Geospatial simulation steering for adaptive management

Highlights

• Computational steering of stochastic geospatial simulations poses unique challenges.

• Steering geospatial simulations enables exploring adaptive management scenarios.

• With intuitive interfaces steering can be successfully used in participatory modeling.

Abstract

Spatio-temporal simulations are becoming essential tools for decision makers when forecasting future conditions and evaluating effectiveness of alternative decision scenarios. However, lack of interactive steering capabilities limits the value of advanced stochastic simulations for research and practice. To address this gap we identified conceptual challenges associated with steering stochastic, spatio-temporal simulations and developed solutions that better represent the realities of decision-making by allowing both reactive and proactive, spatially-explicit interventions. We present our approach, in a participatory modeling case study engaging stakeholders in developing strategies to contain the spread of a tree disease in Oregon, USA. Using intuitive interfaces, implemented through web-based and tangible platforms, stakeholders explored management options as the simulation progressed. Spatio-temporal steering allowed them to combine currently used management practices into novel adaptive management strategies, which were previously difficult to test and assess, demonstrating the utility of interactive simulations for decision-making.

Introduction

Spatio-temporal simulations provide a powerful way to study complex spatial phenomena, develop spatial theories, and even forecast the future, especially when traditional experimental methods to reveal patterns and processes are difficult or impossible to implement (Sullivan and Perry, 2013). Accordingly, substantial research efforts have been devoted to developing dynamic, spatio-temporal models of large-scale, socio-ecological phenomena, such as biological invasions (Meentemeyer et al., 2011; Miller et al., 2017) or sustainable urban growth (Meentemeyer et al., 2013). These models are particularly useful for simulating the efficacy of interventions—such as strategies to curb the spread of invasive species—which may have delayed impacts, cost too much, or become controversial (Garner and Hamilton, 2011).

Given the complexity of socio-environmental problems, researchers increasingly use participatory methods to incorporate diverse stakeholder perspectives into problem-solving. Participatory modeling has been shown to help researchers develop relevant questions, construct better models, and generate solutions that can be easily translated into decisions (Voinov and Bousquet, 2010). Spatio-temporal simulations have proven effective in participatory modeling studies dealing with land use (Lagabrielle et al., 2010), flood hazards (Becu et al., 2017), and disease spread (Hossard et al., 2013; Gaydos et al., 2019), but there is still a need to better integrate these models into the decision-making process (Vukomanovic et al., 2019; Gaydos et al., 2019). Decision support poses a new challenge to modelers, requiring them to make models more interactive and reflective of the realities of decision-making. Most spatio-temporal simulations are not interactive, i.e., they are initialized with a set of inputs that cannot be adjusted while the simulation is running. Such a non-interactive workflow pairs well with Monte Carlo techniques that allow researchers to capture uncertainties associated with stochastic models and model ensembles, and to run calibration or sensitivity analyses by simulating large numbers of model realizations (Yang, 2011; Rubinstein and Kroese, 2016). However, a non-interactive simulation can obscure cause-effect relationships and is impossible to adjust in response to new information or to its own intermediate results. Moreover, most spatio-temporal models do not have interactive, visual interfaces, which are known to facilitate communication of results and their uncertainties, as well as help elicit user input (Voinov et al., 2016). Given decision-makers’ need to quickly explore interventions and their consequences across space and time, these model limitations can exacerbate the knowledge-practice gap, a common challenge in modeling wherein model insights do not directly inform actionable on-the-ground decisions (Voinov et al., 2016; Cunniffe et al., 2015).

Outside of a participatory modeling context, interactive modeling has been studied in computer science and related disciplines for several decades (McCormick et al., 1987). Computational steering refers to a mechanism for interactively controlling the variables of a simulation as the computation is in progress, and is often used to better understand parameter space and simulation behavior (Mulder et al., 1999; Matkovic et al., 2008). In addition to efficiency, computational steering also improves communication and discussion by providing immediate visual representation of the model and results (Van Wijk et al., 1997). Computational steering has been used to advance research in a variety of fields, including atmospheric and weather science, physics, and medical research dynamics (Jean et al., 1995; Walker et al., 2007; Johnson and Parker, 1995) and has proved especially important in computational fluid dynamics simulations (Marshall et al., 1990; Wright and Hargreaves, 2013). Additionally, certain agent-based modeling frameworks provide a form of computational steering for model exploration (Rossiter, 2015; Cordasco et al., 2013) or simulation coupling (Jaxa-Rozen et al., 2019).

Steering can open up new possibilities to explore geospatial ”what if?” questions collaboratively with stakeholders. Although the term “steering” can be used in participatory modeling literature to mean interactive adjustments of key input model variables (Niño-Ruiz et al., 2013; Voinov et al., 2016), we are specifically concerned here with spatio-temporal steering, i.e., allowing users to spatially intervene at any step of the simulation. This type of steering can be critical for strategizing the management of dynamic systems. Computational steering is one of several possible implementations of spatio-temporal steering. Some researchers have demonstrated how, with the help of interactive environments, computational steering can help explore complex spatio-temporal decision-space; the prime example is World Lines (Waser et al., 2010; Ribičić et al., 2013), which combines computational steering of a flooding simulation with versatile, interactive scenario visualization. Waser et al. (2010) demonstrated the approach with a levee-breach scenario, exploring possible methods for closing the breach by simulating the strategic positioning of sandbags in different spatial configurations. Another example of what-if scenario modeling was presented by Afzal et al. (2011) in the context of infectious disease modeling. These authors developed a decision-support environment on top of a mathematical, epidemiological spread model to interactively evaluate scenarios with different mitigating measures Afzal et al. (2011).

Despite general agreement about the advantages of computationally steering simulations, this methodology is still the exception rather than the rule, especially outside of computer science (Pickles et al., 2005), because there are several barriers to its broader usage. One is the increased technological complexity of model implementation, leading to high code maintenance costs and possibly more error-prone code. Another is a lack of user-friendly interfaces that facilitate steering for users with different technical backgrounds. Furthermore, high-performance computing platforms typically associated with computational steering often lack the necessary visualization capabilities and interactivity. Technological advances, such as GPU computing, allowed researchers to make many simulations more interactive and accessible through desktop interfaces (Linxweiler et al., 2010; Afzal et al., 2011; Ko et al., 2014). However, the increased need to provide simulation steering capabilities to analysts and stakeholders has necessitated the use of web-based solutions (Deodhar et al., 2014; Shashidharan et al., 2017) and alternative interfaces offering more natural user interactions (e.g., virtual reality environments (Mulder et al., 1998; Wenisch et al., 2005) or touch-table and tangible interfaces (Mittelstädt et al., 2013; Tonini et al., 2017)).

Spatio-temporal steering also poses conceptual challenges when dealing with stochastic models. Given that there are multiple realizations of a simulation running at the same time for a stochastic model, it is not obvious which realization to use to make steering decisions. Visualizing several stochastic runs using an aggregate representation—such as a probability or an average of model results (Ribičić et al., 2013)—can inform users about the potential range of outcomes. However, real-world decisions are based on observations best represented as a single stochastic run. Applying steering to stochastic spatio-temporal simulations is therefore challenging to inform strategies used in adaptive management, which bases decisions on evaluation of past actions, current observations, and future forecasting.

We encountered these challenges when designing a participatory modeling workshop focused on the spread of an invasive forest disease, sudden oak death (SOD), in Oregon. SOD spread poses serious environmental and economic risks, but because treatments are costly at large scales, decision-makers must strategically target treatments across time and space (Cunniffe et al., 2016). During a prior participatory modeling workshop we conducted (Gaydos et al., 2019), stakeholders expressed the need to explore yearly treatment interventions, which led us to incorporate spatio-temporal steering into our modeling framework. In this paper, we detail how we overcame several challenges associated with steering a stochastic simulation and identify three conceptual approaches to spatio-temporal steering in a participatory modeling context. We present a novel adaptive management approach that better represents the realities of decision-making by allowing both reactive and proactive spatially-explicit interventions. We also suggest simpler, alternative ways to design steerable simulations that do not require the implementation of computational steering, to reduce associated technological complexity.

The paper is structured as follows: Section 2 identifies several conceptual and implementation challenges associated with steering of spatio-temporal simulations and develops methods to address them. In Section 3 we apply the methods in an epidemiological simulation and describe our steering implementation and interfaces developed for our participatory modeling case study. Using this case study, Section 4 demonstrates how workshop participants applied the novel adaptive management approach to interact with the simulation and develop relevant management scenarios. Sections 5 Discussion, 6 Conclusions highlight the importance of using the adaptive management approach during the workshop and discuss the limitations and future work.