuitive measure of climate response 51 and one that has been used in previous biodiversity footprint studies 29 .' We do not think that adding a comparison using GWP would add value, because GWP has a unit of W m -2 yr kg -1 and does not directly translate into a well-known climate response, unlike GTP with its unit of deg kg -1 .Moreover, previous studies estimating biodiversity footprints associated with GHG emissions have also used GTP (e.g.Wilting et al 2017).
Line 241 -The choice of 20 years period should be better justified.If methane accounts for only a small percentage of the total gases I would argue that you should choose a different time horizon.
Methane's contribution to GTP is 80 times that of CO2 over a 20 year horizon and, as we now show, it constitutes 70% of the GHG-driven footprint so it is important to account for its effects in shortterm warming.A 100 year time horizon was originally chosen arbitrarily (Shine 1990) and does not marry well with the short-termism of most political decisions.We have added more justification for our choice of 20 years (lines 266-270 (285-289)): 'Studies often use a time horizon of 100 years following the Kyoto Protocol, but this choice was originally made on an arbitrary basis and is not the most appropriate for shorter term continental climate responses as it masks methane's potency 11,52 .Instead we chose a 20-year time horizon, to capture warming due to the relatively short-lived methane emissions, and to represent a time that is tangible to today's policy and decision-makers.'We have added information about the species distribution modelling algorithms we used to the SI.The section that we changed now reads as follows (page 2-3): 'Projected changes in the distributions of species' (and thus changes in local species richness across terrestrial areas) as a result of climate change were derived from 10 .The response of species' distributions to climate was captured using five different species distribution modelling algorithms (BIOCLIM, DOMAIN, Maxent, Random Forests and Generalised Linear Models).These models each related species' observed distributions according to the IUCN Red List 11 and Birdlife International 12 , to four climatic variables shown in previous studies to be strong correlates of animal distributions: minimum temperature of the coldest month, total annual precipitation, growing degree days and water balance, derived from the Worldclim Version 1.4 database 13 , which captures average climatic conditions for the period .BIOCLIM fits relationships between distribution records and climatic variables using a bounding-box approach in niche space, DOMAIN by comparing the climatic similarity between observed occurrence points and potentially inhabitable areas, random forests using a machine-learning approach to identify climatic patterns in species' occurrence records, while generalized linear models and Maxent use classical parametric statistics or a maximum-entropy approach, respectively, to fit linear and quadratic relationships between species' occurrences and the climatic variables.'Line 265 -266 -you mentioned before emissions from final demand, but here you mention emissions from food production of each product p in each region k.It is not clear what emission were used, please clarify.
Thanks, indeed we have now corrected this and changed 'final demand' to 'production'.
Line 269 -please provide more information on the relationship used.
We have added the following sentence (lines 299-301 (318-320) 'The characterisation factors for species richness have units of number of species × km 2 and can be thought of as the count of the species lost, with this loss extending over the area of land required to produce €1M of product, as given in EXIOBASE' Line 285 -I would say that considering land use as a one-off impact is not compatible with the analysis of the economic flows of one year.
We appreciate we have caused some confusion here.Let's take an example of paddy rice.Let's say an area of forest is converted to paddy rice in the year 2000.If rice is grown in this same area, year on year, the cost to biodiversity from land use is assumed not to increase after the initial conversion of that land to rice paddy.However, the cost from GHG emissions will increase year on year as emissions are produced annually.In our analysis, we are calculating the impacts of food production on biodiversity in 2011 so we are using the land-driven costs of all land that was used in agriculture in 2011 and the GHG costs of all emissions produced in 2011.
To clarify potential confusion due to our term 'one-off cost', we have amended the paragraph accordingly (lines 317-324 (336-344)): 'We calculate the biodiversity change associated with all land used in food production in 2011, regardless of the year of conversion.Our measure of GHG-driven biodiversity change is associated with emissions produced in 2011.We consider the impact of land use to be reversible and view land conversion as a 'one-off' cost, i.e. once land is converted biodiversity change is immediate and does not increase through time.In contrast we view GHG emissions as irreversible, repeated annual costs that occur 20 years after emission.We would expect the global biodiversity loss caused by a single year of emissions to be much lower than that caused by the conversion of the total amount of agricultural land used in 2011.' Line 343 -Does this mean that the countries in the top 10 positions are the same for all the different metrics analyzed?
No -some regions are in the top 10 for one footprint but not another.To clarify, for each relevant figure legend, we have changed it to "Regions which are in the highest ten for one or more footprints are shown." Line 370 -372 -Why does a ratio of 100 indicates that reducing GHG biodiversity footprint is a priority?
Following the correction of our code, the ratios have lowered.We have re-written and clarified this sentence accordingly (lines 362-369 (386-393)): 'The ratio of land-driven to GHG-driven biodiversity loss varied by region from 16 for rarity-weighted richness production footprint in Russia to 855 for production in RoW C&S America, with several regions, including China, India and RoW Asia, having ratios around 50. Finding ratios of 50 or lowers is concerning as it shows that direct emissions from a single year of a region's food production will cause biodiversity loss equivalent to 2% or more of the biodiversity loss caused by that region's total historic land use.Furthermore, we substantially underestimate biodiversity losses from GHG emissions since our analysis does not include emissions from land clearance.'Line 390 -I wonder if this is a real result of somewhat related with uncertainties of the MRIO table.Please check this papers to better understand if this is the case for this result: https://onlinelibrary.wiley.com/doi/10.1111/jiec.12833https://journalofeconomicstructures.springeropen.com/articles/10.1186/s40008-020-0182y#Sec20 We thank the reviewer for bringing our attention to the Giljum et al paper, with which we were not familiar.The discrepancy between the estimates stemming from different MRIOs for Taiwan's material flow is very interesting.EXIOBASE and ICIO's estimates are over double that of Eora's.Obviously at least one of the MRIOs is giving an inaccurate estimate but it's impossible to say which one.
The footprints calculated in the Giljum et al paper are consumption-based footprints per capita whereas the footprint we are commenting on is a production-based footprint per area.However, as we show in Figure 2, regions' production and consumption-based footprints tend to be similar so it seems fair to assume that if Taiwan's consumption-based footprint has been over-estimated then so has its production-based footprint.
We have added the following bracketed phrase to the end of the results paragraph (lines 448 (482-483)) '(although see Giljum et al 2019 regarding possible over-estimation of Taiwan's material footprint).' We now also similarly refer the reader to Giljum et al 2019 in the Discussion (lines 680-681 (735-736)) Line 418 -420 -I would expect Brazil to have a high per capita rarity-weighted footprint.Why is that not the case?Can you please explain further?
Brazil has the 6 th highest per capita RWR footprint so it's not low, we are just pointing out that it is lower than RoW C & S America's.Brazil is consuming products which on average have a higher species richness footprint than RWR footprint.If you look at Supplementary Figure 8 you can see that Brazil's per capita consumption-based species richness footprint is much higher than RoW C&S America's for cattle, processed cattle, dairy, vegetable oils and processed sugar.
We have added the following bracketed sentence to explain this point (lines 474-475 (509-511): '(Brazil has particularly high per-capita species richness footprints relative to RoW CS America for cattle, processed cattle, vegetable oils and dairy, see Supplementary Figure 8).' Line 470 -471 -If I am not mistaken in all figures for all the footprint the top 10 or 5 countries with highest footprints are the same, irrespective of the type of footprint.If this is true, doesn't it make it less important which footprint to use?
The regions with the highest footprints are not the same for each metric, hence why there are more than 10 regions in the figures (i.e., in all panels, we show all metrics that were in the top 10 for any of the metrics).We clarified the captions to make this clearer: "Regions which are in the highest ten for one or more footprints are shown." Line 497 -You don't mention the following studies: https://www.sciencedirect.com/science/article/abs/pii/S1470160X19302687https://www.sciencedirect.com/science/article/abs/pii/S1470160X19302687https://conbio.onlinelibrary.wiley.com/doi/full/10.1111/con4.12321 Thank you for pointing this out -we were familiar with the work of Marquardt et al 2019 and it was an oversight in not citing this work.We cannot compare our results to Marquardt et al's since their study covers all products and not just food.However, we do now cite their work (line 602 ( 643)) when explaining that our discrepancies with Chaudhary and Kastner are likely in part due to differences in biodiversity metrics.
Similarly, much of Kitzes et al's study is based on total goods, not just food related products.However, they do one analysis that breaks goods down by product so we have included this in our comparison section (lines 592-594 (632-634)) ' Kitzes et al. (2017), using metrics based on birds and a greater spatial disaggregation than our study, also find particularly high impacts of bovine products and, in contrast to our results, of processed Thank you for pointing this out.We had explained that differences between analyses are likely to result, at least in part, from different trade models (lines 598-601 (638-642)) but we have now explained that Chaudhary and Kastner used bilateral trade data (lines 579-580 (617-618)).
'Chaudhary and Kastner 16 use bilateral trade data in combination with a metric of the number of species committed to extinction.' Line 519 -%21 -I don't think you can state this without a proper analysis.Also check the literature on the differences of the results obtained with physical and monetary approaches.For example: https://www.sciencedirect.com/science/article/abs/pii/S0921800913003583https://www.sciencedirect.com/science/article/abs/pii/S0921800915000932 We have now cited Marquardt et al 2019 who show that using the PREDICTS data (which we use, in our case, making the important addition of allowing for biome and region sensitivity) versus using the PDF metric, which Chaudhary and Kastner use, does lead to different results (lines 74-76 (84-85).We therefore feel it is fair to suggest that the different biodiversity metrics that we use are likely to account for some of the differences in results.However, we are very aware of how easy it is for differences to arise solely due to trade models and do also mention that trade models will also account for some of the differences in the results.We now cite Kastner et al 2014 andBruckner et al 2015 (lines 598-599 (638-639)).We have revised the sentence as follows and moved it to the end of the paragraph so that it refers to the differences between all the studies we mention: lines 579-580 (617-618)).
'The discrepancies between studies will in part result from differences in trade models 61,62 but are also likely to result from differences in the biodiversity metrics used 28 , adding support for ours and Marquardt et al.'s 28 findings that different biodiversity metrics lead to different conclusions.'Line 530 -536 -See my previous comments on the methods.I suggest you follow the same approach (at least as sensitivity) so that the comparison is possible.
Thank you.As we explain above, we do not think that adding a comparison using GWP would add value.Previous studies (e.g., Wilting et al., 2019) have also used GTP, as we do.More fundamentally, GWP has a unit of W m -2 yr kg -1 and does not directly translate into a well-known climate response unlike GTP with its unit of deg kg -1 .Wilting et al 2017 use integrated GTP (which is essentially calculating the warming due to sustained emissions over 100 years as opposed to a single pulse of emissions from one year.) We don't feel that integrated GTP (as used by Wilting) is the appropriate measure to use in our analysis (we are interested in the impact of a single pulse of emissions, not sustained emissions over a period of time) but were we to use it, as we explain in the discussion (lines 608-609 (651-652)), we would expect the values to be higher than ours.It has hard to estimate exactly how much higherintegrated GTP over a 20 year horizon would be approximately 100 times our measure since it is simply GTP (which we use) integrated over a period of 100 years.Wilting et al used a time horizon of 100 years meaning their measure will not have captured all of the warming from methane and so we would expect their values to be less than 100 times our measure, which indeed they are.
Line 566 -I think you cannot really make this statement since you focus only on food products.
This was careless wording on our part -we have changed the statement to 'different decisions with respect to the sustainable trade of food products' (line 635 (679)).

Reviewer #2 (Remarks to the Author):
I reviewed the analysis title "Impacts of global food supply on biodiversity via land use and climate change".The submitted manuscript investigates new methods by combining different datasets and using an EEMRIO to calculate land-and GHG-driven biodiversity footprints globally.Overall, the analysis is interesting, novel, and well-structured, however, there are a couple of issues and open questions that need to be addressed.Line 46: But how climate change affects biodiversity globally is regionally still different, right?! In some regions with average colder temperatures, like Canada or Siberia, there might be more biodiversity evolving due to climate change?! Yes, absolutely.This is a really good point -in some locations species richness is actually increasing due to climate change.In terms of the structure of the paper we think it's best to keep things general at this early stage in the introduction but we draw attention to the regional differences in climate change further on (lines 283-285 (302-304)).With regard to line 46 we have changed 'biodiversity loss' to 'biodiversity' since as you point out, it is not always a loss.We have also changed 'biodiversity loss' to 'biodiversity change' (line 243 (261)) and made it clear we allow for non-uniformity of warming across the globe (lines 283-285 (302-304)).
Line 49: Either here or later at line 89 you should differentiate and describe production vs consumption footprints.
Thank you for suggesting this.At line 89 (now lines 95-98 (105-108)) we have added 'Production-based footprints are based on the total impacts associated with the products produced within a region, whereas consumption-based footprints are the total impacts associated with the products consumed within that region.'Lines 86-88: For clarification, it would be helpful for the reader if you could state that your data are building on spatially explicit information but is aggregated and averaged to reflect heterogeneity but, in the end, do not represent spatially explicit information any more.
We have changed the sentence accordingly Iines 88-92 (98-102): 'We base our biodiversity metrics on local measures of biodiversity averaged across the relevant agricultural areas as opposed to a value averaged across an entire exporting region, meaning that we better account for the wide variation in species richness that occurs within regions.
Nevertheless, there will still likely be substantial variation in biodiversity responses within our agricultural aggregations.'Lines 95-97: Do you refer to both production and consumption footprints here In the majority of cases, yes, we are referring to both production and consumption footprints here.
We have now made it clear where we refer to production or consumption footprints only (lines 103-105 (113-115)).
'We also explore biodiversity footprints per km 2 (production) and per capita (consumption) for each region and look at the proportion of regions' consumption footprints that are imported.'Line 117: When using Y from EXIBOASE the matrix is 7 x i, right?Because there are 7 different final demand categories.However, not sure if you need to mention that when you aggregate into one final demand category for each region.You could say "associated with the final demand of households, governments, capital formation, … of each region."aThank you -this is a really good suggestion.We have amended the sentence to (lines 134-136 (146-148) 'Y is the ( × ) matrix of final demand (associated with households, non-profit organisations, governments, capital formation, changes in inventories and valuables and exports) given in monetary terms (€1M).' Line 130: "spare vector populated in the entries for …" --Sounds a bit odd.It shows which sectors/production activities require direct land-use.The same allocation principle applies to GHGs, right?
There is a difference between the vectors for land use and GHGs.GHGs are associated with production in all sectors but there are only values of land use impacts for sectors which directly involve cropland or pasture e.g.values for paddy rice but not for processed rice.Hence the use of https://onlinelibrary.wiley.com/doi/10.1111/jiec.12833https://journalofeconomicstructures.springeropen.com/articles/10.1186/s40008-020-0182y#Sec20Line 418 -420 -I would expect Brazil to have a high per capita rarity-weighted footprint.Why is that not the case?Can you please explain further?Line 470 -471 -If I am not mistaken in all figures for all the footprint the top 10 or 5 countries with highest footprints are the same, irrespective of the type of footprint.If this is true, doesn't it make it less important which footprint to use? Line 497 -You don't mention the following studies: https://www.sciencedirect.com/science/article/abs/pii/S1470160X19302687https://www.sciencedirect.com/science/article/abs/pii/S1470160X19302687https://conbio.onlinelibrary.wiley.com/doi/full/10.1111/con4.12321Line 512 -518 -These conclusions are misleading since Chaudhary and Kastner did not use an MRIO model which might have more influence on the results than the biodiversity metric used.Please revise this.

Line 46 :
But how climate change affects biodiversity globally is regionally still different, right?! In some regions with average colder temperatures, like Canada or Siberia, there might be more biodiversity evolving due to climate change?! Line 49: Either here or later at line 89 you should differentiate and describe production vs consumption footprints.

Line 251 -
is the use of RCPs consistent with the use of GTP? Yes, it is.GTP gives us the expected temperature increase associated with each unit of product.The projections of species distributions under the RCP scenarios allow an estimation of the biodiversity change associated with that unit of product by putting temperature change in context with wider climate variables.We have now clarified this in the text (lines 278-281 (297-300)): 'We estimated the sensitivity of biodiversity to climate change based on future projections of changes in the distributions of terrestrial vertebrates under the Representative Concentration Pathways (RCP) climate-change scenarios 5 , thus putting our projected temperature change in context with wider climate variables.'Line 255 -In the SI file you mention that "Species' projected responses to climate change were derived from10".What was the exact relationship?Can you add this information in SI?
): 'This grid is based on projected changes in the distributions of species under climate change, and describes the expected change in local species richness in any terrestrial location associated with a temperature increase of 1°C.' Line 272 -please clarify the unit (see comment referring to line 221) See earlier comment: The unit is species * km^2 and there is no need to further multiply it by area.Lines 238-242 (256-260): conclusions are misleading since Chaudhary and Kastner did not use an MRIO model which might have more influence on the results than the biodiversity metric used.Please revise this.