Bayesian decision network modeling for environmental risk management: A wildfire case study

Highlights

• Landscape management requires tools to cope with complex decisions.

• Bayesian Decision Networks (BDNS) were used to examine trade-offs in fire management.

• Least cost options were identified for decisions considering across multiple assets.

• BDNs are effective tools for multi-criteria decision analysis of environmental management.

Abstract

Environmental decision-making requires an understanding of complex interacting systems across scales of space and time. A range of statistical methods, evaluation frameworks and modeling approaches have been applied for conducting structured environmental decision-making under uncertainty. Bayesian Decision Networks (BDNs) are a useful construct for addressing uncertainties in environmental decision-making. In this paper, we apply a BDN to decisions regarding fire management to evaluate the general efficacy and utility of the approach in resource and environmental decision-making. The study was undertaken in south-eastern Australia to examine decisions about prescribed burning rates and locations based on treatment and impact costs. Least-cost solutions were identified but are unlikely to be socially acceptable or practical within existing resources; however, the statistical approach allowed for the identification of alternative, more practical solutions. BDNs provided a transparent and effective method for a multi-criteria decision analysis of environmental management problems.

Introduction

Effective environmental decision-making requires an understanding of complex interacting systems across scales of space and time. Managers often are required to make decisions in the face of high uncertainty (Fackler and Pacifici, 2014; Thompson and Calkin, 2011) with limited budgets and multiple competing interests. There is increasing public pressure for environmental management agencies to quantify costs and benefits of decisions, yet there is little guidance on the best methods to achieve this.

A number of statistical methods, evaluation frameworks and modeling approaches have been applied for conducting structured environmental decision-making under uncertainty (e.g. Gregory et al., 2012; e.g. Soltani et al., 2017; Williams and Hooten, 2016). Environmental managers need to ensure that the method they adopt incorporates the utility or cost of the action and the impact of actions, and that a decision-advisory model captures the complexity of the system without losing predictive capacity. Adkison (2009) demonstrated that managers implementing simpler decision models outperformed those using overly complex decision models. It is therefore a delicate balance to achieve the necessary level of model simplicity without compromising the predictive capability and key interactions of the model.

Wildfire management is an area typically wrought with uncertainties. These uncertainties stem from the effects of both the wildfire and fire management on biological, social, cultural and economic values (Ager et al., 2015; Finney, 2005; Roloff et al., 2012; Tedim et al., 2016). Decisions made now need to account for the shifting fire regimes that we are already experiencing (Nolan et al., 2020) and predicted future regimes (Brown et al., 2004; Westerling and Bryant, 2008). Fuel treatments are used throughout the world to alter fuel loads in an attempt to reduce future fire impacts (Fernandes and Botelho, 2003). One of the more controversial approaches is prescribed fire (Penman et al., 2020) which is mainly used to protect people, property and infrastructure. Evidence suggests that prescribed burning regimes designed to reduce risk to people and property generally increase the extent of fire in the landscape (King et al., 2006; Price et al., 2015a) and a change in fire season (Penman et al., 2011a) which can impact negatively on environmental assets, such as biodiversity, water and carbon (Bradstock et al., 2012a; Fernández et al., 2006; Ooi et al., 2006).

Various applications of decision-science methods and tools have been developed for wildfire management. For example, Dunn et al. (2017) suggested a decision-support framework for large-fire management that includes consideration for financial, social, and ecological factors. Daniel et al. (2017) developed a stochastic, spatially-explicit state-and-transition simulation model for forest management planning that addresses timber harvest, wildfire, and climate change. Approaches to balancing tradeoffs among fire risk, management of fire-prone vegetation, and social costs in structured decision-theory frameworks have been suggested by Dunn et al. (2017), Roloff et al. (2012), Daniel et al. (2017) and others.

One construct that is useful for addressing uncertainties in environmental decision-making that can provide an intuitive and relatively simple structure is that of Bayesian decision networks (BDNs). Bayesian networks (BNs) are statistical tools that are ideal for risk analysis of complex environmental systems (Johnson et al., 2010; Kelly et al., 2013; Pollino et al., 2007; Sierra et al., 2018). BDNs are extensions of BNs that explicitly include decision structures and utility costs or benefits of those decisions weighted by outcome probabilities. BDNs have been successfully used in management of privately-owned forests (Ferguson et al., 2015), to assess adaptation strategy responses to sea-level rise (Catenacci and Giupponi, 2013), and other applications.

BNs have previously been applied to various aspects of fire management including modeling wildfire behaviour (Dlamini, 2010; Hanea et al., 2012; Penman et al., 2011b), response of vegetation (Liedloff (Liedloff and Smith, 2010), effects on wildlife (Hradsky et al., 2017), and impact on people and property (Cirulis et al., 2019; Papakosta and Straub, 2011; Penman et al., 2015a). However, the application of BDNs to explicitly evaluate expected values of wildfire management costs has seldom been applied to wildfire management.

In this paper, we apply a BDN approach to decision making around wildfire management. We present a real-world case of using BDNs to support prescribed burning decision management in southeast Australia, and then expand our findings to a broader application context. In doing so, we ask what is the value, general efficacy and utility of the BDN approach for further use in resource and environmental decision-making.