Environmental indicators

This section shortly describes the environmental indicators used in the SOL-model. For further details, we refer to the literature²⁶.

Land occupation: Land occupation measures the cropland and grassland areas utilized in agricultural production. For cropland, land occupation combines areas harvested and cropping intensities. The latter indicate how many times a hectare is harvested on average. Cropping intensities are usually less than one (due to fallow areas) and therefore land occupation reports higher values than areas harvested⁷.

In all scenarios, grassland areas are assumed to stay constant²⁶. It has to be mentioned that many grasslands and grazing lands currently face high environmental and societal pressures. Focusing global ruminant production on those areas would thus necessitate to adequately address those challenges⁵⁶.

Changes in land use are thus between arable and non-agricultural land only (e.g. forests). The indicator cropland occupation captures the total land demand in the scenarios, irrespective of where this may be sourced, while the indicator deforestation captures the pressure from this land demand on forests in countries where deforestation is an issue, assuming similar land sourcing patterns as in the baseline (cf. further down).

N-surplus: The N-surplus describes the difference between N-inputs and N-outputs. N-inputs for crops are mineral N-fertilizers, N-fixation, organic fertilizer, crop residues and seeds. N-inputs are derived based on available N (from mineral fertilizers, fixation, crop residues and manure), assigned to the various crops in relation to their relative N-demand as share of total N-demand of all crops. N from atmospheric deposition is not included in the N-surplus. Total N output of a crop equals the amount of nitrogen that is taken up by a crop during the growing period, i.e. the amount of N in yields and crop residues, as well as emissions (NO₃, NH₃ and N₂O). For organic crop activities only organic N-inputs are possible (manure, N-fixation, crop residues), as no mineral N-fertilizers are allowed. For animals, N-inputs are feed and N-outputs are yields, manure and emissions from manure (NO₃, NH₃ and N₂O). Input and output sources for N thus cover all relevant flows and compounds, in particular direct emissions, volatilization and leaching of N₂O, NH₃ and NO₃ from manure management and fertilizer application of any kind. Emission factors are according to IPCC 2006 Guidelines (Tier 1). SOL-model results for the global aggregate N-surplus and for aggregates of sub-categories, such as N-Fixation, etc. in the base year are consistent with the literature (Supplementary Table ¹)^{47, 57, 58}.

Besides the yield gap, we did not assume any systematic differences between organic and conventional livestock activities that would affect N-surplus.

N-surplus is displayed as a global per-hectare average. This thus covers all N-inputs and outputs from croplands and grasslands and takes an average over all those areas. It thus cannot be directly compared to values for cropland reported in the literature. The use of per-ha numbers illustrates our focus on assessing the viability of organic production from an agronomic point of view, as N-supply is often seen as a challenge to organic production^19–21. The choice of a surplus of between 10 kg/ha and 5 kg/ha as optimal is motivated from the literature that reports that N-surplus could be decreased by 50–70% for various cereals without affecting yields and as based on the numbers on potential N-input reduction and shares of excess N in relation to inputs, even higher reduction rates can derived^{59, 60}; the optimal range chosen in the model signifies a reduction of 60–80% with respect to the reference scenario and we thus chose this somewhat higher range of 60–80% reduction of N-surplus as an illustrative optimal level to be aimed at in the assessment of how viable changes in N-surplus in the scenarios are from an agronomic perspective.

The potentially challenging situation regarding N-supply in organic agriculture has also been taken up in the literature²¹. They suggest that this challenge could be met when cropping intensities were to increase and fallow land and intercropping were to be systematically used for legume production. We did not incorporate this in the model as it would necessitate a range of additional uncertain assumptions, such as on water availability, overall adequacy of areas for off-season legume cropping and yields. The corresponding assessment²¹ is highly contested, as they assume legumes between the main crops on all areas, resulting in additional 1360 million ha legumes, and assuming a very high nitrogen fixing rate of about 100 kg N/ha. For this, we also refer again to the critical assessment of this analysis^{19, 20}.

P-surplus: As for N, P-surplus is defined as the difference between P-inputs and outputs. P-flows are expressed as P₂O_5. Inputs and outputs are mineral P fertilizer, P₂O₅ in feed, manure, crop residues and yields. When assessing P-surplus, it has to be considered that large quantities of P are fixed in soils and the surplus thus rather expresses a ‘loss potential’, that can be realized, e.g. through erosion, than actual losses to the environment. SOL-model results for the total P-balance in the base year are consistent with literature values⁵⁸.

Non-renewable energy use: The life cycle impact assessment methodology ‘cumulative energy demand’ (CED)⁶¹ is used to calculate non-renewable energy use. Renewable energy components are disregarded. The share of non-renewable energy for fuels and electricity was assumed to stay constant in all scenarios and no technical progress in energy efficiency was assumed.

Inventory data for each activity, including the differentiation between energy use in conventional and organic activities, were taken from LCA-databases, i.e. the ecoinvent 2.0 database and other sources^62–64. Energy use is linked to farm activities and includes energy use for seeds, crop protection, fertilization, mechanization, organic fertilization, fences, stables and depots for roughage. Data for animal production were taken from the ecoinvent 2.0 database and other sources^{64, 65}. Energy use for fertilizer production is modelled specifically for the fertilizer quantities used for each crop in each country. Energy carriers inputs were modelled according to the ecoinvent 2.0 and SALCA inventories⁶⁶. Due to lack of data for trade, transportation energy use was disregarded. We emphasize that absolute numbers on energy use may be biased due to the data quality behind ecoinvent 2.0, which is partly rather old, but relative differences between scenarios are much less sensitive to such data problems.

Greenhouse gas emmissions: GHG emissions are based on Tier 1 and 2 approaches from the IPCC-Guidelines from 2006. Emissions for agricultural inputs and infrastructure are taken from the ecoinvent 2.0 database and LCA studies^62–64. Emissions from deforestation and from agriculturally managed organic soils are taken from FAOSTAT (2). We did not differentiate emission factors between organic and conventional production systems and assumed the same feeding rations and the same shares in different manure management systems for organic and conventional production systems.

For the GHGs, Global Warming Potentials (GWP) from the IPCC2006 100a Tier 1 methodology were used, i.e. 25 t CO₂e/t for CH₄ and 297 t CO₂e/t for N₂O. IPCC Tier 1 methods were used to calculate the emissions from manure management and fertilizer application. The Tier 2 methodology was used for enteric fermentation, in order to capture the impacts of different feeding regimes. ecoinvent 2.0 and other data⁶⁷ were used to calculate the GHG emissions from the production of mineral fertilizers and pesticides. GHG emissions from processes and buildings were derived from the respective CED-values and application of process-specific conversion factors derived from ecoinvent 2.0. When aggregating over the common emission categories only, SOL-model results for total GHG emissions in the base year are similar to the values reported in the literature (Supplementary Table ²)^{68, 69}. These two literature references differ substantially in the values for emissions from enteric fermentation; SOL-model results are more similar to the values based on the emissions data in FAOSTAT⁶⁸.

Water use: Water use was calculated from AQUASTAT data⁷ on consumptive irrigation water use per ton of irrigated production and data on irrigated areas for various crops and crop categories. We assumed similar irrigation values per ton of irrigated production for organic and conventional production. Differences between the systems then arise due to different yields and different area shares for different crops with different crop-specific irrigation values as reported in AQUASTAT.

Pesticide use: There is no consistent data set on pesticide use covering different countries, and we thus developed an impact assessment model for assessing pesticide use incorporating three factors: pesticide use intensity per crop j and farming system k (PUI_j,k), pesticide legislation in a country i (PL_i), and access to pesticides by farmers in a country i (AP_i) (Supplementary Table ³).

This model has been described in the supplementary information to Schader et al. (2015)²⁶ and we quote from this description in the following, as we used the same model in this work.

Each factor was rated on a scale from 0 to 3 by FAO-internal and external experts (Jan Breithaupt, FAO, involving experts from regional FAO offices; Frank Hayer, Swiss Federal Office of the Environment; Bernhard Speiser, Research Institute of Organic Agriculture, FiBL) with experience in different countries and with different methods of calculating pesticide impacts (life cycle assessment, risk assessment). The descriptors for each scale have been designed so that risks from pesticides are 0 if only one of the three model parameters is equal to 0. For instance, if there are no harmful pesticides used in a crop, or if in a country legislation completely bans harmful pesticides or farms do not have access to pesticides at all, the impact factor for a crop-country combination will be 0.

As an example, the pesticide use impact factor (IF_i,j,k) for coffee production in Ghana was 6 as: PUI was rated as 3, PL was rated as 2 and AP was rated as 1. Values for PL and AP are shown in the Supplementary Table ⁴ below, values for PUI can be found in Supplementary Table ⁵. To calculate crop and country-specific pesticide use IF, the three factors were multiplied together (Eqn ¹).

Thus, for each crop in each country, a value between 0 and 9 has been assigned on a per-hectare basis, serving as an indicative proxy for overall pesticide use per crop and country (and per-ha). Aggregate values per country were derived by multiplying this pesticide use indicator with the respective crop areas and summing over all crops. The pesticide use intensity of organic activities was evaluated to be zero throughout all activities and countries. This neglects certain aspects of plant protection in organic agriculture, such as the use of copper.

Deforestation: FAOSTAT deforestation values for the base period 2005–2009 are set in relation to the change in agricultural land areas over this period in each country. This ratio is then used to derive deforestation values from area changes in the scenarios. Thereby, we have attributed 80% of deforestation to agriculture⁷⁰. Assuming the same deforestation pressure by country in 2050 as in the baseline is a very strong assumption, but due to the lack of better data on a global level, we decided to uses those rates for a first assessment of deforestation pressure.

In some cases, no data on change in agricultural land area has been available for 2005–2009. Then, the ratio between deforestation areas (multiplied by 0.8) and total agricultural land area has been built for the base year. This ratio is then multiplied with the total agricultural land areas in the scenarios to derive values for deforestation. In these cases, we thus used total agricultural area (instead of the change in agricultural area) as a proxy for the pressure of agriculture on forests. In cases where total forest area increased, deforestation values have been set to zero.

Modelling deforestation in relation to agricultural area changes explains the drop in deforestation rates in the reference scenario, as annual land expansion rates to 2050 are projected to be lower than the observed rates for the base years 2005–2009.

Modelling deforestation in this way thus captures the pressure of land increase on forests in countries, where deforestation is an issue, by assuming a similar dynamics as in the baseline. It thus complements the land occupation variable that captures the land demand, irrespective of where it may be sourced from in a specific country. In particular for larger increases in land occupation, this approach may rather underestimates deforestation, as a large part of this additional land likely would have to be sourced from forests, given the assumption of constant grassland areas. A more detailed assessment of the deforestation dynamics with increased cropland demand would necessitate combination with data on the suitability of grassland, forest and other areas for crop production of various kinds, which is however beyond the scope of this paper.

Soil erosion: Soil erosion was based on data for soil quantities lost (tonnes soil per-ha per year) via water erosion on a per country basis²⁶. Due to lack of data, wind erosion has not been included. Per-ha soil erosion rates were then combined with a soil susceptibility index for different crops to differentiate between crops with lower and higher soil erosion risks. This index was set 0 for permanent grasslands, 1 for crops with a short period of bare fallows and 2 for crops with longer periods of bare fallows such as maize or beets. This classification is based on expert consultations and literature^71–77. The on average higher soil organic matter contents in organic agriculture are likely to reduce soil erosion rates^{78, 79}. However, in order to produce conservative estimates at global level, we did not consider this impact in our model. Potential differences in soil erosion between organic and conventional systems thus arise from the different area allocation to different crops, thus changing relative shares of crops more or less susceptible to erosion.