ARTICLE IN PRESS

Land Use Policy 24 (2007) 531–545

www.elsevier.com/locate/landusepol

Combining top-down and bottom-up modelling approaches of land use/

cover change to support public policies: Application to sustainable
management of natural resources in northern Vietnam
Jean-Christophe Castellaa,, Suan Pheng Kamb,1, Dang Dinh Quangc,
Peter H. Verburgd, Chu Thai Hoanhb,2
a
Institut de Recherche pour le Développement (IRD), B.P. 64501, 34394 Montpellier cedex 5, France
b
International Rice Research Institute (IRRI), DAPO 7777, Metro Manila, Philippines
c
Vietnam Agricultural Science Institute (VASI), Thanh Tri, Hanoi, Vietnam
d
Department of Environmental Sciences, Wageningen University, P.O. Box 37, 6700AA Wageningen, The Netherlands

Received 23 February 2005; received in revised form 12 August 2005; accepted 4 September 2005

Abstract

Over recent years, the scientific community has developed different modelling methodologies of land use/cover change (LUCC)
depending on their intended use, and also on the scale of investigation, disciplinary background and scientific tradition of the research
teams. Consequently, each LUCC model has its own potential and constraints with respect to the needs and expectations of land use
planners and policy makers. Faced with the issue of integrated management of natural resources from local to regional scales,
agricultural research can benefit from reconciling top-down and bottom-up approaches to LUCC modelling.
Three modelling methodologies belonging to these two LUCC approaches were applied in the province of Bac Kan in northern
Vietnam. The multi-agent model SAMBA was developed through an adaptive, bottom-up process while LUPAS and CLUE contributed
to a top-down process. Applying these three methodologies at the same research site allowed a critical evaluation of their respective
utility for land use analysis and planning. They played complementary roles in bridging knowledge gaps and increasing interactions
between stakeholders along the continuum from research to development and policy formulation. Combined use of these modelling
methodologies should be promoted when complex natural resource management issues at multiple scales need to be tackled.
r 2006 Elsevier Ltd. All rights reserved.

Keywords: Land use/cover change; Natural resources management; Modelling; Participatory simulation; Methodology combination; Multiple stakeholder
platforms; Vietnam

The point is not to predict the future, but to make it one site to larger areas are often limited. For this reason the
possible [Antoine de Saint-Exupéry] conceptualization, implementation, and interpretation of
NRM research require integration of different levels of
Introduction: achieving impact of research on management of biophysical, socioeconomic and political organization
natural resources within defined geographical areas (Harrington et al.,
2001; Sayer and Campbell, 2004). Two major trends
Natural resource management (NRM) tends to be site- emerge in addressing regional-level NRM for agricultural
specific, and the ability to extrapolate research results from development: (i) towards more integrated and eco-
logically oriented agriculture, and (ii) towards the need
Corresponding author. Tel.: +33 4 67 63 69 80; fax: +33 4 67 63 87 78.
for increased institutional cooperation (Bouma et al.,
E-mail address: [email protected] (J.-C. Castella). 1995). The main challenge for research to support these
1
Current address: World Fish Center, P.O. BOX 500, GPO, 10670
trends is to avoid producing more academic models for
Penang, Malaysia.
2
Current address: International Water Management Institute (IWMI), NRM but rather to strive for widespread acceptance and
P.O. Box 2075, Colombo, Sri Lanka. use of NRM approaches that have real and lasting impact

0264-8377/$ - see front matter r 2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.landusepol.2005.09.009
 ARTICLE IN PRESS
532 J.-C. Castella et al. / Land Use Policy 24 (2007) 531–545

on development (ISNAR, 2004). For this to happen, NRM The pilot ecoregional project for the RRB focused on
research requires new partnership mechanisms that allow Bac Kan province in the upper catchment of the RRB; the
better integration of the scientific—technical perspectives provincial capital Bac Kan is located about 150 km north
of the researchers with the local—indigenous perspectives of Hanoi (Fig. 1). Bac Kan province was chosen for the
of other stakeholders (Castella et al., 2001). following reasons: (i) it is recognized as one of the poorest
The Red River Basin (RRB) was selected as one of the pilot provinces in the country, relying mainly on subsistence
sites of the ecoregional initiative for the humid and sub-humid agriculture, and as such it provides the challenge for NRM
tropics of Asia (EcoR(I)Asia), convened by the International research to make real impact on poverty alleviation; (ii) it
Rice Research Institute for the implementation of integrated has the attention of national and provincial policy makers
NRM research for agricultural development. The RRB on difficult issues of rampant shifting cultivation practices,
depicts the accelerated changes in land use widely encountered growing population pressure, and the related risk of
in the Southeast Asia ecoregion, driven mainly by market environmental degradation in the fragile upland ecosys-
forces, i.e. rapid population increase and economic growth tems such as soil erosion and depletion of natural resources
(Teng et al., 1995). The RRB features a wide range of agro- due to intense deforestation; and (iii) the expectation that
ecological and socio-economic situations, which call for the local authorities and communities would be readily
development of specific tools and methodologies for inte- mobilized around these important environmental issues
grated management of natural resources across multiple (Castella et al., 2004).
scales. Reconnaissance meetings and a number of workshops In this paper we describe a specific component of the
(Kinh et al., 1999; Kam et al., 1999) were organized with NRM research carried out in Bac Kan, with regard to
concerned institutions to identify key NRM issues, take stock analyzing LUCC to support agricultural planning and
of past and ongoing NRM research, and identify appropriate development within an integrated NRM context. We
methodologies that could be easily implemented or developed adapted three different but complementary methodologies
jointly, and to strengthen inter-institutional partnerships for LUCC analysis with the intention of assessing their
(Castella et al., 2001). It was agreed that comprehensive respective roles within an integrated framework to guide
analysis of land use/cover change (LUCC) to understand past NRM efforts involving multiple scales and stakeholders.
and ongoing changes was needed as a first essential step for We illustrate its application to the specific situation in Bac
integrated, multi-scale NRM research (Kam and Hoanh, Kan, focussing on issues relating to land changes
1999). This need for common baseline information on land consequent to the economic reforms of the 1990s and the
use changes also provided the ground for collaboration associated risks of resource depletion and land degrada-
among researchers and institutions. tion. On a more general note, we discuss the relevance of

W E

BA BE
NGAN SON

BACH THONG

TX BAC KAN
CHO DON

NA RI

Commune boundary CHO MOI

District boundary
CHO DON District name 10 km
Research site of the SAM project

Fig. 1. Location of the research sites in Bac Kan province, Vietnam.

 ARTICLE IN PRESS
J.-C. Castella et al. / Land Use Policy 24 (2007) 531–545 533

this approach to land use planning and policy making and was hampered by the lack of financial and human
its applicability in other contexts. resources. On the other hand, regional studies based on
remote sensing and census data enabled wide geographic
Method: knowledge and methodology integration across coverage but were limited in their understanding of
scales underlying mechanisms of land use change. In regional
approaches discrete time series data were analysed statis-
Bridging gaps in research on changes in land use to achieve a tically to explore regional variability and identify associa-
higher impact on NRM tions between variables. They did not satisfactorily address
the complex cause-effect relationships between variables
The integrative framework arrived at in this study was that drove the observed changes in land use, and did not
not pre-planned at the start of the project but was incorporate the decision making of land users. Conse-
gradually and jointly constructed by all stakeholders quently it was essential to bridge these two approaches to
through an iterative and adaptive process (Jiggins and LUCC in order to understand mechanisms of change over
Röling, 2000). The development of the research framework large geographic areas. As the two approaches corre-
is presented here as successive stages that reflect our sponded to different entry points in the study of land
learning pathway. changes and were based on different scientific tradition and
The objective of developing an operational approach for research communities, our objective of knowledge integra-
regional-level NRM challenges traditional scientific ap- tion entailed a process of methodology combination.
proaches in three main dimensions of knowledge integra-
tion: from disciplinary to interdisciplinary research, across Combining modelling approaches to LUCC
scales from field/farm to district/province, and along the
continuum from research to development and policy Because no single research tool would be able to address
making (Kam et al., 2002). Tackling these three dimensions all interrelated issues of sustainable management of natural
makes the operational approach much more complex than resources, three different methodologies for LUCC analy-
the conventional discipline-oriented approach. Depending sis were developed and applied in Bac Kan by three
on the level of investigation, the interactions between research teams.
biophysical and socio-economic components of the
coupled human-environment are analyzed as farming  The SAMBA3 methodology was developed by the
practices (field/farm level), as collective actions (village/ Mountain Agrarian Systems Program4 in the specific
community level), or as governance systems (district/ context of Bac Kan province to understand the
province level). The abundant literature on action research mechanisms of land use change that accompanied the
shows that integrating these three levels of management of transition from centrally-planned to family-based agri-
natural resources requires: (i) comprehensive understand- culture in the 1980s at the village scale (Castella et al.,
ing of land use change (diagnosis), (ii) the active participa- 2005a). Scenarios of land change are jointly constructed
tion of stakeholders sharing multiple views through by researchers and local stakeholders in a participatory
negotiation platforms (mediation), and (iii) a framework manner using multi-agent models and role playing
for action that takes into account the cause-effect relation- games (Bousquet et al., 2002; Castella et al., 2005b).
ships across scales (i.e. the relative impact of direct actions Simulations of changes in land use are run to
of land users versus the institutional and societal structures incorporate multiple representations of the different
that define their choices) (Röling, 1994; Steins and groups of stakeholders within a common frame of
Edwards, 1999; Meinzen-Dick et al., 2004). reference, which is a prerequisite for negotiation and
Although complementary, the methods and tools devel- concerted action on NRM (Boissau and Castella, 2003).
oped at each scale by different scientific communities may  The land use planning and analysis system (LUPAS)
be difficult to reconcile (Rindfuss et al., 2004). In Bac Kan, methodology (Roetter et al., 2005) was developed as
the combination of bottom-up and top-down approaches
to data collection and analysis of land use systems was 3
The model’s name is formed from SAM, the French acronym for
particularly challenging (Castella et al., 2004). On the one mountain agrarian systems and the name of the project in which the
hand, local case studies provided a refined understanding methodology was developed, and BA, which means ‘‘three’’ in Vietna-
of local issues based on long term investigation at specific mese. The model is considered a link between two components of the SAM
project; the first deals with identifying land-use changes and the problems
locations. Information from the historical profiles of land that are a priority with local stakeholders, and the second involves
use changes and in-depth household surveys was used to intervention through the design and testing of technical and organiza-
identify and rank the pertinent NRM issues (Castella and tional innovations, in collaboration with farmers.
4
Quang, 2002). However, the possibility to generalize the A joint research project between the Vietnam Agricultural Science
findings was limited by the geographic coverage of the Institute (VASI), Institut de Recherche pour le Développement (IRD),
Centre de Coopération Internationale en Recherche Agronomique pour le
studies and the locality-specific conditions, especially given Développement (CIRAD), and the International Rice Research Institute
the heterogeneous mountain environment of the study (IRRI); for more information see: http://www.knowledgebank.irri.org/
area. Coverage of such detailed studies over a larger area sam/home_en.html.
 ARTICLE IN PRESS
534 J.-C. Castella et al. / Land Use Policy 24 (2007) 531–545

part of the systems research network (SysNet) for the development objectives, identify the typology of land
ecoregional land use planning in tropical Asia (Van use, etc.), through refinement of the model and examina-
Ittersum et al., 2004), and was evaluated in four case tion of model outputs—mainly involved the regional
study regions located in India, Vietnam, Malaysia and ‘‘governing’’ authorities rather than farmers. On the other
the Philippines. This methodology is based on a multiple hand, SAMBA provides an explanatory model of changes
goal linear programming (MGLP) model implemented in land use based on observation of interactions among
together with stakeholders (e.g. local authorities, land farmer-agents. Local stakeholders are involved in a
use planners, policy makers) to explore land use options bottom-up process to elicit individual decision-making
that are optimized for a given set of stakeholder defined rules and the gradual formulation of human-induced
objective functions and existing resource constraints. drivers of LUCC. The three methodologies play comple-
LUPAS represents the latest generation in a long mentary roles in operational research as they usually
tradition of land use models based on land evaluation involve different groups of stakeholders.
and land use optimization with linear programming (De Prescriptive, descriptive and explanatory approaches to
Wit et al., 1988; Van Ittersum et al., 1998). LUCC: The three modelling approaches differ considerably
 The conversion of land use and its effects (CLUE) in both concept and formalism as a result of the scientific
methodology (De Koning et al., 1999; Verburg et al., traditions of the research communities they originate from.
2002) was developed at Wageningen University (The LUPAS belongs to the tradition of prescriptive methods
Netherlands) to project and visualize the spatial patterns while CLUE is a descriptive tool of LUCC (Verburg et al.,
of changes in land use as they are expected to develop 2004a; Briassoulis, 2000) and SAMBA can be considered as
under a set of conditions that are specified in scenarios. an explanatory tool (Parker et al., 2003). LUPAS does not
The methodology does not provide optimal solutions attempt to explain (the driving factors of) land use change
but rather projects plausible developments. The model is but explores plausible alternatives for land/resource use on
based on a combination of empirical analysis of the the premise of optimization given the wishes of the
(proximate) driving factors of land use and the dynamic stakeholders/modeller and the constraints of the system.
modelling of competition and land use dynamics. Local In LUPAS, there is no explicit identification of driving
knowledge and policy views are explicitly incorporated forces, except through the selection of factors that define
in the scenario development phase. the constraints and the influence of inputs and outputs.
CLUE-type models describe development pathways to see
Top-down versus bottom-up approaches to LUCC: These what happens to the system ify while multi-agent models
three methodologies belong to different scientific tradi- attempt to explain underlying mechanisms of change.
tions. LUPAS and CLUE analyse LUCC in a top-down These different domains of application also justify the
manner (Fig. 2). The CLUE methodology links land use complementary use of these three methodologies.
change to its causal ‘driving forces’, searching for correla- In addition, the methodologies have characteristics in
tions that explain the emergence of spatially variable land common that facilitated their combined use in Bac Kan
use patterns. LUPAS uses optimization algorithms province. Spatially explicit computer models are central to
(MGLP) to generate a model representing inputs, outputs land use studies. As dynamic models they are useful tools
and interaction between different land use types specific to to collect and organize data, synthesize knowledge, and
the study area. The participatory mode of operation—from build consensus about the management of complex land
the early stage of developing the model (i.e. to help define use systems (Costanza and Ruth, 1998). Indeed, all three

LUPAS CLUE
Country
Regional policy Driving forces LUCC

Province Regional policy Hot spots

LU options
Land allocation
District Extrapolation through
participatory simulations
Village
Farm management
NRM communication
platforms
Farm

Participatory trials
Field
SAMBA

Fig. 2. Combining complementary top-down and bottom-up approaches across scales.

 ARTICLE IN PRESS
J.-C. Castella et al. / Land Use Policy 24 (2007) 531–545 535

LUCC methodologies were used independently: (i) to the minimum set of variables that would explain land use
integrate multi-disciplinary knowledge on complex land changes observed at the village scale (Castella et al., 2005a).
use systems, (ii) to facilitate interaction between stake- A farmer-agent would allocate its resources to different
holders around scenarios of land change towards a shared production activities on the hillsides according to the level
representation of the problems at stake, and (iii) as decision of rice self-sufficiency obtained in the rice paddies. The
support tools to explore and discuss potentially feasible simulations helped understand the rapid deforestation of
options for land/resource use in a problem-solving process. the mountains following the allocation of the lowlands to
Their combined use could only reinforce their respective individual households in the 1980s. Then, the SAMBA
strengths. multi-agent model was translated into a role-playing game
to: (i) ease communication barriers between researchers
Results: an adaptive action-research framework in Bac Kan and local stakeholders on burning issues such as land rights
province and land allocation, and (ii) elicit and incorporate local
knowledge into the SAMBA model to improve its
In this section we present the research sequence as it representation of the most recent changes in land use.
unfolded from initially geographically limited case studies The main rules of the initial computer model were
(village/watershed/farming community) to a regional ac- subjected to role play using a wooden game board and
tion-research framework involving different groups of land played by ten local farmers selected to represent a diversity
users and policy makers. The results obtained through this of conditions and production strategies (Boissau et al.,
adaptive research design depended to a large extent on the 2004). Repeating these participatory simulations at differ-
quality of the interactions between scientists from different ent locations and in different contexts enabled us to elicit
disciplines and between researchers and stakeholders. decision rules that were further incorporated into a new
generic version of the SAMBA model (Boissau and
A learning path towards participatory management of Castella, 2003, Fig. 3). We checked that the resulting
natural resources model: (i) correctly simulated the most recent changes in
land use following the allocation of forestland to individual
The diagnostic study started with the collection of farmers that started in 1993, and (ii) could generalize the
empirical data at two different scales: the village and the findings from the successive sessions of participatory
province. A geographic information system (GIS) database simulation to the entire province of Bac Kan. The multi-
was developed at the province level. Variables pertaining to agent model was validated firstly by comparing simulated
biophysical and socioeconomic aspects of the land use and observed land use maps, and secondly by submitting
system were subjected to statistical analysis that revealed its results (i.e. simulated scenarios of LUCC) to the
correlations between land use patterns and trajectories of judgement of local stakeholders in two communes which
land changes, and their driving factors. These results were had previously not been exposed to the SAMBA process
used for two related purposes: (i) to characterize regional (Castella et al., 2005b). This participatory assessment was
land use systems and the changes that took place during crucial for confirming that the model developed was
the 1990s, and (ii) to select seven case study sites along a applicable to the entire province without needing to repeat
gradient of agro-ecological diversity and level of integra- the intensive data collection effort over a large number of
tion to market economy (Castella and Quang, 2002). At the sites.
village level, researchers spent months living and interact- The main steps and model formalisms constituting the
ing with the local communities. Household surveys and SAMBA methodology (i.e. narratives, graphic, multi-agent
interpretation of chronological series of aerial photographs model, role playing game) provide a common frame of
(1954, 1977, and 1998) provided the quantitative data to reference for all participants in the participatory simula-
support a conceptual model of LUCC. A very simple tions conducted at different locations (Fig. 4). In all
model of household decision making was incorporated into versions of the SAMBA model the agents—household,
a computational model named SAMBA in order to identify buffalo—and the environment are the same. The colors

Field studies Participatory simulations

Data Land use

collection study Participatory
Farming Hypothesis rejected
modelling
systems Multi-agent Role playing
Household surveys
typology model game
Hypothesis
Analysis Validation Decision
Remote sensing data Communication
Indicators / parameters making platform
GIS support

Fig. 3. Design and procedure for use of multi-agent-based participatory simulations with SAMBA.
 ARTICLE IN PRESS
536 J.-C. Castella et al. / Land Use Policy 24 (2007) 531–545

used to represent different land use types are similar, and

so is the grain (grid cell size) and extent (village) of the
simulation interface. The size of the grid cell (1000 m2) is
the unit of area measurement commonly used by the
highland communities in northern Vietnam. The village
was selected as a coherent entity in terms of its biophysical
(watershed), social (community) and the geographic
dimensions (Papin and Tessier, 2002). Natural resources
are managed by the farm households at the village level
(Castella and Quang, 2002). While this rule-based model of
LUCC proved useful in capturing the farmers’ decision
process at household and village levels, it has several
limitations for land use planning and policy-making at the
regional level. The methodology can easily deal with
scaling up from household to village level, but generalizing
to the regional level involves scaling out (i.e. replication in
many villages) and not scaling up (i.e. aggregating results
to higher hierarchical level) of the local models (Harrington
et al., 2001). In out-scaling, interactions between villages
and through markets were not taken into account in the
simulations. This type of organizational design is good for:
(i) two-directional communication between farmers, local
authorities, and scientists, and (ii) facilitating negotiation
between land users about topics and at scales that they are
comfortable with. The role playing game was specifically
designed to elicit from local stakeholders/land users their
actual decisions and strategies in managing their resources,
as opposed to official positions reflecting top-down
decision by government authorities and policy makers
(Castella et al., 2003). In the process, regional land use
planners were bypassed. However, it is recognized that
policy makers do have a key role to play in the platforms of
negotiation on NRM as no development plan can be
implemented without their agreement.

Application of top-down LUCC modelling methodologies in

Bac Kan province

The LUPAS methodology was applied in Bac Kan

province as a complementary approach to SAMBA. While
SAMBA is a people-centered approach, LUPAS analyses
different options for change based on the potential of the
land and other available resources. The first step of the
LUPAS methodology involves a qualitative evaluation of
biophysical resources to determine the suitability of crops
by season, and combinations of crops into existing
and promising land use types (LUT) specific to Bac Kan
(Fig. 5). Data were collected from field surveys in all 122
communes. The next step involves quantification of the
main inputs and outputs for all LUT; the data were
organized into input–output tables (Roetter et al., 2005).
Land evaluation and resource balance were carried out for
Fig. 4. The different formalisms of the SAMBA model: (A) narrative homogeneous land units (LU) delineated by GIS map
storylines and graphics, (B) computer interface in the multi-agent model, overlay of thematic layers (soil, slope, climate, adminis-
(C) the wooden board for the role playing game, and (D) the computer
interface of the coupled multi-agent model—geographic information
trative boundaries). Different scenarios of total available
system under the CORMAS platform. area for agricultural land use allocation were developed by
considering options to include or exclude non-arable land
 ARTICLE IN PRESS
J.-C. Castella et al. / Land Use Policy 24 (2007) 531–545 537

Land evaluation
Constraints concerning resources:
Delineation of
area, water, labour
land units,
resource
assessment,
Land use optimisation
Optimization
Land suitability Input/output
Model Land use options &
analysis & Yield tables of
Goal achievements
production Interactive Multiple
estimation Goal Linear
activities
Programming
Data Maps

Objective functions

Policy views

Scenario construction

Fig. 5. Operational structure of the LUPAS methodology.

Fig. 6. Allocation of land use types (1–3 cropping cycles per year) with LUPAS.

and areas not intended for agriculture but already (Fig. 7). A land use allocation map can be generated for
encroached (e.g. shifting cultivation in legally protected each point along the trade-off curve representing the
land). For each LU, all LUT were rated according to their continuum between maximizing food production or
suitability. A MGLP model then optimized the geographic regional income. A series of trade-off curves can be
allocation of land use types according to two contrasting generated to represent situations whereby existing con-
policy perspectives defined in collaboration with policy straints are progressively removed. The uppermost curve,
makers (maximizing food production or maximizing where land availability is the only remaining constraint,
regional income), subjected to a number of constraints delineates the least-constraints limits, subject to the defined
e.g. available land, capital, water, labor (Yen et al., 2002). objective functions. Presented to governing authorities,
The outputs can be maps representing the optimal spatial these trade-off curves explicitly quantify the implications of
land allocation to satisfy a particular objective (Fig. 6), policy decisions and interventions on land/resource use
or a graph plotting the trade-offs between two objectives scenarios. Throughout the study period, the team of
 ARTICLE IN PRESS
538 J.-C. Castella et al. / Land Use Policy 24 (2007) 531–545

800 scenarios, even for maximizing regional income, could be

generated by fixing some existing land use types as
700 constraints (e.g. present lowland rice areas cannot be
Regional income (billion VND)

Share
labor changed to other crops). The provincial policy makers were
600
also interested in the intermediary products of the research
500
(e.g. suitability maps for different crops or for the number
Remove
capital
of cropping cycles per year).
400
constraint Second, the local participants were comfortable with the
research data used, which were obtained from the same
300 sources that they use for their planning activities. Third,
No capital constraint
policy makers trust the sound, scientific assessment of the
200 regional potential for production. Their main concern was
With capital constraint
how to translate the results of the LUPAS into concrete
100
0 50 100 150 200 250 300 350
policies.
The CLUE methodology was tested in Bac Kan in
Food production (thousand tons)
response to some of the shortcomings of the two
methodologies presented above. A CLUE model was
Current situation Land, capital, water, labor w/o sharing
developed for Bac Kan province using empirical data that
Land & labor w/o sharing Land constraint only were already available in the provincial GIS. By projecting
land changes in the near future (5–15 years) from a recent
Fig. 7. Trade-offs between food production and regional income.
land use map (1998), the map outputs could serve as a basis
for the discussion of alternative development pathways.
researchers5 consulted with the provincial and district The resolution (250 m grid) and extent (Bac Kan province)
authorities on many occasions to get their inputs on of the CLUE study was the same as for LUPAS. The
determining the objectives, their cooperation in data model is made of four components as shown in the middle
collection and their feedback in discussing the model column of Fig. 8. Together they create the set of conditions
results. on which the calculation of the changes in land use pattern
A number of points raised by the provincial authorities is based (Verburg et al., 2002). (1) The characteristics of a
on these occasions are to be noted for further improvement location determine the relative suitability of that location
of the methodology and implementation of the model. for the different LUT. The relation between the probability
First, they found the approach to be relevant for exploring of finding a land use type at a particular location and the
the biophysical limits of the agro-ecosystem under different biophysical (e.g. soil, geology, topography, climate, acces-
constraints. ‘‘When setting goals based on the knowledge of sibility) and socio-economic (e.g. population, rural labor
the current situation it is important to know where the overall force, poverty, illiteracy) characteristics of that location
limits of the system are as we cannot increment growth was defined in a logit model (Willemen et al., 2002). The
objectives indefinitely’’ said a participant from the provin- suitability maps accounted for projected changes in
cial Department of Planning and Investment. The trade-off population, accessibility etc. and reflected suitability based
analysis (Fig. 7) defines the domain of viable policy options on an analysis of current land use configurations captured
instead of a single optimum, and this corresponds better to by logistic regression rather than the biophysical potential
the frame of reference of land use planners. The LUPAS that is reflected in the suitability maps produced by the
model explicitly takes into account possible interactions LUPAS methodology. (2) Conversion settings specific to
and resource sharing among communes. The stakeholders each land use type determine the temporal dynamics of the
were quick to point out that some land use allocation simulations. The conversion elasticity defines the reversi-
scenarios are unrealistic. For example Fig. 6, which depicts bility of change of a LUT, for example perennial
the scenario of attaining maximize regional income, implies plantations will not be easily converted to other uses
the need to convert existing paddy land to fruit tree because of their high capital investment. For each LUT,
cultivation. Such unrealistic results, which generated lively the land use transition sequence indicates what other LUT
discussion among the stakeholders, effectively highlight the it can change into in the next time step. Examples include
point that certain development objectives were just not fallow land and forest regrowth following swidden
feasible, and that compromise situations somewhere along cultivation (Verburg et al., 2004b). (3) Spatial policies
the trade-off curve (Fig. 7), rather than at its extremes, may and restrictions specify areas where land use changes are
be more feasible. On the other hand, more realistic restricted through land use policies or tenure status (e.g.
protected areas). (4) Land use requirements constrain the
5 simulation by setting the total allowable change in land use
The project was conducted by the National Institute of Soils and
Fertilizers (NISF) in partnership with IRRI, VASI and IRD. It involved to be allocated by the model at the grid cell level. Three
soil scientists, agronomists, hydrologists, GIS specialists, biometrician and demand scenarios were calculated at the provincial level
modellers. based on policy views of the provincial authorities: (1)
 ARTICLE IN PRESS
J.-C. Castella et al. / Land Use Policy 24 (2007) 531–545 539

Fig. 8. Structure of the CLUE methodology.

livestock/pasture land development in response to an with a high probability of land conversion in the near
increased demand for meat, (2) forest expansion to reach future) under different development pathways that deserve
the objective of 50% of the province covered with forest in closer attention from policy makers. The identification and
2010, and (3) increasing population putting too much characterization of land use drivers based on logistic
pressure on the upland land resources. regressions was considered by the provincial authorities
Once all inputs were provided the CLUE model to be a scientific demonstration that would be easy for
computed the most likely changes in land use that fulfilled them to use in the future. In contrast, patterns of land
the demand projections based on the specific scenario change defined by the local farmers in SAMBA that
conditions. The initial land use map of 1998 and the results underlie NRM strategies specific to each village were more
of the simulation after 12 time steps (2010) under the three difficult to incorporate in their planning practices. Two
scenarios are presented in Fig. 9. The simulations resulted main constraints were identified in relation with the
in complex (i.e. non-linear) changes in land use patterns implementation of CLUE in the context of Bac Kan
that were studied in detail and discussed with policy- province. First, in complex, heterogeneous mountain
makers at the district and province levels (Willemen et al., landscapes the accuracy of the current land use map is
2002). crucial because it constitutes the initial condition and
The CLUE methodology added new perspectives in land therefore has a large influence on the outcome of the
use planning by identifying hotspots of change (i.e. areas simulation as well as the implications on development
 ARTICLE IN PRESS
540 J.-C. Castella et al. / Land Use Policy 24 (2007) 531–545

Fig. 9. Results of the simulation of the scenarios with the CLUE model (initial land use configuration and scenario results for 2010).

pathways (Verburg et al., 2004b). For example in shifting Vietnam (Castella et al., 2005a). A rule-based approach
cultivation systems it is difficult to distinguish between the similar to the one developed in SAMBA, may thus need to
different stages of fallow and secondary vegetation on be combined with the CLUE approach to make it more
degraded soils, and this may lead to different simulation adaptable to the changing circumstances that may be
results. Second, the empirical specification of the suitability expected under some of the simulated scenarios. The third
map reflects the current decision making in land use in constraint we identified regarding the CLUE methodology
relation to the (proximate) land use drivers. In the current is that it does not explicitly determine the resource limits
application it is assumed that the influence of the land use (both biophysical and socio-economic) within which the
drivers remain unchanged throughout the simulation projections of future land use are made; although implicitly
period. This condition may not hold true in situations of these may be incorporated in the demand scenarios
very rapid changes driven by different factors in time, as (assuming that the specified demands already take the
encountered in the recent period of agrarian transition in capacity and limits of the resource base into consideration).
 ARTICLE IN PRESS
J.-C. Castella et al. / Land Use Policy 24 (2007) 531–545 541

Discussion: reconciling top-down and bottom-up approaches quired to bridge the gaps between these three methodol-
to LUCC ogies and highlight their complementary roles in addressing
the overriding needs of the land users, planners, and policy
The implementation of the SAMBA, LUPAS and CLUE makers. Fig. 10 illustrates the conceptual framework we
methodologies in Bac Kan province revealed their relative suggest.
strengths and weaknesses with respect to the different The LUPAS methodology provides an envelop curve
objectives and expectations of policy makers. Implement- that delineates a domain of possible options within the
ing three distinct LUCC methodologies in the same target resource limits of the system. Instead of providing a single
area required that the stakeholders gain sufficient insight optimum, the model simulations define a viability domain
into these methodologies, what they can do, and the subtle situated under each trade-off curve. The positioning of the
differences between their outputs. A likely stumbling block current situation within this domain can be identified (the
(and risk) is that they become bewildered and fail to circle in Fig. 10). Awareness of this viability domain helps
understand why these different methodologies ought to be policy makers identify policy interventions that would
applied in tandem, particularly if the outputs seem to expand the domain by alleviating specific constraints (e.g.
say different things and in doing so, complicate their releazing capital constraint, allowing exchange of labor
choices and decision making. A conceptual framework of between communes), or modify development goals that
methodological integration is presented in the next section cannot be achieved if constraints cannot be removed (i.e.
based on the lessons drawn from our action-research settle for intermediate rather than the extreme positions
project. We believe it can help facilitate and structure along a particular trade-off curve). However, the LUPAS
stakeholders’ interactions in the situation of the real methodology does not help chart the pathway to be taken
implementation of the operational design in a self- from the existing situation (i.e. from the circle in Fig. 10) to
reinforcing positive loop. the preferred situation—the target point along the trade-off
curve.
CLUE allocates land use types according to different
Combination of the three methodologies for a comprehensive
scenarios based on projections of current land use. The
understanding of land change
CLUE methodology does not explicitly determine the
resource limits (both biophysical and socio-economic)
These studies were carried out relatively independently
within which the projections of future land use are made.
by separate groups of researchers through interactions with
However, resource limits that are determined indepen-
different groups of stakeholders. Despite the original
dently (e.g. using the LUPAS methodology) can be
intention of generating complementarities between the
incorporated in the demand scenarios to run the CLUE
methodologies, for a number of reasons actual interactions
model. Because the CLUE model explicitly incorporates
among the research groups went little beyond sharing
the time dimension in projecting land use change, running
experiences and data sets. Faced with the common
CLUE with resource limitations implicitly incorporated in
challenge of describing and understanding a complex
the demand scenarios would generate outputs that chart
social–ecological system, each group sought its own entry
the temporal progression of land use changes from the
point, concepts, preferred partners, and modus operandi.
existing to a preferred scenario (Fig. 10). By comparing
Therefore, a conceptual framework is consequently re-
these alternative developments of near-future land use with
the optimized land use situations it can be determined
which scenario conditions favor most the development
towards the optimized outcome of the LUPAS methodol-
Biophysical limits of the agro-
ecological system as ogy. The CLUE model works on the assumption that
determined by LUPAS specific driving factors operate and influence changes from
Collective learning
one LUT to another, within user-specified rules of
path toward desired permissible change and rates of change. As mentioned
Regional income

local development above, a more realistic projection of temporal land use

target with SAMBA change could be achieved if the time-invariant influence of
each driving factor over the simulation period was replaced
by varying relative weights of different land use drivers to
capture changes in land use decisions made by farmers and
communities. If these changes in CLUE model can be
Scenarios of plausible LU accommodated, the decision rules that are revealed during
developments with CLUE the participatory simulations—with the role plays and the
multi-agent modelling of the SAMBA family of methodol-
ogies—would then provide more realistic estimates for the
Food production
varying influence of drivers to be used in the CLUE
Fig. 10. Combination of the three modelling methodologies of LUCC. modelling.
 ARTICLE IN PRESS
542 J.-C. Castella et al. / Land Use Policy 24 (2007) 531–545

This framework will make it possible for land use in the multi-agent model increased gradually when they
planners: (i) to explore different anticipated developments realized that the rules they had played in the game were
by changing the land use requirements (i.e. demand faithfully reproduced in the computer model. By applying
scenarios in CLUE) or land use drivers affecting the land the SAMBA methodology at the same sites where the field
allocation procedure in CLUE, (ii) to assess the impact of surveys were conducted, we were able to check that the
different policy instruments (e.g. spatial restrictions, properties of the social–ecological system that emerged
changes in regulations concerning land rights, subsidies), from the interactions between individual behaviors during
and (iii) to check if the system develops towards the the role playing game were consistent with the results of the
optimal conditions or could lead to non-optimal pathways more classical case studies. The multi-agent model was not
of change and finally the break-down of the system, scientifically validated in the traditional sense (i.e. by
resulting in social and environmental degradation. SAM- refutation based on statistical comparison between simu-
BA can then be used to test different local pathways that lated and observed data: Parker et al., 2003; Bousquet and
would be compatible with the goals that have been defined Le Page, 2004) but it gained sufficient legitimacy with all
and scenarios that have been selected at the regional level participants to be used as a learning device (Barreteau
through coupled LUPAS-CLUE analysis. Anticipated et al., 2001). As the objective of such processes is not to
local changes induced by regional policy interventions predict the future but to accompany the changes in land
could be visualized, and target villages identified for use with concerted actions, emphasis is placed on the
adaptation of specific policies. heuristic value of the proposed methods, and their capacity
to stimulate interactions between different groups of
Towards a platform for analysis of local–regional LUCC stakeholders (Castella et al., 2005b). Descriptive models
scenarios like CLUE can and should be thoroughly validated to test
their ability to make realistic projections of the changes in
Operationalizing such an integrative platform for land use pattern (Pontius et al., 2004). The LUPAS model
analysis of LUCC requires a collaborative mode of cannot be validated in this sense since it does not produce
communication. Different scenarios need to be discussed, realistic land use changes but provides optimization under
compared, and negotiated within and between social a set of conditions (Van Ittersum et al., 2004; Roetter et al.,
hierarchies of stakeholders with different views and 2005).
interests, who interact at different scales. While it is The quality of the data was a matter of greater concern to
recognized that such deliberations between diverse stake- policy makers than the quality of the models. Indeed,
holders would not necessarily lead to a consensus or the knowledge about the likely consequences of each decision
‘‘best solution’’, they would nevertheless provide opportu- option depends to a large extent on the available database.
nities to articulate diverse viewpoints and would foster Stakeholders involved in the LUPAS and CLUE modelling
better understanding and shared representation of the processes invariably questioned the quality of the input
problems at hand. A new mode of participation was data. Since both models used the same data sources as
proposed to support such interactions between several those used by decision-makers at the provincial level, it
groups of stakeholders, including scientists. It was built on increased their legitimacy for policy-makers. However,
the participatory design of the three methodologies: co- because the data were gathered from diverse sources, there
construction of scenarios with local farmers using SAMBA were problems of reconciling the disjointed datasets. The
and with district and province decision-makers around research teams constructed preliminary models with
LUPAS and CLUE models. In order to reconcile LUCC incomplete datasets (working on sub-regions or with
models with users’ expectations, a compromise was found missing variables) in order to elicit interest and response
by deploying scientifically sound approaches to support from provincial partners in the simulation exercise. The
non-scientific debate involving all parties (Castella et al., outcomes of the early versions of the models encouraged
2004; Van Paassen, 2004). ‘Non-scientific’ does not mean them to provide more data as well as data of better quality
that the deliberation process was not structured. The for improving the model, hence reinforcing positive feed-
integrative framework helped organize the interactions back.
and assigned appropriate roles to researchers, technical Participation quality was identified as a key issue for the
experts, local stakeholders, policy makers and members of effective implementation of the integrative framework. We
civil society (Renn, 2006). The issues related to the believe that in combination with the multi-agent model, the
practical implementation of our multi-stakeholder plat- role playing game encouraged genuine participation and
form in Bac Kan province are summarized below (Castella trust building, and consequently facilitated the convergence
et al., 2004). of views, values, and frames of reference of all participants,
Validation of the LUCC models prior to their use in a culminating in achieving a shared representation of the
negotiation platform was considered to be essential to system (Boissau and Castella, 2003). As an extension of
build the trust of all participants in the collaborative this logic, the combined use of the SAMBA, CLUE and
process. In the case of farmers confronted for the first time LUPAS methodologies involving different groups of
in their life with a computer interface, their individual trust stakeholders in the social hierarchy could potentially
 ARTICLE IN PRESS
J.-C. Castella et al. / Land Use Policy 24 (2007) 531–545 543

increase awareness of issues and problems at the different they are complementary in terms of scale, outputs and local
scales. Through our attempt to set up a collaborative partners. As members of such collaborative platforms,
platform involving different groups of stakeholders researchers should provide a basket of tools and guidance
in Bac Kan, we found that the quality of the partici- to stakeholders in their choices. Depending on their specific
pation largely depended on the selection of participants needs and objectives, the stakeholders may seek the
(Ravnborg and Westermann, 2002), their degree of assistance of researchers in combining results from
awareness of problems and their sense of need for solutions different modelling methodologies or to integrate the
(Castella et al., 2004; Van Ittersum et al., 2004; Van models themselves. Such a vision may seem utopian and
Paassen, 2004). the feasibility of a grand-scale concerted action may be
legitimately questioned (Jiggins and Röling, 2000). How-
Relevance to land use planning and policy making ever, our preliminary results show that concrete achieve-
ments accomplished from pilot studies can illustrate and
In trying to implement such a negotiation platform for convince key stakeholders of the benefits that they would
regional-NRM in Bac Kan province, the institutional gain from embarking on such a joint venture.
dimension of knowledge integration came rapidly to the
fore. The relationships among stakeholders largely deter-
mine the nature of information exchanged and knowledge Conclusions
generated through the interactive process (David, 2001).
We realized that farmers’ worldviews were disconnected With this paper we attempt to provide a concrete
from the worldviews of the provincial decision-makers, example of how abstraction of science and related
which made it difficult to link knowledge representations technologies can interact with laypeople’s knowledge in
yielded by the three modelling methodologies. Village support of collective action for NRM. Three modelling
communities enjoy relative independence in managing their methodologies (SAMBA, LUPAS and CLUE), based on
local resources and they systematically adapt policy two different yet complementary approaches of LUCC (i.e.
messages or regulations imposed at national or provincial bottom-up and top-down), were applied to the province of
levels to suit their own circumstances. Although there is Bac Kan in northern Vietnam. Our experience of
some degree of information transmission between groups, implementing the three models in the same target area
we observed tensions and a general lack of communication provided valuable insight into designing a multi-stake-
among stakeholders managing resources at different scales. holder platform for the analysis of LUCC. The combined
Each group of stakeholders concentrated on the methodol- use of these three methodologies was made possible by: (i)
ogy they initially contributed to without showing interest the well-defined development issues coming from the field
or relating to others’ models. The main impediment in and shared by all participants, (ii) a number of interrelated
effecting a multi-stakeholder NRM negotiation platform scientific questions on the LUCC process at multiple scales,
that would actually influence policy design lies in the and (iii) the availability of a portfolio of complementary
interaction between different stakeholder groups and not in modelling approaches (Castella et al., 2004). We draw three
technical issues related to reconciling findings of comple- main lessons from our studies: (i) more than one tool/
mentary LUCC models. method is needed to address the different aspects of
The next steps toward implementing a more integrative ecoregional analysis and to address different stakeholders’
framework for NRM will inevitably involve organizational issues at different scales, as no single method can capture
changes that may give a key role to the extension agents all levels adequately, (ii) for better interaction between the
(Shanks, 2002; Hoang et al., in press) as mediators between different groups of stakeholders more integration between
farmers, researchers and policy-makers. In Bac Kan the methods is needed, and (iii) methodological develop-
province, they play an important role in implementing ment should focus on the complementarities of the
the project of a ‘‘Regional Resource Center’’ (Castella approaches rather than on the models themselves.
et al., 2004). This center provides physical (i.e. infrastruc- Our story in Bac Kan is unique; it is one pathway among
ture), organizational and institutional support to the many other potential paths. There is consequently no
networking activities and collective learning on NRM. guarantee that it can be replicated in another province or
Data from different sources and collected for different even at the national or ecoregional scale. However, the
purposes would be made available for the use of different lessons learned from our experiment converge with other
stakeholder groups. Different tools and methods would be ecoregional research around the world (ISNAR, 2004;
tested, compared and combined in a flexible fashion and Sayer and Campbell, 2004), which suggests extrapolation
deployed for developing concerted plans for action by well- may be possible. The interactive use of simulation models
informed partners. The three models presented in this may contribute to translate the understanding gained
paper are only three such tools that were selected for through LUCC modelling into concrete policies, provided
practical reasons (i.e. because a large database was already that transparent procedural processes, previously agreed
available, the provincial authorities were interested, and upon by all participants, are set up. Our example in Bac
the research teams could be mobilized) and also because Kan only gives a brief glimpse of the major changes
 ARTICLE IN PRESS
544 J.-C. Castella et al. / Land Use Policy 24 (2007) 531–545

required to institutionalize such an integrative framework northern Vietnam: application of a multi-agent simulation model.
to LUCC analysis. Agricultural Systems 86 (3), 312–332.
Castella, J.C., Trung, T.N., Boissau, S., 2005b. Participatory simulation of
land use changes in the northern mountains of Vietnam: the combined
Acknowledgments use of an agent-based model, a role-playing game, and a geographic
information system. Ecology and Society 10 (1), 27 (online URL:
/http://www.ecologyandsociety.org/vol10/iss1/art27/S).
The research presented in this article was developed Costanza, R., Ruth, M., 1998. Using dynamic modelling to scope
under the umbrella of the EcoR(I)Asia, convened by the environmental problems and build consensus. Environmental Manage-
International Rice Research Institute. The authors would ment 22, 183–195.
like to express their sincere gratitude to all people and David, A., 2001. Models implementation: a state of the art. European
Journal of Operational Research 134, 459–480.
agencies that have contributed to this initiative and
De Koning, G.H.J., Verburg, P.H., Veldkamp, A., Fresco, L.O., 1999.
especially to all partners who made its implementation Multi-scale modelling of land use change dynamics in Ecuador.
possible in Bac Kan province, Vietnam. Agricultural Systems 61, 77–93.
De Wit, C.T., Van Keulen, H., Seligman, N.G., Spharim, I., 1988.
Application of interactive multiple goal programming techniques for
References analysis and planning of regional agricultural development. Agricul-
tural Systems 26, 211–230.
Barreteau, O., Bousquet, F., Attonaty, J.M., 2001. Role-playing games for Harrington, L., White, J., Grace, P., Hodson, D., A. Hartkamp, D.,
opening the black box of multi-agent systems: method and lessons of Vaughan, C., Meisner, C., 2001. Delivering the goods: scaling out
its application to Senegal River Valley irrigated systems. Journal results of natural resource management research. Conservation
of Artificial Societies and Social Simulation 4 (2), 5 (online URL: Ecology 5 (2), 19 (online URL: /http://www.ecologyandsociety.org/
/http://www.soc.surrey.ac.uk/JASSS/4/2/5.htmlS). vol5/iss2/art19/S).
Boissau, S., Castella, J.C., 2003. Constructing a common representation of Hoang, L.A., Castella, J.C., Novosad, P, in press. Social networks and
local institutions and land use systems through simulation-gaming and information access: Implications for agricultural extension in a rice
multi-agent modeling in rural areas of northern Vietnam: the SAMBA- farming community in northern Vietnam. Agriculture and Human
week methodology. Simulations & Gaming 34 (3), 342–347. Values.
Boissau, S., Anh, H.L., Castella, J.C., 2004. The SAMBA role play game ISNAR, 2004. Method in Our Madness: Making Sense of Ecoregional
in northern Vietnam. An innovative approach to participatory natural Research with Modelling Tools and Processes. International Service
resource management. Mountain Research and Development 24 (2), for National Agricultural Research, Hague, The Netherlands.
101–105. Jiggins, J., Röling, N., 2000. Adaptive management: potential and
Bouma, J., Kuyvenhoven, A., Bouman, B.A.M., Luyten, J.C., Zandstra, limitations for ecological governance. International Journal of
H.G. (Eds.), 1995. Ecoregional Approaches for Sustainable Land-use Agricultural Resources, Governance and Ecology 1 (1), 28–42.
and Food Production. Kluwer Academic Publishers, Dordrecht, The Kam, S.P., Hoanh, C.T., 1999. In: Proceedings of a Workshop: Scaling
Netherlands. Methodologies in Ecoregional Approaches for Natural Resources
Bousquet, F., Le Page, C., 2004. Multi-agent simulations and ecosystem Management, 22–24 June 1998, Ho Chi Minh City, Vietnam. IRRI,
management: a review. Ecological Modelling 176 (3–4), 313–332. Los Banos, Philippines.
Bousquet, F., Barreteau, O., d’Aquino, P., Etienne, M., Boissau, S., Kam, S.P., Hoanh, C.T., Castella, J.C., Kinh, N.N., 1999. Summary
Aubert, S., Le Page, C., Babin, D., Castella, J.C., 2002. Multi-agent Proceedings of a Technical Workshop: Ecoregional Approaches for
systems and role games: an approach for ecosystem co-management. Natural Resource Management in the Red River Basin, Vietnam, 9–10
In: Janssen, M.A. (Ed.), Complexity and Ecosystem Management: The November 1998, Hanoi. IRRI, Los Banos, Philippines.
Theory and Practice of Multi-Agent Systems. Edward Elgar Publish- Kam, S.P., Castella, J.C., Hoanh, C.T., Trebuil, G., Bousquet, F.,
ers, Cheltenham, UK and Northampton, MA, USA, pp. 248–285. 2002. Methodological integration for sustainable natural resource
Briassoulis, H., 2000. Analysis of land use change: theoretical and management beyond field/farm level: lessons from the ecoregional
modeling approaches. In: Loveridge, S. (Ed.), The Web Book of initiative for the humid and sub-humid tropics of Asia. International
Regional Science (www.rri.wvu.edu/regscweb.htm). Regional Re- Journal of Sustainable Development and World Ecology 9 (4),
search Institute, West Virginia University, Morgantown, WV. 383–395.
Castella, J.C., Quang, D.D. (Eds.), 2002. Doi Moi in the Mountains. Land Kinh, N.N., Teng, P.S., Hoanh, C.T., Castella, J.C., 1999. Towards an
Use Changes and Farmers’ Livelihood Strategies in Bac Kan. The Ecoregional Approach for Natural Resource Management in the Red
Agricultural Publishing House, Hanoi, Vietnam. River Basin of Vietnam. The Agricultural Publishing House, Hanoi,
Castella, J.C., Kam, S.P., Hoanh, C.T., 2001. New partnership mechan- Vietnam.
isms for integrated natural resource management: the Red River Basin Meinzen-Dick, R., DiGregorio, M., McCarthy, N., 2004. Methods for
(Vietnam) pilot region for operational ecoregional approach. Interna- studying collective action in rural development. Agricultural Systems
tional Journal of Sustainable Development 4 (4), 351–367. 82 (3), 197–214.
Castella, J.C., Boissau, S., Anh, H.L., 2003. Enhancing communities’ Papin, P., Tessier, O. (Eds.), 2002. Lang o vung chau tho song hong: Van
adaptability to a rapidly changing environment in Vietnam uplands: de con bon go. The Village in Questions. EFEO/Trung tâm khoa hoc
the SAMBA role-play. In: Serrano, R.C., Aggangan, R.T. (Eds.), xa hôi và nhân van quôc gia, Hanoi.
Sustaining Upland Development in Southeast Asia: Issues, Tools & Parker, D.C., Manson, S.M., Janssen, M.A., Hoffmann, M., Deadman,
Institutions for Local Natural Resource Management. PCARRD, Los P., 2003. Multi-agent systems for the simulation of land-use and land-
Banos, Philippines, pp. 203–236. cover change: a review. Annals of the Association of American
Castella, J.C., Quang, D.D., Thévenot, P., 2004. Toward new modes of Geographers 93, 314–337.
governance of the research-development continuum to facilitate the Pontius, R.G., Huffaker, D., Denman, K., 2004. Useful techniques of
dissemination of agricultural innovations in a mountainous province validation for spatially explicit land-change models. Ecological
of northern Vietnam. International Journal of Agricultural Resources, Modelling 179 (4), 445–461.
Governance and Ecology 3 (1–2), 77–94. Ravnborg, H.M., Westermann, O., 2002. Understanding interdependen-
Castella, J.C., Boissau, S., Trung, T.N., Quang, D.D., 2005a. Agrarian cies: stakeholder identification and negotiation for collective natural
transition and lowland-upland interactions in mountain areas in resource management. Agricultural Systems 73, 41–56.
 ARTICLE IN PRESS
J.-C. Castella et al. / Land Use Policy 24 (2007) 531–545 545

Renn, O., 2006. Participatory processes for designing environmental Van Ittersum, M.K., Roetter, R.P., Van Keulen, H., De Ridder, N.,
policies. Land Use Policy 23 (1), 34–43. Hoanh, C.T., Laborte, A.G., Aggarwal, P.K., Ismail, A.B., Tawang,
Rindfuss, R.R., Walsh, S.J., Turner II, B.L., Fox, J., Mishra, V., 2004. A., 2004. A systems network (SysNet) approach for interactively
Developing a science of land change: challenges and methodological evaluating strategic land use options at sub-national scale in South and
issues. Proceedings of the National Academy of Sciences of the USA 101 South-east Asia. Land Use Policy 21 (2), 101–113.
(39), 13976–13981 /www.pnas.org/cgi/doi/10.1073/pnas.0401545101S. Van Paassen, A., 2004. Bridging the gap: computer model enhanced
Roetter, R.P., Hoanh, C.T., Laborte, A.G., Van Keulen, H., Van learning for natural resource management in Burkina Faso. Tropical
Ittersum, M.K., Dreiser, C., Van Diepen, C.A., De Ridder, N., Van Resource Management Paper 49. Wageningen University and
Laar, H.H., 2005. Integration of systems network (SysNet) tools for Research Centre, Wageningen, The Netherlands.
regional land use scenario analysis in Asia. Environmental Modelling Verburg, P.H., Soepboer, W., Veldkamp, A., Limpiada, R., Espaldon, V.,
& Software 20 (3), 291–307. Mastura, S.A., 2002. Modelling the spatial dynamics of regional
Röling, N., 1994. Platforms for decision-making about ecosystems. In: land use: the CLUE-S model. Environmental Management 30 (3),
Fresco, L.O., Stroosnijder, L., Bouma, J., Van Keulen, H. (Eds.), The 391–405.
Future of the Land: Mobilising and Integrating Knowledge for Land Verburg, P.H., Schot, P., Dijst, M., Veldkamp, A., 2004a. Land use
Use Options. Wiley, Chichester, UK, pp. 385–393. change modelling: current practice and research priorities. GeoJournal
Sayer, J., Campbell, B., 2004. The Science of Sustainable Development: 61 (4), 309–324.
Local Livelihoods and the Global Environment. Cambridge University Verburg, P.H., Veldkamp, A., Willemen, L., Overmars, K.P., Castella,
Press, Cambridge, UK. J.C., 2004b. Landscape level analysis of the spatial and temporal
Shanks, E., 2002. Agriculture and Forestry Extension and Sustainable complexity of land use change. In: De Fries, R., Asner, G., Houghton,
Livelihoods in the Uplands. Swiss Agency for Development Coopera- R. (Eds.), Ecosystems and Land Use Change. Geophysical Mono-
tion, Hanoi, Vietnam. graph Series, vol. 153. American Geophysical Union, Washington,
Steins, N.A., Edwards, V.M., 1999. Synthesis: platforms for collective DC, USA, pp. 217–230.
action in multiple-use common-pool resources. Agriculture and Willemen, L., Verburg, P.H., Castella, J.C., Vu, N., 2002. Modelling of
Human Values 16, 309–315. land cover changes with CLUE-S in Bac Kan province, Vietnam. SAM
Teng, P.S., Hossain, M., Fischer, K.S., 1995. Developing a R&D model Paper Series, Vol. 17. Vietnam Agricultural Science Institute, Hanoi,
for the humid tropical ecoregion in Asia. In: Bouma, J., Kuyvenhoven, Vietnam.
A., Bouman, B.A.M., Luyten, J.C., Zandstra, H.G. (Eds.), Ecoregio- Yen, B.T., Kam, S.P., Ha, P.Q., Hoanh, C.T., Hien, B.H., Castella, J.C.,
nal Approaches for Sustainable Land Use and Food Production. Duc, H.Q., Tuan, V.D., Nguyen, V., Son, C.K., 2002. Exploring land
Kluwer Academic Publishers, Dordrecht, Netherlands, pp. 305–332. use options for agricultural development in Bac Kan province,
Van Ittersum, M.K., Rabbinge, R., Van Latensteijn, H.C., 1998. Vietnam. Paper No. 1217. In: Proceedings of the 17th World Congress
Exploratory land use studies and their role in strategic policy making. of Soil Sciences, 14–21 August 2002. Soil and Fertilizer Society of
Agricultural Systems 58, 309–330. Thailand, Bangkok, Thailand.