Agrarian transition and lowland–upland interactions in mountain areas in northern Vietnam: application of a multi-agent simulation model

AGRICULTURAL SYSTEMS Agricultural Systems 86 (2005) 312–332 www.elsevier.com/locate/agsy

Agrarian transition and lowland–upland interactions in mountain areas in northern Vietnam: application of a multi-agent simulation model Jean-Christophe Castella a,b,c,*, Stanislas Boissau Tran Ngoc Trung c, Dang Dinh Quang c a

a,c

,

Institut de Recherche pour le De´veloppement (IRD), 213 rue Lafayette, 75480 Paris Cedex 10, France b International Rice Research Institute (IRRI), DAPO Box 7777, Metro Manila, Philippines c Mountain Agrarian Systems Program, Vietnam Agricultural Sciences Institute (VASI), Thanh Tri, Hanoi, Vietnam Received 18 November 2003; received in revised form 13 August 2004

Abstract The agrarian transition that accompanied the decollectivisation of agriculture in the northern Vietnam uplands challenges traditional approaches to land use analysis. In the 1980s, a series of reforms progressively returned means of production to individual farmers. Over the same period the mountains in the study area lost 50% of their forest cover. This paper provides an overview of the impact of changes in government policies on farmersÕ practices, on land use, and on environmental dynamics. It applies a multi-agent simulation model to validate the hypotheses derived from household survey and remote sensing data about the mechanisms linking the allocation of paddyland to farm households with shifting cultivation on the hillsides, and deforestation. The model explains the diversity of the current land use systems by analysing the interactions between land tenure policies in the lower part of the toposequence and the dynamics of land use in the upper part. The findings are based on an extensive diagnostic survey of a number of representative sites, on generation of primary data from an intensive survey of * Corresponding author. Present address: Institut de Recherche pour le De´veloppement (IRD), B.P. 64501, 34394 Montpellier Cedex 5, France. Tel.: +33 4 67 63 69 80; fax: +33 4 67 63 87 78. E-mail address: [email protected] (J.-C. Castella).

0308-521X/$ - see front matter
2004 Elsevier Ltd. All rights reserved. doi:10.1016/j.agsy.2004.11.001

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

313

50 randomly selected households, and on insights gained by the authors during their long association with an action research project using a participatory observation method. The authors also explore whether the modelling approach can capture the main features of complex humanenvironment interactions of a site and can help apply the findings to the broader agroecosystem the site represents.
2004 Elsevier Ltd. All rights reserved. Keywords: Agrarian transition; Uplands; Multi-agent simulations; Land tenure; Differentiation; Vietnam

1. Introduction The agricultural collectivisation implemented in Vietnam in the early 1960s temporarily put an end to individual ownership of land. Land was declared to be the common property of all Vietnamese people and a system of cooperative farms was set up. Between the end of the 1970s and early 1980s, this cooperative system entered a crisis phase characterised by a dramatic decrease in rice production (Donnell, 1980; Kerkvliet and Porter, 1995; Kerkvliet and Selden, 1998). Two successive reforms were implemented in 1982 (Decree 100) and 1986 (Resolution 10) to gradually shift control over agricultural production from cooperatives to individual households (Beresford, 1990; Dao, 1997). During the same period, massive deforestation occurred in the mountains of northern Vietnam that was linked to the rapid expansion of upland cultivation (Donovan et al., 1997; Jamieson et al., 1998; De Koninck, 1999). Although it appears that reforms in land-ownership policy caused the rapid changes in land use, this causal relationship must be questioned for two reasons: first, other factors may have interacted with the land policy reform to trigger such rapid environmental changes (e.g. opening to market economy, improvement in transportation infrastructure, etc.); second, the effects of the policy reform varied considerably between agroecological zones depending on local interpretation and implementation of national policies (Le and Rambo, 2001; Sikor, 2001; Castella and Dang, 2002). Surveys conducted in northern Vietnam showed relative homogeneity of agricultural practices and livelihood systems during the collectivist period as compared to the post collectivist period, especially in the early 1990s (Fforde, 1987; Dao, 1997). This led to the hypothesis that most of the current diversity in farming systems emerged during the transition period between cooperative and household-based agriculture. To test this hypothesis, and to understand the mechanisms of changes in land use that occurred in the northern mountains during this transition period, we developed the SAMBA 1 multi-agent model, which simulates the interactions between the processes 1 The name ‘‘SAMBA’’ is formed from SAM (the French acronym for Mountain Agrarian Systems), the name of the project in which the methodology was developed, and ‘‘ba’’ meaning ‘‘three’’ in Vietnamese. The SAMBA model is considered as a link between two components of the project: (i) one about diagnosis of land use changes and problem prioritisation with local stakeholders, (ii) the other about intervention through design and the testing, in collaboration with farmers, of technical and organisational innovations (for further information see project web site at: http://www.knowledgebank.irri.org/sam/ home_en.html).

314

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

of allocation of lowlands and land use in the uplands. Beside its heuristic function as a tool to integrate interdisciplinary knowledge, the model was designed to elucidate mega-trends (i.e., the major forces driving changes in land use), and the diverse expressions of these mega-trends as influenced by highly variable local contexts. In recent years, agent-based models have been used by an increasing number of researchers to simulate changes in land use/land cover (Berger, 2001; Lambin et al., 2003; Parker et al., 2003). The complex interactions between agents and between agents and their environment are simulated by coupling a cellular model that represents the biophysical components of the environment with an agent-based model that represents human decision making (Ferber, 1999). Agent-based simulations can be used both as part of explanatory approaches or descriptive approaches to land use changes (Parker et al., 2003). The model presented in this paper pertains to the first category. It does not attempt to reproduce actual land use systems but to generalise the findings from highly diverse research sites. It builds upon assumptions about agentsÕ/farmersÕ behaviour generated through deductive analysis of empirical evidence. The combined behaviours of autonomous agents in interaction with the simulation environment generate a macro-phenomenon (i.e. new land use patterns, deforestation trend) as an emergent property of the human-environment system. The emergent properties that result from repeated simulations are then checked against empirical data (e.g. land use maps) using an inductive approach. The modeller can thus demonstrate that a set of rules can lead a particular phenomenon at a higher hierarchical level, explore other possible causes that could lead to the same outcome, and discover outcomes that were not originally anticipated (Parker et al., 2003). The SAMBA model was developed from comprehensive surveys conducted in a network of field research sites located in Bac Kan province. After a short presentation of the overall methodological framework we describe the main findings of our empirical research. We then show how we identified a minimum set of variables from our survey data that we believed was sufficient to explain recent changes in land use. We incorporated these variables and rules into a rule-based model to test our hypotheses through simulations of the forces driving differentiation trajectories in farming systems and changes in land use. Finally, we discuss the behaviour of the model and possible further improvements. Indeed, this first model has already been further developed into a more descriptive one that was used to understand more recent changes in land use (i.e. forest regeneration associated with forestland allocation during the 1990s) and to assist local stakeholders in decision making about management of their natural resource as part of a companion modelling approach (Bousquet et al., 2002; Boissau and Castella, 2003).

2. Methods Bac Kan province was selected because of the rapid expansion of shifting cultivation systems in recent years and the concomitant risk of environmental crisis, which regional and national decision-makers considered a priority (Do, 1994;

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

315

Fig. 1. Map of Bac Kan province showing research sites.

Morrison and Dubois, 1998). Bac Kan province is divided into seven administrative districts, each of which is sub-divided into 5–20 communes (Fig. 1). Six commune-level research sites were selected along a gradient of integration into the market economy, and according to the agro-ecological diversity observed at the provincial scale (Castella and Dang, 2002). The diagnostic studies conducted in each of these pilot sites typically included one year of field survey and spatial analysis of remote sensing data. In this paper, one of the case studies, conducted in Xuat Hoa commune is used to illustrate the whole research process (Sadoulet et al., 2002). An exhaustive survey investigated the family structures and the combined production of each of the 450 farm households in the commune. Households were first classified according to their place in the family cycle (i.e. young couples, families with school age children, parents living with the family of the oldest son), wealth, and dominant production activities. Then a stratified random sample of 50 representative households was selected from three villages that are representative of the main ethnic groups in the region: i.e. Tay, Dao and Kinh. Semi-structured interviews with these 50 families examined their history, resource endowment, and production strategies. Key informants (officials, village elders, etc.) were also interviewed about the villagesÕ history; in particular about changes in land tenure and agricultural practices over the last forty years. The spatial dimension of land use dynamics was examined by interpreting a chronological series of aerial pictures from 1954, 1977 and 1998, and satellite images from 1990, 1995, and 2001. A systems approach combined these data from differ-

316

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

ent sources to investigate the forces that drove past changes in land use and their impact on the transformation of agricultural landscapes and on the differentiation of farming systems (Mazoyer and Roudart, 1997). The method and its results (i.e. farm typology, quantitative data about production strategies, land use maps and maps of changes in land use, etc.) are described in Sadoulet et al. (2002). A conceptual farming-system differentiation model was derived from the analysis of the household surveys. This differentiation model, which is based on the ratio ‘‘household labour force to number of mouths to feed’’ and on access to the lowland fields suitable for paddy, was first described through narrative story lines. It describes land use change at a specific locality, Xuat Hoa commune, based on recall information from the surveys (Lambin et al., 2003). The SAMBA rule-based model was then designed to test the influence of the demographic composition of households on the differentiation of farming-systems during the 1980s. Multi-agent simulations led to emergent properties at the village scale from very simple behavioural rules of individual household access to land, allocation of the labour force, and the use of production surpluses (Ferber, 1999; Bousquet et al., 2002). These behavioural rules were then validated by comparing the outputs of the simulation (i.e. changes in land use, household types, and aggregated data) with empirical data.

3. Changes in land use and differentiation of farming systems in Xuat Hoa commune 3.1. Impact of land policy reforms on management of the agro-ecosystem Agricultural production in the study area relies on the combined use of the two main agro-ecological units: irrigated paddy land in the valley bottoms and rainfed steep hillsides. The historical review presented here can be better described as a succession of four different land use systems (LUS) corresponding to three main stages (i.e. pre-independence, collectivist, family-based) in the evolution of the agrarian system (Fig. 2).  Before collectivisation (i.e. before 1960), agricultural pressure on the ecosystem was relatively low (LUS A in Fig. 2). Paddy fields producing one rice crop per year were combined with swiddens on the hillsides that alternated upland crops (upland rice, maize, and cassava) with long fallow periods. Household labour capacity was the main determining factor of the size of cropped areas. The agrarian system was characterised by a preferential use of lowland by the Tay and upland by the Dao ethnic groups. Farmers were free to open new upland fields in the forest and could retain rights of use of the land even during extended fallow periods.  In the mid-1970s, collective agriculture concentrated on the intensification of paddy fields through introduction of Green Revolution innovations (LUS B in Fig. 2). The increase in production in the lowlands through double cycle cropping, new varieties, fertilisation, etc. lowered the pressure on the uplands. Lowland rice

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

317

Fig. 2. Chronological succession of agrarian systems in Xuat Hoa commune.

became more productive and less risky than upland rice because it was better protected from climatic vagaries. On the other hand, the forest protection policy became a disincentive to slash and burn agriculture.  By the late 1970s, the increasing population steadily reduced mean per capita rice production, collective work yielded decreasing returns and the cooperatives were in a state of crisis (LUS C in Fig. 2). Natural resources were preserved in the uplands, but people became hungry. Individual upland fields rapidly expanded, because labour productivity was much higher than on paddy fields. Households raised increasing numbers of livestock on their own, leaving the collective buffalo herd to stagnate. Lowland fields were under-exploited.  By the early 1980s, most of the hillsides were deforested and the swiddens reached the limit of their exploitable area (LUS D in Fig. 2). Decreasing yields on the slopes, combined with the new policy of secure land tenure in the lowlands, became major incentives first to higher labour investment, then to capital investment in the paddy fields. Agricultural production increased rapidly in the lowlands. New, more-sustainable land management practices emerged on the slopes, including agroforestry systems. However, not all families were able to implement such practices as these required high capital investment. Comparative analysis of our complementary study sites confirmed that the trends described here above pertained to the whole province of Bac Kan (Castella and Dang, 2002). Each change in land tenure policy triggered considerable changes in land use systems and in their impact on the environment (Fig. 3). However, the transition periods, which usually occurred during a crisis of the previous system (Mazoyer and Roudart, 1997), were not well documented. This is the case for both the collectivisation and decollectivisation processes in the mountains of northern Vietnam. This paper focuses on the latter transition period because it has shaped the current rules in natural resource management and because we believe that it is largely responsible for the current diversity in farming systems.

318

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

Fig. 3. Agricultural transitions and ecological dynamics in Xuat Hoa commune.

3.2. Impact of land policy reforms on differentiation in farming systems The forces driving the differentiation process of the 1980s can be grouped in three categories: (i) the inequity of the land allocation system under ‘‘Resolution 100’’ set up in 1982, (ii) the privileged situations of some groups that resulted from the groupsÕ access to exploitation of forest resources, and (iii) reclaiming of ancestral lowlands, resulting in the re-emergence of former inequalities (Sadoulet et al., 2002). 3.2.1. Decree 100 (1982–1986) Decree 100 marked the beginning of a new distribution system for lowland fields and paddy production among members of cooperatives: (i) allocation of lowland areas as a function of the number of members in each family in order to secure food self sufficiency; (ii) deduction of a specified output quota for the land,

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

319

2

Paddy field area per worker (in m ) 1400

Maximum area per worker for two-cycle paddy

350

Ratio worker to mouth to feed minimum: 1/4

maximum: 1/1

Fig. 4. Distribution of paddy areas under Decree 100. An average of 350 m2 was distributed per family member.

intended to pay for inputs and cooperative labour; and (iii) payment of cooperative labour based on labour points. Such a system seemed simple and equitable. Households with a lot of mouths to feed would be able to obtain a bigger rice surplus. But the implementation of this policy led to the opposite effect. Families with a small labour force relative to the number of mouths to feed had to cultivate bigger areas per worker, sometimes larger than the maximum area that its workers could manage with two-cycle rice (Fig. 4). Given their shortage of available time for lowland work, families with large paddy field endowments could not crop the slopes even though this would have provided higher return on labour than paddy production. During this period, some labour-rich families were able to clear fertile forested land and expand maize cultivation needed for pig raising. Clearing and cultivating forested land was primarily a strategy to acquire land rather than to generate income, because market opportunities were still limited. By concentrating control of forested land in the hands of labour-rich families, the implementation of Decree 100 led to a rapid social differentiation of the households that intensified in subsequent years (Fig. 5). 3.2.2. The ‘‘adjusted contract’’ (1986–1989) After only four years, the agricultural system combining collective management of the lowlands and private shifting cultivation in the uplands had entered a major crisis. By 1986, all land suitable for slash-and-burn cultivation had already been cleared. Only marginal land remained or formerly fallowed, upland fields, the production capacity of which had already been depleted. Labour productivity was decreasing on the hillsides and the overall agricultural production would have fallen if the production-contract system had been maintained on collective land. Therefore the Xuat Hoa cooperative decided, more or less autonomously, to change its rules. Control over means of production shifted from the cooperative to the households. Only actual ownership of the land remained collective. Paddy fields were redistrib-

320

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

Increasing paddy area per household Large inherited paddy field area

Paddy field purchase or creation TYPE Ia

TYPE IIIa

TYPE IIIb

TYPE IIa TYPE IIIc

Large inherited paddy field area High value of ratio “labour force / mouths to feed”

TYPE IIb High value of ratio “labour force / mouths to feed” TYPE Ib

TYPE IIId Large inherited paddy field area

TYPE IIc

TYPE IIIe

TYPE IIIf

Time 1980

1985

1990

1995

2000

Typology I: after Decree 100

Typology II: Adj. contract

TYPE Ia: Families rich in labour force: food surplus (stocks) and large area under slash and burn cultivation on the slopes TYPE Ib: Families suffering from regular food shortage, often indebted with the cooperative (short term loans)

TYPE IIa: Capital available since TYPE IIIa: Combination of extensive paddy fields and fruit the implementation of Decree 100. tree plantations. Very limited annual crops on the slopes. Pig raising, savings = buffaloes TYPE IIIb: Medium sized paddy area, crops on slopes with TYPE IIb: Favorable ratio labour long-term economic return (fruit trees) force to mouths to feed. TYPE IIIc: Extensive paddy area, gradual capitalization and Capitalization during this period investment in fruit tree plantations, evolution towards (timber exploitation, plantation of TYPE IIIa fruit trees) TYPE IIId: Serious rice shortage. Upland crops (maize, TYPE IIc: Labour force lacking,no cassava) valorised through intensive pig husbandry surplus. TYPE IIIe: Medium sized paddy area, short-term return on slopes (cassava, taro) TYPE IIIf: Small paddy area, low capital, off-farm activities and indebtedness

Typology III: Resolution 10

Fig. 5. Farming system differentiation tree and typology in Xuat Hoa commune.

uted according to the number of workers per family. The cooperative sold buffaloes equitably among all families. The end of cooperative interventions had an immediate effect on rice production. Families again started to invest time in paddy fields. During this period, labour-rich families were able to diversify agricultural production and invest in fruit tree plantations (especially apricots) that became profitable in the early 1990s when the market developed. 3.2.3. Resolution 10 (from 1990 to the present) In 1988, the government endorsed a new re-organisation of the agricultural sector, named Resolution 10. This new national policy triggered a new wave of spontaneous changes in the mountainous region. In 1990, Tay families in Bac Kan Province claimed back the land that their ancestors had contributed to the cooperatives in the 1960s. They got back their former paddy fields and distributed them among their heirs. This led to very unequal land distribution that reproduced the inequalities between ethnic groups of the pre-independence system. However,

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

321

compared to pre-independence, the population density was much higher and the forest had been almost completely cleared (Fig. 3). The scarcity of alternatives for those who were left without paddy fields helped widen the gap between the different household groups (Fig. 5). Since 1994, Xuat Hoa has implemented the governmentÕs new forestland allocation policy. After lowland allocation, official land titles to the hillsides were distributed to families. But the rules for the distribution of forestland were derived from the traditional land-right system that allocated to families the same areas that they had cultivated in the past (Morrison and Dubois, 1998). Consequently, forestland distribution reinforced the inequalities created among families between 1982 and 1986 based on the amount of access they had to upland cultivation. 3.3. Hypotheses on mechanisms for redistribution of means of production Fig. 5 shows a differentiation tree of farming systems since the early 1980s based on data from Xuat Hoa commune. This conceptual model of differentiation emphasises the following two factors: the ratio ‘‘labour force to mouths to feed’’ in the 1980s, especially between 1982 and 1986, and the extent of paddy fields inherited from ancestors in the early 1990s. At each intersection on the evolutionary tree, differentiated access to lowland linked to the demographic composition of the family under Decree 100 (typology I) and Adjusted Contract (typology II) then because of lowland appropriation by the Tay (typology III) resulted in a particular set of production systems and production strategies, i.e. a combination of cropping and animal husbandry systems (Fig. 5). A detailed description of each household type and corresponding production strategies is provided in Sadoulet et al. (2002) and supported by quantitative data. Nevertheless, the household differentiation model is based on strong assumptions and relies on locality-specific data. The mechanisms for the redistribution of means of production were poorly documented and the implementation of the land reforms differed substantially between provinces in northern Vietnam (Le and Rambo, 2001; Sikor, 2001) and even between communes within Bac Kan Province (Castella and Dang, 2002). Several reasons can be put forward to explain the lack of documentation and thus the difficulty in generalising these findings through a comparative analysis between research sites. Local decisions often preceded official policies. Official policies were clearly documented, but in a context of crisis (especially regarding production cooperatives), decisions made by the local authorities were not well documented. Changes were extremely rapid with a new land policy appearing every four years. Local authorities had difficulty coping with these rapid changes and implemented the policy in a very flexible manner that depended on local circumstances. The extreme diversity of local situations and characteristics of the different villages including their natural environment, resource endowment, accessibility, and ethnic groups, thus led to a diversity of land use trajectories that are impossible to document systematically. Last, some of the land tenure conflicts are still not settled today and discussing these issues remains taboo in many places.

322

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

To open the back box of the transition period, we had to formulate a series of hypotheses regarding the past impact of land rights and access to resources on (i) agricultural and environmental dynamics and (ii) differentiation of farming systems. These hypotheses are summarised as a set of driving forces that may explain changes in land use and provide a scenario of the agrarian transformations. 3.3.1. Driving forces of changes in land use After Lambin et al. (2003), we listed ‘‘five high level causes of land use change’’ that contributed to the changes observed in the 1980s: 1. resource scarcity due to the increasing population and poor productivity of the cooperative system in the lowlands; 2. changing opportunities created by an emerging market and poor enforcement of the regulations for forest protection; 3. changing land policies: i.e. succession of four major reforms in 10 years (Decree 100, Adjusted contract, Resolution 10 and Forestland allocation under the 1993 land law); 4. loss of capacity to adapt linked to the rapid pace of change and to the absence of historically constructed technical referential, and increased vulnerability caused by a food deficit that came close to famine; 5. changes in social organisation with the end of the cooperative system, and in attitudes with the emergence of very individualistic modes of production as a result of the malfunctioning of the collectives (Rambo et al., 1995).

3.3.2. Scenario of the agrarian transformations As a consequence of land policy reforms, the use of the agroecosystem was gradually intensified in three successive stages. First, the distribution of land use rights for the lowlands associated with free access to the uplands, and higher labour productivity in the uplands compared to the lowlands, led to agricultural expansion in the uplands through swidden cultivation systems. Secondly, once the limits of suitable land for swiddening were reached, paddy rice systems were intensified in the lowlands and livestock herds were expanded. Last, secure land tenure in the lowlands and later in the uplands became strong incentives to increase labour investment in food crops for those households who were not self-sufficient in rice, and, for the others, to diversify agricultural production with cash crops and animal production. At the local scales (i.e. household, village, commune), such a narrative approach was unable to reveal any regularities in land use/cover changes resulting from differentiated farmersÕ strategies; quite the contrary, it told as many stories as the number of case studies (Castella and Dang, 2002). The SAMBA multi-agent model was consequently developed as an integrative framework for multidisciplinary and multi-site empirical knowledge. Its main objective was to validate our hypotheses by identifying the minimum set of variables that would generate a coherent scenario of land use change from comparative analysis of our six commune-level research sites in Bac Kan Province.

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

323

4. The SAMBA multi-agent model 4.1. Model structure The SAMBA model has been programmed under the CORMAS multi-agent platform (Bousquet et al., 1998). It includes two entities: agents ‘‘HouseHold’’ and cells ‘‘Cell_Samba’’ that provide the environment in which the agents make decisions. The entities, as well as their attributes, parameters, and behavioural methods are derived from the field surveys. Each ‘‘HouseHold’’ agent is characterised by its demographic composition: i.e. number of individuals that make up the household (mouthsToFeed) of which a subset are workers (labourForce); land endowment (land) expressed as the number of plots of 1000 m2 ; the number of buffaloes the household owns (buffalo); and the amount of available cash (wallet). Each ‘‘HouseHold’’ agent can implement the following actions: compute its paddy production: computePaddyProduction, compute its rice surplus or deficit by comparing its paddy production (4 t/ha by worker cultivating a maximum area of 2000 m2), and the food requirements of the family (300 kg/pers.). Next, the agent decides how best to distribute the remaining labour force between two production systems: upland rice and cash crops (including fruit tree plantations): balanceProductionNeed. It computes the income generated by rice production: computeIncome. Rice production exceeding family needs is completely sold at the average price of 2.500 VND/kg (USD1 = VND14.000). Wallet content increases from selling rice and/or cash crop production: increaseWallet. A buffalo is bought at the price of 2.000.000 VND each time there is enough money in the wallet: buyBuffalo. The possibility to invest in other consumer goods (motorcycles, hand tractors, etc.) was not introduced in this version of the model, because at the time, access to market was very limited in the mountains. The environment is made up of a 2.500 cell grid (50 · 50). Each cell represents a plot of 1000 m2. Each cell is characterised by its distance from the village (remoteness), and its land use (state) that can have six values: paddyField, uplandRice, fallow, pasture, cashCrop, and forest. At the beginning of the simulation, cell attributes ‘‘remoteness’’ and ‘‘state’’ are initialised as follows: ‘‘remoteness’’ is an index ranging from 1 (cells close to the grid centre) to 10 (for cells located farther from the village). A group of cells at the centre of the grid are initialised as ‘‘paddyField’’ and others cells as ‘‘forest’’. We started with such a simplified environment to keep the model as generic and theoretical as possible at the beginning of the simulation. The proportion of the different resources and their distance to the village should influence the simulation outcome, not the actual position of the resources on the grid. In addition, the land cover at the beginning of the simulation is also a proxy for the geomorphology of the mountain village, with the irrigated flat land at the valley bottom represented as paddyField and the surrounding hillsides as forest. Cells can display the following dynamics: land use change to fallow when the plot is abandoned after a cultivation period: turnToFallow, gradual forest regeneration when the plot is under fallow: regeneration. The model algorithm for one time step corresponding to a one-year simulation is presented in Fig. 6.

324

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

balance Production Need

need := 300 * mouths To Feed production := compute Paddy Production

production > need

True

False

surplusLabor > 0

if suitable land available

True

growUR := true

False

if suitable land available

grow Cash := true

Fig. 6. Diagram showing the method of allocating labour force to different activities.

4.2. Simulations The SAMBA model can simulate different situations by changing: - The size of the environment (grid size) and the number of cells initialised as ‘‘paddyField’’, - The human population (number of individual ‘‘HouseHold’’ agents) and household composition (mouthsToFeed and labourForce within each HouseHold), - The rules for the allocation of paddy fields: depending on the number of mouths to feed or on the number of workers per household, equal distribution among households or only to some of the households. Simulations presented here were parameterised using the case study conducted in Xuat Hoa commune (Sadoulet et al., 2002). Simulations were implemented on a 2.500 cell grid (50 · 50) with 144 cells initialised as ‘‘paddyField’’. A family

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

325

composition database derived from the actual Xuat Hoa commune birth registry was used to initialise the ‘‘HouseHold’’ agents. The village model includes 50 ‘‘HouseHold’’ agents representing a total population of 246 virtual individuals. At initialisation, ‘‘paddyField’’ cells are distributed according to the number of mouths to feed, and after four time steps, they are re-distributed according to the labour force of each ‘‘HouseHold’’ agent. This simulation corresponds to the calendar of events as they occurred in Xuat Hoa commune in 1982 and 1986 during the transition period between cooperatives and family-based agriculture (Fig. 2). The simulation then runs for four more time steps. The total duration of one simulation is eight time steps, representing eight years (1982–1990), during which ‘‘HouseHold’’ agents and cells evolve according to the rules described above. The results are exported to a database from which simulation graphics are generated. During the first four time steps, ‘‘HouseHold’’ agents assign their labour force surplus to upland rice cultivation and thus bring about a decrease in forest cover. A small surplus in rice production allows some capital accumulation and therefore some buffalo purchase. After the fifth time step, the new distribution of ‘‘paddyField’’ cells allows the development of cash crops and a decrease in land under upland rice. Fig. 7 shows changes in land use over the eight time steps. The changes in forest cover between the beginning and the end of the simulation were consistent with the actual changes in forest cover in Xuat Hoa commune (50%) observed from remote sensing data between 1977 and 1990. As the simulation grid was too abstract for a geographic comparison of land use changes, we studied the trends such as a decrease in forest area, an increase in agricultural land and shrub land, changes in the ratio of forest area to shrub area, etc. between key dates. The simulated dynamics of changes in land use (Fig. 7) were consistent with those reported from field surveys, which provided the first stage of validation. The increase in the number of buffalo was also observed in reality as attested by field surveys and secondary data (Castella and Dang, 2002). The model also allows investigation of the individual behaviours of ‘‘HouseHold’’ agents that are hidden behind these aggregated results. Three ‘‘HouseHold’’ types emerged from the simulation: Number of cells of 1000m2 2500 2000

uplandRice cashCrop buffalo forestCover treeCover

1500 1000 500 0

Years 1

2

3

4

5

6

7

8

Fig. 7. Simulated land use changes in the uplands. Cumulative result for all households.

326

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332 paddy

uplandRice

cashCrop

buffalo

Number of cells of 1000m2 4.5 4

A

3.5 3 2.5 2 1.5 1 0.5 0

Years 1

2

3

4

5

6

7

8

4

5

6

7

8

5

6

7

8

Number of cells of 1000m2 9 8

B

7 6 5 4 3 2 1 0

Years 1

2

3 2

Number of cells of 1000m 2.5

C

2 1.5 1 0.5 0

Years 1

2

3

4

Fig. 8. Simulation results for household types A, B and C.

Type A households (Fig. 8A): The first distribution of paddy fields as a function of the number of mouths to feed was not sufficient to cover family rice needs. Labour force surplus was then assigned to upland rice cultivation. Lowland re-allocation at the fifth time step as a function of the family labour force allowed most/all households to meet their rice needs. The remaining labour force was allocated to growing cash crops. Type B (Fig. 8B): Regardless of which land allocation procedure was applied, rice production in the lowlands did not cover household needs. The labour force was

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

327

Table 1 Demographic composition of the three household types identified through SAMBA simulation Mouths to feed

1

2

3

4

5

6

7

8

9

10

A

B A

B B A

C B B A

C C B A A

C C B B A A

C C C B B A A

C C C B B B A A

C C C B B B A A A

C C C C B B B A A A

Labour force 1 2 3 4 5 6 7 8 9 10

concentrated on upland rice production. Upland rice surpluses were saved and invested in buffaloes. Type C (Fig. 8C): These households lacked an adequate labour force. The number of paddy fields allocated at the first time step was not high enough for the family to secure food sufficiency, even though all family workers concentrated on the production of paddy. Labour shortages prevented these households from producing upland rice. The second wave of land allocation (as a function of the available labour force) allowed additional lowland rice production and thereby freed some of the labour force to be assigned to upland rice cultivation. Even so, these households still showed a rice deficit. Household demographic composition corresponding to these three types is summarised in Table 1. The household typology that emerged from the SAMBA simulations is consistent with the one derived from field surveys over the period 1980–1990 (Sadoulet et al., 2002). The three types A, B and C presented above correspond to types IIa, IIb and IIc, respectively, in Fig. 5. The comparison of simulated and observed patterns of household differentiation provided the second stage of validation of the SAMBA model. 4.3. Model response to initial conditions Initialisation parameters have a considerable impact on the results produced by the householdÕs behaviours and land use strategies. The demographic composition of the households and the total number of ‘‘paddyField’’ cells at the beginning of the simulation determine to a large extent land use dynamics in the uplands. If the land available for rice is insufficient, then none of the households, regardless of their demographic composition, will be able to produce cash crops (Fig. 9A). On the other hand, when the rice area per mouth to feed is too big, cash crops can appear as soon as the first paddy land is allocated (Fig. 9B), allowing the emergence of a new kind of behaviour that was not observed in real conditions. In fact, the market economy was also undergoing transition during the period under simulation and mountain farmers had only limited opportunities to market their agricultural products.

328

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

A

Case where the paddy field area is not sufficient to cover rice needs (paddy rice production < 300kg x village population) Number of cells of 1000m2 3000 2500

uplandRice cashCrop buffalo forestCover treeCover

2000 1500 1000 500 0 1

B

2

3

4

5

6

7

Years

8

Case where the paddy field area largely covers rice needs (paddy rice production > 300kg x village population) Number of cells of 1000m2 2500 2000 1500 1000 500 0 1

2

3

4

5

6

7

8

Years

Fig. 9. Simulated land use changes in the uplands. Cumulative result for all households.

These figures show how important the ratio between the population size and village resource endowment is for the emergence of the different land use systems. The emergence of village trajectories depended on a shortage of land for the production of lowland rice. Insufficient land for paddy prevented the development of cash crops, whereas a large paddy area would have favoured this development as soon as the first lowlands were distributed. Field surveys showed that the real situation of Tay villages in Xuat Hoa commune lay between the two. Most farmers only engaged in cash crop production once they had secured rice self-sufficiency from the paddies after the second allocation of lowland under the Adjusted Contract. As expected, the villages where shifting cultivation persisted and even increased after the successive land distributions are those with a low paddy area per household. These simulations thus confirmed certain empirical findings. For example, they showed that unlike in other provinces in Vietnam (e.g. the Central Highlands with coffee, rubber, sugarcane, etc.), mountain farmers in Bac Kan would not engage in cash crop production unless they had secured rice self-sufficiency from the paddies (Sadoulet et al., 2002).

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

329

5. Discussion: limits of the model and possible future development The results of the simulation show that the SAMBA model can satisfactorily mimic agricultural dynamics and differentiation of farming systems in the 1980s, using only very simple household-based rules. In this model, agents do not interact directly with each other but indirectly via the changes they trigger in their environment. This modelling algorithm corresponds to the hypothesis of an absence of coordination among farmers during the period under investigation. Instead, agents display reactive behaviour. Each agent acts independently from the others, and always responds to a given stimulus in the same way. These very simple rules are sufficient to trigger the emergence of behaviours that are very similar to those observed in real conditions. The collective work system managed by the cooperatives was followed by a land appropriation ‘‘rush’’ that was driven by individual strategies and relied to a large extent on the labour force available in the different households. The same kind of behaviours has been already observed in the case of rapidly advancing agricultural pioneer fronts (Angelsen, 1995; Perz, 2002). Field surveys and further action research conducted in Bac Kan Province (Castella et al., 2003) confirmed the limited interactions between farmers regarding cropping practices and decision making on land use during this transition period, thereby confirming our choice of reactive agents for the sake of the simplicity of the model. So far, the SAMBA model has been used to reveal the mechanisms underlying the changes in land use during the transition period of decollectivisation. It was also possible to investigate local variations in changes in land use as a function of household demographic composition and paddy field endowment. The advantage of such a model is its simplicity and thus its adaptability to the very diverse situations that we observed through field surveys. We do not pretend that such a ‘‘reactive’’ model takes into account all the factors involved in the observed changes and we do not deny the strong coordination patterns that existed, and still exist, among farm households with respect to traditional kinship and other social networks. The model was designed with only the degree of complexity necessary to test hypotheses derived from agronomic surveys and spatial analyses of remote sensing data. It serves to identify ‘‘sufficient conditions’’ to explain the emergence of the different land use strategies that were observed at the end of the cooperative period. The absence of coordination between reactive ‘‘HouseHold’’ agents limits the usefulness of the model as a tool for understanding the more recent period, which is characterised by much more complex processes (Castella et al., 2003). Including these other factors, such as social interactions, relations of production among households, was very challenging given the very rapid pace of change and the extreme diversity in the study area. To face this challenge, we designed a role-play using the main features of the SAMBA model. Each player represented a household entity and wooden cubes on the game board each represented a grid cell. The coloured sides of the cubes corresponded to six different land uses displayed on the grid of the computer model. Initially, rules and parameters of the game were the same as those in the model (Boissau et al., 2004). We used the role-play to feed the model with the new rules given to us by the farmer-players during the game. The SAMBA

330

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

model thus gradually evolved toward a descriptive model by introducing only the rules and parameters that seemed important and meaningful to local stakeholders to simulate the most recent changes (Bousquet et al., 2002). This new version of the model is more realistic as it takes into account interactions between agents; it incorporates the biophysical characteristics of the environment through coupling with GIS and initialisation with actual village maps; and it allows validation by comparison between the simulated map output and the actual land use map based on remote sensing data. However, it has lost its simplicity, its more general applicability, and its heuristic value (Parker et al., 2003). The first sessions of the SAMBA role-play were dedicated to the validation of the original version of the SAMBA multi-agent model by local stakeholders. In addition to the comparative analyses of simulated and observed farm household typologies and trends in changes in land use, the role-play provided a third level of validation of the forces driving change in the 1980s, and of assumptions such as using reactive agents (Castella et al., 2003). Through the game we verified that the SAMBA model could satisfactorily capture the changes that had occurred in other communes than Xuat Hoa. Seven role-play sessions in a representative sample of communes confirmed that the SAMBA model was relevant for the whole province of Bac Kan for the period considered (Boissau and Castella, 2003). This implies the model can be used anywhere in the province without going through the lengthy process of data collection once again.

6. Conclusion The SAMBA model explains the diversity of current land use systems by enabling better understanding of the interactions between land tenure policies in the lower part of the toposequence and agricultural dynamics in the upper part during the post-collectivist period. It links the outcome of individual land use decisions and measures of landscape change. A more classic monographic approach would have ended with the formulation of hypotheses based on results of field surveys. However, such an approach would have been unable to capture the extreme diversity of local situations at the provincial scale. To avoid this limitation, the SAMBA model was developed to understand regional changes in land use from the analysis of local agricultural dynamics. In a context of very rapid changes, this type of modelling approach can capture the main features of the local conditions and identify the key parameters necessary for its generalisation to larger geographic areas.

References Angelsen, A., 1995. Shifting cultivation and ‘‘deforestation’’: A study from Indonesia. World Development 23, 1713–1729. Beresford, M., 1990. Vietnam: socialist agriculture in transition. Journal of Contemporary Asia 20, 466– 486.

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

331

Berger, T., 2001. Agent-based spatial models applied to agriculture: A simulation tool for technology diffusion, resource use changes, and policy analysis. Agricultural Economics 25, 245–260. Boissau, S., Castella, J.C., 2003. Constructing a common representation of local institutions and land use systems through simulation-gaming and multi-agent modeling in rural areas of Northern Vietnam: the SAMBA-Week methodology. Simulations and Gaming 34, 342–347. Boissau, S., Hoang, L.A., Castella, J.C., 2004. The SAMBA role play game in northern Vietnam. An innovative approach to participatory natural resource management. Mountain Research and Development 24, 101–105. 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. Bousquet, F., Barreteau, O., dÕAquino, P., Etienne, M., Boissau, S., Aubert, S., Le Page, C., Babin, D., Castella, J.C., 2002. Multi-agent systems and role games: an approach for ecosystem co-management. In: Janssen, M. (Ed.), Complexity and Ecosystem Management: The Theory and Practice of Multiagent Approaches. Edward Elgar Publishers, Cheltenham, UK and Northampton, MA, USA, pp. 248–285. Castella, J.C., Boissau, S., Hoang, L.A., 2003. Enhancing communitiesÕ adaptability to a rapidly changing environment in Vietnam uplands: the SAMBA role-play. In: Serrano, R.C., Aggangan, R.T. (Eds.), Sustaining Upland Development in Southeast Asia: Issues, Tools and Institutions for Local Natural Resource Management. PCARRD, Los Banos, Philippines, pp. 203–236. Castella, J.C., Dang, Q.D. (Eds.), 2002. Doi Moi in the Mountains. Land Use Changes and FarmersÕ Livelihood Strategies in Bac Kan Province, Vietnam. The Agricultural Publishing House, Hanoi. Dao, T.T., 1997. The agrarian transition in Vietnam: institutional change, privatisation and liberalisation. In: Spoor, M. (Ed.), The ÔMarket PanaceaÕ: Agrarian Transformation in Developing Countries and Former Socialist Economies. Intermediate Technology, London, pp. 156–169. De Koninck, R., 1999. Deforestation in Vietnam. International Development Research Centre, Ottawa, Canada. Do, S.D., 1994. Shifting Cultivation in Vietnam: Its Social, Economic and Environmental Values Relative to Alternative Land Use. International Institute for Environment and Development, London. Donnell, J.C., 1980. Vietnam 1979: year of calamity. Asian Survey 20, 19–32. Donovan, D., Rambo, A.T., Fox, J., Le, C.T., Tran, V.D., 1997. Development Trends in VietnamÕs Northern Mountain Region. vol. 1. An Overview and Analysis. National Political Publishing House, Hanoi. Ferber, J., 1999. Multi-Agent Systems: An Introduction to Distributed Artificial Intelligence. AddisonWesley, Reading, MA. Fforde, A., 1987. Socio-economic differentiation in a mature collectivised agriculture: North Vietnam. Sociologia Ruralis 27, 197–215. Jamieson, N., Le, C.T., Rambo, A.T., 1998. The Development Crisis in VietnamÕs Mountains. East-West Centre, Honolulu. Kerkvliet, B.J., Porter, D.J. (Eds.), 1995. VietnamÕs Rural Transformation. Westview Press, Boulder. Kerkvliet, B.J., Selden, M., 1998. Agrarian transformations in China and Vietnam. The China Journal 40, 37–58. Lambin, E.F., Geist, H., Lepers, E., 2003. Dynamics of land-use and land-cover change in tropical regions. Annual Review of Environmental Resources 28, 205–241. Le, C.T., Rambo, A.T., 2001. Bright Peaks, Dark Valleys: A Comparative Analysis of Environmental and Social Conditions and Development Trends in Five Communities in VietnamÕs Northern Mountain Region. National Publishing House, Hanoi. Mazoyer, M., Roudart, L., 1997. Histoire des Agricultures du Monde. Editions du Seuil, Paris. Morrison, E., Dubois, O., 1998. Sustainable Livelihoods in Upland Vietnam: Land Allocation and Beyond. Forestry and Land Use Series No. 14. Forestry and Land Use Programme, International Institute for Environment and Development, London. Rambo, A.T., Reed, R.R., Le, C.T., DiGregorio, M.R. (Eds.), 1995. The Challenges of Highland Development in Vietnam. East West Centre, Honolulu.

332

J.-C. Castella et al. / Agricultural Systems 86 (2005) 312–332

Parker, D.C., Manson, S.M., Janssen, M.A., Hoffmann, M., Deadman, P., 2003. Multi-agent systems for the simulation of land-use and land-cover change: a review. Annals of the Association of American Geographers 93, 314–337. Perz, S.G., 2002. Household demography and land use allocation among small farms in the Brazilian Amazon. Human Ecology Review 9, 1–16. Sadoulet, D., Castella, J.C., Vu, N.H., Dang Q.D., 2002. A short history of land use changes and farming system differentiation in Xuat Hoa Commune, Bac Kan Province, Vietnam. In: Castella, J.C., Dang, Q.D. (Eds.), Doi Moi in the Mountains. Land Use Changes and FarmersÕ Livelihood Strategies in Bac Kan Province, Vietnam. The Agricultural Publishing House, Hanoi, pp. 21–46. (http://www.knowledgebank.irri.org/sam/sam/pdf/chap01_E.pdf). Sikor, T., 2001. The allocation of forestry land in Vietnam: Did it cause the expansion of forests in the Northwest?. Forest Policy and Economics 2, 1–11.