An agent-based simulation model of humanenvironment interactions in agricultural systems
Pepijn Schreinemachers and Thomas Berger
Dept. of Land Use Economics in the Tropics and Subtropics,
Universität Hohenheim, 70593 Stuttgart, Germany
This is a post-print accepted manuscript, which has been published in the journal
Environmental Modelling & Software. Please cite this publication as:
Schreinemachers, P., T. Berger. 2011. An agent-based simulation model of humanenvironment interactions in agricultural systems. Environmental Modelling & Software 26:
845-859.
You can download the published version at: http://dx.doi.org/10.1016/j.envsoft.2011.02.004
1
Abstract
This paper describes an agent-based software package, called Mathematical Programming-based
Multi-Agent Systems (MP-MAS), which builds on a tradition of using constrained optimization
to simulate farm decision-making in agricultural systems. The purpose of MP-MAS is to
understand how agricultural technology, market dynamics, environmental change, and policy
intervention affect a heterogeneous population of farm households and the agro-ecological
resources these households command. The software is presented using the Overview, Design
concepts, and Details (ODD) protocol. Modeling features are demonstrated with empirical
applications to study sites in Chile, Germany, Ghana, Thailand, Uganda, and Vietnam. We
compare
MP-MAS
with eight
other
simulators of human-environment
interactions
(ABSTRACT, CATCHSCAPE, ECECMOD, IMT, LUDAS, PALM, SAM, and SIM). The
comparison shows that the uniqueness of MP-MAS lies in its combination of a microeconomic
modeling approach and a choice of alternative biophysical modules that are either coded as part
of the software or coupled with it using the Typed Data Transfer (TDT) library.
Keywords: mathematical programming, integrated modeling, multi agent systems, land use cover
change, technology adoption, impact assessment
2
Software availability
Name of software:
Mathematical Programming-based Multi Agent Systems (MP-MAS)
Version 3.0 (April 2010)
Developers and
contact address:
Thomas Berger, Pepijn Schreinemachers, Thorsten Arnold, Tsegaye
Yilma, Lutz Göhring, Chris Schilling, Dang Viet Quang, Nedumaran
Swamikannu, Christian Troost, Evgeny Latynskiy, Prakit
Siripalangkanont, and Germán Calberto
Dept. of Land Use Economics in the Tropics and Subtropics
Universität Hohenheim (490d), 70593 Stuttgart, Germany
Phone: +49 711 459 24116; Fax: +49 711 459 24224
Email: [email protected] (Thomas Berger).
Hardware required:
32-bit PC with Windows OS or Linux
1-10 GB of free disk space (depending on application)
We recommend using a high-speed processor and at least 4 GB of RAM.
Software required:
MS Office Excel for constructing input files. IBM OSL1 as optimization
software. STATA SE is recommended for analyzing the simulation
output.
Program language:
C++
Availability and
cost:
Freeware downloadable from: https://mp-mas.uni-hohenheim.de/
including a user guide. A default data set is available online for testing
the software. Most of the model features presented here, except the
coupling with external software, can be tested by switching specific
features on or off in this default version.
1
IBM has stopped its development of OSL and transferred the software code to an open source community, called
COIN. Work is ongoing to implement COIN in MP-MAS so that it can run be run without OSL.
3
1
Introduction
Simulation models have become important tools for assessing, ex ante, the impact of new
technologies, policy interventions and climate change on agricultural systems (e.g. Doglioni et
al. 2009, Gilfedder et al. 2009, Wang et al. 2010). Among model developers, a consensus has
grown that such simulation models should ideally give a balanced representation of the
economic, environmental, as well as social dimensions of a given system (Parker et al. 2002,
Schaldach and Alcamo 2006).
While many simulation models combine economic, environmental and social components at fine
resolution, only a few incorporate the dynamic interactions between them (Argent 2004).
Incorporating these interactions is required to better capture management options for agricultural
systems in which the economic decisions and social interactions of farmers and other land users
partly depend on environmental change, and in which the environment changes in response to
the decisions and interactions of all land users (Walker 2002). A special case are so-called
Social-Ecological Systems in which land users create environmental externalities that affect other
resource users, for example through water return-flows in irrigation systems or through soil
erosion and sedimentation in agricultural landscapes (Janssen and Ostrom 2006).
This paper presents a software package, called Mathematical Programming-based Multi Agent
Systems (MP-MAS), that couples dynamic models of farmer decision-making and interactions
with various dynamic models of environmental processes, notably water flows and soil fertility
changes. The purpose of MP-MAS is to understand how agricultural technology, market
dynamics, environmental change, and policy intervention affect a heterogeneous population of
farm households and the agro-ecological resources these households command.
4
The paper has three objectives. The first is to describe MP-MAS using the ODD (Overview,
Design concepts, and Details) protocol developed by Grimm et al. (2006). Recent papers
reported on selected aspects of MP-MAS (Berger and Schreinemachers 2006, Schreinemachers
and Berger 2006) but a comprehensive description of the software is missing; with this paper we
therefore aim to close this gap.
The second objective of the paper is to discuss modeling choices in MP-MAS with regard to
empirical research questions and system characteristics. Previous papers presented results from
applications of the software (Schreinemachers et al. 2010, 2009, 2007, Berger et al. 2007) but a
general discussion of the empirical modeling choices within MP-MAS has not been undertaken
so far. The overview of applications to six study sites shows that the optional components in MPMAS allow users to apply the software to a wide range of research questions, spatial scales, and
levels of data availability without having to change the software code.
The third objective of the paper is to position the MP-MAS software package in the recent
literature on simulation models of human-environment interactions in agriculture. We compare
MP-MAS to eight simulators: ABSTRACT (van Oel et al. 2010), CATCHSCAPE (Becu et al.
2003), IMT (Letcher et al. 2006a, 2006b), SIM (Krol and Bronstert 2007), ECECMOD (Vatn et
al. 1999, 2006), LUDAS (Le et al. 2008, 2010), PALM (Matthews 2006) and SAM (Belcher et
al. 2004). The comparison shows that the uniqueness of MP-MAS lies in it being an agent-based
approach with a strong foundation in microeconomics and farm management theory and optional
modules for simulating biophysical dynamics. Because of this, the software is able to assess ex
ante how farm households would adapt if economic incentives changed, for example through
policy intervention, market dynamics and resource depletion. Another unique feature of MPMAS is its ability to simulate the diffusion of agricultural innovations by combining a frequency-
5
dependent communication effect at population level and an economic assessment of costs,
benefits, and resource constraints at farm household level.
The remaining four sections of the paper are organized as follows. Section 2 provides
background information about the software development and highlights its relationship to bioeconomic modeling. After describing the software using the ODD protocol in Section 3, the
paper proceeds to give an overview of applications to six study sites in Section 4. This is
followed in Section 5 by a comparison of MP-MAS to eight other simulators of humanenvironment interactions in agricultural systems. In the conclusion we then bring together the
proceeding sections by highlighting the novel aspects of the MP-MAS software package.
2
Background
2.1
Development of MP-MAS
The simulation of land use change and adaptation in agricultural systems has a tradition spanning
more than forty years. Starting in the 1960s, scholars such as Richard Day and Theodor
Heidhues were using recursive linear programming models for farm policy analysis, and
implemented the precursors of today‘s agent-based models in agriculture – so-called adaptive
macro and micro systems (see, for example Day and Singh 1975; an update of this work can be
found in Day 2005).
After a decline in interest for simulation modeling during the 1980s, Balmann (1997) developed
an agent-based linear programming model and was followed by Berger (2001), who drew on
Balmann‘s work and began writing the first lines of code for MP-MAS. These two agent-based
models have diverged since then, although the underlying concept of representing agent
decisions through mathematical programming (MP) remains the same.
6
Balmann, together with Happe, turned their code into the Agricultural Policy Simulator
(AgriPoliS). AgriPoliS has been used to ex ante simulate structural change in European
agriculture, in response to proposed changes to the Common Agricultural Policy (Happe et al.
2009, 2008, 2006). Berger, on the other hand, developed his code into a Multi-Agent System
(MAS) of human-environment interactions (Berger et al. 2006), which later got the name MPMAS. This multi-agent software has been applied to the ex ante impact assessment of
agricultural innovations and policies, especially in developing countries. Hence, the chief
difference is that MP-MAS includes endogenous biophysical dynamics and is tailored to the
assessment of agricultural innovations with most applications to developing countries, whereas
AgriPoliS is a more economic simulation model tailored to assessing structural change with all
applications to European countries.
Both MP-MAS and AgriPoliS fall into the class of Multi-Agent Systems applied to LandUse/Cover Change (MAS/LUCC), defined as simulation models that couple a cellular
component representing a geographical landscape with an agent-based component representing
human decision-making (Parker et al. 2003, Robinson et al. 2007).
2.2
Extensions to bio-economic modeling
MP-MAS shares features with bio-economic farm household models, such as those developed by
Schuler and Sattler (2010), Roetter et al. (2007), Holden and Shiferaw (2004), and Kuyvenhoven
et al. (1998). Akin to these bio-economic models, MP-MAS uses whole-farm mathematical
programming, which captures human-environment interactions by updating the yield coefficients
in the MP decision matrix, and has strengths in ex ante impact assessment. Yet, MP-MAS
extends these conventional bio-economic models by implementing three major features: agent
and landscape heterogeneity, spatial interactions, and social interactions.
7
First, it has been increasingly realized that the distributional effect of change is as important as
its quantity and direction. Climate change, technology diffusion and agricultural policies such as
conservation payments create winners as well as losers (e.g. Reidsma et al. 2010, Happe et al.
2009, Henseler et al. 2009). Whereas conventional bio-economic models are based on a limited
number of representative farms (usually four or five) MP-MAS captures more diversity in
economic and biophysical conditions by individually representing all plots and all farms.
Second, by representing the spatial aspects of agricultural production, MP-MAS allows for a
coupling of decision models with spatially-explicit biophysical models, such as hydrological
models of water flows and biophysical models of soil erosion. This is not possible with
conventional bio-economic models, as they do not fully cover the spatial set-up of a landscape,
creating only sketchy land use maps.
Third, by representing farm households individually, direct interactions between households can
be modeled. This may include information sharing about new technologies, as well as the
bilateral exchange of scarce resources (e.g. water and land) and certain forms of collective action
(e.g. irrigation water use). Including these features in conventional bio-economic models requires
iterative file-based procedures (see, for example Roebeling et al. 2000).
3
Description of the software
3.1
Purpose
The purpose of MP-MAS is to understand how agricultural technology, market dynamics,
environmental change, and policy intervention affect a heterogeneous population of farm
households and the agro-ecological resources these households command. The modeling
approach employs a scenario-based analysis to explore the impact of these changes.
8
The software is complex as the resource decisions and decision outcomes of each model agent
depend on the quality of its resources, which is endogenous, as well as on the decisions of other
model agents. MP-MAS is particularly suited to simulate empirical situations in which (a) there
is upstream/downstream competition for irrigation water; (b) there is resource degradation such
as soil fertility depletion; or (c) new technologies are introduced into the system.
Scientists are the main target users of MP-MAS as the parameterization and validation requires a
strong background in mathematical programming and statistics. Resource managers, such as
watershed managers and extension officers, can use a calibrated version with a Visual Basic
front-end to test new management scenarios (Latynskiy et al. 2010).
3.2
State variables and scales
Within the agent-based component, the MAS has four hierarchical levels:
 Agents represent farm households, with as many agents modeled as there are farm
households in reality. State variables of the agents include the location of the agents‘
farmsteads, the location of their fields, the individual household composition (age, sex, and
labor supply), available resources such as cash, livestock, tree orchards, farm equipment, and
certain agent characteristics (e.g. membership in a population cluster and in an innovation
segment).
 Population clusters are groups of agents with similar resource endowments used for the
initialization of the agent-based model component (see initialization below).
 Innovation segments represent groups of agents with similar communication behavior when
adopting innovations on their farms. Agents in the most innovative segment consider
innovations in their decision-making right after introduction, while agents in the less
innovative segments postpone deciding about innovation until many of their peers have
9
already adopted, i.e. sufficient information contagion has taken place. Early adopters thereby
create a positive, frequency-dependent externality that lowers the hurdles for innovation for
late adopters.
 Populations reflect social communication boundaries. Agents only interact with other agents
in the same population.
Within the cellular component, the MAS has two hierarchical levels:
 Pixels represent the smallest landscape unit. The size of a pixel depends on the spatial
resolution of the biophysical modules used in MP-MAS. The pixel size should ideally be set
to the smallest observed field size in order to represent each field in the MAS. Multiple
pixels can represent larger fields. State variables depend on the biophysical modules used and
can include separate soil nutrients, organic matter, slope, slope length, and location in the
hydrological system.
 Soil types are groups of pixels with similar soil fertility. The agent decision module refers to
these groups as agents expect the same average crop yield for all pixels within one soil type.
This reflects the empirical fact that farmers do not have perfect information about the state of
each small landscape unit.
MP-MAS can be parameterized at various spatial scales, ranging from a village community to a
large region within a country, depending on model purpose and data availability. If including
hydrological processes then the spatial extent is usually a watershed or river basin.
3.3
Process overview and scheduling
The MAS proceeds in annual time steps. Within each time step, the agent decision module goes
through three phases of investment, production, and consumption as shown in Figure 1. The
10
biophysical module is executed between the production and the consumption phase and can
follow any phases from daily to annual, depending on the particular biophysical features used.
Agent decision
module
Investments in livestock, soil
or water conservation
Biophysical
module
Investment
decisions
Initial resource
conditions
Production
decisions
Updating cash
and assets
Crop yields
Consumption
decisions
Final resource
conditions
Updating resource
conditions
Figure 1: Human-environment interaction in MP-MAS
Updating expectations and savings
Investment:
Fixed
Fixed
Production:
Take
Consider
Take
Fixed
Consumption:
Consider
Consider
Take
Timing of
decisions:
Expectations:
Resources:
Decisions at start of the period
Expected yield, prices, and water supply
Expected longterm average
Decisions after
harvest
Actual yield
and prices
Current resource supply
Investment refers to the acquisition of assets such as land, equipment, or the planting of
perennial crops that give returns beyond the current year. Investment decisions are therefore
based on long-term expected yields and prices, and on expected future resource supplies. After
investing, cash and asset endowments of each agent are updated. Although the MAS proceeds at
annual time steps, seasonal or monthly constraints in the labor supply or water availability are
taken into account by specifying these as constraints in the agent decision module.
11
The agent subsequently allocates its available resources to production in the current time step,
according to the actual resource supply, as well as short-term expected prices and yields. Actual
crop yields are then simulated for each plot from the production decision (crop choice, variable
input level) and the resource condition (which, is optionally, affected by investments in
livestock, soil, or water conservation). Finally, based on the actual simulated crop yields, the
MAS calculates the agent household revenue. Part of the revenue is consumed or used to repay
debts, and the remainder is added to savings that can be used for investment or production in the
next time step.
At each time step, the age of farm equipment, livestock, and orchards is updated. Using an
optional demographic module, also the age of individual household members can be updated.
When using this module, each sex and each age has a pre-defined probability to give birth or to
die. At each time step, the demographic module uses these probabilities to update the household
composition and recalculate the labor supply. In the current model version, neither the change in
family composition nor the founding of new households are endogenous agent decisions. As a
consequence the number of households can decline over the simulation period, and their plots
remain idle. Work is ongoing to implement the founding of new households and the reallocation
of land in MP-MAS. For now, the demographic module should only be used for relatively short
simulation periods.
3.4
Design concepts
Emergence: The diffusion path of an innovation (speed and saturation rate) emerges from the
innovativeness of agents, the relative profitability of the innovation, and the frequencydependent information contagion across agent segments. Land use patterns emerge in response to
12
changes in the relative scarcity of resources. Water flows result from water distribution in and
abstraction from irrigation canals.
Adaptation: Agents can adjust their resource use (land, labor, water, cash, and livestock) in
response to changes in the relative scarcity of these resources as affected by policies, climate,
technologies, and markets. For instance, if the water supply declines then agents can switch to
crops that use less water (e.g. grow soybean instead of rice) or adopt new technologies that save
water (e.g. apply drip irrigation instead of sprinkler irrigation).
Objectives: Agents maximize their expected net farm and non-farm income while, optionally,
satisfying their households‘ consumption needs. For investment decisions, agents maximize their
expected long-term average levels of net farm and non-farm income, while for production and
consumption decisions they optimize current short-term levels.
Prediction: Agents form expectations about crop yields, water supply and market prices based on
past experience, following the theory of adaptive expectations. Agents revise their expectations
for a crop activity j periodically in proportion to the difference between actual (Y j,t-1) and
expected values (Y*j,t-1), as follows:
Y*j,t = Y*j,t-1+ λ * [Yj,t-1 – Y*j,t-1], 0 < λ ≤ 1
(1)
in which λ is the coefficient of expectations. If λ=0 then previous prices are expected to persist
(static expectations). If 0<λ<1 then agents make proportional (downward or upward) adjustments
to their expectations. 2
Agent-agent interaction: MP-MAS contains three options for including interactions between
agents: local land markets, local water markets, and technology diffusion. Agents transact land
2
This is the basic functional form used for applications with only little inter-annual variability. In our most recent
applications to semi-arid climates in Ghana and Chile we use more elaborate functional forms with 3 to 5 years
weighted averages.
13
and water rights through an auction mechanism based on shadow prices. If an agent‘s shadow
price for a resource is below the average price then the agent will attempt to rent out a unit of the
resource; otherwise it will try to acquire a unit by placing a bid in the auction. The resource is
temporary exchanged with the highest bidding agent for an amount of rent. The auction
considers internal transport costs, which favors resource exchanges among neighboring agents.
Agent interactions in technology diffusion were implemented as frequency-dependent contagion
effect; the more agents adopt a technology, the more it becomes accessible to others.
Agent-environment interaction: Crop yields in the MAS capture the two-way interaction between
farm decision-making and agro-ecology (Figure 1). For MP-MAS applications that simulate
water flows, crop yields are a function of the water supply, irrigation method, and waterunconstrained yield potential that captures land quality factors constant during the simulation.
For MP-MAS applications with soil nutrient flows incorporated, crop yields are a function of the
nutrient-unconstrained yield potential, management decisions and available soil nutrients.
Agent decisions also have a direct impact on resource dynamics, through crop choice,
investments in irrigation infrastructure and the use of fertilizers, but it is only through impacts on
crop yields and irrigation water that agents receive feedback about changes in resource
conditions. Agents learn about environmental changes through a reduction in crop yields, which
creates an incentive to adjust crop choice and input levels, or to adopt soil and water
conservation methods.
Stochasticity: Random numbers can be used for initializing the MAS if empirical data are
incomplete to parameterize all state variables individually for each agent (see next section). If the
MAS includes rainfall, then stochasticity is also used to simulate rainfall events from historical
14
weather data. If the MAS includes mortality and fertility of agent household members, random
numbers are combined with empirical probabilities to simulate demographic change.
3.5
Initialization
Berger and Schreinemachers (2006) described the Monte Carlo method implemented in MPMAS to generate alternative agent populations from a random sample of farm households.
Following this method, agents are grouped into more homogeneous clusters, and probability
distribution functions are quantified for each cluster and for each resource. A randomization
procedure is then used to generate as many agents as there are farm households in the cluster.
The method has been shown to take correlations between resources into account and to create
agent populations with a close statistical fit to the actual farm population.
Spatial input data include a raster map indicating which pixel belongs to which agent. This map
can be created from a farm survey, cadastral map, or randomly if empirical data are incomplete.
Other spatial data depend on the biophysical component used.
3.6
Input
MP-MAS input data are organized by modules in a set of 14 Excel workbooks (or an SQL
database, see Berger et al. 2010) listed in Table 1.
The generic MP matrix is organized in an Excel workbook called Matrix.xlsx. This workbook
defines the structure of the MP matrix, the signs of the constraints, and contains all matrix
coefficients that do not change during the simulation. Matrix can also be used as a stand-alone
MP model that can be solved independently from the other input files using a spreadsheet solver.
All other input files in MP-MAS, listed in Table 1, tailor the MP matrix to each agent‘s decision
space. Berger and Schreinemachers (2009) provided details about each module and input file.
15
The data requirements of MP-MAS depend on which modules are selected for an empirical
application. At a minimum, it requires farm survey data to parameterize resource endowments
and gross margins for all farm activities. It also requires spatial data on the location of
farmsteads and agricultural fields as well as spatial data to differentiate fields into different
qualities, such as slope, elevation, soil nutrients—depending on the application.
16
Table 1: MP-MAS input files and their function
Input file
Explanation
Function
Standard modules
1. ScenarioManager Defines the scenarios and converts Excel files to plain
Scenario setup
text input files
2. Basic Data
Contains basic parameters used in several modules of
the software (e.g. pixel size)
Model parameters
3. Matrix
Defines the structure and coefficients of the
mathematical programming matrix
To simulate the
decision-making
4. Population
Basic characteristics of the agent populations: e.g.
members, assets, access to technologies
5. Map
All spatial information including the location of agents
and plots, land quality and water flows
6. Network
Defines the characteristics of innovation networks and
specifies technical details of all innovations
7. Demography
Specifies the available labor hours and, optionally,
fertility and mortality rates and food requirements
8. Market
Market prices of purchased inputs and farm outputs.
Optionally specifies parameters of expenditure models
Optional modules
9. Perennials
Input and output of perennial crops by type and age
10. Livestock
Input and output of farm animals by species, age, and
sex
11. Soils
Parameters of the Tropical Soil Productivity Calculator
or coupling data for external software (e.g. LUCIA)
12. Routing
Precipitation levels and the distribution of surface and
sub-surface water flows
13. Water Rights
Agents‘ access to water sources for irrigation
14. CropWat
Specifies crop yields as a function of crop water
supply
Defines the initial
characteristics of
the agents
Define the changes
over time (e.g.,
human ageing, tree
and livestock
growth, price
changes)
Endogenous
biophysics (soil,
water, crop yield)
17
3.7
Submodels
Agent decision module: At the core of the MAS is the MP matrix that simulates the decisionmaking of individual farm households by repeatedly solving a constrained optimization problem.
MP-MAS uses recursive MP as it tailors the matrix to each agent by updating at each time step
its resources, such as land, labor, cash and machinery, as well as its expectations about water,
crop yields, and market prices. MP-MAS uses optimization in a descriptive (―positive‖) and not
in a prescriptive (―normative‖) way. The optimization routine simulates the actual decisions of
farm households; it does not prescribe individual farmers how to produce more with the same
resources.
Following Hazell and Norton (1986: 11), the general optimization (or MP) problem is formulated
as:
n
max Z =
𝑐𝑗 𝑋𝑗
(2)
j=1
Subject to:
n
𝑎𝑖𝑗 𝑋𝑗 ≤ 𝑏𝑖
all 𝑖 = 1 to 𝑚
j=1
𝑋𝑗 ≥ 0
all 𝑗 = 1 to 𝑛
(3)
(4)
In which the farm household objective Z is a linear function of the alternative farm activities X j
(j=1 to n), such as growing crops, raising livestock, hiring labor, consuming food, and adopting
new technologies, and the expected returns per unit of the activity cj. Depending on the empirical
situation, the function Z can be defined in MP-MAS as expected net farm income, net household
income, or a multi-dimensional utility function that includes income and consumption. The
optimization algorithm will find values of Xj that give the highest value for Z without violating
18
the resource constraints (Eq. 3) or involving negative activity levels (Eq. 4). The technical
coefficients aij in Eq. 3 are the quantities of an available resource bi required to produce one unit
of an activity (Xj); for instance, man-days of labor to cultivate one hectare of rice. These
technical coefficients can be derived from empirical production functions, estimated using field
experiment data or farm survey data. Resource constraints can include hard constraints (land,
labor supply, and cash) as well as soft constraints (limited information and market access). They
can also include minimum consumption requirements to capture the fact that in some empirical
settings farm households give, for instance, greater priority to fulfill their own rice consumption
needs than to sell cash crops to the market. The inclusion of soft constraints creates a fissure
between the actual farm decisions and the economic optimum decisions. The use of adaptive
expectations—using expected rather than actual values for crop yields, water supply and market
prices—creates an additional such fissure.
Technology diffusion (optional module): Technology diffusion in MP-MAS follows a two-stage
process. The first stage includes a frequency-dependent contagion effect. Agents are grouped
into five innovation segments reflecting their willingness or ability to adopt new technologies
(that is, their innovativeness). Only when agents in the higher segment of innovativeness have
adopted, does the technology become accessible to agents in the lower segment. Once accessible,
the technology is inserted as one activity in the specific MP matrix of the agent. Solving all MP
matrices for all agents then simulates the sequence of adoption decisions and the emerging
innovation diffusion. Membership to innovation segments can be empirically estimated from
social network analysis (for more details see Berger 2001 and Schreinemachers et al. 2009).
Land and water markets (optional): Transactions between agents regarding land and water rights
are based on shadow prices and distance costs. For instance, the shadow price of land is the
19
amount the MP objective value (usually the expected household income) would increase if one
more unit of land was available. This shadow price minus a distance cost, which captures
transportation to and from the plot, represents the agent‘s willingness to pay for an additional
unit of land.
Livestock (optional): The input data specify input requirements (e.g. labor, feed, pasture area,
cash) and output quantities (e.g. milk, meat, eggs, offspring, manure) per head for each animal
species and for each age in its lifespan. The choice of animal species as well as the number and
type of inputs and outputs is flexible. At each time step, these values are updated in the agent‘s
MP matrix, according to an individual-based livestock herd model. Agents then decide whether
to maintain or sell the animal.
Perennial crops (optional): Similar to livestock, the input data also specify the input
requirements (labor, cash, and water) and output quantities per hectare of perennial crops (e.g.
apple trees, coffee, roses) for each time step in their lifespan. Different from livestock, the
current version of MP-MAS does not allow agents to cut their perennial crops, which are
assumed to last till the end of their pre-defined lifespan. Agents can, however, switch
management practices and change the input intensity and yields in their orchards.
Biophysical modules (optional): MP-MAS simulates resource dynamics and crop yields, either
utilizing built-in features or using external software. The decision-making module interacts with
the biophysical module in the following way: First, the decision-making module simulates land
use based on expected conditions (soil, water, prices). The simulated land use is translated to a
raster map and is transferred to the biophysical module, which then simulates the actual resource
conditions and calculates the actual crop yield in each pixel. Finally, the actual crop yield (and
20
optionally, the actual water supply) is transferred back to the agent decision module where it is
evaluated and used to update the yield expectations and, optionally, water expectations.
Two types of resource dynamics are currently incorporated in MP-MAS, one related to water
flows, the other related to soil erosion and soil nutrients. For each of these there are empirical
and process-based implementations available. Empirical here means that biophysical processes
in soils and plants are not made explicit, but are approximated by fitting functions to observed
data.
First, the EDIC (spanish acronym of Civil Engineering Consortium in Chile) simulation model is
an empirical model of irrigation water flows using a node-link network (for parameters and
equations, see Arnold 2010). The landscape is divided into irrigation sectors that receive water
from streams (surface flows) and sub-surface flows. Fitted parameters, defined in the input data,
determine the proportion of the inflows used for irrigation in each sector. The EDIC model
considers the irrigation return-flows, i.e. seepage, bypass and tailwater flows, that can be reused
further downstream. Since return-flows result from the actions of upstream water users, this
particular feature constitutes one important agent-environment interaction. The input data also
defines the total volume of river flow and precipitation on a monthly basis.
Second, WASIM-ETH (Water balance Simulation Model ETH-Zürich) is a process-based
hydrology model developed by Jasper et al. (2002) to capture water flows at high spatial and
temporal resolutions. The model contains features such as a set of evaporation algorithms, an
interception module, a layered soil module based on the Richards equation, a multi-layer
groundwater module, and a glacier module (Schulla and Jasper 2007). WASIM-ETH can be
coupled to MP-MAS as an external software component, using the Typed-Data-Transfer (TDT)
library of Linstead (2004) (see details in Arnold et al. 2008). To capture irrigation flows with
21
relatively low on-field irrigation efficiencies, we extended the original WASIM-ETH source
code as explained in Uribe et al. (2009).
Both EDIC and WASIM-ETH determine the water supply to each irrigation sector in the
landscape. A predefined set of agent water rights then determines the actual distribution of water
among agents and plots in each sector. Here there are three options available in MP-MAS: (i) all
agents have equal access to water, (ii) all plots have equal water supply, or (iii) empirical water
distribution rules (for example, based on the actual water rights registry of water user
associations). As an optional feature, agents can trade water rights amongst themselves. The crop
yield is a function of the deficit between the crop water supply and the crop water requirement,
as based on the FAO56 (also called CropWat) model (Allen et al. 1998). The MAS simulates the
actual yield by multiplying the water-unconstrained yield potential with a reduction factor. The
input data specify the crop water requirements and irrigation methods.
Third, the Tropical Soil Productivity Calculator (TSPC) is an empirical model of soil fertility
changes and resulting crop yields (Aune and Lal 1995). Schreinemachers (2006) specified the
model equations separating nutrient stocks and available nutrients for plants. TSPC simulates
crop yields by multiplying the crop yield potential with a reduction factor, if available nutrients
(N, P, and K), acidity and soil organic matter are below their optimum levels. Available nutrients
are a function of initial soil conditions, harvest removal, erosion and leaching, decomposition of
organic matter, and management decisions.
Fourth, the Land Use Change Impact Assessment (LUCIA) model is a raster-based, spatiallyexplicit dynamic model that simulates watershed functions, soil fertility and plant growth for
small catchment areas up to about 30 km2 (Marohn et al. 2010). LUCIA has hydrological
functions that describe the flows and stocks within each pixel, while flows between pixels are
22
routed according to a local drain direction map. Biomass production in LUCIA follows a
hierarchical approach in which the potential growth—determined by radiation, temperature, leaf
area index, and development-dependent maximum assimilation—is first reduced by the water
supply and then by N, P, and K availability based on CGMS-WOFOST (Supit 2003). LUCIA
can be coupled to MP-MAS as an external software component using the TDT library.
Fifth, EXPERT-N is a process-based model system to simulate crop growth in arable farming. It
builds on state-of-the-art crop growth models such as CERES, SPASS, SUCROS, and GECROS,
and contains various modules for soil water flow, soil heat transfer, soil carbon and nitrogen
turnover, and other soil-plant-atmosphere processes on a daily basis (Priesack et al. 2001, 2006).
Like WASIM-ETH and LUCIA, EXPERT-N can be coupled with MP-MAS as an external
software component using the TDT library (Berger et al. 2010).
4
Modeling choices in MP-MAS
4.1
Empirical applications
MP-MAS has been applied to various research questions related to the impact of policy changes
and the introduction of agricultural innovations, as listed in Table 2, with studies located in six
countries: Chile, Germany, Ghana, Thailand, Uganda and Vietnam. The applications to
Germany, Thailand, and Uganda have detailed associated documentation which can be
downloaded from the software website (see first page).
23
Table 2: Research questions addressed in MP-MAS applications
Project
period
Country
Location of site
Research question
Key references
Chile
Maule river basin
Berger et al. 2007, 2004Berger 2001
2009
Germany
Swabian Jura,
southwest
Germany
What would be the impact of water
pricing on the efficiency of irrigation
water use and land use dynamics?
What could be the impact of climate
change on land use and farm incomes?
Ghana
White Volta basin
What would be the impact of a
transition from rain fed to irrigated
agriculture?
Birner et al. 2010
20042009
Thailand
Chiang Mai
Province in the
mountainous north
What would be the impact of four fruit
tree technologies (fruit drying, artificial
flower induction, extended shelf-life,
and improved irrigation) on land use,
erosion and levels of pesticide use?
Schreinemachers
et al. 2010, 2009
20062012
Uganda
Lake Victoria
crescent in the
southeast
What would be the impact of hybrid
maize varieties and better access to
farm credit on poverty and
environmental sustainability?
Schreinemachers
et al. 2007,
Schreinemachers
2006
20012006
Vietnam
Son La Province
in the country‘s
mountainous north
How would the adoption of soil
conservation measures affect short- and
long-term household incomes?
Quang et al. 2010
20062012
Berger et al. 2010 20092011
Table 3 shows that MP-MAS can be applied to a variety of spatial extents and human population
sizes. In large research interdisciplinary research projects, the spatial extent is often based on
natural boundaries such as river basins or watersheds (the Chile, Ghana, Thailand, and Vietnam
applications). Given availability of empirical data, MP-MAS can use any spatial extent or
population size and can therefore adjust itself to the spatial extent of the biophysical modules
selected. The application to a catchment area in Vietnam had a spatial extent of about 6 km2,
while the application to a river basin in Chile covered 5,300 km2. The number of households
individually represented in MP-MAS ranged from 471 in Vietnam to 34,691 in Ghana. Also the
pixel size can vary between applications, but should ideally be set to the smallest observed field
size to make sure that every field is represented in the MAS.
24
Table 3: Scale of landscape and agent representation in MP-MAS
Model component
Chile
Germany
Extent of study area
Watershed
Landscape Watershed
Watershed Two villages Catchment
Pixels
530,000
129,758
377,900
8,542
2,258
61,965
Pixel size (m)
100x100
100x100
100x100
40x40
71x71
10x10
Agricultural area (km2)
5,300
1,297
3,779
13.67
11.38
6.20
3,592
1,700
34,691
1,309
520
471
0.004
0.23
0.23
0.34
Ghana
Thailand
Uganda
Vietnam
Landscape
Agents
Number
Sampling fraction
1.0
*
1.0
*
Note: * Based on census data and therefore the sampling fraction is 1.0.
Generally, both water and soil conditions determine crop yields but, as Table 4 shows, most
applications either represented hydrology processes or soil processes in MP-MAS. The
identification of the most constraining factor to crop yields mainly determines this selection. The
exception is the coupling to LUCIA, which simulates both hydrology and soil dynamics. It
would also be possible in MP-MAS to combine FAO56 + EDIC + TSPC.
The choice between empirical biophysical models (EDIC, TSPC) vs. process-based biophysical
models (WASIM-ETH, LUCIA, EXPERT-N) largely follows from the organization of the
project and its research objectives. Soil and plant scientists in large multi-disciplinary teams
usually prefer working with state-of-the-art biophysical models and can hardly be convinced of
using simple empirical models. This is especially the case when the project seeks to find new
insights into the agricultural system, which can only be obtained by inclusion of detailed
biophysical processes. In small teams or individual PhD projects, however, selecting processbased biophysical models is clearly out of scope.
Table 4: Main differences between the MP-MAS applications
25
Model feature
Chile Germany Ghana Thailand Uganda Vietnam
Crop-soil
TSPC






EXPERT-N






FAO56 + EDIC






FAO56 + WASIM-ETH












Perennial crop production






Livestock production






Out-migration






Expenditure model (LA-AIDS)






Fertility & mortality (pop. growth)






Endogenous commodity prices






Endogenous land markets






Endogenous water markets






Innovation diffusion in networks






Expected net cash income






Exp. consumption of own farm produce






Minimum food requirements






1,119
1,280
752
1,820
2,350
941
Number of constraints in MP
224
879
250
812
560
488
Number of integers in MP
27
23
60
73
50
43
Crop-hydrology
Crop-soil-hydrology
LUCIA
Optional modules
Agent-interactions
Agent objectives
Size of the decision component
Number of activities in MP
26
Each study tailored MP-MAS to a different agricultural system by specifying livestock, crops,
crop combinations, and crop constraints in the MP matrix, and by selecting the relevant model
features. Table 4 shows for each application which modules were selected. The empirical setting,
the objectives of the research, and the particular project set-up determined each selection. For
instance, land transactions are only included if this is frequent in reality, and population growth
is only included if there is evidence that this is a driver of land use change.
Table 4 also shows the size of the MP matrix. The number of activities is generally large as there
is one separate activity for each combination of crops, input levels, and land quality. The Uganda
application, for instance, included 11 crops and 7 intercrop combinations, each segmented into
90 activities by specifying different combinations of land quality, labor, and fertilizers
(Schreinemachers et al. 2007). Technical coefficients were derived from regression functions,
and the non-linear response of crop yields to alternative combinations of inputs was captured
using a piecewise linear segmentation.
4.2
Selected results related to technology assessment
In the study of the Maule river basin in Chile, Berger (2001) showed that the impacts of new
technologies crucially depend on the type of technology diffusion and the adoption cost this
involves. Berger compared two scenarios. In one scenario, the agents had equal and immediate
access to all technologies, while in the other technologies were diffused through a frequencydependent contagion effect. He found that the latter scenario created significantly more structural
change in farm sizes and land use patterns: early adopters gained higher profits, which increased
the amount adopters would pay non-adopters for their land, and led to more land transactions
from non-adopters to adopters.
27
Schreinemachers et al. (2007) assessed the impact of better access to short-term credit, improved
maize varieties, and mineral fertilizer both on soil fertility and poverty distribution in two
Ugandan villages (Figure 2). They found that improved access was able to reduce poverty by 20
percent, as 31 percent of the agents moved from below to above the poverty line, which was
expressed in terms of calorific intake in the agent households. The ecological sustainability of
the system was, however, not improved as inorganic fertilizers did not yield long-run benefits in
terms of soil fertility.
Figure 2: Distributional effect of improved access to credit and technologies in a Ugandan
village
0.15
Baseline
0.10
0.05
poverty line
Kernel density
With access to
credit & technologies
0.00
0
5
10
15
20
Consumption poverty
[billion joules/capita/year]
Note: Simulated poverty levels for 520 agents in two scenarios averaged over 15 years. Poverty line is 3.3 billion
joules per capita (in male adult equivalents) per year.
In a study to a watershed in northern Thailand, Schreinemachers et al. (2009) compared the
diffusion of greenhouse agriculture for farmers with and without transferable land titles that can
be used as mortgage collateral. They found that if mortgage collateral would not be required,
28
then diffusion among farmers without titles could reach nearly 77% by 2020, as compared to
about 36% under current conditions (Figure 3).
Figure 3: Diffusion of greenhouse agriculture in a northern Thai watershed, 1995-2020
Upper part of watershed
100
90
80
Adoption 1995-2006
Baseline
Credit without collateral
70
percentage
60
50
40
30
20
10
0
1995
2000
2005
2010
2015
2020
year
Note: Graph shows observed values for 1995-2006 and simulated values for 2007-2020.
The results show the importance of capturing agent heterogeneity when simulating technology
adoption in agricultural systems. Due to the heterogeneity of resources, technologies are not
equally attractive to all farm households; and further, because of the heterogeneity in terms of
innovativeness, the benefits of adoption are spread unequally. The combination of these two
factors creates an intricate pattern of technology diffusion, resource use, and environmental
externalities that these software applications are capable of capturing.
4.3
Model validation tests
McCarl and Apland (1986) separated between ‗validation by construct‘ and ‗validation by
results‘. In MP-MAS, validation by construct is tackled by building the MAS on well-established
theories in farm economics and agro-ecology. Production and expenditure functions have a solid
29
footing in economic theory, and validity of their functional parameters can be judged from their
signs, their significance levels, and the predictive power of the function as a whole.
Validation by results is achieved by studying the baseline scenario, which reflects the current
situation and current processes without additional interventions. As a minimum requirement, all
applications of MP-MAS compared the observed land use with the simulated land use. Spatial
validation methods have so far not been used with MP-MAS as all applications have studied land
use at a household or population level and none have looked at the spatial patterns of land use
change. In the following we highlight the validation methods used in the Uganda and Thailand
studies.
The Ugandan study validated the baseline in three ways. It first compared the simulated soil
nutrient balances to secondary data from literature and found the model predictions to be within
the range of what other studies had suggested. Second, the study regressed the observed land use
(as a percentage of the total crop area) on the simulated land use averaged over the first three
years of the simulation run and found a sufficient goodness-of-fit. Third, it compared the
simulated distribution in poverty levels with the actual distribution from survey data and also
found a close resemblance between these distributions.
The Thailand study solely relied on regression of observed land use on simulated land use but
did so at five levels of aggregation from the watershed to the individual household level. It
showed a good fit at the watershed level, a sufficient fit at the cluster level, and a moderate fit at
the household level. This suggests that MP-MAS was not able to replicate the individual land use
decisions of each real-world household, but at higher levels of aggregation (agent clusters,
innovation segments) it was a sufficiently good statistical representation of reality.
4.4
Treatment of uncertainty
30
The Monte Carlo initialization of MP-MAS generates many possible and statistically consistent
agent populations that can then be used for repetitions of simulation experiments. For example,
Schreinemachers et al. (2010) allocated agents randomly to innovation segments to study the
diffusion of tree crop technologies in northern Thailand; they ran the scenario ten times and then
depicted the diffusion curves using confidence intervals.
Comparing impacts from different policy interventions requires some form of uncertainty
analysis. We explain our approach by referring to the Ugandan case study in Berger and
Schreinemachers (2006). As shown in Figure 4, we tested the sensitivity of simulation outcomes
for key policy indicators such as crop yield, productivity, and farm income. Variation of these
indicators was measured by standard deviations (expressed as percentages of the normalized
mean) in fifty different agent populations for the baseline scenario.
Figure 4: Model uncertainty for key policy indicators in Uganda
Note: Variation results from fifty different agent populations for the baseline scenario. Variation is measured by
standard deviations (expressed as percentages of the normalized mean).
31
The sensitivity tests revealed rather low variation for most key policy indicators in the order of
5%; only in the case of farm assets including savings variation was in the order of 10%. This
inherent model ―noise‖ (or uncertainty) has to be considered when comparing various simulation
experiments, for example on policy interventions in markets for credit and fertilizers. Only in
case that the policy indicators differ by more than this ―noise‖, interventions should be
interpreted as having significant impact. More details on uncertainty and sensitivity testing can
be found in Schreinemachers (2006).
5
Comparison to other recent simulators
In this section we briefly compare MP-MAS to eight other agent- and non-agent-based software
packages that simulate human-environment interactions in agricultural systems. The purpose is
to compare decision routines, biophysical dynamics, and the interactions between them as well
as the way these features are implemented in the various software packages. We will not try to
judge model and software quality, which is impossible, as the modeling approaches represent
different agricultural systems, have different purposes and cover different spatial extents.
5.1
Selected simulators for comparison
Table 5 lists the simulators in our comparison and summarizes their objectives and application
areas; the footnote to the table cites the articles that the comparison was based on. Two criteria
were used in selecting the candidates for comparison. The first criterion was that the decisionmaking and the biophysical dynamics were both endogenous and interact with each other.
Simulation models such as DSSAT (Jones et al. 2003), APSIM (Keating et al. 2003), SWAT
(Arnold et al. 1998), and many others, also simulate environmental processes and include a
32
management component, but do not simulate decision-making endogenously. We also have not
included agricultural systems models such as AgriPolis (Happe et al. 2006) or LUPAS (Roetter
et al. 2007, 2005), which incorporate endogenous land use decisions, but do not have
endogenous biophysical dynamics that react to the land use decisions.
The second criterion was that these software packages simulated the actual decision-making of
farm households. We therefore excluded all types of optimal control models that give a
prescriptive (―normative‖) assessment on how to optimize decision-making. All simulators
selected are descriptive (―positive‖) models and use some form of adaptive expectation
formation, which captures the fact that real-world decision-makers cannot predict crop yields
with certainty.
33
Table 5: Basic information the eight simulators used in the comparison
A. Software for simulating water flows
Software
1
Study area
Agricultural system Purpose of the study
(broadly defined)
Decision-making
Resource dynamics
One river basin
in northeastern
Brazil (~74,000
km2)
Commercial agriculture, semi-arid,
with upstream and
downstream water
users
To analyze the dynamics
of water use for irrigation
and its effect on the
distribution of water
resources
Heuristics (agent-based)
with random elements for
crop choice. Separate
heuristics for upstream,
floodplain, and
downstream agents
Semi-distributed water
balance model (nodelink-network) and a
hydraulic model to
regulate water flows on
10 day interval
Crop growth simulated
on 10-day interval as a
function of water stress
based on CATCHCROP
2. CATCH- One sub-catchSCAPE3
ment in north
Thailand (~44
km2) 4
Semi-commercial
with upstream and
downstream water
users and forest use
To facilitate the negotiation between
stakeholders about the
sustainable use of water
resources
Constrained optimization
(agent-based) combined
with heuristics for rice
planting and off-farm
labor.
Same as ABSTRACT
Same as ABSTRACT
3. IMT 2
One subcatchment in
north Thailand
(~44 km2) 2
Semi-commercial
agriculture on
sloping land (rice,
sorghum,
groundnut)
To assist stakeholders in
their land-use planning at
a catchment level
Constrained optimization
(representative farm
approach) of expected
annual income, using
three farms for the study
area
Rainfall-runoff model
(IHACRES), a water
allocation model based
on stakeholder
recommendations, and
an erosion model
(USLE)
Crop growth (10-day
time step) function of
temperature, water
supply, soils, and
fertilizer use as based on
CATCHCROP
4. SIM
Two states in
northeastern
Brazil
(~400,000 km2)
Commercial
agriculture under
semi-arid
conditions
To assess the impact of
climate change on natural
resources and to identify
possible coping strategies
Constrained optimization
of farm income to
simulate land use
decisions at municipality
level (~1200 km2)
Regional-level models
for climate change
scenarios, water balance,
and water use
Crop yields function of
water and aeration stress,
incl. variations in soil
quality and management
1. ABSTRACT
Crop yields
34
B. Software for simulating soil fertility dynamics
Software
1
Study area
Agricultural system Purpose of the study
(broadly defined)
Decision-making
Resource dynamics
5. ECECMOD
Four regions in
southern
Norway (~989
km2)
Commercial
agriculture (grains,
grassland,
livestock)
To estimate the effect of
changes in economic
incentives on emissions
(N, P, and pesticides)
Constrained optimization
(representative farm
approach) of expected
profit
Models of hydrology, N Crop yield is a function of
and P dynamics, weeds, N (from manure, mineral
fertilizer, organic matter)
erosion and leaching,
and ammonia losses
6. LUDAS
One watershed
area in central
Vietnam (~100
km2)
Subsistence
agriculture at forest
margins, including
forest logging
To provide technical
decision support for
stakeholders in land use
planning
Heuristics (agent-based)
combined with random
elements
Natural vegetation
growth using rule-based
transitions in land cover.
A growth equation
simulates the forest
stand basal area
Crop yield is a function of
input and labor use,
physical soil
characteristics and land
use history based on
regression model
7. PALM
Mid-hills of
Nepal (~0.98
km2)
Semi-subsistence
agriculture on
sloping land
susceptible to
degradation
To evaluate the potential Heuristics (agent-based)
combined with mimicking
of various soil fertility
enhancing innovations and behavior for 98 agents
the likelihood of their
adoption
CENTURY model of
organic matter
decomposition 5. Water
and N dynamics based
on DSSAT family of
models 6
Crop growth (1-day time
step) function of climate,
soils, and input use with
separate models for crops,
trees and weeds
8. SAM
South-western
Saskatchewan,
Canada (~1,364
km2)
Commercial
agriculture (wheat,
canola, pea, forage,
and native
vegetation)
To evaluate the agroecosystem sustainability
of agricultural production
systems
Model of soil water
availability to crops, and
models for available N
and P
Crop yield a function of
climate and soil quality;
the latter influenced by
crop management in
previous years
Constrained optimization
of profit with rules for
fertilizer use (N and P) to
simulate land use at the
level of four ―ecodistricts‖
Crop yields
Notes: 1 Some software packages include additional components not listed here. The comparison was based on the following references: CATCHSCAPE: Becu
et al. 2003; ABSTRACT (Agent-based Simulation Tool for Resource Allocation in a Catchment): van Oel et al. 2010; SIM (Semi-arid Integrated Model): Krol
and Bronstert 2007; IMT (Integrated Modeling Toolbox): Letcher et al. 2006a, 2006b; SAM (Sustainable Agroecosystem Model): Belcher et al. 2004; PALM
(People and Landscape Model): Matthews 2006; ECECMOD (Economics and Ecology Modeling System): Vatn et al. 1999, 2006; LUDAS (Land Use Dynamic
Simulator): Le et al. 2008, 2010. 2 Developed within the Integrated Water Resource Assessment and Management (IWRAM) project. 3 Based on the CORMAS
platform (Bousquet et al. 1998) and is used for participatory simulation modeling with stakeholders. 4 CATCHSCAPE and IMT use the same study site (Mae
Uam sub-catchment). 5 Parton et al., 1988; Parton and Rasmussen, 1994. 6 http://www.icasa.net/dssat/
35
5.2
Decision-making
Figure 5 classifies the software packages according to the decision-making modules
implemented. SAM and SIM do not model individual farm households but higher levels of
aggregation (eco-districts and municipalities, respectively); both use constrained optimization to
simulate land use decisions. ECECMOD and IMT also use constrained optimization but, similar
to conventional bio-economic models described above, use a representative farm approach and
do therefore not individually represent each real-world farm household. Of the agent-based
simulation models, ABSTRACT, LUDAS and PALM use heuristics while MP-MAS and
CATCHSCAPE use constrained optimization to simulate crop choice and management
decisions, adding heuristics to capture additional decisions in the MAS. However,
CATCHSCAPE uses a highly simplified MP matrix with most decisions modeled using
heuristics.3 The comparison hence shows that MP-MAS distinguishes itself in its use of a
detailed constrained optimization approach to represent the decision-making of each individual
farm household. It is therefore the only agent-based software package with a strong footing in
microeconomics and farm management theory.
Figure 5: Classification of decision-making components
Agent-based
Household level
Representative farmbased
Aggregate level
(eg. municipality)
3
Heuristics
ABSTRACT, LUDAS, PALM
Constrained optimization
CATCHSCAPE, MP-MAS
Constrained optimization
ECECMOD, IMT
Constrained optimization
SAM, SIM
CATCHSCAPE actually represents only two-fifths of the real-world plots and farm households.
36
Of all simulators in our comparison only PALM and MP-MAS have modules for technology
diffusion. Agents in PALM mimic each other if they see their immediate neighbors use a strategy
that gives higher returns. This is different from MP-MAS, which combines communication
network effects with an economic evaluation of technologies in which agents only adopt if they
expect that adoption would contribute to reaching their individual household objectives.
5.3
Environmental dynamics
The conventional taxonomy for environmental models separates them into deterministic
(process-based and statistical-based) and stochastic models. Integrated software packages,
however, almost always combine all these features into a single modeling approach. As an
alternative, we classify the various software packages in a first step based on the main type of
biophysical process modeled (i.e., hydrology or soil dynamics), and in a second step based on
their temporal resolution (Figure 6). We note that models with fine temporal resolution are
mostly process-based while those using a coarse resolution are mostly empirical-based, relying
on statistical relationships in the data.
Figure 6: Classification based on biophysical modules
Simulating water
flows
Fine temporal resolution
ABSTRACT, CATCHSCAPE, IMT, SIM,
MP-MAS(+WASIM ETH)
Coarse temporal resolution
MP-MAS(+EDIC)
Simulating soil
fertility change
Fine temporal resolution
ECECMOD, PALM, SAM,
MP-MAS(+LUCIA), MP-MAS(+EXPERT N)
Coarse temporal resolution
LUDAS, MP-MAS(+TSPC)
37
Among the software packages simulating water flows, ABSTRACT, CATCHSCAPE, and
ITM—all based on the CATCHCROP model (Perez et al. 2002)—include hydrological
processes at a 10-day time step while SIM simulates water availability at a daily time step. This
fine temporal resolution is comparable to the MP-MAS (+WASIM) specification while MPMAS (+EDIC), on the other hand, uses a coarse resolution of monthly time steps and, unlike the
other simulators, is not process-based. The option using the EDIC module is suitable if water
availability is an important aspect of the system under study, yet hydrological data such as flow
measurements are largely unavailable.
Among the software packages simulating soil fertility changes, we can also distinguish between
simulators using a fine temporal resolution (ECECMOD, PALM, SAM, MP-MAS (+LUCIA or
EXPERT N)), which are based on an advanced understanding of the biophysical processes in the
soil-plant system, and those that rely on statistical correlation in the data (LUDAS, MP-MAS
(+TSPC)).
The comparison hence shows that MP-MAS is relatively flexible in its use of biophysical
modules. It can either simulate water flows or soil fertility changes; and it can either use a fine or
a course temporal resolution, depending on the purpose of the application and the availability of
empirical data.
5.4
Human-environment interactions
In all software packages, agents influence the environment through their land use and input
decisions while the environment influences agents by returning a level of crop yield, which is a
function of input decisions and environmental processes such as weather, water flows, and soil
nutrients. The last column in Table 5 shows how each package simulates crop yields.
38
As farmers cannot know at planting time, how much they will actually harvest at the end of the
year, most software packages contain modules for explicit simulation of expectations, especially
about crop yields. In IMT and CATCHSCAPE the expected water supply and expected yields
are the actual simulated values from the previous year, also called naïve expectations. This is
also possible in MP-MAS, by setting the coefficient λ in Equation 1 to zero. But MP-MAS also
allows for more gradual adjustments by setting λ to greater than zero. MP-MAS is also the only
package in which agents update price expectations.
The other software packages used different approaches to expectation formation. For instance,
ECECMOD has no internal updating of expectations but approximates expectations (about crop
yields, N absorption, weed development, ammonia losses and several other biophysical
conditions) from average simulated values of model pre-runs for each soil type and each
agronomic practice. SAM assumes farmers will apply optimum amounts of N fertilizer, but
reduces the amount if soil moisture at the beginning of a period falls below a threshold level, as
farmers expect drought. In the case of ABSTRACT, agents in the upstream and floodplain areas
form their expectations about the water availability in the dry season, by comparing the actual
water availability in the preceding wet season to a threshold level (defined in the input data);
agents in the downstream area form their expectations from the actual amount of water collected
in a reservoir.
5.5
Model integration and software coupling
Approaches to model integration and software coupling fall into two classes. First, embedded
coupling is an approach in which all model components are incorporated in the same source
code, which often involves rewriting or merging existing code into the code of the integrated
model system. This type of model integration has been applied for CATCHSCAPE (written in
39
SmallTalk), LUDAS (NetLogo), PALM (Delphi), ECECMOD, IMT, and MP-MAS
(+EDIC/+TSPC). For large model components or models developed in interdisciplinary research
teams, this practice is usually impractical and run-time software coupling might be preferable.
This class of external coupling includes SIM, MP-MAS (+WASIM ETH/+LUCIA/+EXPERT
N).
A further distinction can be made in the frequency of interaction between the land use decision
module and the biophysical module in each software. For instance, ABSTRACT loops through
its set of decision heuristics at 10-day time steps, equal to the hydrology module. If the algorithm
finds a date to plant, it evaluates the expected water availability and if enough, chooses a crop
based on a random probability function that reflects the crop preferences of the respective agent
group. Also PALM simulates agent decisions using heuristics at the same time steps as the
biophysical module, and considers farm activities in great detail, including planting, fertilizing,
weeding and milking.
All other simulators execute the land use decision module and the biophysical module only once
per period. For example, MP-MAS simulates the agent land use decisions for a whole year and
then simulates the biophysical processes and crop yields. There is hence no updating of land use
decisions within one period in response to environmental changes. An exception is MP-MAS
(+EDIC/+WASIM ETH) with irrigation, in which agents irrigate their plots at monthly time
steps and change their original land use plans in case they receive less irrigation water than
expected.
40
Figure 7: Classification based on software coupling approach
Integration of model
components in the
software code
Coupling of software
components
5.6
Single human-env. interaction per time step
CATCHSCAPE, ECECMOD, IMT, LUDAS, SAM,
MP-MAS(+TSPC)
Multiple human-environment interactions per time step
ABSTRACT, PALM, MP-MAS(+EDIC/+WASIM
ETH)
Single human-env. interaction per time step
SIM, MP-MAS(+WASIM ETH), MPMAS(+LUCIA), MP-MAS(+EXPERT N)
Limitations of MP-MAS
MP-MAS has several limitations as the software package was developed case-by-case to capture
the particular research questions in the six aforementioned applications. It would be possible to
address these limitations in future versions of the software.
MP-MAS is limited to simulating the decision-making of farmers; other agents such as landowners and water user associations are represented with highly simplified decision rules.
CATCHSCAPE, for instance, models the decision-making of other agents such as government
agencies, industries, and non-farm households. MP-MAS is therefore not suitable for modeling
situations in which farmers compete for resources with non-farmers.
Moreover, agents in MP-MAS cannot yet coordinate their decisions. Agent can, for instance,
exchange water rights if using the optional water market module, but agents cannot collectively
decide to build a new water reservoir or to dam a river. The impact of such changes can only be
assessed by introducing such change exogenously through the input data of the software and then
comparing the simulation outcomes with and without the reservoir or dam. We note that none of
the other simulation packages allows for endogenous collective action to emerge. Work is
41
ongoing to implement group decision-making of user associations in MP-MAS, for example, for
joint investment and marketing activities.
Different from CATCHSCAPE, LUDAS and IMT, agents in MP-MAS cannot encroach into
collective forests or open up new land areas. The current version of the software relies on clearly
defined ownership rights over pixels; only if agents are given ownership over particular forest
pixels can the software simulate their decision to leave it as forest or to convert it to agricultural
land.
6
Conclusion
The strength of the MP-MAS software package is in its combination of the following three
aspects:
1. The use of constrained optimization to simulate decision-making of farm households gives
MP-MAS a strong footing in microeconomics and farm management theory. It is unique in
separating investment, production, and consumption decisions, and by including price
expectations and market dynamics.
2. The use of an agent-based modeling approach allows MP-MAS to capture much economic
and environmental heterogeneity among agents; it also allows capturing interactions between
agents, and facilitates the combination with spatial modules simulating environmental
dynamics. With regard to technology diffusion, MP-MAS is unique among integrated models
of agricultural systems in combining the communication of innovation information with the
microeconomic evaluation of new technologies.
3. Various optional modules for environmental dynamics allow users to tailor their MP-MAS
application to different purposes and different levels of data availability. Modules for water
42
flows and soil nutrient changes are available at both coarse and fine temporal resolutions. In
addition, MP-MAS can be coupled at run-time to process-based external software packages
such as WASIM-ETH, LUCIA, and EXPERT-N.
Acknowledgments
We would like to thank five anonymous reviewers for their thoughtful comments. We also would
like to thank the MP-MAS developer team and all model users who helped to improve the
software package over the years. Funding by the Deutsche Forschungsgemeinschaft for the
projects SFB-564 and PAK-346 is gratefully acknowledged.
References
Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotranspiration - Guidelines for
computing crop water requirements. FAO Irrigation and drainage paper 56. Available at:
http://www.fao.org/docrep/X0490E/X0490E00.htm (Accessed 6 September 2010).
Argent, R.M., 2004. An overview of model integration for environmental applications —
components, frameworks and semantics. Environmental Modeling & Software. 19, 219234.
Arnold, J.G., Srinivasan, R., Muttiah, R.S., Williams, J.R., 1998. Large area hydrologic
modeling and assessment. Part I: Model development. Journal of the American Water
Resources Association. 34, 73-89.
Arnold, T., Berger, T., Uribe, H., 2008. Step by step calibration of an Integrated Model System
for irrigation management. In: Sànchez-Marrè, M., Béjar, J., Comas, J., Rizzoli, A.E. and
Guariso, G., eds., Proceedings of the iEMSs Fourth Biennial Meeting: International
Congress on Environmental Modeling and Software (iEMSs 2008), International
Environmental Modeling and Software Society, Barcelona.
Arnold, T., Uribe, H., Troost, C., Berger, T., 2010. Irrigation Management in Chile: Integrated
Modeling of Access to Water. In: Swayne, D.A., Yang,W., Voinov, A.A., Rizzoli, A.,
Filatova, T. (Eds.), International Environmental Modelling and Software Society (iEMSs)
2010 International Congress on Environmental Modelling and Software, Ottawa, Canada.
Available at: http://www.iemss.org/iemss2010/index.php?n=Main.Proceedings (Accessed 6
September 2010).
Aune, J.B., Lal, R., 1995. The Tropical Soil Productivity Calculator—A Model for Assessing
Effects of Soil Management on Productivity. In: Lal, R., Stewart, B.A. (Eds.), Soil
43
management. Experimental basis for sustainability and environmental quality, Lewis
Publishers, London, pp. 499-520
Balmann, A., 1997. Farm-based modeling of regional structural change: A cellular automata
approach. European Review of Agricultural Economics. 24, 85-108.
Becu, N., Perez, P., Walker, B., Barreteau, O., Page, C.L., 2003. Agent-based simulation of a
small catchment water management in northern Thailand. Description of the Catchscape
model. Ecological Modeling. 170, 319–331.
Belcher, K.W., Boehm, M.M., Fulton, M.E., 2004. Agroecosystem sustainability: a system
simulation model approach. Agricultural Systems. 79(2), 225-241.
Berger, T., 2001. Agent-based models applied to agriculture: a simulation tool for technology
diffusion, resource use changes and policy analysis. Agricultural Economics. 25(2/3), 245260.
Berger, T., Birner, R., Díaz, J., McCarthy, N., Wittmer, H., 2007. Capturing the Complexity of
Water Uses and Water Users within a Multi-Agent Framework. Water Resources
Management. 21(1), 129-148
Berger, T., Schreinemachers, P., 2006. Creating agents and landscapes for multi-agent systems
from random samples. Ecology and Society. 11(2), Art.19. Available at
http://www.ecologyandsociety.org/vol11/iss2/ (Accessed 6 September 2010).
Berger, T., Schreinemachers, P., 2009. Mathematical programming-based multi-agent systems
(MP-MAS). Technical model documentation. Version 2.0 (February 2009). Stuttgart,
Hohenheim
University.
Available
at
https://www.unihohenheim.de/mas/software/Manual.pdf (Accessed 6 September 2010).
Berger, T., Schreinemachers, P., Woelcke, J., 2006. Multi-agent simulation for the targeting of
development policies in less-favored areas. Agricultural Systems. 88, 28-43.
Berger, T., Troost, C., Calberto, G., Ingwersen, J., Warrach-Sagi, K., 2010. Fine resolution
modeling of adaptation to climate change in agriculture, Research documentation.
Stuttgart,
Universität
Hohenheim.
Available
at
https://www.unihohenheim.de/mas/Germany/PAK043Public.zip (Accessed 6 September 2010).
Birner, R., Berger, T., Schiffer, E., 2010. Integrating Knowledge from Computational Modeling
with Multi-stakeholder Governance: Towards More Secure Livelihoods through Improved
Tools for Integrated River Basin Management. CPWF Project Report. Available at
http://www.igm.uni-hohenheim.de/ (Accessed 6 September 2010).
Bousquet, F., Bakam, I., Proton, H., Le Page, C., 1998. Cormas: common-pool resources and
multi-agent systems. Lecture Notes in Artificial Intelligence. 1416, 826-838.
Day, R.H., 2005. Microeconomic foundations for macroeconomic structure, Papers on
Economics and Evolution, 14. Jena, Evolutionary Economics Group, MPI Jena. Available
at ftp://papers.econ.mpg.de/evo/discussionpapers/2005-14.pdf (Accessed 6 September
2010).
Day, R.H., Singh, I.J., 1975. A dynamic model of regional agricultural development.
International Regional Science Review. 1, 27-48.
Doglioni, A., Primativo, F., Laucelli, D., Monno, V., Khu, S.-T., Giustolisi, O., 2009. An
integrated modelling approach for the assessment of land use change effects on wastewater
infrastructures. Environmental Modelling & Software 24, 1522-1528.
Gilfedder, M., Walker, G.R., Dawes, W.R., Stenson, M.P., 2009. Prioritisation approach for
estimating the biophysical impacts of land-use change on stream flow and salt export at a
catchment scale. Environmental Modelling & Software 24, 262-269.
44
Grimm, V., Berger, U., Bastiansen, F., Eliassen, S., Ginot, V., Giske, J., Goss-Custard, J., Grand,
T., Heinz, S.K., Huse, G., Huth, A., Jepsen, J.U., Jørgensen, C., Mooij, W.M., Müller, B.,
Pe'er, G., Piou, C., Railsback, S.F., Robbins, A.M., Robbins, M.M., Rossmanith, E., Rüger,
N., Strand, E., Souissi, S., Stillman, R.A., Vabø, R., Visser, U., DeAngelis, D.L., 2006. A
standard protocol for describing individual-based and agent-based models. Ecological
Modeling. 198(1-2), 115-126.
Happe, K., Balmann, A., Kellermann, K., Sahrbacher, C., 2008. Does structure matter? The
impact of switching the agricultural policy regime on farm structures. Journal of Economic
Behavior and Organization. 67(2), 431-444
Happe, K., Kellermann, K., Balmann, A., 2006. Agent-based analysis of agricultural policies: an
Illustration of the Agricultural Policy Simulator AgriPoliS, its adaptation and behavior.
Ecology
and
Society.
11(1),
49.
Available
at
http://www.ecologyandsociety.org/vol11/iss1/art49/ (Accessed 6 September 2010).
Happe, K., Schnicke, H., Sahrbacher, C., Kellermann, K., 2009. Will They Stay or Will They
Go? Simulating the Dynamics of Single-Holder Farms in a Dualistic Farm Structure in
Slovakia. Canadian Journal of Agricultural Economics/Revue canadienne d'agroeconomie.
57(4), 497-511.
Hazell, P., Norton, R., 1986. Mathematical programming for economic analysis in agriculture.
Macmillan, New York.
Henseler, M., Wirsig, A., Herrmann, S., Krimly, T., Dabbert, S., 2009. Modeling the impact of
global change on regional agricultural land use through an activity-based non-linear
programming approach. Agricultural Systems. 100(1-3), 31-42.
Holden, S., Shiferaw, B., 2004. Land degradation, drought and food security in a less-favored
area in the Ethiopian highlands: a bio-economic model with market imperfections.
Agricultural Economics. 30(1), 31-49.
Janssen, M.A., Ostrom, E., 2006. Governing Social-Ecological Systems. In: L. Tesfatsion, K.L.
Judd (Eds.) Handbook of Computational Economics II: Agent-Based Computational
Economics, Elsevier Publisher, pp. 1465-1509
Jasper, K., Gurtz, J., Lang, H., 2002. Advanced flood forecasting in Alpine watersheds by
coupling meteorological observations and forecasts with a distributed hydrological model.
Journal of Hydrology. 267, 40-52.
Jones, J.W., Hoogenboom, G., Porter, C.H., Boote, K.J., Batchelor, W.D., Hunt, L.A., Wilkens,
P.W., Singh, U., Gijsman, A.J., Ritchie, J.T., 2003. The DSSAT cropping system model.
European Journal of Agronomy. 18(3-4), 235-265.
Keating, B.A., Carberry, P.S., Hammer, G.L., Probert, M.E., Robertson, M.J., Holzworth, D.,
Huth, N.I., Hargreaves, J.N.G., Meinke, H., Hochman, Z., McLean, G., Verburg, K., Snow,
V., Dimes, J.P., Silburn, M., Wang, E., Brown, S., Bristow, K.L., Asseng, S., Chapman, S.,
McCown, R.L., Freebairn, D.M., Smith, C.J., 2003. An overview of APSIM, a model
designed for farming systems simulation. European Journal of Agronomy. 18(3-4), 267288.
Krol, M.S., Bronstert, A., 2007. Regional integrated modeling of climate change impacts on
natural resources and resource usage in semi-arid Northeast Brazil. Environmental
Modeling & Software. 22(2), 259-268.
Kuyvenhoven, A., Ruben, R., Kruseman, G., 1998. Technology, market policies and institutional
reform for sustainable land use in southern Mali. Agricultural Economics. 19, 53-62.
45
Latynskiy, E., Troost, C., Berger, T., 2010. Impacts of Climate Variability and Climate Change
on Agriculture and Rural Poverty in Central Chile. Contributed paper for the 117 th Seminar
of the European Association of Agricultural Economists, Universität Hohenheim, Stuttgart,
Germany, November 25-27, 2010.
Le, Q.B., Park, S.J., Vlek, P.L.G., 2010. Land use dynamic simulator (LUDAS): a multi-agent
system model for simulating spatio-temporal dynamics of coupled human-landscape
system. II. Scenario-based applications for impact assessment of land-use policies.
Ecological Informatics 5(3), 203-221.
Le, Q.B., Park, S.J., Vlek, P.L.G., Cremers, A.B., 2008. Land-Use Dynamic Simulator
(LUDAS): A multi-agent system model for simulating spatio-temporal dynamics of
coupled human-landscape system. I. Structure and theoretical specification. Ecological
Informatics. 3(2), 135-153.
Letcher, R.A., Croke, B.F.W., Jakeman, A.J., Merritt, W.S., 2006a. An integrated modeling
toolbox for water resources assessment and management in highland catchments: Model
description. Agricultural Systems. 89(1), 106-131.
Letcher, R.A., Croke, B.F.W., Merritt, W.S., Jakeman, A.J., 2006b. An integrated modeling
toolbox for water resources assessment and management in highland catchments:
Sensitivity analysis and testing. Agricultural Systems. 89(1), 132-164.
Linstead, A., 2004. Typed Data Transfer (TDT) User's Guide. URL: http://www.pik-potsdam.de.
Marohn, C., 2008. Documentation of the LUCIA model. Stuttgart, Department of Plant
Production and Agroecology in the Tropics and Subtropics, University of Hohenheim.
Available
at
https://www.unihohenheim.de/sfb564/project_area_c/c4/c4_files/manual_lucia.pdf (accessed 2 August
2010).
Marohn, C., Siripalangkanont, P., Berger, T., Lusiana, B., Cadisch, G., 2010. Integrated land use
modeling in an interdisciplinary project: The LUCIA model, Proceedings of the
LANDMOD 2010 Conference, Montpellier, 3-5 February 2010.
Matthews, R., 2006. The People and Landscape Model (PALM): Towards full integration of
human decision-making and biophysical simulation models. Ecological Modeling. 194(4),
329-343.
Parker, P., Letcher, R., Jakeman, A., Beck, M.B., Harris, G., Argent, R.M., Hare, M., PahlWostl, C., Voinov, A., Janssen, M., 2002. Progress in integrated assessment and modeling.
Environmental Modeling & Software. 17(3), 209-217.
Parton, W.J., Rasmussen, P.E., 1994. Long-term effects of crop management in wheat-fallow: II.
CENTURY model simulations. Soil Sci. Soc. Am. J. 58: 530-536.
Parton, W.J., Stewart, J.W.B., Cole, C.V., 1988. Dynamics of C, N, P and S in grassland soils: a
model. Biogeochemistry 5: 109-131.
Perez, P., Ardlie, N., Kuneepong, P., Dietrich, C., Merritt, W.S., 2002. CATCHCROP: modeling
crop yield and water demand for integrated catchment assessment in Northern Thailand.
Environmental Modeling & Software. 17(3), 251-259.
Priesack, E., Achatz, S., Stenger, R., 2001. Parametrisation of soil nitrogen transport models by
use of laboratory and field data, in: Shaffer M.J., Liwang, Ma., Hansen, S. (Eds), Modeling
Carbon and Nitrogen Dynamics for Soil Management. CRC Press Inc., Boca Raton,
Florida, pp. 461-484
Priesack, E., Gayler, S., Hartmann, H.P., 2006. The impact of crop growth sub-model choice on
simulated water and nitrogen balances. Nutrient Cycling in Agroecosystems. 75, 1-13
46
Quang, D.V. Schreinemachers, P, Berger, T., 2010. Integrated assessment of soil conservation:
Quantifying trade-offs between incomes and sustainability in northern Vietnam. Paper
presented at the International Symposium on Sustainable Land Use and Rural Development
in Mountainous Regions of Southeast Asia, Hanoi, 21-23 July 2010
Reidsma, P., Ewert, F., Lansink, A.O., Leemans, R., 2010. Adaptation to climate change and
climate variability in European agriculture: The importance of farm level responses.
European Journal of Agronomy. 32(1), 91-102.
Ritchie, J.T., Singh, U., Godwin, D.C., Bowen, W.T., 1998. Cereal growth, development and
yield. In: Tsuji, G.Y., Hoogenboom, G., Thornton, P.K. (Eds.), Understanding Options for
Agricultural Production. Systems Approaches for Sustainable Agricultural Development.
Kluwer Academic Publishers, Dordrecht, the Netherlands, pp. 79-98.
Roebeling, P.C., Jansen, H.G.P., Schipper, R.A., Enz, F.S., Castro, E., Ruben, R., Hengsdijk, H.,
Bouman, B.A.M., 2000. Farm modeling for policy analysis on the farm and regional level.
In: Tools for Land Use Analysis on different Scales. Bouman, B.A.M., Jansen, H.G.P.,
Schipper, R.A., Hengsdijk, H. and Nieuwenhuyse, A.N. (Eds.). Dordrecht, the Netherlands,
Kluwer Academic Publishers, pp.171-198.
Roetter, R.P., Hoanh, C.T., Laborte, A.G., Van Keulen, H., Van Ittersum, M.K., Dreiser, C., Van
Diepen, C.A., De Ridder, N., Van Laar, H.H., 2005. Integration of Systems Network
(SysNet) tools for regional land use scenario analysis in Asia. Environmental Modeling &
Software. 20(3), 291-307.
Roetter, R.P., van den Berg, M., Laborte, A.G., Hengsdijk, H., Wolf, J., van Ittersum, M., van
Keulen, H., Agustin, E.O., Thuc Son, T., Xuan Lai, N., Guanghuo, W., 2007. Combining
farm and regional level modeling for Integrated Resource Management in East and Southeast Asia. Environmental Modeling & Software. 22(2), 149-157.
Schaldach, R., Alcamo, J., 2006. Coupled simulation of regional land use change and soil carbon
sequestration: A case study for the state of Hesse in Germany. Environmental Modelling &
Software 21, 1430-1446.
Schreinemachers, P., 2006. The (Ir)relevance of the crop yield gap to food security in developing
countries. With an application of multi-agent modeling to farming systems in Uganda.
Cuvillier
Verlag,
Göttingen.
Available
at:
http://hss.ulb.unibonn.de/diss_online/landw_fak/2006/schreinemachers_pepijn/.
Schreinemachers, P., Berger, T., 2006. Land-use decisions in developing countries and their
representation in multi-agent systems. Journal of Land Use Science. 1(1), 29-44.
Schreinemachers, P., Berger, T., Aune, J.B., 2007. Simulating soil fertility and poverty dynamics
in Uganda: A bio-economic multi-agent systems approach. Ecological Economics. 64(2),
387-401.
Schreinemachers, P., Berger, T., Sirijinda, A., Praneetvatakul, S., 2009. The diffusion of
greenhouse agriculture in northern Thailand: Combining econometrics and agent-based
modeling. Canadian Journal of Agricultural Economics/Revue canadienne
d'agroeconomie. 57(4), 513-536.
Schreinemachers, P., Potchanasin, C., Berger, T., Roygrong, S., 2010. Agent-based modeling for
ex-ante assessment of tree crop technologies: litchis in northern Thailand. Agricultural
Economics. Forthcoming.
Schuler, J., Sattler, C., 2010. The estimation of agricultural policy effects on soil erosion--An
application for the bio-economic model MODAM. Land Use Policy. 27(1), 61-69.
47
Schulla, J., Jasper, K., 2007. Model Description WaSiM-ETH, Technical Report, 181. Available
at http://www.wasim.ch/downloads/doku/wasim/wasim_2007_en.pdf.
Supit, I., 2003. Updated system description of the WOFOST crop growth simulation model as
implemented in the crop growth monitoring system applied by the European Commission.
Treemail
Publishers,
Heelsum,
the
Netherlands.
Available
at
http://www.treemail.nl/download/treebook7/start.htm (Accessed 6 September 2010).
Todd, P.M., Gigerenzer, G., 2000. Precis of simple heuristics that make us smart. Behavioral and
Brain Sciences. 23, 727-780.
Uribe, H., Arnold, T., Arumí, J., Berger, T., Rivera, D., 2009. Modificación del modelo WaSiMETH para mejorar su aplicación en áreas regadas. Ingeniería Hidraulica en México. 24 (2),
23-36.
van Oel, P.R., Krol, M.S., Hoekstra, A.Y., Taddei, R.R., 2010. Feedback mechanisms between
water availability and water use in a semi-arid river basin: A spatially explicit multi-agent
simulation approach. Environmental Modeling & Software. 25(4), 433-443.
Vatn, A., Bakken, L., Bleken, M.A., Baadshaug, O.H., Fykse, H., Haugen, L.E., Lundekvam, H.,
Morken, J., Romstad, E., Rørstad, P.K., Skjelvåg, A.O., Sogn, T., 2006. A methodology for
integrated economic and environmental analysis of pollution from agriculture. Agricultural
Systems. 88(2-3), 270-293.
Vatn, A., Bakken, L., Botterweg, P., Romstad, E., 1999. ECECMOD: an interdisciplinary
modeling system for analyzing nutrient and soil losses from agriculture. Ecological
Economics. 30(2), 189-206.
Walker, D.H., 2002. Decision support, learning and rural resource management. Agricultural
Systems. 73, 113-127.
Wang, J., Chen, J., Ju, W., Li, M., 2010. IA-SDSS: A GIS-based land use decision support
system with consideration of carbon sequestration. Environmental Modelling & Software
25, 539-553.
48