Evaluation of the predictive capacity of dead fuel moisture models for Eastern Australia grasslands Miguel G. Cruz A,D , Susan Kidnie B , Stuart Matthews C , Richard J. Hurley A , Alen Slijepcevic B , David Nichols B and Jim S. Gould A A CSIRO, GPO Box 1700, Canberra, ACT 2601, Australia. B Country Fire Authority (CFA), Fire and Emergency Management, PO Box 701, Mount Waverley, Vic. 3149, Australia. C NSW Rural Fire Service, 15 Carter Street, Lidcombe, NSW 2141, Australia. D Corresponding author. Email: miguel.cruz@csiro.au Abstract. The moisture content of dead grass fuels is an important input to grassland fire behaviour prediction models. We used standing dead grass moisture observations collected within a large latitudinal spectrum in Eastern Australia to evaluate the predictive capacity of six different fuel moisture prediction models. The best-performing models, which ranged from a simple empirical formulation to a physically based process model, yield mean absolute errors of 2.0% moisture content, corresponding to a 25–30% mean absolute percentage error. These models tended to slightly underpredict the moisture content observations. The results have important implications for the authenticity of fire danger rating and operational fire behaviour prediction, which form the basis of community information and warnings, such as evacuation notices, in Australia. Received 2 March 2016, accepted 17 May 2016, published online 5 July 2016 Introduction The moisture content of fuels determines the energy require- ments for their ignition, hence exerting a strong effect on fuel availability and fire behaviour characteristics such as fire sustainability, spread rate and intensity. Dead fuel moisture content, or its surrogates, is a primary input of fire behaviour prediction systems such as BehavePlus (Andrews 2014), Farsite (Finney 2004) and Prometheus (Tymstra et al. 2010). Being able to correctly estimate dead fuel moisture, as it varies spatially and temporally, is key to the accurate prediction of landscape-scale fire spread (Sullivan and Matthews 2013). In Australia, these predictions form the base of emergency recommendations and warnings, such as evacuation orders. Further operational and research applications of dead fuel moisture predictions encompass estimating fire danger rating, prescribed burn plan- ning and climate change studies (Sullivan 2010; King et al. 2012; Matthews et al. 2012). Grasses have several features that differentiate them from litter or fine woody fuels. Grass fuels are finer than other common wildfire fuels, resulting in faster moisture response times to changes in the weather (Van Wagner 1972; Viney and Hatton 1989). Typical grasslands have little to no overstorey canopy cover, enabling full fuel exposure to the drying effects of solar radiation and wind. Furthermore, the bulk of the fuel is standing in the grass sward, limiting the effect of rainfall and soil moisture content in increasing dead fuel moisture (Viney 1991; Matthews 2014). The combination of these features results in grass fuels being typically drier than other wildfire fine fuels such as forest litter (e.g. Viney and Hatton 1989; Sullivan and Matthews 2013). In Australia, several tools have allowed the estimation of dead grass fuel moisture content in a typical south-eastern Australian summer (McArthur 1960, 1966, 1977). Noble et al. (1980) used data extracted from the McArthur (1977) MK 5 slide rule to develop an equation describing the overall moisture content of the grass sward from air temperature, relative humidity and curing level. For fully cured grasses, the moisture content is only dependent on the first two variables. Cheney et al. (1989) parameterised another equation from tabular data in McArthur (1960). Both these models are used to estimate dead grass fuel moisture content operationally in Australia (Cruz et al. 2015), although their predictive capacity has not yet been evaluated. There are several other models used to quantify fuel moisture in grassland fuels elsewhere in the world that could have applicability for the Australian environment, namely models within the National Fire Danger Rating System in the US (Deeming et al. 1977; NFDRS) and the Canadian Forest Fire Danger Rating System (CFFDRS; Van Wagner 1987). Models described in Matthews (2006) and Wotton (2009) have also potential to be used in Australia to predict the moisture content of dead grass fuels. The main objective of the present paper is to test the predictive capacity of dead fuel moisture content models for CSIRO PUBLISHING International Journal of Wildland Fire 2016, 25, 995–1001 http://dx.doi.org/10.1071/WF16036 Journal compilation Ó IAWF 2016 www.publish.csiro.au/journals/ijwf Research Note