1 INTRODUCTION

Coral reefs are one of the most biodiverse and valuable ecosystem types in the world (Adey, 2000; Anthony et al., 2017) but are also among the most threatened (Bellwood et al., 2019; Burke et al., 2011; Halpern et al., 2015; Hoegh-Guldberg et al., 2017) due to climate-driven and local pressures increasing in magnitude and frequency (Anthony et al., 2017; Beyer et al., 2018; Costa et al., 2020). Effective assessments are critical to making informed decisions about reef management and conservation (Dixon et al., 2021; Obura et al., 2019; Pressey et al., 2017). A clear priority is identifying which reefs are likely to be more resilient to disturbances and therefore retain their biodiversity, ecosystem function and values (Darling & Côté, 2018; Hock et al., 2017; Macharia et al., 2016).

Resilience-based management takes a dynamic and adaptive perspective to enhancing natural processes that promote resilience and can include social and ecological dimensions (Gibbs & West, 2019; Harvey et al., 2018; Mcleod et al., 2019; Nyström et al., 2008). Ecological resilience is the capacity of an ecosystem to maintain or recover state, functioning and structure following a disturbance (see Box 1). Despite the potential of resilience-based management, it remains underused in reef management and conservation planning particularly in developing countries (Bates et al., 2019; Roche et al., 2018), in part due to difficulties operationalising the concept of resilience (Angeler & Allen, 2016; Maynard et al., 2010; Mumby et al., 2014). A clear definition of ecological resilience and the indicators to quantify it are necessary foundations for operationalisation in management (Box 1), but their interpretation and implementation vary globally (Lam et al., 2017; Maynard et al., 2017; Standish et al., 2014).

Early work to include resilience in management for example by Obura (2005), Obura et al. (2006), Salm et al. (2001), and West and Salm (2003) culminated in the first formalised method to assess the ecological resilience ‘potential’ of coral reefs, based on 61 resilience indicators (Obura & Grimsditch, 2009). Resilience potential assessments have since become the most widely used method for operationalising coral reef resilience globally (McLeod et al., 2021). The assessments can be used as decision-support tools providing reef managers with locally contextualised information to prioritise areas and actions for protection (Bachtiar et al., 2019; Ladd & Collado-Vides, 2013; McLeod et al., 2021).

Science for environmental decision-making and indicator design has developed over the last two decades (Bundy et al., 2019), concurrent with developments in resilience science (Angeler & Allen, 2016; Cumming et al., 2017; Roff & Mumby, 2012). This body of theory could strengthen the capacity of resilience potential assessments and other indices of ecosystem condition and integrity (Grantham et al., 2020; Hill et al., 2022; Karr, 1981) to support management decisions. Multiple frameworks have been developed for designing, selecting, and evaluating indicators, including applications of decision theory (Possingham et al., 2001; Watermeyer et al., 2021), conceptual frameworks and formal selection criteria (Brown et al., 2021; Bundy et al., 2019; Failing & Gregory, 2003; Keith et al., 2013), and quantitative performance testing (Collen & Nicholson, 2014; Nicholson et al., 2012). Key considerations for indicator construction that emerge from these frameworks include having a clear definition of indicator objectives, consideration of uncertainty, transparency about methods for aggregating indicators into composites, selecting indicators based on explicit criteria and evidence, and testing indicator performance to understand their behaviour and predictive capacity. For instance, conceptual models that describe the key features, relationships and dynamics of a system can provide a foundation for indicator selection, ensuring representation of key components and reducing redundancy in indicator suites (Keith et al., 2013). This is particularly important given the push to use fewer variables mainly for practical reasons (Bang et al., 2021; Maynard et al., 2012; McClanahan et al., 2012).