Shrublands are typically composed of a variable live and dead fuel mixture, and thus adequate to the present analysis. Anderson et al. (2015) developed an empirical R model from a comprehensive set of 79 field experiments of fire spread in shrublands worldwide. The M-effect was modelled as an exponential decay based only on dead fuels, given by exp(-0.0721 M _d). To obtain a measure of how the established important influence of wind speed ( U) on R might hinder a proper field evaluation of the M-effect, we assessed the relative effects of U and M _d using daily R simulations for a two-year series (2015–2016) of meteorological observations recorded at 12:00 in Lousã, Central Portugal. M _d was estimated from the Fine Fuel Moisture Code (FFMC) of the Canadian Forest Fire Weather Index System (Van Wagner, 1987). We did not consider days with M _d >35% (because of unlikely fire spread) and used a bulk density of 1.8 kg m ^-3 (mean experimental value). U and M _d relative contributions to R were assessed using classification and regression trees (CART) analysis.

Rossa & Fernandes (2017) obtained an M–damping effect based on total fuel bed water content (live and dead fuels), given by M _w ^-0.6253, using 51 fire spread laboratory tests in fuel beds composed of a litter layer (dead foliage) over–layered by vertically oriented quasi–live fuels, thus approaching the natural fuel structure. M _w can be obtained from:

We compared the laboratory and field-based M–damping effects for fire spread in a mixed Calluna vulgaris (L.) Hull and Pterospartum tridentatum (L.) Wilk. fuel complex, common in N–C Portugal shrubland, for the typical M–evolution throughout the year. We used monthly averages from a long term (1996-2012) M–assessment (Lopes, 2013) of a few selected fuel species, conducted in Lousã. M _d was obtained as the mean value between Pinus pinaster Ait. and Eucalyptus globulus Labill. litter, and M _l as the mean between live C. vulgaris and P. tridentatum, which assumes an equally distributed mixture. M _w was computed for ƒ _d of, respectively, 0.2, 0.4, and 0.6 – i.e., 20, 40, and 60% of fine dead fuels - and we assessed the significance ( p<0.05) and strength ( R ²) of the linear relationship between the two M–damping effects. The range in f _d for Anderson et al. (2015) model development was 0.14–0.91 (mean was 0.49), thus confirming that the selected values of ƒ _d for analysis are realistic.

The U effect dominated over M _d, respectively explaining 61.9 and 38.1% of the variance in R predictions. The prevalence of such effect would probably be higher had we resorted to simulations based on hourly weather data, because of higher variation in U (Beck & Trevitt, 1989). In the year-round experimentally–assessed M evolution, M _d and M _l varied within 12.8–52.6% and 76.0–108.2% (Fig. 1). The linear relationship between laboratory and field-based M-damping effects (Fig. 2) approached significance for ƒ _d=0.2 ( p=0.06) and was highly significant for ƒ _d=0.4 ( p=0.004) and ƒ _d=0.6 ( p<0.0001). Likewise, R ² increased markedly from 0.318 to 0.813 with the rise of ƒ _d.

Laboratory testing allows the control and/or accurate monitoring of the main parameters influencing fire propagation. Rossa (2017) used data from 185 burns under windless conditions in the absence of slope, covering a wide diversity of fuel bed composition, arrangement and M (6-179%) conditions, showing that the M-effect on R does not depend on vegetation condition (live or dead). Extension of the results to real-world fires was confirmed by model validation against field fires. In the case of slope ( e.g., Rossa et al., 2016) or wind-driven ( e.g., Marino et al., 2012; Rossa & Fernandes, 2017) laboratory trials, R is limited by the fire front width (Fernandes et al., 2009). Still, these trials are representative of the initial stage of a point ignition field fire and the M-damping effect, as well as the independence from plant phenology, are expected to be independent of scale and hold under field conditions (Rossa et al., 2016).

Outdoors experimental fires occur in real–world conditions. No fire behaviour model can be completely proven correct until it faces validation against field fires. Their use as a source of development data is also appealing because there are many fire–spread situations that cannot be reproduced in the laboratory without serious scaling limitations. Yet, this option is challenged by lack of control over environmental parameters, heterogeneity in fuel bed properties, and correlated fuel descriptors. These shortcomings have been fostering the dispute over the M _l influence on R, not clarifying if it is eluded by the difficulty in detecting specific effects, or if it is diluted by field-specific spread mechanisms. The latter remains to be verified only for the case of slope and wind-driven fires, since for no–wind or no–slope burns – which also approach backing fires (Rossa et al., 2015) –, the accuracy of M _w–based R predictions was already tested against field fires, confirming the M _l influence (Rossa, 2017).

M _l will be approximately constant during a field experimental program in shrubland fuels conducted over a period of a few weeks, contrasting with M _d that will vary substantially in response to changes in air temperature, relative humidity and solar radiation (Anderson et al., 2015). Other types of live vegetation, like evergreen mature tree foliage without severe soil water deficit, will even usually maintain an approximately constant M _l over the year (Pook & Gill, 1993; Agee et al., 2002). However, for the shrubs addressed by this study, the difference between minimum and maximum M _l was 32.1% for average monthly means (Fig. 1), a value similar to that observed for M _d (39.8%). Nevertheless, because of similar seasonal trends for M _l and M _d, which are evident in Fig. 1, M _d and M _w-based damping effects are correlated (Fig. 2).

Although the M _l evolution in Fig. 1 is not necessarily representative of a general shrubland, several year-round Mediterranean M _l assessments for a great number of shrub species (Piñol et al., 1998; Pellizzaro et al., 2007; Yebra et al., 2008; Yebra & Chuvieco, 2009), report a similar seasonal pattern, with a M _l decrease between spring and summer, followed by an increase until winter. Because dry soils are frequently concurrent with meteorological conditions that yield low M _d values as well, the relationship between M _l and soil moisture content found by Qi et al. (2012) for a shrubland in Northern Utah (USA), also lends support to our hypothesis of correlated M _d and M _w damping effects.

Both cases of: (i) a year–round roughly constant M _l, or (ii) correlated M _d- M _w variations, explain why the M _l-effect is difficult to detect from the statistical analysis of field data, although laboratory studies systematically show that R is influenced by the total amount of water in the fuel complex, independently of vegetative condition. Dominance of R by U also adds to the challenge of properly assessing the M-effect based on field fires. The good agreement between the M-damping effects of field and laboratory-based R models also enlightens why the sole use of M _d is enough to provide a satisfactory R explanation in most fuel-dependent models ( e.g., Cruz et al., 2015). Differences between the combustion mechanisms of live and dead fuels also do not necessarily lead to different R, because if all water is evaporated during the passage of the flame (before or after ignition) the net heat release sustaining fire–spread should approximately be the same (Rossa, 2017).