Are the applications of wildland fire behaviour models getting ahead of their evaluation again?

Abstract

Evaluation is a crucial component for model credibility and acceptance by researchers and resource managers. The nature and characteristics of free-burning wildland fires pose challenges to acquiring the kind of quality data necessary for adequate fire behaviour model evaluation. As a result, in some circles it has led to a research culture that tends to avoid evaluating model performance. Operational fire modelling systems commonly used in western North America have been shown to exhibit an underprediction bias when employed to determine the threshold conditions necessary for the onset of crowning and the associated spread rate of active crown fires in conifer forest stands. This pronouncement was made a few years ago after at least a decade of model misapplication in fire and fuel management simulation modelling stemming from a lack of model evaluation. There are signs that the same situation may be repeated with developing physics-based models that simulate potential wildland fire behaviour; these models have as yet undergone limited testing against observations garnered from planned and/or accidental wildland fires. We propose a broad co-operative project encompassing modellers and experimentalists is needed to define and acquire the benchmark fire behaviour data required for model calibration and evaluation.

Introduction

Wildland fire behaviour is principally concerned with the ignition or inception of a free-burning fire, its spread rate, energy released and associated flame front dimensions, perimeter and area growth, and related phenomena such as torching, crowning, spotting and firewhirl activity (Fig. 1). In recent years, models, modelling systems, and simulation modelling have come to gradually dominate scientific curiosity and practical application in predicting wildland fire behaviour. This has been a process greatly aided more by advances in computer technology (Andrews and Queen, 2003) than by breakthroughs in our understanding of fire dynamics.

Wildland fire behaviour models are typically distinguished into (1) empirical or semi-empirical and (2) physical models (Sullivan, 2009a, Sullivan, 2009b). Empirical or semi-empirical-based models aim to support an operational decision-making process whereas physics-based models are primarily developed with theoretical purposes in mind, aiming to better understand the physical and chemical processes controlling fire propagation.

The use of models has played an integral role in natural resource management decision-making. But model adoption and application should be preceded by an evaluation protocol as well as verification that demonstrates model outcomes represent the processes they aim to describe within acceptable error bounds (Johnson et al., 1985, Randall et al., 2007). The evaluation process is a component of judicious development of environmental models, particularly those aimed at supporting a decision-making process (Jakeman et al., 2006, Robson et al., 2008).

Model evaluation is of critical importance in the research associated with natural hazards phenomena where the use of models or the interpretation of their outcomes has direct implications for the safety of the general public as well as fire and emergency services staff. The court process and conviction of seven experts who provided public advice ahead of the 2009 L'Aquila earthquake in central Italy (Cartlidge, 2012) highlights the issues scientists and model users can be confronted with in the era of modern accountability (Eburn and Dovers, 2012).

We contend that the focus of operationally oriented models used in simulating surface and crown fire behaviour during the past 10–15 years has been misguided (Cruz and Alexander, 2010). Further, we suggest that there is the potential for the application of physics-based fire behaviour models to proceed ahead of the needed model evaluation. In this paper we outline the basis for our déjà vu – like revelation. We also recommend ways to improve the situation. However, it is not our purpose here to provide a detailed evaluation protocol per Jakeman et al. (2006) and Blocken and Gualtieri (2012) for wildland fire behaviour models.