Introduction

We are faced with the challenge of assigning irrigation to soybean crops. The data is quite scarce and does not always match across sources. Some of the data sources we have are:

The objective is to derive the amount of blue water used to irrigate soybean and assign this amount to municipalities that report such irrigation consumption. There are a few things to keep in mind:

  1. Our estimate of soybean ET using the Rockström (2003) curve, provides the water footprint (WF) of soybean according to the yield
  2. In cases where soybean is irrigated, a portion of this WF should be blue
  3. We could estimate the amount of water consumed based on the model they use to define irrigation water use (through a calculate of water deficit) for the first 3 phases of development for the crop.

From the validation data we understand that our estimates of ET and total soy WF are likely greater than previous estimates. So we should maintain 2 estimates here: from ANA and the Rockström (2003) curve.

Ultimately we have opted for the results generated using the Rockström curve due to the non-linear relationship between WF and yield and the ability to calculate the WF from yield. Also some of the assumptions behind the ET values obtained from ANA lead to non-sensical soy ET values in cases where yields are low (which is an issue with IBGE data).

Gather irrigation data

ANA produces an Atlas de Irrigacão which provide irrigation water consumption as well as area of pivot irrigation per municipality. The area is not available for every year (e.g. 2010, 2014, 2017, 2019) and so we need to make assumption across years when looking specifically at 2014-2017.

One big issue with the ANA data is that there are some missing estimates in municipalities that produce soy, e.g.:

Formosa do Rio Preto in 2017 does not have an estimate of water use by ANA. You can check this by filter for NAs in the Deficit column when looking at the et_soy_muni_ana dataset

The table above shows differences between Mapbiomas and ANA irrigation maps, but also shows the total irrigation water consumption (which includes soybean). We therefore need to extract the portion of the irrigation water use that is for soybean production.

To do so, we could take the crop specific modeling information that ANA used to make this type of estimate. We assume that soybean could be planted as early as September and as late as November of each year with irrigation. This means that:

We then:

We could combine this information to the estimate of water use from ANA (water consumption from rainfall), but we actually took the estimates already provided by ANA below for soybean per municipality and per year.

The above method could be refined and used at a later date, it is only kept here for posteriority.

Estimating green and blue water footprints

We then try different estimates of green and blue water for soybean production in 2014-2017 testing 2 approaches:

First estimate: Rockström (2003) curve with ANA’s ratio of blue water to total consumption

We apply the ratio of “deficit” (as irrigation required for a given year) to total water consumption from ANA for soybean (per year and municipality) and apply this ratio to the soybean WF derived using the Rockström (2003) curve.

In the above estimate, there is are about 8 municipalities that have a WF that is at least 500 m3 tonne-1 larger than originally calculate; up to 2000 m3 tonnes-1. See below.

There are no issues reported in the municipal level calculations.

Second estimate: ANA’s green and blue water estimates direcly from the ANA data

We then use ANA data to estimate both the green and blue water, based on their estimates of water consumption as:

  • Effective precipitation (green water)
  • Deficit (similar as above) (blue water)

Here we take the ANA estimates in m^{3} s^{-1} and normalize them per ha of soybean produced in the municipality, before estimating the water consumption per area of either irrigated or rainfed fields and dividing by the total production of the municipality.

There is an issue here because not all municipalities that have soy in IBGE do have a water use estimate in ANA. This makes it pretty un-usable to associate with trade.

Note that we already have a soy WF (total) calculated from ANA through the soy_et_ana.R script, but here we are exclusively using the crop water balance to calculate directly the green and blue soy WF (so differences are expected).

Validation of the water use estimates

There are at least 4 datasets that we can compare our results to:

  • Results from the Water Footprint Network (1996-2005): This is a key reference, but the data range is old and does not overlap our 2015-2017 years of interest. Still interesting to compare to.
  • Results from Figueiredo et al (2021) who simulated water productivity in kg/ha.mm in 16 agro-climatic using a calibrated model (from 5 field sites across Brazil). The simulations are available from different planting dates across the year, we select those that are between day of year 200 and 330.
  • Results from Petrillo et al (2023) who calculated both green and blue WF for soy using a 5x5 arcmin gridded dataset. Results are available for the year 2004 and 2018 (we use 2018 as the closest to 2017).
  • Results from Mialyk et al (2024) who provide an update to the WFN data

Comparison with Figueireido et al

First we compare the Figueiredo et al (2021) data with our results obtained using the Rockström (2003) curves:

We can look at the linear models for the ANA data relationship which seems more in agreement with the Figueiredo et al (2021) data (assuming y = ax).

## 
## Call:
## lm(formula = fig_comparison$WP_rock ~ 0 + fig_comparison$WP_YIELD_KG_HA_MM_MEAN)
## 
## Residuals:
##     Min      1Q  Median      3Q     Max 
## -1.8671 -0.5446 -0.1505  0.5496  4.2617 
## 
## Coefficients:
##                                       Estimate Std. Error t value
## fig_comparison$WP_YIELD_KG_HA_MM_MEAN  0.47624    0.01478   32.22
##                                                  Pr(>|t|)    
## fig_comparison$WP_YIELD_KG_HA_MM_MEAN <0.0000000000000002 ***
## ---
## Signif. codes:  0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## 
## Residual standard error: 1.196 on 63 degrees of freedom
## Multiple R-squared:  0.9428, Adjusted R-squared:  0.9419 
## F-statistic:  1038 on 1 and 63 DF,  p-value: < 0.00000000000000022

The ANA data aligns better with the Figueiredo et al (2021) data, but it is also incomplete overall which is a big conundrum. The alignement with the ANA data might also be an indication of some of ANA’s assumptions.

Then we compare to the WFN data for the 1996-2005 period, even though the timelines do not overlap

Comparison with the results from the Water Footprint Network (1996-2005)

The data seems to align more with the results obtained from the Rockström approach.

Comparison with the results from De Petrillo et al (2023)

We then compare to the latest published data on the green and blue water footprint of soy in Brazil. The data is available per municipality and disaggregated, but not for the same years. We compare 2017 and 2018.

There is basically no agreement on the blue WF likely due to the information used to derive irrigation.

Comparison with the results from Mialyk et al (2024)

Our estimates are larger than the values of Mialyk et al (2024) and maybe due to a few factors worth exploring more:

  • The soy coverage is old (MIRCA2000) and does not include new extent?
  • The yield and crop calendars may be different
  • The ET is obtained using FAo56, so our estimates will be larger

Their values seem closer to the ANA estimates, but we decided not to use those due to missing information and the assumption that WF increases linearly with yield.

## [1] "saved at C:/Users/MICHAE~1/AppData/Local/Temp/RtmpcvyLSS/file4b9012cc793b/soybean_wf_disag_2014_2017.csv"

References

De Petrillo et al (2023) International corporations trading Brazilian soy are keystone actors for water stewardship Communications Earth and Environment 87, 4(1), doi: 10.1038/s43247-023-00742-4.

Figueiredo et al (2021) Impact assessment of soybean yield and water productivity in Brazil due to climate change European Journal of Agronomy 129: 126329, doi: 10.1016/j.eja.2021.126329

Mialyk et al (2024) Water footprints and crop water use of 175 individual crops for 1990-2019 simulated with a global crop model Scientific data, doi: 10.1038/s41597-024-03051-3

Water Footprint Network.