Relationships and Modeling of Cerrado Grassland Fuel in Jalapão, Tocantins state, Brazil

Study area

This study was conducted in the south-central portion of the Serra Geral do Tocantins Ecological Station (EESGT). The EESGT is an integral protection conservation unit located in the Cerrado biome in the Jalapão region, covering 716 306 ha. It is in the northern region of Brazil, under the central coordinates 10°54'09.70"S, 46°41'49.65"W (datum WGS 84). This station covers the municipalities of Almas, Mateiros, Ponte Alta, and Rio da Conceição in the state of Tocantins and the municipality of Formosa do Rio Preto, which belongs to the state of Bahia. According to the Köppen classification, the climate of the region is of the Aw type (tropical savanna climate), with annual precipitations ranging from 1400 to 1500 mm (Seplan, 2012), which is higher than the potential annual evapotranspiration. The summer is rainy and occurs between October and April, and dry winters occur from May to September (ICMBio, 2014).

The predominant phytophysiognomy of the Cerrado is grassland, including dry (‘campo limpo seco’, ‘campo sujo seco’, and ‘campo rupestre’) and wet (‘campo limpo úmido’ and ‘campo sujo úmido’) grasslands (Pereira Junior et al., 2014).

The predominant soil type is quartz sand or quartzarenic neosol, which has a sandy texture down to least 2 m deep, with up to 15% clay. The relief varies from relatively flat to gently undulating, with average elevations between 300 and 550 m (Santos et al., 2013).

Treatments and procedures for collecting variables

The treatments were based on fuel collections spanning four months of the dry season (May, June, August, and September) in areas with four periods without fires (fuel age), ranging from one to four years. Predominantly, the rural areas located in the EESGT are regions with no-burning periods of one up to four years, as a result of integrated fire management actions (MIF).

No-burning periods were determined using remote sensing techniques and Landsat 8 OLI satellite images, analyzing the burn scars reported in the EESGT. Thus, throughout the experiment, 128 plots, called data sampling units, were installed.

Each data sampling unit was defined by two 30 m transects, distributed by drawing lots, containing eight sub-samplings called fuel sampling plots, each with an area of 0.25 m² (0.5 m x 0.5 m) (Silva et al., 2018). The size defined for the fuel sampling plots aimed to ensure a broad sampling and greater variability in the sampled fuel. The transects were separated by 14 m, and the location of the fuel sampling plots in each transect was marked every 6 m. In these markings, it was possible to install up to two plots, one at 1 m to the right of the transect line and the other at 1 m to the left of the transect line (Santos et al., 2020). The samples were taken in both the fuel (before the fire) and fuel consumption (after the fire) stages.

At the end of the study, 1024 fuel sampling plots had been analyzed (128 sampling units with eight fuel sampling plots each). To obtain the fuel load values, the units were converted from kg m² to Mg.ha^-1 for each data sampling unit. Meanwhile, for the quantification of species and total individuals, no conversions were performed, and the original size of the sample was considered (0.25 m²).

Before sampling via the destructive method, measurements of grass height (Htg, cm) and litter height (Htl, cm) were taken, with two measurements for each of these variables in order to determine their average value. The number of species (Nsp) and the number of individuals (Nin) in each 0.25 m² sample were also collected using the aerial part of each plant while avoiding the duplication of individuals, considering that two or more stems could belong to the same individual. If it was unclear how many plants were involved, part of the surface layer of the soil was removed for further study. Species were differentiated based on visual inspection, as detailed identification was not required to meet the objectives and hypotheses. Htg, Htl, Nsp, and Nin were considered to be non-destructive variables in this study.

At the time of collection, information on temperature (°C) and relative humidity (%) was obtained from a portable meteorological station (Kestrel 4200 Pocket Air Flow Tracker), which was installed near the fuel collection site, and recorded in a field form.

Surface fuel sampling

The fuel sampling plots were destructively separated according to their physiological state (live and dead) and their diameter class (time-lag), following the methodologies proposed by Schroeder and Buck (1970). Thus, live fuels were classified according to the following classes: (i) live grass fuel (Lg): annual or perennial grasses and herbs in an active physiological stage; (ii) 1-h live fuel: all live woody material with a diameter of less than 0.7 cm and leaves; (iii) 10-h live fuel: live woody material with a diameter ranging from 0.7 to 2.5 cm; and (iv) 100-h live fuel: live woody material with a diameter ranging from 2.5 to 7.5 cm.

Similarly, dead fuels were classified as follows: (i) dead grass fuel (Dg): physiologically inactive grasses and herbs or perennials; (ii) 1-h dead fuel: all dead woody material with a diameter of less than 0.7 cm and senescent leaves; (iii) 10-h dead fuel: dead woody material with a diameter ranging from 0.7 to 2.5 cm; and (iv) 100-h dead fuel: dead woody material with a diameter ranging from 2.5 to 7.5 cm. However, for the classes 10-h live, 100-h live, 10-h dead, and 100-h dead fuel, the load values were not discussed, as they reported nonsignificant amounts in relation to the others, making it impossible to carry out statistical analyses. Dead leaves deposited on the soil were included in the 1-h dead fuel class. In addition to analyzing these variables, the values for the total live fuel load (Tl), the total dead fuel load (Td), and the total fuel load (Tl+Td) were examined across areas with varying fuel ages and during different months of collection.

A subsample for each fuel class was placed in a numbered kraft paper bag, and its mass was determined using a digital scale. The subsamples were then dried in a laboratory oven until they reached a constant weight for the subsequent determination of their dry mass and fuel moisture content. The fuel moisture was determined from the ratio of the difference between the wet and dry mass on a dry basis, which was transformed into a percentage.

Fuel consumption information

Eight fuel consumption parcels were drawn and demarcated, similarly to the demarcation of the fuel parcels. After sampling the fuel, controlled burns were conducted in each data sampling unit, and basic information was collected to determine the fire behavior. This study focused on analyzing the intrinsic characteristics of the fuel and the time of collection that best contributed to its consumption by fire.

After the controlled burns, the waste materials were collected. Fuel consumption was quantified for plots with an area of 0.25 m², from which all the residual material was collected. The finer residues deposited on the ground were removed with the aid of a brush, a shovel, and a sieve to filter any soil particles that might cause variations in the sample mass.

The consumption samples were placed in plastic bags, identified, and sent to the laboratory for further drying in an oven until they reached a constant mass. The dry sample masses were then measured using a precision scale, and the amount of dry waste material in the fuel was determined.

Statistical analysis of the data

The statistical analyses of the data obtained in this work were carried out using the R Statistical Software (v4.1.0; R Core Team, 2021). To understand the relationship between the moisture of the fuel and its changes according to variations in the temperature and relative air humidity, as well as the relationship between destructive and non-destructive characteristics, Pearson’s linear correlation analysis was performed. This analysis was performed with all fuel variables obtained, aiming to identify those with significant correlations (p<0.05; p<0.01) (Parresol et al. 2012).

Knowing the parameters with the highest correlation, adjustments were made to the regression equations for estimating the load and moisture variables of the fuel classes, which were classified as dependent variables. Since they showed no significant quantities, no statistical analyses of correlation and regression were performed for 10-h live, 100-h live, 10-h dead, and 100-h dead fuels.

A stepwise method was used to adjust the linear regression equations. The best equations were chosen by analyzing the highest values for the adjusted coefficients of determination (R²aj) and the lowest values for both the absolute standard error of the estimate (Syx) and its percentage form (Syx%). To analyze the residuals, standardized residuals with a mean of zero and a variance approximately equal to one were used. Values outside the range of -3 to 3 were considered as outliers.

To assess the percentage (%) of fuel consumed by fire (Cos) and the dry waste material (Dw in Mg.ha^-1), a principal components analysis (PCA) was employed, which was based on the correlation matrix (Battaglia et al., 2010; Santos et al., 2018). Adjustments were also made to the linear regression models to estimate Cos and Dw based on the studied destructive and non-destructive variables. The PCA aimed to determine the fuel variables (before the fire) that most influenced the amount of fuel consumed, such as grass height (Htg), litter height (Htl), dead grass fuel load (Dg), 1-h dead fuel (Df1), and total dead fuel (Td). Thus, we considered separating the fuel by age characteristics (1 to 4 years), by month of collection (May, June, August, and September), and by fuel class.

Linear correlations

The total fuel load, with a coefficient of variation (CV%) = 12.34, ranged from 1.93 to 11.77 Mg.ha^-1. The surface fuel load values ranged from 0.19 to 1.97 Mg.ha^-1 for live grass fuel (Lg; CV%=7.00) and from 0.3 to 3.9 Mg.ha^-1 for 1-h live fuel (CV%=26.15). The estimates of dead grass fuel (Dg; CV%=7.62) ranged from 0.0 to 2.8 Mg.ha^-1 and, for 1-h dead fuel (CV%=27.07), a variation of 0.17-4.1 Mg.ha^-1 was observed.

Table 1 shows the Pearson linear correlation coefficients between the fuel load (total and by diameter class) and the variables of the non-destructive method, i.e., grass height (Htg), litter height (Htl), number of species (Nsp), and number of individuals (Nin).

Table 1: Linear correlations between fuel load and non-destructive characteristics, as obtained during evaluations conducted in the region of Jalapão, Tocantins, Brazil

* p < 0.05; ** p < 0.01

The highest correlations for the non-destructive characteristics were observed between the dead and the total fuel variables. Thus, for dead grass fuel, significant correlations (p<0.01) with all non-destructive characteristics were observed. For live fuels, less significant correlations with non-destructive characteristics were noted when compared to dead fuels. The number of species and the number of individual variables showed significant inverse correlations with certain dead fuel variables, particularly the grass fuel load and the dead fuel subtotal.

Linear regression analysis

Table 2 presents the adjustments of the linear regression equations for the indirect estimation of the fuel load variables in their classes. In general, the dead fuel variables (dead grass fuel, 1-h dead fuel, and total dead fuel) showed better results with respect to the adjusted coefficient of determination (R²aj) and the standard prediction error in absolute and percentage form (Syx and Syx%). This, in comparison with the live fuel variables.

Table 2: Adjusted regression equations for estimating the fuel load during evaluations conducted in the Jalapão region, Tocantins, Brazil

Note: Htg = grass height; Htl = litter height; Nin = number of individuals; Nsp = number of species; R²aj = adjusted coefficient of determination; Syx = standard prediction error; Syx% = percent standard prediction error; and ε = model error

The equation for estimating the total dead fuel (Td, p<0.001) reached the highest value of R²aj, with a Syx% of 28.92. Dead grass fuel and 1-h dead fuel showed similar R²aj values, with Syx% = 39.09 and 36.02, respectively. In the equation adjusted for estimating the total fuel, the value of R²aj was close to those observed for dead grass fuel and 1-h dead fuel, but with the lowest value of Syx% (23.01). Figure 1 shows the graphs with the distribution of the standardized residuals of the equations adjusted to estimate the load of the different fuel variables under study.

Figure 1: Distribution of standardized residuals of the equations adjusted for estimating fuel load variables: a) residuals of the best equation for estimating 1-h live fuel; b) residuals of the best equation for estimating dead grass fuel; c) residuals of the best equation for estimating 1-h dead fuel; d) residuals of the best equation for estimating total live fuel; e) residuals of the best equation for estimating total dead fuel; f) residuals of the best equation for estimating the total fuel load (live and dead)

The least significant values for the coefficient of determination were found in the adjusted equations for estimating live fuels. For the 10-h and 100-h live and dead wood fuel classes, the regression models were not adjusted, since they did not exhibit a significant load.

Relationship between fuel moisture and meteorological variables

Table 3 shows the results of the correlation between the moisture estimates of the main fuel classes (predominant in the area) and the local meteorological variables at the time of collection, namely ambient temperature (°C) and relative air humidity (%). No significant difference was observed between temperature and humidity for live grass fuel (Lg), total live fuel (Tl), and total fuel.

Table 3: Correlations between the humidity of the fuel classes and air temperature and relative humidity, as obtained during evaluations conducted in the region of Jalapão, Tocantins, Brazil

Note: Lg (U%) = percent moisture of live grass fuel; 1-h live fuel (U%) = percent moisture of 1-h live fuel; Dg (U%) = percent moisture of dead grass fuel; 1-h dead fuel (U%) = percent moisture of 1-h dead fuel; Tl (U%) = percent moisture of the total live fuel; Td (U%) = percent moisture of the total dead fuel; Total (U%) = percent moisture of the total fuel. * p < 0.05; ** p < 0.01

Despite the statistical indications in Table 3, the correlations were not significant. Therefore, the humidity of the fuel classes in the campestral Cerrado’s phytophysiognomy cannot be explained by instantaneous variation, i.e., by rapid changes between samplings with regard to temperature and relative air humidity, as a function of the short fuel moisture variation range.

Table 4 shows the adjustments of the linear regression models to estimate fuel moisture for different classes as a function of air temperature and relative humidity. In general, the R²aj values were low and similar, ranging from 0.09 to 0.14.

Table 4: Adjusted regression equations to estimate fuel moisture from relative humidity and air temperature, as obtained during evaluations conducted in the Jalapão region, Tocantins, Brazil

Note: AT = air temperature; RH = relative humidity; R²aj = adjusted coefficient of determination; Syx = standard prediction error; Syx% = percent standard prediction error; ε = model error.

Fuel consumption

The areas reporting one year without burning did not ignite during the experiment, for a total of 32 sampling units in which fire did not spread, considering plots with 0% consumption.

The fuel consumption values (%) were higher in areas reporting four years without burning for collections performed in the first month of the dry season (May), as well as in areas with two, three, and four years without burning for collections conducted in the last month of the dry season (September).

The four principal components were observed when analyzing fuel consumption (Figure 2). Component 1 explained 41.9% of the variation and was best represented by grass height (Htg), dead grass fuel (Dg), dry waste material (Dw), and fuel consumed by fire (Cos). Component 2 explained 21.3% of the variation, represented by litter height (Htl), 1-h dead fuel, and total dead fuel (Td). Component 3, with 11.2%, corresponded to live fuel load variables: live grass fuel (Lg), 1-h live fuel, and total live fuel (Tl). Component 4 explained 9.6% of the variation, represented by the non-destructively obtained fuel variables: number of species (Nsp) and number of individuals (Nin). Inside the boxes, the treatments of the study were identified in the following order: (1) month of fuel collection and (2) years without burning. The fuel collection months were denoted by the following numbers: (1) May, (2) June, (3) August, and (4) September. Burn-free periods were represented by the number of years without burning, ranging from one to four years.

Figure 2: Principal components representation (biplot), considering the consumption of fuel obtained during evaluations conducted in the region of Jalapão, Tocantins, Brazil

Table 5 presents the adjusted linear regression models for estimating fuel consumption as a function of non-destructive (Htg, Htl, Nsp, and Nin) and destructive (Dg, Lg, 1-h dead fuel, 1-h live fuel, Td, Tl, and total fuel load) variables from the fuel load collected before burning.

Table 5: Adjusted regression equations to estimate percent consumption and dry waste material, based on destructive and non-destructive variables associated with the fuel from the Jalapão region, Tocantins, Brazil

Note: Cos = fuel consumed by fire (%); Dw = dry waste material (Mg.ha^-1); Htg = grass height; Htl = litter height; Nsp = number of individuals; Nsp = number of species; Dg = dead grass fuel (Mg.ha^-1); Lg = live grass fuel (Mg.ha^-1); Total = total fuel load (Mg.ha^-1); R²aj = adjusted coefficient of determination; Syx = standard prediction error; Syx% = percent standard prediction error; ε = model error.

For the percent fuel consumption (Cos), the highest adjusted R² value was obtained with the model that considered all variables collected before the fire (both destructive and non-destructive). However, for the dry waste material (Mg.ha^-1), the adjusted R² value was lower than that of the model for estimating consumption, considering that it was adjusted with destructive and non-destructive variables.

Linear correlations

By analyzing the relationship between the different surface fuel types and classes (Table 1), low and null correlations were observed between the variables obtained via non-destructive methods and live fuel classes. This behavior is due to the time of collection, i.e., the dry season, when live fuels exhibit their lowest proportions and lower response capabilities regarding variations in indirect factors. An example of this is the null correlation between live grass fuel (Lg) and grass height (Htg).

As for dead fuels, which included fine materials of 1-h live fuel less than 0.7 cm in diameter, as well as dead grass fuel, good correlations with non-destructive characteristics were noted (Table 1). For dead grass fuel (Dg), inverse and significant relationships were observed between the number of individuals and the number of species at the location. That is to say that, as the shading by grassland individuals increases, the total number of individuals and species decreases due to increased competition for light, water, physical space, and nutrients, in addition to certain pioneer species’ lower tolerance to shading, preventing the development of new individuals (Parresol et al., 2012; Soares et al., 2017). After the fire has passed, the cycle starts again with an increase in the number of individuals, and, as the biomass of grass individuals increases, the number of individuals and the number of species per area decreases as a result of the aforementioned competition, with the consequent shading of intolerant species. The species in this study belonged to the herbaceous subshrub stratum of the Cerrado grassland biome, which is characterized as being predominantly pioneering and intermediate, with little tolerance to shading (Hoffmann et al. 2012).

A statistically significant correlation (p<0.01) between litter height (Htl) and 1-h dead fuel was observed, corroborating the hypothesis of a relationship between these variables (Table 1). This is because Htl is directly associated with the presence of woody debris and litter. We also noted a significant correlation between Htl, the total amount of fuel, and the total amount of dead fuel. In this regard Souza et al. (2002)observed a less significant correlation (r= 0.60) between the average height of the needle layer and the 1-h fuel load. Beutling et al. (2012) showed a correlation of r = 0.56 between 10-h fuel load and litter height. Another significant correlation with a non-destructive characteristic of fuel was that between the dead grass fuel load and grass height, which represents an indirect parameter of the vegetation that can be used to estimate Dg.

Linear regression analysis

The adjusted equations for estimating the fuel load in different classes behaved similarly to the aforementioned correlations, with the highest R²aj values observed when estimating dead fuels (Table 2). In pine crops in southern Brazil, Beutling et al. (2012) obtained R²aj values of 0.59-0.68 from linear and non-linear equations for estimating the total load. Considering the age of shrub fuel in eucalyptus forests, Gould et al. (2011) presented non-linear regression equations with R²aj values between 0.70 and 0.74 to obtain the fuel load. In pine forests in the United States, Parresol et al. (2012) presented R²aj values ranging from 0.026 to 0.97 for load estimation in different types of fuel, while Battaglia et al. (2010), working with mixed coniferous forests, obtained R²aj values ranging from 0.37 to 0.94 . We noted a scarcity of studies in the Cerrado biome with regard to the adjustment of regression relationships in order to estimate the fuel variables of its campestral phytophysiognomy.

Relationship between fuel moisture and meteorological variables

Despite the lack of studies on the influence of temperature and relative air humidity on changes in the fuel moisture content of the Cerrado, the results presented in this work were not satisfactory, as the regression relationships established to estimate fuel moisture had very low R²aj values (Table 4). However, in a similar study in southern Brazil in an area with Pinus plantations, Alves et al. (2009) tested two different methodologies, one with the plastic basket method and the other with independent samples, obtaining a superior and positive correlation with relative humidity (r=0.40 and 0.81) and a higher negative correlation with temperature (r =-0.45 and -0.79) than that of this study. Meanwhile, in Taiwan red pine forests, Lin (2004) found values similar to ours while considering temperature (r=-0.34) and higher values when considering relative humidity (r=-0.73). Among the few studies with dead grass fuel (buttongrass moorlands), Marsden-Smedley and Catchpole (2001) found higher correlations between temperature and relative humidity (r=-0.78 and 0.82, respectively). In all the aforementioned studies, regression models were built to indirectly estimate the moisture content of the fuel. However, these works consider fuel with more homogeneous characteristics than the surface fuel used in our work, which exhibits considerably heterogeneous and variable characteristics. Thus, our regression equations, adjusted to estimate the moisture content of the fuel, showed low R²aj values.

The moisture response of the fuels from the Cerrado (Table 3) can be explained by the small variation range of the fuel’s moisture conditions in response to temperature and relative humidity changes over short periods (between one collection and another). According to Schroeder and Buck (1970), the moisture content of live and dead fuels has different water retention mechanisms and responses to the environment. According to these authors, herbaceous perennial plants have deep and strong root systems and are less sensitive to short-term changes in temperature and soil surface moisture. Schmidt et al. (2017) highlighted that the low intensity of fire in the grassy areas of humid regions in the Cerrado biome (veredas), whose soil water availability is high, allows perennial plants to retain high levels of moisture even during the most critical period of the year, i.e., the dry season. The type of fuel used in this work, which mostly belongs to the herbaceous-shrub stratum, was either physiologically active or at least capable of exchanging moisture with the soil through its roots. Therefore, it may have been less able to respond to rapid changes in temperature and relative humidity.

The studies by Lin (2004)and Alves et al. (2009) used senescent pine needles deposited in the soil, which no longer engaged in any type of physiological activity. Marsden-Smedley and Catchpole (2001) also performed a study on dead grass vegetation, albeit using a homogeneously distributed species that was typical of the area. Another factor that may have contributed to the low ratios shown in Table 4 may be the heterogeneity of the fuel present in the Cerrado countryside (Castro & Kauffman, 1998), in addition to the different physiological states of the plant material, as discussed above. Despite the small fuel moisture variations in the instantaneous response to weather changes, a longer time frame improves the moisture response. An example of this is the moisture behavior of the fuel from the beginning to the end of the dry season. However, there is a need for further studies on surface fuels from campestral areas to reach more accurate conclusions.

Fuel consumption

Regarding the results of the PCA on fuel consumption, component 1 (Figure 2) showed that the highest consumption (Cos) and the lowest amount of dry waste material (Dw) after burning were closely related to the presence of dead grass fuel (Dg) and to grass height (Htg) in the final months of the dry season (August and September) and in longer periods without burning. The percent rate of consumption and the amount of post-burning dry waste material were in opposite positions, i.e., they behaved inversely. The influences of dead grass fuel, grassland height, and various collection times across periods without burning are demonstrated by the high (positive) values of the representative factors.

The results show a segregation between the dead fuels from the herbaceous layer and the other dead fuel classes, wherein dead grass fuel had a greater influence on fire consumption. These results demonstrate the high flammability of grass fuels in the area (Hoffmann et al., 2012; Santos et al., 2018), which are spatially well-distributed and whose leaves have a high surface-to-volume ratio, facilitating heat absorption (Ganteaume et al., 2013; Molina et al., 2017) and moisture exchange with the environment (Schroeder & Buck, 1970; Dios et al., 2015).

As for the consumption of fuel by fire in the Cerrado, Castro and Kauffman (1998) found higher a consumption in phytophysiognomies related to a greater presence of grass fuel. These authors highlighted that grass fuels were consumed more than others and that the high rates of energy released by fire, along with a more complete combustion, were directly related to the predominance of grass fuel. Therefore, it is important to understand the characteristics of the fuel, which, together with meteorological and topographical conditions, define the severity of a fire and, consequently, the amount of fuel consumed (Ottmar, 2014; Lydersen et al., 2014; Prichard et al., 2014).