The impacts of agricultural production on natural conditions are strongly dependent on specific local conditions. Changes in water or nutrient cycles are related to soil conditions, terrain type and local climate conditions. Hence it is necessary to link economic conditions of agricultural production to the place-specific biophysical conditions, in order to better understand their interactions. The key challenge with respect to modelling is to link place-specific models of agricultural production and land use with models representing important elements of the biosphere and hydrology.
A comprehensive analysis of the world food system can draw upon a substantial volume of existing research in the area of integrated assessment and modelling. Issues of climate change and agricultural land use have been covered in the IMAGE1 project and the ICLIPS2 project (Toth et al., 2003), where greenhouse gas emissions of different land use patterns as well as the potential of bio-fuel production on agricultural land as an alternative energy source have been analysed (Sands and Leimbach, 2003). The US Department of Agriculture maintains its FARM3 model, a computable general equilibrium (CGE) model with a focus on the interaction between climate change, economic growth, agricultural production and environmental resource use. The GTAP4 consortium has developed a CGE modelling framework as well as a database for global economic analysis, and is also extending its focus towards agricultural resource use, especially land use issues. The International Institute for Applied Systems Analysis (IIASA) maintains its Basic Linked System (BLS) which has been applied to various questions on global environmental change (Fischer et al., 1988). It has also been linked with the agro-ecological zones (AEZ) model to
1 Integrated Model to Assess the Global Environment. See also: http://sedac.ciesin.org/mva/image-2.0/image-2.0-toc.html.
2 Integrated Assessment of Climate Protection Strategies.
4 Global Trade Analysis Project: www.gtap.agecon.purdue.edu.
assess future changes in global land use and land cover (Fischer, 2001). The International Food Policy Research Institute (IFPRI) has a long tradition of partial equilibrium agricultural trade modelling with its IMPACT5 model (Rosegrant and Ringler, 1997). Recently the IMPACT model has been coupled with the global hydrological model WaterGAP6 in order to come up with more reliable global projections for water demand and supply (Cai and Rosegrant, 2002).
Our starting point to improve the understanding of society-biosphere interactions is the extension of one of the most advanced and comprehensive models of the global biosphere - the Lund-Potsdam-Jena Dynamic Global Vegetation Model (LPJ).7 We suggest a way to integrate human activities into LPJ and come up with a coupled climate-biosphere-economy modelling framework, including the global water cycle. This is an important improvement on existing research, as LPJ endogenously models the linkages between climate and soil conditions, water availability and plant growth in a dynamic way. This yields an advanced representation of global biogeochemical conditions, which can be used to define plausible biophysical constraints to agricultural production, or to human activities in general for that matter.
The Lund-Potsdam-Jena model (LPJ)
LPJ is a coupled non-equilibrium biogeography-biogeochemistry model which combines process-based representations of terrestrial vegetation dynamics and land-atmosphere carbon and water exchanges in a modular framework (Sitch et al., 2003). LPJ explicitly considers key ecosystem processes such as vegetation growth, mortality, carbon allocation, and resource competition, though their representation is of intermediate complexity to allow for global applications. To account for the variety of structure and functioning among plants, 10 plant functional types (PFTs) are distinguished. Leaf phenology of summergreen and of raingreen PFTs is determined daily, depending on temperature and water stress thresholds. Gross primary production is computed based on a coupled photosynthesis-water balance scheme; net primary production is given by subtracting autotrophic respiration. After additional subtraction of a reproduction cost, the remaining carbon is allocated to three pools for producing new tissue. Carbon from dead leaves and roots enters litter; decomposition of litter and soil organic matter is driven by soil temperature and water content. A PFT-specific mortality rate is determined at the end of each year as a result of heat stress, low growth efficiency, a negative carbon balance, light competition, or violation of bioclimatic limits. The presence and fractional coverage of PFTs is thus determined annually according to individual bioclimatic, physiological, morphological, and fire-resistance features (Sitch et al., 2003). The structure and distribution of the PFTs is decisive for the simulated site water balance, since evapotranspiration, soil water content, and runoff generation are modulated by
5 International Model for Policy Analysis of Commodities and Trade.
6 Water - Global Analysis and Prognosis: http://www.usf.unikassel.de/usf/mitarbeit/homepages/doell/ research3.htm
7 For a full documentation see: www.pik-potsdam.de/lpj/
PFT-specific attributes such as interception storage capacity, seasonal phenology, rooting depth, and photosynthetic activity.
The fundamental entity simulated in LPJ is the average individual PFT. This concept provides a simple way for process acting at the level of the plant individual to be scaled up to the 'population' over a grid cell. The grid cell is treated as a mosaic divided into fractional coverages of PFTs and bare ground. It is assumed that the physical environment of the plants is well mixed, i.e., the PFTs do not occupy discrete blocks, but compete locally for resources. The global version of LPJ has a spatial resolution of 0.5°, which is equivalent to a pixel size of about 50 x 50 km at the equator. This implies a total number of about 60,000 grid cells covering the whole terrestrial earth surface.
Overall, LPJ simulates the global terrestrial carbon pool sizes and fluxes, and captures the biogeographical distribution of Earth's major biomes. Recent applications of the model include assessments of the carbon balance of the terrestrial biosphere, the representation of fire regimes, and the simulation of transient vegetation responses to climate warming.8
A typical simulation with LPJ starts from 'bare ground' and 'spins up' for 1,000 model years until approximate equilibrium is reached with respect to carbon pools and vegetation cover. The model can then be driven with a transient climate (i.e. future climate scenarios provided by MPI Hamburg or the Hadley Centre). The standard LPJ simulation is run with the transient CRU data for 19001998.
In addition to the PFTs representing natural vegetation, recently 13 crop functional types (CFTs) have been implemented in LPJ in order to simulate potential agricultural production. These CFTs represent 8 classes of agricultural crops, e.g. temperate cereals (wheat), tropical cereals (millet), rice, maize, pulses (lentil), oil crops (sunflower, soybean, groundnut, rapeseed), roots and tubers (sugar beet, maniok), and fodder crops (C3 and C4 grass). As agricultural crops cover about 40% of the global land area, it has been shown that global carbon pools and water runoff are significantly affected when crops are taken into account in a global vegetation model like LPJ (Bondeau et al., 2003).
Input data required by LPJ are monthly fields of mean temperature, precipitation and cloud cover, which are taken from the Climate Research Unit (CRU) monthly climate data on a 0.5° x 0.5° global grid (CRU05, 1901-1998).9 A data set of historical global atmospheric CO2 concentrations extending from 1901-1995 was obtained from Carbon Cycle Model Linkage Project (CCMLP).10 Soil texture data are from the FAO soil data set. Standard LPJ outputs include changes in net primary production and different fractions of biomass, changes in carbon pools (e.g. vegetation carbon, soil carbon), and changes in water balances (e.g. runoff). Under given climate conditions, soil type and water supply, the CFTs generate crop yields in terms of above-ground biomass as well as harvested organs (like grains, roots etc.). The CFTs are currently specified as to represent observable yields at the end of the 20th century.
8 See LPJ website at http://www.pik-potsdam.de/lpj/lpj_publicvt1.html for a full list of publications.
9 Climate Research Unit (CRU), University of East Anglia, Norwich, UK; www.cru.uea.ac.uk.
10 CCMLP data source: http://eos-webster.sr.unh.edu/data_guides/ccmlp_dg.jsp.
MAgPIE - a management model of agricultural production and its impact on the environment
MAgPIE is set up as a linear-programming optimisation model with a focus on agricultural production, land and water use. The goal function is to produce a required amount of food energy, defined in GigaJoule (GJ), at minimal costs. Food demand is defined for an exogenously given population in three energy categories: crop energy, meat energy, and milk energy (at the moment we abstract from other vital food ingredients like proteins, etc.). Energy can be produced by choosing from 8 cropping activities (food grains, feed grains, oil crops, sugar crops, roots/tubers/pulses, vegetables/fruits/nuts, rice, fodder crops) and 3 livestock activities (ruminant meat like beef, veal, sheep and goat meat; non-ruminant meat like pork and poultry meat; milk).
Input factors of production are labour, chemicals, and other capital (measured in US$), land and water (measured in physical units, ha and m3, respectively). Labour, chemicals and capital are in unlimited supply at a given price. Land and water are available in fixed amounts and are implemented as physical constraints to production. Available land is divided in crop land and pasture.
Given a certain yield per hectare for each cropping activity, the corresponding energy delivery is calculated with standard energy content parameters. Livestock energy is produced either with feed grains (non-ruminant meat) or with a mixture of pasture, green fodder and feed grain (ruminant meat, milk), in addition to labour, chemicals and capital. Currently we are looking only at one region without external trade. That means, the regional demand for intermediate inputs like feed grain and green fodder has to be met by regional production. In the model, the region is forced to be self-sufficient in food production.
Water supply is currently entirely from precipitation inflows. There are no managed water stocks like groundwater reservoirs, lakes or water storages. Water demand from production activities is calculated using fixed coefficients per unit of crop or livestock output. Water balances are calculated in the hydrological subsystem of LPJ.
In order to keep the cropping mix within plausible bounds we introduce rotational constraints. In our sample region Germany, for instance, it seems plausible to limit grain production to a maximum of 66% of total crop area, as on average every third year a different crop will be planted for reasons of crop management. For the same reason, sugar beets have been limited to 25% and oil crops to 33% of total crop area.
Even though in this chapter we are looking at Germany as an example, we use only data sources that are also available on a global scale. Crop yields are taken from the CFTs in LPJ and are checked for consistency with average regional yields according to FAO statistics. Average cost structures for production activities are calculated on the basis of FAO production and land use statistics and national social accounting matrices (SAMs) from GTAP.11
11 At the time of writing this chapter, SAMs are available in the GTAP database for 78 regions with up to 57 economic sectors. See: http://www.gtap.agecon.purdue.edu/databases/v5/default.asp.
Output generated by MAgPIE includes crop and livestock energy produced, shares of different crops in total use of arable land, purchases of variable inputs, and shadow prices for inputs in limited supply and other constraints, like rotational limits. The generation of shadow prices (or 'opportunity costs') for land and water is probably the most useful feature of this model. It facilitates the assignment of internal use values to factors of production for which no proper markets and, hence, no observable prices exist. This can be particularly useful for the systematic valuation of ecosystem services, like water supply, as this model provides a rigorous economic framework for the use of these services.
Of course, this model is a strong over-simplification of real agricultural production. It is, for instance, not at all clear whether or not actual producers always act as strict cost minimisers or profit maximisers. Hence, the optimised mix of production and resource use generated by the model almost certainly differs from empirical observations. Moreover, the current version is a static model with a lot of exogenous inputs. However, this type of model can be easily scaled down to a single farm and scaled up to the world as a whole, thus providing the opportunity for nested modelling structures. It can also in principle be coupled to a food demand model or an economy-wide model, in order to make markets and prices for outputs and inputs endogenous. Here we will build upon recent developments in the area of model coupling and meta-optimisation at PIK (Jaeger et al., 2002).
Spatial scaling, model coupling and information flow
Several challenges have to be overcome in coupling a biosphere model like LPJ with an economic model like MAgPIE. First, thematic scales have to be matched. CFTs in LPJ, which are defined according to plant-physiological properties, have to be matched with groups of crops which provide a similar type of output for human consumption. Oil crops, for instance, comprise a wide variety of plant species (e.g. rapeseed, groundnuts, sunflowers, oil palms etc.), but they all deliver similar types of oil, which are almost perfectly substitutable in the processing of agricultural products. Currently our 8 cropping activities in MAgPIE match sufficiently well with the CFTs defined in LPJ.
Second, temporal scales have to be made consistent. Standard LPJ runs into the future covering a period up to the year 2100. Most economic forecasting exercises do not go beyond a time frame of 10-20 years. As they run into the longer-term future, they usually get more aggregated and lack structural detail. One of the reasons is that changes in technology and input use are very hard to predict in the long run. At the moment, we abstract from technical change and restrict ourselves to stylised scenarios in a comparative static manner. That means, we take 'time slices' out of certain LPJ runs, and couple them with static MAgPIE scenarios.
Third, and most importantly for the illustrative purpose here, we have to bridge the gap between the national (or even larger) scale in MAgPIE and the 0.5°-grid scale in LPJ. This is the most challenging aspect in coupling these two models. On the one hand, LPJ provides information on climate, soil, biomass and crop yields, carbon and water balances for about 60,000 grid cells on a global scale. On the other hand, information on food demand, agricultural cost structures, input use, crop shares and many crucial economic indicators are usually only available from official statistics for whole nation states. While it is obviously impossible to model economic activity on a 0.5°-grid, it does not make much sense either to model environmental impacts on the national level.
In order to bridge this gap, we developed a procedure to group the grid cells in LPJ into a small number of 'productivity zones', according to the normalised level of crop yields in each grid cell. These zones do not have to form compact geographic regions. However, once the zones are established, they are taken as homogeneous and in MAgPIE all cells within a certain zone are treated in the same way. The zones can differ with respect to climate conditions (temperature, precipitation), crop yields, share of crop land in total area, and their total size (i.e. number of grid cells belonging to the zone).
For the case of Germany we have 185 grid cells grouped into 6 different zones. Effectively this means, that MAgPIE can choose among 8 cropping activities and 3 livestock activities in 6 different zones, yielding in total 66 ( i.e. 8 x 6 + 3 x 6) different production activities in the given region. With this procedure we are able to generate considerable differences in regional cropping patterns without being too demanding with respect to required data, especially on the economic side of our modelling framework.
We are also able to distinguish between constraints to be fulfilled in each zone and constraints to be fulfilled at the regional level. This introduces aspects of trade between zones. For instance, feed grain produced in any zone is pooled across all zones and can be used in the whole region, as long as the overall balance is maintained. In contrast, green fodder realistically has to be used locally (usually even on the same farm) and, hence, we impose a separate constraint for each zone. Land and water are also constrained in each zone, as they cannot be easily moved around. For the moment we abstract from the possibility of water transport through rivers, canals and pipes.
Having separate constraints for different zones implies that MAgPIE generates different shadow prices for each zone. Moreover, we get different patterns of specialisation and land use shares for each zone. The sequence of our joint modelling exercises runs as follows:
• Run LPJ for all crops separately with one CFT at a time, in order to determine potential crop yields for all crops in each grid cell at a certain point in time.
• Group the grid cells into productivity zones, according to normalised crop yields. Here, we should note that depending on climate and soil conditions, some crops are more productive than others - and vice versa - in different zones.
• Deliver information on zones (number of grid cells, average fraction of arable land, average precipitation) and crop yields (ton/ha, average for each crop in each zone) from LPJ to MAgPIE.
• Optimise production pattern and resource use for the whole region in MAgPIE.
• Deliver land use shares for all crops in all zones from MAgPIE to LPJ.
• Calculate impacts of different land use patterns on carbon and water balances in LPJ.
Was this article helpful?