year <- 2017
irr  <- 90   # mm/y of irrigation 
tocc_m <- 0.34 # occupation time based on planting (+/- 0.04 y)
tocc_l <- 0.30
tocc_u <- 0.38
save_loc <- "C:/Users/michael.lathuilliere/Desktop/Paper/"

Purpose of this document

This document summarizes the results to be published in a manuscript describing the commodity supply mix (CSM) in life cycle assessment (LCA) based on Trase results for Brazil soy (v.2.5.0). The CSM is derived for land occupation (FU = 1 tonne of soybean harvested and exported) for soy destined to China (Mainland), the EU, France, the rest of the world (RoW), and Brazil (domestic consumption) in 2017. Results were generated using 2 classification schemes of municipalities of origin into respective ecoregions and river basins:

There were 7 % of “unknown trade flows” meaning that we were unable to identify the municipality of production for export (and hence, the ecoregion, or river basin). In cases of “unknown” flows, we propose to:

Trade flows with unknown municipality of production represent 7% of total export in 2017.

Note: the results are only for soybean, not soy cake or oil, so expect the results to be different from what can bee seen on trase.earth.

Step-by-step procedure to derive the CSM

  1. Derive the total exports of each trade partner (China, EU, France, Brazil, RoW) in tonnes (from the export data) and hectares (using yield at the municipality level from which soybean was sourced)
  2. Derive the total exports of each trade partner per ecoregions, in tonnes and hectares (following 1.)
  3. Obtain hectares per tonne for each destination country and source ecoregion (from 2.) - the result is in hectares(ecoregion) per tonne(ecoregion)
  4. Obtain the percentage of soy exported by each ecoregion for each country of destination - the result is tonne(ecoregion) per tonne(export)
  5. Derive the portion of ecoregion per tonne of soy exports multiplying 3. and 4. to obtain the life cycle inventory of land occupation (converting ha into m2)

The above is also carried out for Brazil-wide soybean production considering the CSM for the overall municipalities in Brazil and the corresponding characterization factors (see below).

The total exports of soybean in 2017 was 98.855 Mtonnes in 2017.

Results

Land use

Life cycle inventory of land occupation (LCI_OCC)

First, we determine the LCI and potential biodiversity impacts of the Brazilian soybean production mix which considers Brazilian soybean production in each municipality and assume equal probability of sourcing. This result is what could have been derived by an analyst without knowing anything about trade to carry out the LCA (so not specific to the respective consumption centers). This calculation is carried out as follows:

  • Obtain a production mix considering the tonnes of soybean produced in each municipality compared to total production in 2017 (assumes equal probability of sourcing, based on production data)
  • Obtain the inverse yield of soybean production in each of Brazil’s municipalities to derive the LCI
  • Apply the characterization factors of Chaudhary and Brooks (2018) based on either Majority soy or Majority area classification rules.

The values were also used to fill in the trade flows that were not identified as having a source ecoregion (labeled as “Unknown”).

We obtain impacts to biodiversity (note values should be the same under majority soy or area):

  • Majority soy LCI = 1006 m2 y tonne-1 and biodiversity impact of 2.08e-10 PDF y tonne-1
  • Majority area LCI = 1006 m2 y tonne-1 and biodiversity impact of 2.11e-10 PDF y tonne-1

Average conditions in Brazil: 3076.9230769 m2 tonne-1 (from IBGE) and impact to biodiversity of 2.55^{-10} PDF y tonne-1 obtained using the country weighted characterization factor from Chaudhary and Brooks (2018).

We obtain impacts to water scarcity (note values should be the same under Majority soy or Majority area classification):

  • Majority soy WFI = 266 m3 tonne-1 and water scarcity footprint of 292 m3 tonne-1
  • Majority area WFI = 266 m3 tonne-1 and water scarcity footprint of 292 m3 tonne-1

Average conditions in Brazil: 304 m3 tonne-1 and a water scarcity footprint of 807 m3 ton-1 obtained using the country weighted characterization factor from Boulay et al (2018).

Results for “majority soy” case LCI are shown in the table below:

And differences in life cycle inventory from “majority soy” and “majority area” classifications:

The values of the LCI of land occupation for each country (considering the range of tocc) are:

  • Brazil: 884 – 1120 m2 y
  • China (Mainland): 885 – 1120 m2 y
  • EU: 929 – 1176 m2 y
  • France: 919 – 1164 m2 y
  • RoW: 900 – 1140 m2 y

Biodiversity impact using characterization factors from Chaudhary and Brooks (2018)

Note: Graphs show the “Majority soy” results only, data also available for “Majority area” (used in uncertainty analysis).

At this point, we are using the Crop Intensive characterization factors from Chaudhary and Brooks (2018).

We also compare the results against the Brazil soybean production mix results which are based on production data in municipalities (weighted by production volume) as they relate to individual ecoregions and river basins (see dashed line below). Results are the sum of the LCI and biodiversity damage using the mean characterization factors:

## # A tibble: 5 x 4
##   COUNTRY  BIO_L95 BIO_MEAN  BIO_U95
##   <chr>      <dbl>    <dbl>    <dbl>
## 1 BRAZIL  1.73e-10 2.28e-10 3.03e-10
## 2 CHINA   1.58e-10 2.10e-10 2.80e-10
## 3 EU      1.10e-10 1.48e-10 2.01e-10
## 4 FRANCE  1.20e-10 1.61e-10 2.18e-10
## 5 RoW     1.43e-10 1.91e-10 2.56e-10

A reviewer suggested that we compare the results between CSM and production mix, aggregated by ecoregion and river basin. With respect to the above figure, this means taking the weighted average for each coloured bar and comparing against the production mix.

The results cannot be the same because the weighting is different in both calculations:

  • In the production mix, weighting is based on municipal production
  • In the CSM, weighting is based on the municipal export with respect to total export to market

As an example we can show both results for the Auracaria Moist Forest ecoregion:

  • Production mix gives an LCI of 133 m2 tonne-1 and impact score of 6.5^{-11} PDF m2 tonne-1
  • CSM for Brazilian market gives an LCI of 164 m2 tonne-1 and impact score of 7.99^{-11} PDF m2 tonne-1

Water use

Water scarcity footprint using characterization factors from Boulay et al (2018)

We assume 90 mm of irrigation applied in the first month of the soybean development cycle and calculate the water scarcity footprint following Boulay et al (2018) using river basin specific AWARE factors.

The water footprint inventory (WFI) is the amount of water (m3) required per tonne of soybean exported from a river basin and is calculated as:

  • WFI = 90 mm x 0.001 (m/mm) x 10,000 (m^2/ha) x (1/yield) (h tonne-1)

Results for the “Majority soy” case:

The values of the WFI for each country are:

  • Brazil: 265
  • China (Mainland): 265
  • EU: 279
  • France: 276
  • RoW: 270

And differences in life cycle inventory from “Majority soy” and “Majority area” classifications:

Then, the water scarcity footprint (WSF) is obtained by multiplying the WFI by the river basin-specific Aware factors, weighted according to the share of each river basin contributing to the total exports to markets (China (Mainland), the EU and France, RoW and Brazil).

We use the AWARE_AGRI factor representing scarcity in the event of crop irrigation. Note that the name of the river basins have been aggregated for clarity in the Figure below.

Similar to the biodiversity damage we compare results against a the results obtained from Brazili production mix average water footprint inventory (dashed line):

## # A tibble: 5 x 2
##   COUNTRY   WSF
##   <chr>   <dbl>
## 1 BRAZIL   250.
## 2 CHINA    301.
## 3 EU       421.
## 4 FRANCE   259.
## 5 RoW      329.
## # A tibble: 5 x 2
##   COUNTRY   WSF
##   <chr>   <dbl>
## 1 BRAZIL   248.
## 2 CHINA    301.
## 3 EU       406.
## 4 FRANCE   261.
## 5 RoW      335.

Uncertainty analysis

We consider 3 different sources of uncertainty in the presented results:

  • The uncertainty in the attribution of the characterization factors based on the classification of each municipality into distinct ecoregions/river basins (Majority soy vs Majority area);
  • The uncertainty in the impact based on the 95% confidence interval for each characterization factor (specific to biodiversity impacts);
  • The uncertainty in the LCI and, by extension, the biodiversity impacts based on the occupation time windows based on soybean planting practices in Brazil (Flach et al 2020)

Spatial variability and probability density

We derive 5 graphs to show changes in impact assessment results based on the weighting decisions (probability). These results are plotted in sequence to identify the effects of the modeling decisions of a LCA Analyst for potential biodiversity impacts and water scarcity footprint. The graphs are described as such in the manuscript:

  1. Production: (P) Impact scores are derived assuming an equal probability of soybean supply to markets. In mathematical terms, the values of the LCI are obtained in each of the Brazilian municipalities and then divided by the number of municipalities (2275). This case can be interpreted as a typical option available for an analyst who would only have information on Brazilian soybean production.
  2. Production mix: (PM) The probability of impact scores is distributed according to the contribution of each municipality to total Brazilian soybean production. This case is similar to the P case, but for which a weighing factor accounts for the variability in soybean production across Brazil, assuming that markets have a greater probability of supplying soybean in municipalities with greater production. This can also be interpreted as a typical option available for an analyst seeking to include more information in the LCI.
  3. Production mix to market: (PMM) is the same at the PM case, but is augmented by the information provided by the trade connections between municipalities of soybean production and each market. In other words, the impact scores are obtained using information on municipal soybean production, but reduced to those municipalities identified as suppliers of each of the markets. This case can be interpreted as an option resulting from in-country research or specific supply chain research carried out for the LCA.
  4. Supply mix (CSM): (CSM) results identified above using the commodity supply mix.
  5. Consumption boundary: (B) for which the probability of impact scores is derived assuming an equal probability of sourcing from the ecoregion or river basin boundaries identified in Trase. This case can be interpreted as a option resulting from general tendencies of supply focused specifically on ecoregion or river basin boundaries without additional knowledge on the amount of soybean sourced from these geographic boundaries.

These graphs are shown in sequence (from left to right) for potential biodiversity damage (green) and water scarcity footprint (blue).Note: the code only shows one country at a time. Results are also only shown for the Majority soy classification case.

Weighted average impact scores considering uncertainty in LCI and characterization factors

Uncertainty from time of occupation:

Flach et al (2020) derive a short (110 days), medium (125 days) and long (140 days) soybean cycle based on Abraham and Costa (2018), we use those 3 windows as possible planting dates for soybean harvested and exported in 2017 for the purposes of estimating uncertainty:

  • Short planting season tocc = 110/365 = 0.30 y
  • Medium planting season tocc = 125/365 = 0.35 y
  • Long planting season tocc = 140/365 = 0.38 y

We use a mean tocc of 0.34 y (+/- 0.04 y). The differences in occupation time affect the LCI.

The differences in biodiversity impact assessment results include the change in occupation time (tocc):

  • Biodiversity impact at L95 includes tocc of 0.30 y
  • Biodiversity impact at U95 includes tocc of 0.38 y

Majority soy

  • Brazil: lower: 1.73e-10 PDF y tonne-1, mid: 2.28e-10 PDF y tonne-1, upper: 3.03e-10 PDF y tonne-1
  • China: lower: 1.58e-10 PDF y tonne-1, mid 2.1e-10 PDF y tonne-1, upper: 2.8e-10 PDF y tonne-1
  • EU: lower: 1.1e-10 PDF y tonne-1, mid: 1.48e-10 PDF y tonne-1, upper: 2.01e-10 PDF y tonne-1
  • France: lower: 1.2e-10 PDF y tonne-1, mid: 1.61e-10 PDF y tonne-1, upper: 2.18e-10 PDF y tonne-1
  • RoW: lower: 1.43e-10 PDF y tonne-1, mid: 1.91e-10 PDF y tonne-1, upper: 2.56e-10 PDF y tonne-1

Majority area

  • Brazil: lower: 1.77e-10 PDF y tonne-1, mid: 2.35e-10 PDF y tonne-1, upper: 3.12e-10 PDF y tonne-1
  • China: lower: 1.6e-10 PDF y tonne-1, mid 2.13e-10 PDF y tonne-1, upper: 2.84e-10 PDF y tonne-1
  • EU: lower: 1.1e-10 PDF y tonne-1, mid: 1.48e-10 PDF y tonne-1, upper: 2.01e-10 PDF y tonne-1
  • France: lower: 1.2e-10 PDF y tonne-1, mid: 1.61e-10 PDF y tonne-1, upper: 2.18e-10 PDF y tonne-1
  • RoW: lower: 1.42e-10 PDF y tonne-1, mid: 1.89e-10 PDF y tonne-1, upper: 2.54e-10 PDF y tonne-1

We then calculate the mean results of the spatial variability for all countries, using the mean characterization factors from Chaudhary and Brooks (2018), the mean occupation time from Flach et al (2020) together with their standard deviation and 95% confidence interval (CI):

Majority soy only

  • Brazil: 2.61e-13 PDF y tonne-1 (sd = 4.11e-13, 95% CI = 2.1e-14)
  • China: 1.61e-13 PDF y tonne-1 (sd = 7.82e-13, 95% CI = 4.24e-14)
  • EU: 2.33e-13 PDF y tonne-1 (sd = 8.5e-13, 95% CI = 6.59e-14)
  • France: 4.14e-12 PDF y tonne-1 (sd = 6.14e-12, 95% CI = 1.93e-12)
  • RoW: 1.86e-13 PDF y tonne-1 (sd = 7.08e-13, 95% CI = 4.33e-14)

Similarly for the water scarcity footprints:

  • Brazil: 2.86e-01 m3 tonne-1 (sd = 1.16e+00, 95% CI = 5.93e-02)
  • China: 2.31e-01 m3 tonne-1 (sd = 1.33e+00, 95% CI = 7.23e-02)
  • EU: 6.62e-01 m3 tonne-1 (sd = 3.33e+00, 95% CI = 2.58e-01)
  • France: 6.65e+00 m3 tonne-1 (sd = 1.85e+01, 95% CI = 5.81e+00)
  • RoW: 3.21e-01 m3 tonne-1 (sd = 1.64e+00, 95% CI = 1e-01)

Uncertainty from characterization factors and classification of municipalities into ecoregions

We now want to check the differences in results based on the choices made to calculate biodiversity damage, using the upper and lower 95% values of the characterization factors from Chaudhary and Brooks (2018) combined to the shorter and longer occupation times from Flach et al (2020). The results are as follows:

Majority soy

Lower estimate:

  • Brazil: 1.97e-13 PDF y tonne-1 (sd = 3.11e-13, 95% CI = 1.59e-14)
  • China: 1.21e-13 PDF y tonne-1 (sd = 5.88e-13, 95% CI = 3.19e-14)
  • EU: 1.73e-13 PDF y tonne-1 (sd = 6.29e-13, 95% CI = Inf)
  • France: 3.08e-12 PDF y tonne-1 (sd = 4.61e-12, 95% CI = 1.45e-12)
  • RoW: 1.39e-13 PDF y tonne-1 (sd = 5.31e-13, 95% CI = 3.25e-14)

Upper estimate:

  • Brazil: 3.46e-13 PDF y tonne-1 (sd = 5.45e-13, 95% CI = 2.79e-14)
  • China: 2.15e-13 PDF y tonne-1 (sd = 1.04e-12, 95% CI = 5.67e-14)
  • EU: 3.17e-13 PDF y tonne-1 (sd = 1.16e-12, 95% CI = Inf)
  • France: 5.59e-12 PDF y tonne-1 (sd = 8.22e-12, 95% CI = 2.58e-12)
  • RoW: 2.49e-13 PDF y tonne-1 (sd = 9.47e-13, 95% CI = 5.79e-14)

Majority area:

Lower estimate:

  • Brazil: 2.03e-13 PDF y tonne-1 (sd = 3.31e-13, 95% CI = 1.7e-14)
  • China: 1.23e-13 PDF y tonne-1 (sd = 5.95e-13, 95% CI = 3.23e-14)
  • EU: 1.73e-13 PDF y tonne-1 (sd = 6.32e-13, 95% CI = Inf)
  • France: 3.09e-12 PDF y tonne-1 (sd = 4.62e-12, 95% CI = 1.45e-12)
  • RoW: 1.38e-13 PDF y tonne-1 (sd = 5.33e-13, 95% CI = 3.26e-14)

Upper estimate:

  • Brazil: 3.56e-13 PDF y tonne-1 (sd = 5.9e-13, 95% CI = 3.02e-14)
  • China: 2.18e-13 PDF y tonne-1 (sd = 1.06e-12, 95% CI = 5.73e-14)
  • EU: 3.17e-13 PDF y tonne-1 (sd = 1.16e-12, 95% CI = Inf)
  • France: 5.59e-12 PDF y tonne-1 (sd = 8.24e-12, 95% CI = 2.58e-12)
  • RoW: 2.48e-13 PDF y tonne-1 (sd = 9.49e-13, 95% CI = 5.8e-14)

We then test differences between the distribution of mean biodiversity damage and water scarcity footprints following the majority soy and majority area classification of municipalities into ecoregions and river basins:

For biodiversity damage:

  • Brazil: means are different with p-value = 0.7242838
  • China: means are different with p-value = 0.9416464
  • EU: means are different with p-value = 0.9989129
  • France: means are different with p-value = 0.9993137
  • RoW: means are different with p-value = 0.96946

For water scarcity footprint:

  • Brazil: means are different with p-value = 0.9653209
  • China: means are different with p-value = 0.9986124
  • EU: means are different with p-value = 0.8994861
  • France: means are different with p-value = 0.9939953
  • RoW: means are different with p-value = 0.9445036

Check embedding process (QA)

## [1] "QA CF_OCC Majority soy: ALL PASS"
## [1] "QA CF_OCC Majority area: ALL PASS"
## [1] "QA AWARE Agri Majority soy: ALL PASS"
## [1] "QA AWARE Agri Majority area: ALL PASS"

Map land occupation life cycle inventory per ecoregion and country of destination

### Map biodiversity impact per ecoregion and country of destination

Map water footprint inventory per river basin and country of destination

Note: River basin names are missing at this point (only IDs are shown)

Map water scarcity footprint per river basin and country of destination

References Cited

Abraham G and Costa MH (2018) Evolution of rain and photoperiod limitations on the soybean growing season in Brazil: The rise (and possible fall) of double-cropping systems. Agriculture and Forest Meteorology 256-257: 32-45. doi: 10.1016/j.agrformet.2018.02.031.

Boulay et al (2018) The WULCA consensus characterization model for water scarcity footprints: assessing impacts of water consumption based on available water remaining (AWARE). The International Journal of Life Cycle Assessment 23(2): 368-378, doi: 10.1007/s11367-017-1333-8

Chaudhary and Brooks (2018) Land Use Intensity-specific Global Characterization Factors to Assess Product Biodiversity Footprints. Environmental Science and Technology 52(9): 5094-5104, doi: 10.1021/acs.est.7b05570

Flach et al (2018) The effects of cropping intensity and cropland expansion of Brazilian soybean production on green water flows. Environmental Research Letters 29(27), in press. doi: 10.1088/2515-7620/ab9d04.