Potential Environmental Impacts of a Trade Agreement

Get Complete Project Material File(s) Now! »

Exploring different agricultural developments

Baseline assumptions

We draw our analysis from a reference situation describing a plausible future up to 2050 for the diﬀerent regions of the world. Population and gross domestic product (GDP) changes follow the assumptions from scenario SSP2 (“Middle of the Road”) of the Shared Socioeconomic Pathways (SSPs) developed by the climate change community (O’Neill et al., 2012). Under this scenario, the world population reaches 9.2 billion people by 2050, whereas the average world GDP per capita increases from US$ 6,700 in 2005 to US$ 16,000 in 2050. The food demand projections in GLOBIOM are based on income elasticities calibrated on trends from the Food and Agriculture Organization of the United Nations (FAO) (Alexandratos and Bruisma, 2012) and the world food consumption grow in our model by 68% in kcal terms between 2000 and 2050 and reaches 3,045 kcal/cap/day at that horizon (see complementary Table 2.6 at the end of this Chapter for details). The share of animal products in diet only slightly increases, from 16% in 2000 to 17.3% in 2050 because the increase in developing countries are partly compensated by some decrease in developed regions. Other competitive uses of agricultural products are increasing with bioenergy demand. First generation biofuels are assumed to continue in line with current commitments levels until 2030 and are later stabilised. Bioenergy and biomass use for heating and cooking are fixed exogenously following scenarios from the POLES model (Russ et al., 2007).

Future yield development

Yield growth for crops and livestock are assumed to follow in the baseline recent historical trends, which are extended linearly to 2050. In the case of crops, such trends are derived from the analysis of past FAOSTAT yields between 1980 and 2010. Fertiliser use is assumed to increase with crop yield with an elasticity of 0.75, following the world average trend observed over the last 30 years. For livestock, we rely on the feed conversion eﬃciency information from Bouwman et al. (2005) and apply them to the diﬀerent grass-based and mixed systems in the model. For both crops and livestock, we consider in the baseline that other input and factors than land and feed are increased and production costs per unit of output are only marginally aﬀected.
Four diﬀerent yield scenarios and three productivity pathways are considered around the baseline (scenario “TREND” with pathway “High-Input”). Yield scenarios only modify crop yields projections (Figure 2.1) and ruminant feed eﬃciencies projections (Figure 2.2) in developing countries and economies in transition, according to assumptions in Table 2.1. The productivity pathways distinguish how these yield changes are attained (Table 2.2).
Yield scenarios
The first alternative scenario to the baseline considers that yield improvements cannot keep on the present trends and stall over the next decades at half the currently observed growth rate (“SLOW”). This scenario is a stylised representation for the interplay of many factors that could aﬀect yield diﬀerently, such as failure in technology adoption, increase in rural poverty due to resource scarcity, land degradation, pressure from climate change, or lack of investments or access to credit.
We contrast this perspective with a convergence scenario (“CONV”), where eﬃcient rural development policies improve cropping and herd management practices. As a result, we assume that 50% of the estimated yield gaps in the baseline are bridged for crops. These yield gaps are calculated comparing current observed crop yields from FAO with potential yield for rain-fed and irrigated systems estimated with the EPIC model. In the case of livestock, developed regions are used as the benchmark for feed conversion eﬃciency for ruminants and 25% only of the distance to this frontier is bridged, to avoid too strong structural breaks for this sector. For non-ruminant animals, we do not consider any change from the baseline trend because productivity gaps for industrial systems, where most of the future production will take place, appear much more limited across regions (see Appendix A).
To better understand the contribution of the diﬀerent sectors on the results, the convergence scenario is further decomposed into two additional variants: “CONV-C” which corresponds to a convergence in crop yield only, while livestock feed eﬃciency remains unchanged; and “CONV-L” which considers the opposite situation where only ruminant eﬃciency is increased.
Productivity pathways
Pathways describe how yield scenarios are reached. The reference pathway considered for the baseline and all scenarios is a conventional intensification of practices (“High-Input” pathway). We increase for this pathway all inputs requirements and factor costs associated with yield improvement. For crops, such scenario implies additional fertilisers, pesticides, and irrigation, as well as investment in machineries and equipment, which are still limiting in many developing regions (Neumann et al., 2010, Mueller et al., 2012). Given the possibilities of increasing yield through more sustainable practices, we also consider a pathway where these crop yield improvements are obtained without additional synthetic fertilisers, mainly through more eﬀorts on optimised rotation, crop-livestock systems integration, and precision farming (“Sust-Intens” pathway). On the livestock side, these two previous pathways are considered similar: they rely on investment in adequate capital and labor and a better management of herd, to decrease mortality, improve feeding practices and hence increase meat and milk output per head (McDermott et al., 2010). Last, a third pathway is explored, relying much more on innovation and total factor productivity gains. For this pathway, all input and factor requirements are kept constant, and the extra production is reached through new technologies adoption and public expenditures towards R&D and infrastructure investments, bringing substantial yield boost overall without extra cost for farmers (“Free-Tech”).

Modelling framework

We analyse the eﬀects of the previous scenarios using a linear programming (LP) model of agriculture, forestry, and land use change: the GLOBIOM, developed atIIASA by a collective team (Havlík et al., 2011, 2013).
GLOBIOM is a global partial equilibrium model allocating land-based activities under constraints to maximise the sum of producer and consumer surpluses. It is grounded in a well-established tradition of LP models (Takayama and Judge, 1971, McCarl and Spreen, 1980), and similar in structure to the US-FASOM model (Schneider et al., 2007, Beach et al., 2012). The model operates at two levels: on the supply side, a detailed resolution grid based on a 0.5 0.5 degree cells structure; on the demand and trade side, a representation of the world into 30 markets, separated by trade costs and tariﬀs. The model is used under a recursive dynamic approach, and for the current work is run 10-year time steps over the 2000-2050 period. The main characteristics of this model are presented below1 and the modifications to implement the scenarios are described in Subsection 2.2.3.

Supply side representation

On the supply side, crop, livestock and forestry activities in the model are described at the grid-cell level. Data are supplied at the most detailed resolution, called Simulation Units (SimUs, Skalský et al., 2008), that are latter aggregated at the model resolution, the production unit (PU). Each production unit can be used to supply a combination of up to 18 crops, seven livestock products sourced from eight animal types and five primary wood products.
Simulation and Production Units
SimUs are the most detailed units, delineated at the 5 minute of arc resolution as the intersection of zones of same altitude, slope, and soil class, 0:5 0:5 degrees grid, and the country boundaries. These units are used as the base architecture to input all geographically explicit data to the model (land use, crops and animal location, etc.). It is also the level at which productivities for crops can be calculated using the EPIC model, as well as the biomass net primary productivity (NPP) to estimate grazing potential for livestock, and forest productivity.
The total number of SimUs being greater than 200,000, raw data can be aggregated in the model to reduce the computation time. For this study, we use a 2 2 degree resolution to run the model, only keeping a layer of diﬀerentiation across three agro-ecological zones to distinguish livestock systems (Seré and Steinfeld, 1996). This leads to a total of 10,894 diﬀerent production units distributed across the 30 regions.
Crop production
Each activity at the PU level u can produce various products i according to diﬀerent Leontief technologies or management m. Input and output characteristics of these technologies are calculated with a specific biophysical model – the crop productivity model EPIC, the digestibility model RUMINANT for livestock, and the forest model Global Forestry Model (G4M).
GLOBIOM represents up to 18 crops.2 Production Si;u of each crop i in unit u can then be written as: X Su;i = Yi(Au;i;m; Ku;q1;i;m; :::; Ku;qn;i;m); (2.1) m where Yi is the production function of product i, Au;i;m is the land area demand for this production and Ku;q;i;m a list of required inputs q 2 (q1; :::; qn). Because we assume fixed technologies (i.e. fixed shares of K and A) for each management system m, we can use the crop activity model to calculate land productivity for crops in all PUs as yu;i;m = Yi(Au;i;m; Ku;q1;i;m; :::; Ku;qn;i;m)=Au;i;m.
Four diﬀerent management systems m are considered for each crop: subsistence, low input, high input, and irrigated, when water resource is available (see below). Crop yields yu;i;m are generated with EPIC for all locations u on the basis of soil, slope, altitude and climate information from the SimU, as well as climate information. Each management is associated with a fixed cost per unit of
area cAu;i;m, calculated as a function of yield yu;i;m, initial market price pi;r in the region r, and the cost of the diﬀerent inputs.
At the market level in region r, supply can simply be defined as Si;r = yu;i;mAu;i;m: (2.2)
Livestock production
Livestock production depends, among other inputs, on feed crops and, for ruminants, on consumption of grass g, whose production function is the same as for crops. We represent production here with a function depending on number of animals Nu;a;m, where a is animal species. Productivities of animals yr;i;a are defined at the regional level by animal species and management system. The production of product i for unit u in region r can be expressed as:
Su;i = yr;a;i;mNu;a;m = Yr;a;i;m(DG ; DF ; :::; DF ); (2.3)
u;a;m u;a;i1;m u;a;in;m
a m a m
2 cNr;a;m
where Du;a;mG and Du;a;i;mF are the demand for grass and for feed product i 2 (i1; :::in), respectively, from animal a in location u and system m. The corresponding production functions Yr;i;a;m therefore directly depends on feed consumption.
The model incorporates seven animal types – dairy and other bovines, dairy and other sheep and goats, laying hens and broilers, and pigs – that produce four meat types, milk, and eggs. These animals are distributed across diﬀerent management systems (Seré and Steinfeld, 1996) using spatial distribution data from Robinson et al. (2011).3 Feed is distinguished between: grass, stover, and feed crops.4 Feed and stover are provided by the regional market, whereas the constraint on grazing demand for ruminants can simply be written for each unit u in function of production of grass g: Du;a;mG Su;g;m (2.4)
Feed composition and eﬃciency yr;a;i;m are calculated according to Herrero et al. (2013) who process regional data on animal diet with the RUMINANT model. Eﬃciency could be diﬀerentiated across all management systems m and animal types i, and for 28 diﬀerent regions.
Production cost depends on feed market prices, but also of a fixed cost of animal maintenance
specific to the animal and management type of each region.
Forestry and bioenergy
Although not the focus of this study, the forestry sector also participates to land use dynamics in GLOBIOM. Five forest primary products – pulp logs, saw log, biomass for energy, traditional fuel wood, and other industrial logs – are consumed by industrial energy, local population for traditional use, or processed as final wood products (pulp and sawn wood). These products are supplied by two diﬀerent land use types (managed forests and short rotation plantations) that follow the same production specifications as for crops. Yields yu;i;m and harvesting costs cAu;i;m are sourced from the G4M model (Kindermann et al., 2006) on the basis of information on species selection, variation of thinning and choice of rotation length.
Forestry products, as well as some biofuel crops can be processed in the model through sawmill and bioenergy supply chains. These transformation chains z can process a quantity Dr;z;iT of product i into Tr;z;i0 of product i0 with a linear processing cost cTr;z per unit produced.
Land use competition
Each PU has an initial allocation of land cover Lu;l within its total land area Lu, where l defines six
possible land use types: cropland, grassland, managed forest, short rotation plantations (SRPs),
primary forest, and other natural land. Land dynamics Lu;l is subject to several constraints related to supply and demand for land: 8 i2X X Lu;cropland where we call WoodM the group of products specifically produced in managed forest and WoodP the group of products specifically produced in SRPs.
Managed land use is associated an initial land rent wr;l, that increases with land supply with an elasticity « Lr;l. Additionally, land conversion LUC r;l;l0 from land use type l to another l0 is associated a specific conversion cost with a constant term r;l;l0 and a linearly increasing one r;l;l0.
Resource constraints
The supply side is also subject to several input constraints. The most notable is water for irrigation,
but a constraint on fertilisers can also be considered. For a resource q supplied in quantity Qr;q in region r, the model integrates the constraint on demand Ku;q;i;m for input of resource q simply as:
Ku;q;i;m Qr;q: (2.6)
u2r i m
The constraint on resource q for this study is defined for water at the regional level. The resource
price varies around its initial level pr;q and is linked to supply level Qr;q through an elasticity of supply « Qr;q.
Total production costs and producer surplus
All the relations above allow us to express the production cost PC r for each region of the model. If
we define S and L the production and land use levels at calibration, we then obtain:

READ dust and surface chemistry

Demand and trade

Product processing, trade and consumption occur at the level of the region r. The world is split into 30 economic regions for which all products i are homogeneous.
Demand
Users of agricultural and wood products in the model are households, livestock for intermediate consumption of feed, industrial demand for sawn wood and wood pulp, and bioenergy demand. Livestock demand directly enters the production function of livestock products as input requirements. Bioenergy demand is treated as an exogenous constraint, according to the scenario assumptions (see Section 2.1.1).
Food demand Dr;i is in the model endogenous for any period t and depends on product prices. To clarify how exogenous trends are taken into account over time, let’s add a price index and note for this subsection demand and price of period t as Dr;i;t and pi;r;t , respectively. Population and GDP from the socio-economic scenario (Section 2.1.1) are taken into account to redefine along the baseline the initial level of demand Dr;i;t. The change in final demand can then be expressed as a function of base year price pr;i;2000 and elasticities, as:
Dr;i;t pr;i;t « r;i;tPr Popr;t Cap ! « r;i;tInc
= where Dr;i;t = GDPr;t Dr;i;2000 (2.9)
Dr;i;t pr;i;2000 Popr;2000 GDPCap
r;2000
For each product i in region r and period t, the prior demand quantity is calculated as a function of population Popr;t, GDP per capita GDPCapr;2000 adjusted by the income elasticity « Incr;i;t, and the base year 2000 consumption level as reported in the food balance sheet (FBS) of FAOSTAT.5 The final demand quantity is however also aﬀected by the change in price pr;i;t when compared to the base year price pr;i;2000, through a price elasticity « Prr;i;t sourced from Muhammad et al. (2011). Because food demand in developed countries is more inelastic than in developing ones, the value of this elasticity is assumed to depend on time and decrease with the level of GDP per capita. We consider that the elasticity values of developing countries converge to US elasticity values in 2000 at the same pace as GDP per capita are catching up US GDP per capita. This representation allows us to capture the eﬀect of change in relative prices on food consumption taking into account heterogeneity of responses across regions, products and over time. It however also brings some limitations as cross-price eﬀects are not represented, which means that no substitution can occur at the final demand level between food products. This could induce some overestimation in the final demand response, which highlights the importance of including a sensitivity analysis on the demand response in our analysis.6
Using the previous formula, we can express the final consumer surplus of region r with respect
to the initial price pr;i;2000 and demand Dr;i;t:
i Z pr;i;2000 1+ »r;i;tPr
Dr;i;t dp = Dr;i;t pr;i;2000 1 pr;i;t : (2.10)
CS r = X » # pr;i;t 1 + « r;i;tPr pr;i;2000
Bilateral trade and market clearing
Bilateral trade is represented in the model assuming some linear but also non-linear transportation costs. Because all products are considered homogeneous, no cross flows are present but only a net trade flow. Each region pair r and r0 can trade a quantity Xi;r;r0 = Xi;r0;r. When such trade occurs, market equilibrium imposes:
pi;r0 = pi;r + i;r;r0 Xi;r;r00 ; (2.11)
Xi;r;r i;r;r0
demand of households (Food supply to households or Food availability), i.e. it includes eﬀective food consumption by households but also domestic waste. Therefore, an increase or decrease in food demand does not only inform on change in food ingestion by individuals, but also on change in consumer habit with respect to food handling.
where i;r;r0 is the linear cost, representing fixed trade costs and tariﬀs, and i;r;r0 is a non-linear cost parameter paired with an elasticity i;r;r0.
Total trade costs can be written:
X i;r;r Xi;r;r 1+ i;r;r0
Xi;r;r
0 0 0 (2.12)
TC = r;r0;i :
1 + i;r;r0 Xi;r;r0
Market clearing conditions impose the constraint that, for all products i and for all regions r, market supply should exceed final demand and input demand:
Dr;i + Dr;z;iT + Du;a;i;mF Sr;i + Tr;z;i Xi;r;r0 + Xi;r0;r (2.13)
z 2 r a;m z r0 r0
Objective function
The partial equilibrium of the model is determined by maximising the sum of the producer and consumer surplus, once trade costs are taken into account under all the production and consumption constraints above, i.e.
Max Xr (CS r + PS r) TC ! ; (2.14)
(S;T;Q;X)

Implementation of the productivity scenarios

Yield in GLOBIOM are usually endogenously determined at the PU level by the composition in management systems. Additionally, spatial reallocation introduces an additional composition eﬀect that can aﬀect the final yield observed at the regional level (see equation (2.2)). These mechanisms are useful to assess the response of the supply side to various types of shocks. However, for the purpose of the current study, we want to test the response of production to a predetermined yield trajectory that can be easily characterised and compared to other assessments. We therefore implement two additional constraints to the model to neutralise the composition eﬀects identified above:
where Au;i;m
Au;i;m u;i;m Au;i;m with u;i;m = Au;i;m (2.15)
m Au;i;m
X Pu Au;i;m
m P
u u;m
X X with i;m = (2.16)
Au;i;m i;m Au;i;m u;m Au;i;m ;
is the initial value of Au;i;m. The first constraint (2.15) allows to neutralise the management eﬀect, whereas the second constraint (2.16) manages the reallocation eﬀect.
The long term projections on yield associated to the diﬀerent scenarios are then managed through an exogenous shifter applied to the yield yu;i;m in each unit. This simplified representation means that the yield gap closing is not implemented for the scenarios in a geographically explicit manner but applied homogeneously in the region r.7 Input requirements and production costs are also varied exogenously according to assumptions from section 2.1.2. For the “High-Input” pathway, the crop requirements in Ku;q;i;m are increased for all inputs q, as well as the fixed production cost cA . u;i;m
For the “Free-Tech” scenario, only yu;i;m is increased, but Ku;q;i;m and cAu;i;m are unchanged. Switches between livestock production systems in each PU and at regional level are neutralised the same way as for crops to implement the productivity scenarios.8 Because yield increase is not per unit of area but per unit of feed, we do not change the input requirements of livestock Du;a;i;mF and Du;a;i;mG but only increase the fixed cost per animal cNr;a;m.

GHG emission accounts

Each PU in GLOBIOM is associated a location specific emission factor per sector. The diﬀerent activity models used for input data allow for a precise estimation of these GHG emission factors, based on Tier 1 to Tier 3 methodologies from the IPCC agriculture, forestry and other land use (AFOLU) guidelines (IPCC, 2006). For crops, the emission factor eAu;i;m depends first on rates of synthetic fertiliser use, calculated using the output from the EPIC model, after harmonisation with the consumption statistics from the International Fertilizer Industry Association (IFA). Methane emissions from rice cultivation are based on area harvested and emission factors from FAO (Tubiello et al., 2013). Livestock emissions factors eNu;a;m are sourced from the RUMINANT model which has been applied in each country to the diﬀerent livestock systems from the Seré and Steinfeld (1996)
classification. Three GHG sources are considered: enteric fermentation (CH4), manure management (CH4 and N2O), and manure left on pasture or applied to cropland (N2O). The values of these emission factors are highly influenced by the management system m.
Land use change emissions are calculated as changes in carbon stocks, based on data from the G4M model, consistent with FAO inventories (FAO, 2010b). Forest conversion to agricultural land or plantation is considered to release all the carbon per hectare fu;l contained in above- and below-ground living biomass into the atmosphere. Carbon stocks for land use types other than forests are sourced from the Ruesch and Gibbs (2008) database. Soil organic carbon stocks are not considered in our standard accounting.

Table of contents :

1 Introduction
2 Agricultural Productivity and Greenhouse Gas Emissions
2.1 Exploring different agricultural developments
2.1.1 Baseline assumptions
2.1.2 Future yield development
2.2 Modelling framework
2.2.1 Supply side representation
2.2.2 Demand and trade
2.2.3 Implementation of the productivity scenarios
2.2.4 GHG emission accounts
2.3 Results
2.3.1 Patterns of GHG emissions in developing countries across scenarios
2.3.2 Trade-offs and synergies between GHG mitigation and food availability
2.3.3 Sensitivity analysis
2.4 Discussion and conclusions
2.A Complementary tables and figures
3 Potential Environmental Impacts of a Trade Agreement
3.1 Environmental impact of a trade agreement
3.1.1 Welfare gains from a trade agreement with environmental externality
3.1.2 Emission changes from economic activities
3.1.3 Emissions from change in carbon stock
3.2 The EU-MERCOSUR trade agreement as an application framework
3.2.1 Trade patterns between MERCOSUR and the European Union
3.2.2 Protection structure
3.2.3 Environmental challenges of production reallocation in a EU-MERCOSUR agreement
3.3 Modeling framework for assessing the trade agreement
3.3.1 A CGE model for trade policy analysis
3.3.2 A partial equilibrium model for impacts on agriculture and land use change
3.3.3 Model linkage for an integrated assessment
3.4 Analysis of different trade liberalisation scenarios
3.4.1 Trade scenarios
3.4.2 Economic impact of trade scenarios without environmental externalities
3.4.3 GHG emissions from the trade agreement
3.4.4 Decomposition of effects and role of technological change
3.5 Conclusion
3.A Complementary tables and figures
4 Modelling Land Use Changes Impacts of EU biofuels
4.1 An innovative database for a consistent representation of agricultural sectors in CGE
4.2 MIRAGE-BioF: a model dedicated to land use and bioenergy policy analysis
4.2.1 General features
4.2.2 Agricultural production function
4.2.3 Land use substitution and expansion
4.3 Land-use change from three potential scenarios for EU biofuel policies
4.3.1 Description of baseline and scenarios
4.3.2 Results
4.3.3 Sensitivity analysis
4.4 Fuel versus feed versus food: Domino effects on the demand side
4.4.1 Disappearing food: the role of final demand
4.4.2 Role of coproducts: the first feed retroaction
4.4.3 Role of pasture land: the second feed retroaction
4.4.4 Testing the possibility of higher NDF: a worst case scenario
4.5 Conclusion
4.A List of sectors and regions in the model
4.B Main characteristics of the MIRAGE-BioF database from GTAP7 and FAOSTAT
4.C Emission factors related to land use conversion in the MIRAGE-BioF model
5 Exploring Uncertainty in Indirect Land Use Change Estimates
5.1 Overview of ILUC uncertainty treatment in the economic literature
5.2 A simplified economic formulation of ILUC
5.2.1 Decomposition of a biofuel shock
5.2.2 Economic responses and price elasticities
5.2.3 Algebraic expression of ILUC and NDF
5.2.4 Emissions and ILUC factor
5.2.5 Illustration with a two regions three crops simplified case
5.3 Sensitivity analysis around usual models results
5.3.1 Sensitivity analysis around mean value
5.3.2 Uncertainties interactions
5.3.3 Role of parameter heterogeneity
5.4 Exploring the full range of parameters uncertainty
5.4.1 Overview of parameter uncertainty from literature
5.4.2 Sensitivity of ILUC to literature values
5.4.3 From indirect land use changes to ILUC factors
5.5 Conclusion
6 General Conclusion
A Yield Projection Scenarios for Chapter II
A.1 Crop yield
A.1.1 Yield extrapolation in scenarios
A.1.2 Identifying crop yield gaps
A.1.3 Comparison of crop yield gap results with the literature
A.2 Livestock productivity
A.2.1 Feed conversion efficiency calculation
A.2.2 Baseline and convergence scenario for feed efficiency
B GHG Emissions in GLOBIOM
B.1 GHG emission accounts in GLOBIOM
B.1.1 Livestock sector
B.1.2 Crop sector
B.1.3 Land use change
B.2 Comparison with the literature
B.2.1 Agriculture
B.2.2 Land use change emissions
B.3 GHG emissions uncertainties
C Parameters and modelling for Chapter V
C.1 Elasticities from literature
C.1.1 Demand elasticities
C.1.2 Land supply elasticities per crop
C.1.3 Yield elasticities
C.1.4 Agricultural land expansion elasticities
C.1.5 Marginal yield elasticities
C.2 Co-product substitution
C.3 Land use allocation and emission factors
C.3.1 Forest emission factors
C.3.2 Pasture conversion emission factors
C.4 Summary of parameters and indicator formulas
C.5 Model code in R language
Bibliography
Author index