Advertisement
Browse Subject Areas
?

Click through the PLOS taxonomy to find articles in your field.

For more information about PLOS Subject Areas, click here.

  • Loading metrics

Exploring Agricultural Livelihood Transitions with an Agent-Based Virtual Laboratory: Global Forces to Local Decision-Making

  • Nicholas R. Magliocca ,

    nmag1@umbc.edu

    Affiliation Department of Geography and Environmental Systems, University of Maryland, Baltimore County, Baltimore, Maryland, United States of America

  • Daniel G. Brown,

    Affiliation School of Natural Resources and Environment, University of Michigan, Ann Arbor, Michigan, United States of America

  • Erle C. Ellis

    Affiliation Department of Geography and Environmental Systems, University of Maryland, Baltimore County, Baltimore, Maryland, United States of America

Exploring Agricultural Livelihood Transitions with an Agent-Based Virtual Laboratory: Global Forces to Local Decision-Making

  • Nicholas R. Magliocca, 
  • Daniel G. Brown, 
  • Erle C. Ellis
PLOS
x

Abstract

Rural populations are undergoing rapid changes in both their livelihoods and land uses, with associated impacts on ecosystems, global biogeochemistry, and climate change. A primary challenge is, thus, to explain these shifts in terms of the actors and processes operating within a variety of land systems in order to understand how land users might respond locally to future changes in broader-scale environmental and economic conditions. Using ‘induced intensification’ theory as a benchmark, we develop a generalized agent-based model to investigate mechanistic explanations of relationships between agricultural intensity and population density, environmental suitability, and market influence. Land-use and livelihood decisions modeled from basic micro-economic theories generated spatial and temporal patterns of agricultural intensification consistent with predictions of induced intensification theory. Further, agent actions in response to conditions beyond those described by induced intensification theory were explored, revealing that interactions among environmental constraints, population pressure, and market influence may produce transitions to multiple livelihood regimes of varying market integration. The result is new hypotheses that could modify and enrich understanding of the classic relationship between agricultural intensity and population density. The strength of this agent-based model and the experimental results is the generalized form of the decision-making processes underlying land-use and livelihood transitions, creating the prospect of a virtual laboratory for systematically generating hypotheses of how agent decisions and interactions relate to observed land-use and livelihood patterns across diverse land systems.

Introduction

Land-use change and its effects on land cover are major contributors to global environmental change [1][4]. Agricultural land-use is the most extensive and environmentally consequential land-use on Earth [5], [6]. Until recently (i.e., until the twentieth century), agriculture in most regions has been managed by smallholders: rural households with small amounts of land, who produce at least some of their own subsistence using family labor, and are only partially integrated into markets that are often inefficient or incomplete [7][9]. Smallholders remain important agents of land change [10], as their land-use practices have consequences for local food security, environmental sustainability, and economic development [11]. As economic globalization produces more and stronger teleconnections between urban and rural land systems, local land-use change is increasingly influenced by transitions from subsistence to commercial agriculture and non-farm livelihoods in response to changing regional/global market forces [12], [13].

Current understanding of the dynamics of agricultural land-use by smallholders has been built in part from a rich case-study literature related to ‘induced intensification’ theory [14]. Agricultural intensification was first described by Boserup [15] and Chayanov [16] as a process through which smallholders were forced to increase the labor intensity of cultivation through techno-managerial innovations to meet increasing production demands from rising population density. Case studies from a wide range of disciplines expanded on these insights to consider the roles of environmental suitability [17] and commercial agricultural activities [18][20] in driving agricultural intensification, which became more broadly labeled as ‘induced intensification’ theory [14]. Empirical support for this theory has been established by strong, positive correlations between agricultural intensity and population density, with environmental and economic pressures as mediating factors [14], [17], [21]. However, such methods cannot provide direct, mechanistic explanations of how smallholder behavioral responses to these factors lead to observed patterns of land-use and -cove change (LUCC). Furthermore, empirical evidence for the linkages between livelihood strategies and land-use patterns is based on fragmented literatures of local case studies [22], and is thus subject to the ‘one place, one time’ syndrome [23]. While synthesis methods, such as meta-analysis, can identify common patterns across empirical case studies, they cannot provide mechanistic explanations of how such empirical patterns emerge from underlying processes. Thus, current methods are insufficient for explaining the processes through which agricultural intensification arises within and across sites, and thus lack the means to form hypotheses about how patterns of LUCC result from the responses of actors to a range of global to local conditions.

Rindfuss and colleagues [24] propose that structured cross-site comparisons and syntheses can be facilitated by simulation models of LUCC, and agent-based models (ABMs) in particular because of their explicit representation of human decision-making processes. Parker and colleagues [25] attempted such a cross-site comparison with a set of ABMs of land-use change in frontier regions, but found that the same processes were often represented in different ways across models, which prohibited a clear synthesis of findings across studies. Here, we use a generalized ABM of land-use and livelihood decision-making as a virtual laboratory to investigate relationships between variations in agricultural intensity and the contexts within which agent decision-making occurs. The model’s simple structure links agents’ livelihood decisions to resulting land-use changes and enables generation of hypotheses about how land-use patterns are affected by varying local and global economic, environmental, and demographic conditions. Such a generative modeling tool, employed to productive effect in other fields [26][29], can be used to systematically explore the implications of agent decision-making for land-use and livelihood outcomes, as well as formulate new hypotheses that could lead to more nuanced theoretical explanations of land change processes than the inductive methods common in land change science [30].

We first ground the generalized ABM to induced intensification theory by evaluating outcomes against the empirical relationships derived in Turner and colleagues’ [17] meta-analysis of agricultural intensification and agent-level behavioral patterns described in the livelihoods and development literatures. Our aim is to verify that agents’ assumed decision-making models respond to changing economic, environmental, and demographic forces in realistic ways subject to heterogeneous risk preferences and environmental endowments. We then use a virtual laboratory approach to explore how livelihood strategies and land-use outcomes vary under a wide range of environmental and social conditions that are impossible to control for and observe in the field. As a result of this analysis we identify the effects of increased market integration on land-use and livelihood transitions, beyond those observed in the earlier work on induced intensification, as an area for further empirical testing.

Methods

General characteristics of the model relevant for discussion of results are presented below. Detailed model description using a complete Overview, Design concepts, and Details (ODD) Protocol is provided in File S1. The ODD protocol is an accepted standard for presenting ABMs (http://www.openabm.org/page/standards), which describes the model’s purpose, design, and process overview and scheduling.

2.1. Landscapes

A stylized landscape of 100 by 100 square grid of cells (Fig. 1a) is created in MATALB with each cell representing one hectare (total area = 100 km2). An artificial landscape is used so that: 1) the full range of land suitability classes defined by GAEZ [31], [32], can be simulated within the same landscape; 2) the landscape does not represent any particular place and is thus generic; and 3) the landscape can be easily implemented and reproduced within any simulation environment. Turner and colleagues [17] found that the relationship between population density and agricultural intensity was dampened in favorable agricultural conditions, but exacerbated when significant constraints on agricultural productivity were present. Here, land suitability for agriculture is represented by a combination of slope and precipitation constraints. Slope is a proxy for soil suitability for agriculture with reductions in potential agricultural yields based on slope constraint classes [31]. Precipitation constraints are related to the length of the growing season [32], which impose additional reductions in potential agricultural yields evenly across the entire landscape. Experimental variations in agricultural suitability, which are a combination of slope and precipitation constraints, are implemented by turning slope and/or precipitation constraints ‘on’ or ‘off’ to investigate the influence of environmental conditions on agricultural intensity.

thumbnail
Figure 1. Hypothetical landscapes.

Components of the hypothetical landscapes: (a) artificial topography and land suitability for agriculture; b) simulated land uses with agents 7 and 59 indicated (Section SI-5).