An agent-based platform for simulating complex human–aquifer interactions in managed groundwater systems
Introduction
There is an increasing recognition that groundwater resources generate complex socio-ecological issues (Zellner, 2008). Effective and fair solutions to groundwater management thus require a holistic approach based on the knowledge and expertise of many disciplines. This approach, however, is not always mirrored in classical groundwater modelling tools. What may be described as a complex problem is often compressed into a model that describes a well-defined problem with simple cause-effect relationships (Pahl-Wostl, 2007).
Agent-based models (ABMs) have come forth as a way to model, as opposed to simplify, the complexity of socio-ecological systems (Aulinas et al., 2009, Bousquet and Le Page, 2004, Letcher et al., 2013, Parker et al., 2003). ABMs use the concept of an ‘agent’ (a computational representation of real-world actor) to simulate behaviours and interactions of decision-making entities, including feedbacks between human and environmental processes, physical and institutional constraints, and the different spatiotemporal scales in which these dynamics unfold (Miller and Page, 2009). ABMs allow framing socio-ecological issues based on a set of agent behaviours, and from these simple rules complex system behaviour ‘emerges’ (Mitchell, 2009). In practice, this is the basis for efficient groundwater-resource management (Foster and Garduño, 2012, Foster and Perry, 2010).
In groundwater management, ABMs show significant potential to design policies and incentives that may help balancing the need to produce crops, to provide drinking water and to ensure the long-term sustainability of aquifers. ABMs can also help detect key human and institutional actions that may lead to the sustainable exploitation of aquifers in real-world scenarios. For example, (Blomquist and Ostrom, 1985, Ostrom, 1990) give empirical evidence of cases where self-monitoring has led to efficient management of shared groundwater resources over long periods of time, with little or no intervention of a regulator. Using ABM we can ask: what other mechanisms may have such positive impacts on groundwater use?
Answering this question can be challenging using conventional modelling tools. Tools such as simulation-optimisation (Barlow et al., 2003, Bredehoeft and Young, 1983, Bredehoeft and Young, 1970, Morel Seytoux, 1975, Raul and Panda, 2013, Sedki and Ouazar, 2011, Young and Bredehoeft, 1972), evolutionary algorithms (Babbar-Sebens and Minsker, 2010, Babbar-Sebens and Minsker, 2012, McKinney and Lin, 1994, Mirghani et al., 2009), econometric models (Brozovic et al., 2010, Katic and Grafton, 2012, Wan et al., 2012), game theory (Negri, 1989, Raquel et al., 2007, Saak and Peterson, 2007), and Bayesian networks (Henriksen and Barlebo, 2008, Henriksen et al., 2007, Portoghese et al., 2013) focus on equilibrium states (e.g., a global optimum, a Nash equilibrium), and describe social processes in an aggregate manner (e.g., using an optimisation function, a differential equation, a payoff matrix, etc.) based on the concept of a ‘typical’ agent assumed to be on average rational i.e. making optimal and fully informed decisions. This is unlikely to represent individual variations (heterogeneity) and random influences (stochasticity) in human decisions and interactions (Bonabeau, 2002, Rounsevell et al., 2011). These assumptions undermine the representation of complexity, which is critical for understanding the dynamics of coupled human-groundwater systems (Zellner, 2008). In contrast, ABMs can simulate large cohorts of independent, heterogeneous agents with clearly defined rules in systems that evolve (Miller and Page, 2009). ABMs thus can model groundwater systems affected by human activities, as these activities adapt to changes in socioeconomic and environmental factors.
The difficultly of coupling social-economic ABM models to existing groundwater flow modelling environments (e.g., MODFLOW) is reflected in the few publications where this has been achieved (Barthel et al., 2008, Mulligan et al., 2014, Reeves and Zellner, 2010). This contrasts with the large number of publications coupling ABMs with other biophysical models (An, 2012, Aulinas et al., 2009, Balbi et al., 2009, Bousquet and Le Page, 2004, Gunkel, 2005, Heath et al., 2009, Letcher et al., 2013). Early work on GW-ABMs reports lumped aquifer models implemented on an ABM grid (Carlin et al., 2007, Dray et al., 2006, Feuillette et al., 2003, Guilfoos et al., 2013, Heckbert et al., 2006, Moglia et al., 2010, Perez et al., 2003, Smajgl et al., 2009). Recent work is based on linked GW-ABMs (Barthel et al., 2005, Miro, 2012, Mulligan et al., 2014, Reeves and Zellner, 2010), where an ABM generates groundwater stresses (i.e. pumping rates) that are exported to a groundwater code e.g., MODFLOW (Harbaugh et al., 2000). The groundwater code then updates the physical state variables of the model and the new conditions inform or change the behaviour of agents in the following iteration.
We identify four main limitations in previous attempts to couple agent-based and groundwater flow models. First, when lumped models have been used, assumptions of homogeneous geology and infinite transmissivity constrain the analysis to steady-state conditions, and underestimate pumping costs and damages to linked ecosystems (Brozovic et al., 2010, Esteban and Albiac, 2011, Katic and Grafton, 2012, Koundouri, 2004). Second, linked GW-ABMs can be computationally expensive (Matthews et al., 2005) as they require communication via data files and libraries to synchronize both codes. Their implementation requires expertise in multiple programming languages and demands maintenance, given that ABM and groundwater software is under continuous development. This provides little insight on the actual development process of a GW-ABM and the independent replication its results. Third, linked GW-ABMs offer less flexibility for developing and adapting scenarios. For example, consider a model designed with agents responding to groundwater heads. If one wanted to explore scenarios where agents react to other variables (e.g., the stage of a river or the flow in a spring) one would not only need to modify the ABM, but also the data exchange library. Fourth, sensitivity analyses on a linked GW-ABM can be impractical in real-world management situations. For instance, if one wanted to explore the impacts of geological heterogeneity (i.e., the spatial distribution of hydraulic parameters and their uncertainty) on model output, it would be helpful to generate multiple alternative geological models, run batch simulations, visualise and analyse the output within a single software, without the detour via input/output files and revisions to both codes. The above issues suggest that an integrated simulation environment would facilitate the development and subsequent analysis of GW-ABMs. We propose FlowLogo as a first step towards a common language and standard procedures or templates to build GW-ABMs.
The aim of this work is to present FlowLogo, a new GW-ABM environment based on a finite-difference approximation to the governing equation of groundwater flow and is written in NetLogo, a widely-used and open open-source ABM environment (Heath et al., 2009, Railsback et al., 2006, Wilensky, 1999). FlowLogo is a research tool aimed at interdisciplinary groundwater studies and policy making at the basin scale, targeting researchers from a wide range of fields such as economics, social science, law, and hydrology. We show the application of FlowLogo's main features using a hypothetical example based on simple agent rules that yet lead to complex collective behaviours. Rather than a case study making specific policy recommendations, this example is intended as a guided tutorial for the typical stages of developing an agent-based groundwater model. Similarly, we analyse scenarios representing different combinations of policy levers and agent learning mechanisms to illustrate the platform's potential as a decision-support tool. We discuss the advantages of FlowLogo with respect to general-purpose programming languages, as well as prospective contributions of the platform for decision-support in a selection of groundwater depletion hotspots around the world.
Section snippets
FlowLogo architecture
FlowLogo is implemented in the NetLogo platform, an agent-based programming environment for simulating complex phenomena. NetLogo has become a popular research tool to explore the connections between micro-level behaviours of individuals and macro-level patterns that emerge from their interactions (Tisue and Wilensky, 2004). NetLogo is open-source and it is available for most operating systems. A models library and online repositories (e.g., https://modellingcommons.org, https://www.openabm.org
GW-ABM implementation example
In this section we present a tutorial that illustrates the process of developing GW-ABMs in the FlowLogo modelling environment. It is based on a synthetic model of an expanding agricultural region that depends exclusively on groundwater for irrigation. The scenarios below are not presented to make specific management recommendations beyond our synthetic example, nor do they necessarily represent assumptions based on empirical data. They are given to illustrate how coupled social-groundwater
Discussion and conclusions
This paper describes FlowLogo, a coupled agent-based and a groundwater simulator offering a new environment for modelling complex human–aquifer interactions. We demonstrate its functionalities by implementing a hypothetical aquifer system, a set of agents representing groundwater users and their interactions. To the best of our knowledge, FlowLogo is the first integrated software offering a straightforward way to represent agent behaviours that evolve with groundwater conditions. It allows
Acknowledgements
This work was funded by scholarships from UNSW Australia, the Chilean Government, the National Centre for Groundwater Research and Training (NCGRT) and the Gary Johnston fund. The authors are grateful to The University of Lausanne (UNIL), the Connected Waters Initiative research centre (CWI) and CSIRO for their support in this research. The authors thank the Editor and the anonymous reviewers for providing valuable comments and suggestions to improve the quality of the paper.
References (96)
Modeling human decisions in coupled human and natural systems: review of agent-based models
Ecol. Model.
(2012)- et al.
A Case-Based Micro Interactive Genetic Algorithm (CBMIGA) for interactive learning and search: methodology and application to groundwater monitoring design
Environ. Model. Softw.
(2010) - et al.
Interactive genetic algorithm with mixed initiative interaction for multi-criteria ground water monitoring design
Appl. Soft Comput.
(2012) - et al.
An integrated modelling framework for simulating regional-scale actor responses to global change in the water domain
Environ. Model. Softw.
(2008) - et al.
Large-scale water resources management within the framework of GLOWA-Danube. Part A: the groundwater model
Phys. Chem. Earth, Parts A/B/C
(2005) - et al.
Multi-agent simulations and ecosystem management: a review
Ecol. Model.
(2004) - et al.
On the spatial nature of the groundwater pumping externality
Resour. Energy Econ.
(2010) - et al.
Groundwater and ecosystems damages: questioning the Gisser–Sánchez effect
Ecol. Econ.
(2011) - et al.
SINUSE: a multi-agent model to negotiate water demand management on a free access water table
Environ. Model. Softw.
(2003) - et al.
Transboundary water management Game-theoretic lessons for projects on the US-Mexico border
Agric. Econ.
(2000)