In the Western United States and elsewhere, the growing damages from wildfires lead to calls for increased investment in hazardous fuel reduction to protect the wildland–urban interface (WUI), among other objectives. Limited resources and operational constraints, among other factors, limit the areas where treatments can be implemented. Therefore, there is a clear need for strategic prioritization based on information about what fire and fuel management scenarios are most likely to be effective.

However, the informational basis regarding fuel treatment effectiveness is mixed. Empirical analyses have found that treatments can be effective at reducing fire intensity and severity within treated areas, but evidence of their effectiveness at landscape scales is limited (Fernandes, 2015; Kalies and Kent, 2016; McKinney et al., 2022; Ott et al., 2023; Urza et al., 2023). This gap is critical as many existing or proposed treatments are not directly located in the WUI but rather in proximal wildlands, based on the premise that treatments will interrupt fire spread and reduce WUI exposure. Simulation modeling is widely used to fill that gap, showing how treatments can affect metrics such as landscape burn probability (Ott et al., 2023). However, in almost all cases, simulation modeling is used to evaluate hypothetical future treatments (e.g., Benali et al., 2021; Alcasena et al., 2022), with few examples that analyze how prior treatment interactions with wildfires may have altered landscape outcomes (e.g., Cheney, 2010; Cochrane et al., 2012). As a result, it can be difficult to highlight successful risk reduction interventions and quantify the historical return on investment.

In this study, motivated by research in disaster risk reduction and climate science, we outline a potential pathway for estimating avoided wildfire impacts that blends counterfactual probabilistic modeling with extreme event attribution. Counterfactual reasoning allows events that were not realized to be examined. Large wildfires have numerous risk factors due to both climate change and on-the-ground fuel conditions. Understanding whether a wildfire would or would not have exceeded thresholds for extremes based on these antecedents can help examine the influence of key variables on wildfire outcomes. Counterfactual analyses take into consideration the probability of an event (i.e., a fire of a particular size) occurring in an observed world and an imaginary world where causes are present or absent, unlike the real one (Lott and Stott, 2016).

One central idea is that modeling unrealized events in the absence of mitigation interventions can shed light on the often hidden benefits of risk reduction programs (Lin et al., 2020; Rabonza et al., 2022); that is, successful interventions can be made invisible by their very success, which could reduce awareness or continued investment. Importantly, this shifts the baseline for evaluation from realized outcomes to averted outcomes and characterizes management efficacy in risk-based terms.

Another central idea is that event attribution logic can provide a common language and framework for probabilistic analysis of counterfactual events. Typical applications examine the influence of climate change in extreme weather events, where the counterfactual scenarios experience less severe events due to the absence of climate forcing (Hannart et al., 2016; Otto, 2017; Naveau et al., 2020). Several studies have used these methods to attribute growing wildfire hazards to climate change (Kirchmeier-Young et al., 2017, 2019; Tan et al., 2018; Van Oldenborgh et al., 2021). In this study, we modify the event attribution framework to consider counterfactual scenarios without risk reduction interventions where the unrealized wildfire outcomes could have been more severe, as well as hypothetical scenarios where the current landscape conditions exhibit exacerbated hazards due to a legacy of fire exclusion and forest management practices and could experience more severe outcomes relative to treated conditions.

To demonstrate the key concepts, we reanalyze previously published results from two simulation studies that modeled large fire growth on landscapes in New Mexico, USA (Thompson et al., 2016, 2022). These studies address salient fire and fuel management challenges in fire-prone forests, simulate fire on counterfactual untreated and hypothetically treated landscapes, and reanalyze modeling results in a stylized manner for illustrative purposes. Note that all relevant fire size data are either provided in this study or are available in online data archives (Vogler et al., 2022). We first review the basic framework, then present applications of landscape scenario evaluation for post-event counterfactual analysis and pre-event hypothetical analysis, and finally, discuss ideas for future expansions in theory and methods.

The major contribution of this study is to introduce counterfactual probabilistic analysis and event attribution metrics for quantifying avoided outcomes and analyzing the historical performance of fuel treatments. The absence of such frameworks can perpetuate uncertainty about the return on investment from risk mitigation and lead to a reliance on anecdotes over quantitative metrics. Moreover, examination of the probabilities individually without a FAR or RR framework can lead to cognitive errors when interpreting the risks and benefits of fuel treatments. We illustrated how these event attribution metrics can be complementary to other simulation-based approaches to compare hypothetical treatment scenarios as well. One key reason to introduce the framework for prospective evaluation is to have a common language and logic for analyzing post-event and pre-event modeling exercises. In this process, attribution analysis should be conducive to localized hazards and their use in wildfire assessments.

Useful and interpretable explanatory tools for examining model outcomes can be derived from a counterfactual analysis (Kenny and Keane, 2021). The FAR and RR provide metrics for comparing the probabilities of events with or without antecedents. This was demonstrated in the interpretation and comparison of real-world and simulated outcomes for the two case studies here. Our reanalysis yielded FAR and RR values, suggesting a significantly higher potential for extreme fire growth on both actual and counterfactual untreated landscapes. Other case studies have used the RR and FAR to examine the influence of climate change variables on outcomes of interest, such as wildfires and heat waves. For example, two case studies in China used counterfactual RR analyses to clearly identify a more than doubling in probability for the occurrence of a heavy precipitation event like a record-setting one in 2017 (Sun et al., 2019) and heat waves as intense as the maximum observed index values in multiple regions from anthropogenic climate change influences (Sun et al., 2017). Although both the FAR and RR provide highly interpretable metrics in these and other studies, they may still be supplemented with other communication approaches that rely on more qualitative, context-relevant interpretations for communicating risks to varied audiences (Lloyd and Oreskes, 2018). In turn, these communication approaches, such as storylines and stakeholder co-developed modeling approaches, can assist in iteratively building attribution studies that are well-defined and useful in terms of potential risk factors and events.

The approach taken with the examination of RR and FAR across variable percentiles of fire size also illustrates how interpretations can be affected by how a fire size threshold is defined. Event definition is a major influence on the assessment of attribution (Lott and Stott, 2016). Single thresholds may provide a clear delineation but can be misleading if assessed alone. Large changes in the risk statistics were observed across distinct zones of fire sizes spanning tens of thousands of hectares, and much smaller changes were observed across much larger spans of fire sizes. Thus, understanding the implications across the full range of fire sizes would be useful in future applications. Hannart et al. (2016) also note the importance of the definition of the events to the outcome of counterfactual analyses for heat waves and climate change attribution.

The current case study assumed random treatments for demonstration purposes, but this may not be useful for all applications. Counterfactual analyses can also be useful for more complex modeling contexts. Future methodological explorations may include additional causal considerations in an attribution framework. Structural causal models help facilitate the identification and use of variables that can confound causal interpretations of risk metrics. Confounders are variables that causally influence both the impact and the treatment (Pearl and Mackenzie, 2018). In wildfire treatment, this may include variables related to fuel buildup that would causally influence the treatment locations and the wildfire magnitude. The depth of the relationships can also be expanded to account for cascading impacts through a multi-step attribution framework (Burger et al., 2020). The current study explored a single-step attribution framework that linked treatments to wildfires. However, a multi-step attribution framework may be useful for many applications (Burger et al., 2020). This would facilitate the evaluation of additional causal relationships between factors such as atmospheric climate change, fuels, and wildfires, along with resulting impacts on costs and benefits to human health, society, and the environment. There is often a causal web of influences that may need to be accounted for to understand the impacts of fuel treatments on wildfire outcomes. The development of structural causal models can provide the conceptual and quantitative basis useful for evaluating confounders in a multi-step attribution framework (Pearl and Mackenzie, 2018), necessitating a deeper examination of their potential in future research.

Future research may also examine how representing causal relationships can improve the interpretation and evaluation of counterfactuals. In addition to the metrics examined in this study, counterfactual analyses permit more detailed examinations of causality through probabilities of necessary, sufficient, and necessary and sufficient causation. The FAR is equated with necessary causation under certain assessment conditions (Hannart et al., 2016). However, utilizing the necessary causation over the FAR provides a more in-depth interpretation of the causal relationships between variables. Sufficient causality and necessary and sufficient causality, which were not addressed in this study, can capture additional conditions for causality not recognized by the FAR and RR metrics. Coupling attribution analysis calculations with causal Bayesian networks will allow greater complexity in metrics and compound events to be calculated (Carriger et al., 2021). In a causal framework, the counterfactuals provide a means for looking at how changing causal antecedents could have influenced or changed the observed effects and provide actionable insights.

Beyond advances in theory on the horizon, there are practical modeling considerations to address as well. We reanalyzed existing published data to illustrate the key points using fire size as our basis of comparison, but there is a clear need to expand the framework beyond area burned to other fire impacts. Furthermore, existing models may poorly capture the efficacy of fire suppression operations or how they are enhanced by landscape fuel management (Plucinski, 2019), which gets to the broader point that all modeling exercises are subject to issues of data quality and uncertainty. Caution is warranted when aiming to credibly recreate past fire events, and attention to detail on data quality control is essential. With that said, advances in remote sensing, geospatial, and computing technology provide greater opportunities for more refined analysis compared to what was performed for the Las Conchas Fire. Furthermore, given that agencies such as the USDA Forest Service prioritize investments in the order of billions of USD based on simulation models and hypothetical treatment scenarios, it stands to reason that the same family of models could be used to support analysis of the benefits of those investments in terms of outcomes from real interactions with real fires. We hope this article catalyzes additional interest and research into quantifying avoided wildfire impacts and revealing often unseen mitigation effectiveness.

The original contributions presented in the study are included in the article/supplementary material, further inquiries can be directed to the corresponding author.

MT: Conceptualization, Data curation, Methodology, Writing—original draft, Writing—review & editing. JC: Conceptualization, Methodology, Writing—review & editing.