Ever since terrestrial vegetation created an oxygen rich atmosphere and provided fuel for combustion, wildland fires are an inseparable part of the Earth system [1
]. The Earth is intrinsically flammable, and most of the terrestrial plants can burn under appropriate climatic conditions [2
]. Nevertheless, similar climatic conditions and vegetation types can support different fire regimes, depending on the vegetation flammability [3
]. Even though definitions of flammability and metrics used to describe it differ between authors [2
], evidence implies that plant species do differ in their tendency to ignite, to support fire and to facilitate fire spread (e.g., [2
]). This evidence has been gathered at scales ranging from laboratory testing of small leaf fragments [8
] to field experiments [9
] and large-scale field surveys [10
], encompassing modelling studies [11
]. Furthermore, the relationships between plant functional traits and flammability are extensively investigated, with affirmative results [3
Even though substantial work has been devoted to the leaf litter, it is mostly conducted on constructed leaf litter beds (sensu [15
]) which are tested immediately after preparation. This approach eliminates several aspects of fuel bed dynamics, among which short-term structural changes are the focus of this work. Litter beds are dynamic fuels and their properties are continuously changing. Decomposition [16
] and particle fragmentation [18
] are recognized as important processes altering the fire behavior of the leaf litter beds. Another process appears important as well: compaction (i.e., an increase in the bulk density and packing ratio with a decrease in the amount of empty space within the fuel bed) of fuel beds under exposure to local weather condition was previously observed [18
], but only recently quantified [20
]. To quantify the effects of leaf compaction on fire behavior of the leaf litter beds a novel exposure and testing system was developed. This system enables exposure of standardized fuel beds in environments potentially altering its structure and chemical composition (e.g., in-stand exposure, exposure to outside weather conditions, greenhouse experiments with simulated precipitation events), their retrieval and fire behavior testing without disturbing the sample structure.
Our initial work [20
] revealed species specific and potentially large effects of a 30 days winter exposure to outside weather conditions on fuel bed structure and fire behavior of deciduous tree species. Species with high initial bulk density (BD) were less affected by the exposure than species with low initial BD. Due to unexpectedly large effects detected in our initial work (e.g., the most affected species showed a 124.3% increase in BD), we were interested in further examining the matter.
Here, we compared effects of two different exposure periods (i.e., a simulated single precipitation event (treatment ”10”) and a two months exposure to outside weather conditions (treatment “60”)) to the common approach of testing fire behavior immediately after fuel bed preparation (treatment “0”). We tested three species common in the fire prone regions of the Mediterranean Basin: Aleppo pine (Pinus halepensis
Mill.), carob (Ceratonia siliqua
L.), and downy oak (Quercus pubescens
Willd.). Whereas our initial work [20
] focused solely on deciduous trees, here we included species belonging to different plant groups (evergreen conifers, evergreen broadleaved species and deciduous broadleaved species, respectively) and provide the first information on compaction of evergreen leaf litter beds. Based on the established relationships between leaf traits and fire behavior of litter beds [2
] we expected different fire behaviors and different responses to the treatments for the tested species.
Ormeño and others [21
] demonstrated that an increase in the terpene concentration of leaf litter results in higher, faster spreading flames and shorter combustion duration. The same effects were attributed to an increase in particle size, both for conifers [12
] and broad-leaved species [13
]. For oaks, lobed leaf edges were associated with higher flames and shorter flaming duration, compared to species with entire margins [13
]. Furthermore, an increase in tissue density was related to slower flame spread and longer flaming duration [23
]. Based on these relationships between plant traits and fire behavior, and the presumption that increasing thickness and density of leaf litter particles result in a more rigid fuel bed structure which would be less prone to compaction, the following hypotheses were established:
Resin rich, relatively long and thick (compared to broadleaved species) needles of P. halepensis would create leaf litter beds with low initial BD which would burn with high, fast spreading flames, low flaming duration and high fuel consumption, regardless of the treatment.
Due to a large proportion of dense and thick leaf stems in the leaf litter and leaflet lamina thicker than those of Q. pubescens leaves, C. siliqua would create leaf litter beds with high initial bulk density. These fuel beds would burn with small, slow spreading flames, long flaming duration and low fuel consumption regardless of the treatment.
Larger, thinner, lobed leaves of Q. pubescens would initially (“0” treatment) create leaf litter beds with low BD, which would burn similarly to P. halepensis litter beds.
Due to low particle thickness and density Q. pubescens leaf litter would be susceptible to compaction, resulting in a pronounced increase of the BD in the treatments “10” and “60” compared to the treatment “0”.
The expected increase in BD of Q. pubescens would translate into alterations in the fire behavior, resulting in small, slow spreading flames and long flaming duration (i.e., in a fire behavior comparable to that of C. siliqua) in the treatment “60”.
Confirmation of our hypotheses would indicate that the presumed relationship between particle thickness, density and fuel bed compaction is correct. This would further imply that structural alterations of the leaf litter beds can be predicted based on the particle characteristics. Rejection of our hypotheses would imply that other characteristics have a higher influence on the short term alterations in the fuel bed structure.
Individual authors differently translate the measured fire behavior characteristics into flammability [2
]. Here, we follow the suggestion given by [7
], i.e., we focus on the metrics itself and discuss our results in terms of different fire behavior rather than in terms of a single flammability value.
2. Materials and Methods
2.1. Leaf Litter Handling
Leaf litter was collected by the officers of the Mljet National Park Public Institution, island of Mljet, Croatia (hereafter “Park”), within the territory of the National Park Mljet. P. halepensis litter was randomly collected in a mature forest stand. C. siliqua litter was gathered in old, abandoned, or extensively managed groves. Q. pubescens can be found on the island only as mature, solitary trees. For all species of interest freshly abscised leaf litter of mature trees, with no visible signs of decomposition, was collected, air dried, and initially stored within the Park’s facilities. Samples were sent to the Institute of Landscape and Plant Ecology, University of Hohenheim, Germany, where further processing and testing took place. After arrival at Hohenheim, leaf litter was dried for 72 h at 60 °C and then stored in open plastic boxes in a cold and dry room. Litter was dried once more (24 h at 40 °C) just before individual samples were weighted. After weighing, samples were randomly assigned to a treatment. To characterize fuel particles, 50 random leaf litter particles of P. halepensis and Q. pubescens were taken for detailed morphological measurements, while 50 random leaflets and 10 petioles were measured for C. siliqua.
2.2. Experimental Set-Up
Within the experiment two fuel loads of monospecific leaf litter and different combinations of two-species mixtures were tested (yielding a total of 115 tests). Reporting about all gathered data would exceed the scope of one single paper. Therefore, here we present the results of testing monospecific samples with a weight of 40 ± 0.5 g which were subjected to three different treatments: (i) treatment “0” in which samples were tested immediately after fuel bed construction: (ii) treatment “10” where constructed fuel beds were exposed to a single, simulated precipitation event and left in the greenhouse for 8–12 days before being tested; (iii) treatment “60” in which constructed fuel beds were tested after 59–63 days of exposure to outside conditions. All samples were constructed in the same manner. A small amount of material was dropped into a container; the container was consequently gently shaken to ensure uniform distribution of the litter. Once the whole sample had been transferred, leaf litter was gently pressed. Samples in treatment “0” were constructed directly in the testing cage, while samples in treatments “10” and “60” were established in the exposure construction and transferred to the testing cage prior to testing. Details on the exposure construction, testing cage and transfer process are given in [20
]. In brief, the exposure construction was composed of a 20 cm × 20 cm aluminum mesh bottom with an edge 2.5 cm in height, and a fabric enclosure 13 cm in height. The fabric enclosure of the exposure construction was made of fiberglass reinforced PVC and rigid enough to stand on its own. Thus, no additional support was required in the treatment “10”. Treatment “10” samples were established in the greenhouse on planting tables with a mash surface (openings 25 × 25 mm, 2.75 mm wire diameter) (Figure S1
). Once all samples were in place, a watering hose with a shower nozzle was used to simulate a single precipitation event. Water was gently and evenly sprinkled over the experimental surface until dripping was observed on the lower side of all samples.
Samples for treatment “60” were established on the grounds of the University of Hohenheim. To suppress vegetation growth a black plastic fabric was spread on a lawn. Washed quartz sand was spread on top of the fabric to ensure drainage and prevent excessive heating of the experimental area. Exposure constructions for samples in treatment “60” were placed on top of the sand and each construction was fixed to the ground with four skewer tent pegs 23 cm in length. To prevent disturbance by animals and foreign material falling into the samples, a protection net covered the whole experiment (Figure S1
). Even though established in spring, treatment “60” simulated effects of the prolonged autumn exposure, as spring weather conditions in Stuttgart roughly correspond to autumn/winter weather condition on the island of Mljet (Figure S2
). Treatment “10” simulated the effects of a single precipitation event, whereas treatment “0” represented the common approach of testing constructed leaf litter beds immediately after construction (e.g., [12
Biological interactions and fuel drying are site specific [27
]. Thus, to eliminate the possibility of overestimating changes which would occur under Mediterranean conditions, treatments “60” and “10” were designed to promote fast fuel drying and minimize biological interactions. This reduced potential effects of decomposers and detritivores on the exposed material slowing down changes in the fuel bed structure and chemical composition. Consequently, significant effects of the treatments applied here would suggest that even larger effects could be expected if samples were exposed in the native stands under similar weather conditions.
Treatment “60” was established on 15 March 2017, and treatment “10” on 5 May 2017. Samples assigned to treatment “0” were stored in paper bags in a cold, dry room until two days before testing. All treatments were arranged in blocks. Two days before testing one random block of each treatment was placed in the drying cabinet at 40 °C for 24 h. After drying, samples were left in the fire behavior testing facility for 24 h to equilibrate. Mass differences between initial mass and pre-testing mass were low (4.25 ± 0.07%, average ± standard error) and no observable patterns were detected (ANOVA showed no significant treatment or species effects). Thus, we did not include this data into further analysis. Furthermore, as samples were exposed under the same conditions, both decomposition rate [30
] and moisture adsorption/desorption [32
] are governed by the intrinsic characteristics of the fuel and can be considered characteristics of the material itself. Position of the samples in the drying cabinet, in the fire behavior testing facility, and testing order were random. One full replicate (all treatment and materials) was tested each day. Fire behavior testing was performed on five consecutive days, starting on 15 May 2017. Temperature and relative humidity were logged (Tinytag TGU-4500 and Tinytag TGP-4017, Gemini Dataloggers Ltd., Chichester, UK) at hourly resolution in all facilities (Figures S2 and S3
2.3. Fire Behavior Testing
To enable fire behavior testing of the samples formed in the exposure construction without disturbing their structure, a special combustion chamber was developed. For details on the combustion chamber refer to [20
]. In brief, the stainless steel combustion chamber has an open front side providing air supply and an unobscured view of the burning samples. It is partially insulated with 2.5 cm thick vermiculite insulation board (V-1100 (700), Skamol, Nykøbing Mors, Denmark). A closable side door allows careful and precise handling of the sample. Within the combustion chamber a frame 2 cm in height was filled with washed and size calibrated dry quartz sand at room temperature, creating a surface of 20.5 cm × 21 cm. The sand surface was flattened and samples were carefully positioned on top of the sand. Samples were tested in a testing cage with an area of 21.0 cm × 21.5 cm and a height of 14.5 cm. The testing cage was covered with a fine stainless steel mesh. The back side of the testing cage could be flipped open, allowing samples to be gently slid inside. Before testing of samples from treatments “10” and “60”, the fabric enclosure of the exposure construction was carefully removed, and samples were placed in the testing cage together with the aluminum bottom of the exposure construction. Thus, the sample structure was preserved. To ensure uniformity of the testing procedure, the same type of aluminum bottom was placed in the testing cage before samples from treatment “0” were constructed. An IR lamp (1000W, 1.4–1.7 µm, 90 kW/m², with golden reflector, SRSystems GmbH, Bruchköbel, Germany), aligned with the lower edge of the litter sample, acted as a standardized linear ignition source. Ignition was piloted using a handheld spark generator. A digital camera (Power Shot SX280HS, Canon, Tokyo, Japan) was positioned in front of the combustion chamber, providing video records for all tests. The testing chamber was placed underneath a hood. No air movement was detected on top of the combustion chamber.
A stopwatch was started simultaneously with turning on the IR lamp. The spark generator was held in the smoke convection plume close to the fuel surface until flames appeared. Once flaming was initiated, the IR lamp was turned off. Times until flame appearance, flame reaching the rear side of the fuel bed, flame extinguishment and last ember extinguishment were recorded. Unburned residues were weighed and heated in the muffle furnace for 8 h at 550 °C to determine the ash content.
To ensure comparability with previous work, we strived to include only commonly used and previously well-defined characteristics [12
]: flaming duration (FD), i.e., the time for which flames are visible, smoldering duration (SD), i.e., the time between flame extinguishment and extinguishment of the last ember, maximum flame height (FH), and rate of spread (RoS), i.e., the length of the fuel bed (20 cm) divided by the time required for the initiated flames to reach the rear side of the fuel bed. FH was determined by frame by frame analysis of the video records. VideoPad video editor software (NCH Software, Greenwood Village, CO) was utilized to extract the frame of interest, and Image J [34
] to measure the maximum flame height (Table S1
Due to the distribution characteristics of our data, we report “unconsumed” (UC), i.e., the percentage of fuel remaining after burning (100%–% consumed), instead of the commonly reported consumed percentage of the fuel. Furthermore, we corrected UC for the ash content [20
]. Time was recorded with an accuracy of 0.01 s, masses with an accuracy of 0.01 g, FH was determined with an accuracy of 0.01 cm, and RoS is expressed in cm s−1
2.4. Fuel Characteristics Measurements
As the chosen species are markedly different, we consider it unnecessary to discuss in detail the differences between morphological characteristics measured at the particle level. Nevertheless, in order to confirm that the relations between species presumed in our hypotheses (i.e., Q. pubescens
having thinner, larger, and less dense litter particle that remaining two species, P. halepensis
having relatively long needles, C. siliqua
having dense and thick petioles and leaflets) are true, details on the measuring procedure and the measured values are provided in the supplementary material (Table S2)
Sample height was measured at three random positions within the fuel bed just before testing. Sample volume was calculated by multiplying sample area (400 cm2
) with sample height, and bulk density (BD) as dry mass divided by sample volume. Packing ratio (PR), i.e., the fraction of the fuel array volume that is occupied by fuel, was calculated as BD divided by the average particle density [35
2.5. Data Analysis
Since fire behavior characteristics tend to be highly correlated [7
], principal component analysis (PCA) was used to investigate the association between FD, SD, RoS, FH and UC. The purpose of the PCA was to explore the structure of the fire behavior data, thus fuel bed characteristics (BD, PR) or measurements taken on the particle level were not included. PCA was performed on scaled parameters (mean = 0, standard deviation = 1). If necessary, parameters were transformed before scaling.
Fire behavior parameters with high loadings to the same PC exhibited similar behavior, thus further analyses were conducted on PC scores. K-means clustering was performed to investigate the association between individual samples when scores on both PC-s are taken into account. To check for significance of the effects of treatment, species and their interactions, Two-Way ANOVA with species and treatment as fixed factors was performed on the PC scores, BD and PR. ANOVA was followed by pairwise contrasts with Tukey adjustment for multiple comparisons. Effect size was expressed as partial eta-squared [36
Finally, to investigate to which extent differences in the fuel bed aeration can explain fire behavior of the leaf litter beds, regression analyses with BD or PR as independent and PC scores as dependent variable were performed. BD and PR both relate negatively to fuel bed aeration [22
]) and exhibited similar relationships with PC scores (Figure S4
). Thus, only the relationships with BD as independent variable are presented in the main article. Regression analyses were performed on the whole data set and separately for individual species. All measured values were included in all analyses (N = 45 for overall and N = 15 for species-based regression analyses). Results of the same analyses (ANOVA with post hoc test and effect size (Table S3
), regressions with BD or PR as independent variables (Figure S5
)) conducted on the measured fire behavior characteristics are reported in the supplementary material
. The fuel bed structure used and the fire behavior data are also given (Table S4