Field sampling

Initial sampling, residual fuel sampling, and fuel harvesting

In February 2016, the study area was selected: a tropical oak woodland in the rural community of Villahermosa, Municipality of Villaflores, Chiapas (Figure 1). The zone has a sub-humid warm climate (Aw₂), with 22 to 24 °C isotherms and 1200- to 1500-mm isohyets. The terrain is uneven, mostly dominated by sedimentary and volcano-sedimentary rocks; it belongs to the regosol soil group and the dominant vegetation is the pine-oak association (^{Ecosur, 2018}). Stands of oaks can be found where the study was carried out and the most common species are Quercus peduncularis Née, Q. skinneri Benth. and Q. magnoliifolia Née. Six 0.5-ha plots were established. The measurement of forest fuels involved the establishment of triangular sampling sites in the center of each plot, with stakes marking the vertices (Figure 2). The triangular design had the purpose of concentrating fuel sampling as much as possible around the sampling lines utilized to measure fire behavior. For woody fuels, the ^{Brown’s Planar Intercept Method (1974)} was used, and for leaf litter, fermentation layer, forbs, grasses, shrubs, and regeneration, the measuring and harvest method was used. The 12-m sides of the sampling triangle -and the base following the contour line- were used as sampling lines for woody fuels. The slope of each of the triangle’s sampling lines was measured with a Suunto^© clinometer. Starting from each vertex and working in clockwise direction, 2-, 4-, and 12-m long overlapping lines were established. The 2-m lines were used to record the number of fuel intersections with 1 h of time lag (TR) (<0.6 cm diameter), as well as for those with 10 h of TR (0.6 to 2.5 cm). The 4-m lines were used to record the quantity of fuel intersections with 100 h of TR (2.6 to 7.5 cm). The 12-m line was used to record the intersections with materials with 1000 h of TR, both sound and rotten (˃7.5 cm). In this last case, the diameter of each piece was also recorded. For each TR, 12 specimens of woody materials were collected in order to determine their specific gravity in the laboratory.

Figure 1 General (A) and specific (B) location of the study area. 

Figure 2 Cluster sampling for forest fuels. Site 1 is shown in detail, but sites 2 and 3 (the other sides of the triangle) had the same components and were also sampled. 

Next to each vertex square sampling sites were established on the outside of the triangle, without trampling on the area. The sites had the following dimensions (fuel type indicated in parentheses): 0.3X0.3 m, metal frame (leaf litter and fermentation layer); 1X1 m, rigid plastic frame (PVC) (forbs and grasses), and 4X4 m, with measuring tape (shrubs and regeneration). The leaf litter and fermentation layer coverage (%) was estimated visually, while its layer depth was measured in four points with a measuring tape. The coverage of the remaining fuel was estimated (%) and six height measurements were made. Additionally, all the fuels were sampled again 2 m ahead of these sites and on each side of the triangle. In these last sitios, after they were measured, the materials were harvested, put in brown paper bags, labeled, and placed in plastic bags. As it is explained below, this sampling and harvest were carried out to determine dry weight.

In order to sample the different fuels, 11 sampling sites were established per side, resulting in 33 sites per triangle, and 198 sites for the six triangles. The sampling aimed at harvesting and obtaining dry weights included 6 sites per side, 18 sites per triangle, and 108 sites in total. The three fuel samplings (initial, harvest and residual) involved 504 sampling sites.

Fire behavior

Flame length (LL) and fire spread rate (VP) were measured as fire behavior variables. For this purpose, a 2 m long stake was fixed in the center of the base of the triangle to measure forest fuels, while red and white bands were painted every 0.2 m. Three stakes were placed upwards the slope and one downwards, 5 m apart from each other. The flame length was measured when it reached the stake. The VP was estimated based on the time that the fire required to cross each 5 m section. Fire behavior was measured in 18 sampling sites (Figure 3).

Figure 3 Measurement of fire behavior. 

Laboratory and desk research

Statistical analysis

The variables studied were compared between heading and backing fire treatments, using the Student’s t-test (p≤0.05), with a variance inequality test. Likewise, the graphic method was used to establish normality. If normality was lacking, the Wilcoxon test (its exact value) was used. For all analyses, the Proc ttest and Proc nPar1way procedures of the SAS v. 9.4 statistical software for microcomputers were used.

Weather

During the prescribed heading fires, the means and their standard deviation (in parentheses) of wind speed, temperature, and relative humidity were 4.77 km h^-1 (2.85), 27.4 °C (5.61), and 60.2 % (19.83), respectively. Meanwhile, during the backing fires, the averages reached 3.97 km h^-1 (2.69), 31.2 °C (2.05), and 52 % (29.36), respectively. In the case of the above-mentioned variables, the inequality of variances test had the following p-values: 0.8275, 0.0006, and 0.1543. As a result, only temperature showed inequality of variances. Only this last variable showed significant differences between both treatments (p=0.027); wind speed (p=0.445) and relative humidity (p=0.338) did not show significant differences. The three atmospheric variables under consideration lacked normality; therefore, the p of the exact value of the Wilcoxon test was used (0.523, 0.169, and 0.406). In other words, there were no differences between the medians of the variables of the applied treatments. However, although there were no differences in the wind speed, the wind blew against the fire in backing fires, so the test was also applied with negative wind speed values for this variable. In this case, the mean and its standard deviation was -3.97 km h^-1 (2.69), there was no inequality of variances (p=0.827), and there were significant differences with the t-test (p≤0.001). There was no normality and the Wilcoxon test showed differences between medians (p≤0.001). In both burn methods, the means of the weather parameters were below the limits that ^{Wade and Lunsford (1989)} established as an adequate prescription (up to 5-16 km h^-1 wind speed, 30 to 55 % relative humidity). That is not the case of temperature (18-27 °C), although the said authors mention that it recorded in cold temperate climate or temperate climate zones.

Fire behavior

In the heading fire, the VP variable reached a mean and standard deviation (in parentheses) equal to 3.63 m min^-1 (3.41), while for the backing fire, it was 0.59 m min^-1 (0.26). There was inequality of variances (p≤0.001) and significant differences for this variable with the t-test (p=0.028). There was no normality, and -with the Wilcoxon test- there were differences between medians (p=0.003). Due to safety and accessibility issues, no fire VP measurements were obtained, but they were estimated at ˃30 m min^-1. LL reached 0.91 (0.28) and 0.72 m (0.30), in both heading and backing fires. The variances were similar (p=0.837) and there were no differences between both treatments (p=0.188). In this case, there was no normality and the Wilcoxon test produced no significant results (p=0.129).

In the forest fire, LL reached a mean of 1.94 m (1.06). Variances were unequal (p=0.016) and the t-test showed no differences (p=0.063). There was no normality either and the Wilcoxon test showed significant differences between the medians (p=0.033).

The figures for weather and fire behavior matched those of other researches. Experimental prescribed burns were carried out in a tropical oak woodland of Q. magnoliifolia in Guerrero. The weather had the following characteristics: temperature between 25 and 28 °C, with 33 to 25 % relative humidity, wind speed of 7 to 8 km h^-1, and gusts up to 25 km h^-1. LLs ranged from 0.5 to 3 m and there was a direct logarithmic relationship in which LL is a function of wind speed (^{López et al., 2015}). In pine-oak forests in Chihuahua where Pinus duranguensis Martínez and Q. sideroxyla A. J. A. Bonpland can be found, prescribed burns -carried out with a relative humidity around 60 %, winds of 2 to 6.1 km h^-1, and temperatures of 1.5 to 14.5 °C- showed LL of 1 to 6 m, and PV of 0.8 up to 5 m min^-1 (^{Flores et al., 2010}).

The largest VP and LL in heading fires are known. The wind provides oxygen to the combustion, tilts the convective column and the flame, and brings them closer to the unburned materials, drying them and making them available quicker; in addition, it generates flying sparks. As the fire rages on, the fuels found uphill dry quicker than those found downhill, because they are closer to the convective column and the radiation and the fire in the front line is more intense; this situation accelerates the combustion and VP (^{Scott et al., 2014}). The few differences between the heading and backing fires are based on the fact that (for safety purposes) both were carried out under moderate weather conditions (Figure 4). Regarding the topography, ^{Franklin et al. (1997)} established that the slope is of particular importance for the intensity of the fire in leaf litter fuel beds and the fermentation layer of North American oak woodlands.

Figure 4 Prescribed A) heading fire, and B) backing fire. 

Initial and residual loads of forest fuels

In this case, the following variables were compared for the two types of burns: the partial loads for each type of forest fuel, the sum of woody fuels, and the total load. Likewise, the residual load, the load consumed, the proportion of load consumed, and the consumption factor were compared. Most of the variables showed a lack of normality. The Wilcoxon test showed that there were only differences in medians between the backing and the heading fires for the following elements: residual load of leaf litter, consumption factor of leaf litter, initial grass load, and grass consumption. Comparing total load, residual load, total consumed load, and their consumption factor, there were no differences between any of these variables for these types of burn. This analysis is shown in Table 1 (leaf litter and fermentation layer), Table 2 (grasses, forbs, shrubs, and regeneration), Table 3 (woody materials with TR 1, 10, and 100 h), Table 4 (sound and rotten woody materials with TR 1000 h), Table 5 (total load of woody fuels), and Table 6 (total load of forest fuels). No differences were found between forest fire and heading fires, and then between forest fire and backing fires, when comparing the following variables: total and residual load, consumption, and consumption factor.

In the fire area, the loads reached 32.647 t ha^-1 (19.3049), residual loads reached 1.0662 t ha^-1 (0.9558), consumption reached 31.5808 t ha^-1 (20.1262), and consumption factor reached 95.02 % (4.38). The load is similar to the 31.726 t ha^-1 mentioned by Bonilla et al. (2012) for cold-temperate stands dominated by Q. crassifolia Humb. & Bonpl, in Chignahuapan, Puebla, and by ^{Alvarado et al. (2008)}, for subtropical oak woodlands in western Mexico, where species such as Q. magnoliifolia and Q. resinosa Liebm can be found.

Fuel consumption is related to fire behavior and its immediate and long-term effects. The load and type of fuel influence the level of consumption. The grasses are almost entirely consumed; however, the humidity content, density, and mineral content come into play in organic materials beds. In woody materials, the diameter has an inverse ratio to its consumption (^{Scott et al., 2014}).

The consumption factors obtained in our study match those of other researches in different ecosystems. ^{Urbanski (2013}) determined 88.3 % factors for mixed-conifer forests after forest fires north of the Rocky Mountains. ^{Brose (2016)} obtained a load of 25.9 t ha^-1 (divided approximately in equal parts between light and heavy fuels) and a load of up to 61.8 t ha^-1 (mostly as medium and heavy fuels), in untreated oak forests that were subject to silvicultural cutting. According to the said author, intense prescribed burns during the growing season eliminated almost all light materials, and half of medium and heavy ones. In conifer forests of the Sierra Nevada mountain range, USA -home to seven species, including Q. kelloggii Newb.-, the load was 123.1 t ha^-1, the consumption factor was 65.7 %, and the consumption by type of fuel was rotten woody materials -with 1000 h TR (86.6 %), fermentation layer (84.8 %), leaf litter (63.4 %), and firm woody materials (41.6 %) (^{Kobziar et al., 2006}). In our study, the heavy woody fuels accounted for a smaller proportion of the total load, both in heading and backing fires (30.6 and 26 %), and their consumption factors reached 77 and 100 %. More heavy woodfuel was consumed as a result of its low load and smaller dimensions.

Emissions

For the backing and heading fires, 44.53 and 51.14 t ha^-1 CO₂ emissions were estimated, while the forest fire had a 53.78 t CO₂ ha^-1 value. Since the emission is obtained multiplying the total load consumed by a constant, emissions showed no differences. However, this analysis was focused on superficial fuels and, even though the crown was slightly affected by the prescribed burns, that was not the case at all when fires were involved (See: Severity). In other words, more aerial forest fuels are consumed than those included in this study, only were consideres as affected crown; therefore, the emissions produced by fires are greater than those produced by prescribed burns. ^{Wiedinmeyer and Hurteau (2010)} point out that prescribed burns in western USA -although it is a different ecosystem- reduce fire emissions by 18-25 %, due to their lower intensity and consumption factor.

Severity

The trees of the studied plots had average heights of 9 to 10.8 m and diameter at breast height between 18.3 and 20.8 cm. The CHC and CT reached higher values in backing fires -26.8 % (35.7) and 1.17 m (1.03) (p=0.0054)- than in heading fires -9.3 % (16.3) and 0.36 m (0.42) (p≤0.0001)-, because fire residence time was longer in the former case (with a similar flame length). However, averages CHC did not exceed ⅓ of affectation. In the case of pines in which one third of their crowns are affected, fire even favors the diameter growth of the trees, provided that there is abundant rainfall during the next rainy season. This is caused by the elimination of foliage from the lower part of the crown -which is less photosynthetically efficient- and from the low branches -which demand more photosynthates than those produced by their foliage-, and the fertilization provided by the cation-rich ashes resulting from the combustion (^{Ryan and Reinhardt, 1988}; ^{Rodríguez et al., 2007}). The height of affectation on the trunk is also acceptable in both treatments. In the fire, 96.7 % (7.8) of the crown area was affected; meanwhile, CT reached 2.57 m (1.70), surpassing both backing and heading fires, in CHC (in both cases, p≤0.0001), as well as in CT (p=0.0172 and p=0.0009, respectively). ^{Ryan and Reinhardt (1988)} established that such level of affectation increases the mortality probability among North American conifers. However, ^{Juárez et al. (2012)} point out that oaks (including Q. crassifolia) have a low mortality and regrowth rate, even when their crown is completely burned. This was also the case of young oaks (5 to 6 years of planting) of Q. lobata Née, after spring and summer prescribed burns in California. A low mortality was implied (3-4 %), despite the death of 66-72 % of the aerial part of the oaks, although the trees resprouted (^{Holmes et al., 2011}). In the tropical forests of Q. magnoliifolia in Guerrero, prescribed burns with 0.5 to 3 m LL were more severe than in this study, because 40 % of CT heights reached between 1 and 2 m, and 50 % of the CHC frequency of CHC faced a 25 to 50% affectation (^{López et al., 2015}).

In Sierra Nevada, USA, eight months after the application of prescribed burns, Q. kelloggii (one of seven species found in the area) had a survival rate of 42.6 %. Live trees had a greater average diameter (26.1 cm) than dead trees (7.6 cm), while their average heights reached 11.4 and 4.4 m, respectively. In the same order, the maximum CT height reached 1.4 and 1.7 m (^{Kobziar et al., 2006}).

The information generated in our study provides a fuel model for this type of vegetation in the region and is useful to model fire emissions and danger. In addition, it serves as a guide for prescribed burns in the area, with the following objectives: reducing fuels and CO₂ emissions, the latter in comparison to forest fires and, in general, supporting the integral fire management.

Within the prescription hereby recommended, prescribed fires using heading fire have the advantage of covering more ground with greater speed, as although they have the disadvantage of requiring even more care, since it is easier for burns of this type to escape, than during backing fires.

There are several ways in which prescribed burns help to reduce emissions. The first, because they involve lower emission per unit area, given their lower intensity and severity. The second, because they have a preventive effect: if a prescribed burn is agreed with a producer, a forest fire can be avoided. And the third, because prescribed burns have smaller area than forest fires. All these factors reduce greenhouse gas emissions. Therefore, prescribed burns are a useful tool for that and other purposes.

According to the results of the present experiment, prescribed burns can potentially be used to reduce the greenhouse gas emissions caused by forest fires in the study area. The multiplicity of additional land management objectives that can be achieved with this tool should be also considered.