The study was conducted in the mixed-conifer forests of Sequoia and Kings Canyon National Parks, California, USA (36°36′N, 118°42′W), between 1600 and 2400 m elevation. Leaf litter material was collected during summer (mid-June to mid-July) 2010 from eight tree species: Pinus jeffreyi Grev. & Balf., P. lambertiana Dougl., P. ponderosa Dougl. ex Laws., Abies concolor (Gord. & Glend.) Lindl. ex Hildebr., A. magnifica A. Murr., Calocedrus decurrens (Torr.) Florin, Quercus kelloggii Newb. and Sequoiadendron giganteum (Lindl.) J. Buchholz. These species were chosen because they are representative of the mixed-conifer forest in the parks and constitute the dominant overstorey species found across the elevational range of the study area. In this study, leaf litter refers to the superficial layer of the forest floor composed of mostly undecomposed leaves from this year’s and the previous year’s leaf fall, and small twigs <0.625 cm (1-h fuels, Pyne, Andrews & Laven 1996). We collected from 21 different sites across the parks, with a minimum of four sites (=populations) per species, to capture potential spatial variation in litter traits within a species (see Table S1 in Supporting information). Due to logistical constraints and occurrence, A. magnifica litter was collected from only one location but collection still included variation across 10 individuals. At each site, litter was collected from 2–4 individual trees at least 10 m apart. To obtain more uniform fuel samples representative of the forest floor, we gathered litter at c. 2 m from the individual tree chosen to avoid sampling bark and twigs, which tend to fall closer to the trunk. This was performed because we were interested in assessing the flammability of uniform 1-h fuels. Bark and twigs are denser and have higher residence time, which changes flammability. After being brought back to the lab, the material was dried for 48 h at 60 °C to bring the relative humidity to levels <5% for all samples. This allows us to control for different moisture retention capabilities of the different species, which would become a confounding factor in the analysis.
The flammability tests were performed both on monospecific litter beds and in litter beds composed of mixtures of litter from three different species at a time (each species accounting for a third of the total mass). The eight species were used in 55 litter arrangements (eight single-species tests and 47 different combinations of three species, see Table S2), replicated 3–5 times, bringing the total number of experimental trials to 219. The number of combinations is lower than the total possible with all 56 possible mixtures. Unfortunately, due to limitations in the amount of material brought from the collection sites, it was impossible to burn five replicates of all mixtures. To encompass as much variability as possible, we decided to include the mixtures that offered the greatest range of leaf sizes in the mixture, balancing naturally existing and non-occurring mixtures.
The burn table was built to approximate a one-dimensional cross-section of a flaming front through a litter bed, with a minimum length that would allow steady-state flame spread and a minimum depth and width that permitted ‘natural’ fuel arrangement, thus creating a bulk density similar to that observed in the field. If the table was built too narrowly, the larger leaves would arrange themselves along the trough, an arrangement that would not be expected in the field. Nevertheless, these are recreated litter beds, and we did not expect to obtain a replica of a natural litter bed. We conducted a preliminary set of experiments with tables of different widths to determine the minimum width that did not influence the random arrangement of leaf litter by constraining leaf orientation. As a result, the width was set at 15 cm (results not shown). The litter for burn trials was placed in a 15 × 150 cm channel made of 0.3-mm steel sheeting 12 cm high, and it was closed at both ends with 6.5-mm grid wire cloth to hold the material in and still allow for ventilation and access for the ignition source. The thin steel sheeting was backed by 7-mm-thick ceramic insulation (McMaster-Carr, Los Angeles, CA, USA) surrounded by a wood frame. Ignition was provided by a propane torch, which can achieve a maximum adiabatic flame temperature of 1899 °C (http://www.bernzomatic.com/products/fuel.aspx). We visually assessed maximum flame height using two rulers positioned at 50 and 100 cm from the ignition end of the table. Temperature (in °C) was measured with K-type thermocouples connected to HOBO U12 dataloggers (Onset Corporation, Cape Cod, MA, USA) and placed at 50, 100 and 150 cm along the table. At each position, we placed three thermocouples, one at the bottom of the litter bed, one on its surface and one at 25 cm above the surface. In the data presented here, we chose to only consider the bottom thermocouple, as is the one that can give a more stable measurement of temperature, as well as being the one that better correlates with soil heating. Given the high correlation of temperature measurements at 50 and 100 cm, we chose to average the values and use 75 cm as our reference. We established 100 °C as our temperature threshold since at this temperature we have already reached cellular death, and it is the temperature at which water evaporates. Temperature was recorded every second during each trial. For each burn trial, the height of the litter bed was measured at four points along the table and its average taken to calculate bed depth and from that litter bulk density.
We standardized the samples by mass, with each trial set at 450 g of leaf litter, which provides a large range of variation in litter depth, from 3 to 11 cm, mimicking the natural variation in the field. All weight measures were taken using a balance sensitive to 0.1 g (model XS16001L; Mettler Toledo, Columbus, OH, USA).
We determined such flammability parameters as ignitability, calculated as the time from moment of exposure to a heat source to production of flame; combustion time (sustainability), measured as the time flame was visible from ignition until fire extinction; spread rate, calculated as the ratio between the length of the burned surface and the residence time; maximal flame height (combustibility), calculated by means of marks on two stainless steel metre rulers, positioned at 50 and 100 cm along the table, taken as reference; and percentage of mass loss (consumability), calculated as the difference between fuel mass before and after flame extinction (Anderson 1970; Cornelissen et al. 2003; Ormeño et al. 2009). We measured duration above 100 °C, which is the length of time the fire burned hotter than 100 °C, temperature above 100 °C, which is the average temperature the fire burned at after it reached 100 °C, and temperature integration, which integrates duration of combustion and temperature above 100 °C and serves as a proxy for heat release.
A stopwatch was used to measure (i) time to ignition (ignitability, in s), which is the time it takes for the fuel to catch fire once exposed to the propane torch; (ii) time for the flaming front to reach the end of the table (rate of spread, in cm s−1); and (iii) time until flames are extinguished (which, added to the previous, accounts for sustainability, in s).
The burn trials took place in a cement structure used to simulate house fires at the Fire Department of the City of Lubbock. This eliminated wind and helped regulate temperature and relative humidity. Temperature and relative humidity were measured every two hours using a Kestrel 3000 (Nielsen-Kellerman, Boothwyn, PA, USA) to examine possible covariates of flammability measures. The trials were conducted from 2 October 2010 to 9 December 2010, and 15–22 trials were conducted each day. No more than one replicate of a mixture type was burned on any day.
We measured eight flammability parameters that were likely to co-vary, and therefore, flammability parameters were first studied using principal components analysis (PCA) to explore such covariance and to guide selection of key flammability parameters for further analyses (Mardia, Kent & Bibby 1979; ‘prcomp’ function, R Development Core Team 2011). We used analysis of variance (anova) to test for species differences in each flammability parameter (using only the monoculture trials). We tested whether three leaf traits (leaf area, leaf length, SLA) predicted litter flammability and which had the strongest effects. To accomplish this, we explored three linear models for each flammability parameter with leaf traits as the explanatory variables with species as a random nesting factor. We then determined the strength of the effect based on the P-values.
We assessed non-additivity in mixtures by comparing the flammability parameters to a null model based on the average of the monoculture flammability values of the three individual species that made up a mixture. Under the null model, the expected difference between the measured parameter and the predicted one is zero. We tested for a significant departure from zero using a mixed effects linear model (‘lme’ function, from the ‘nlme’ package in r, Pinheiro et al. 2011). Such a departure was indicated by a significant intercept term in a model with the flammability parameter as the response variable and with mixture type as a random factor and with no fixed effects (Pinheiro et al. 2011). Spread rate, time to ignition and percentage mass loss were log-transformed to meet the assumption of normality of the linear models. We conducted an additional analysis of non-additivity in which the null expectation was the average flammability parameters weighted by volume rather than by mass to explore effects of weighting method on our results. We also used an additional method to predict expected litter density: the expected density of a mixture was calculated as the total mixture mass divided by the sum of the volumes of the constituent species – this null expectation assumes no physical mixing of particles.
To further explore which species drove the mixture effects studied above, we investigated the contribution of each individual species to the behaviour of a mixture. The average flammability contribution for each species in a mixture represents how close the behaviour of a mixture is to the behaviour of the species individually. To do this, we calculated an ‘average effect in mixture’ for each species across all mixtures. We averaged the differences between the observed value of a mixture and that of each of the three species contributing to it, for all species and all parameters under investigation. This provided us with an average value for each species for each parameter, which indicates the contribution of each species to a mixture. Values of low magnitude indicated that the species’ flammability in monoculture was similar to that of mixtures in which it occurred.
The building eliminated wind, but air temperature and relative humidity were measured every two hours. Average temperature for the trials was 18.8 °C (±4.8), and average relative humidity 31.3% (±8.9). For each flammability parameter, we compared the main model to two additional models (one that included relative humidity and one that included temperature in addition to the main predictor of interest). We used Akaike’s information criterion (Burnham & Anderson 2002) to determine whether including these climate covariates improved model fit (reduced AICc).