Fire is a widespread seasonal phenomenon in savanna ecosystems (Kugbe et al., 2012). In most African savannas, natural or human-induced fires (Archibald et al., 2012) have long been recognized as beneficial for the maintenance of the tree/grass balance (Frost et al., 1986).   Fires  control  vegetation  structure,  composition, succession and productivity (Midgley et al., 2010), and nutrient budgets and cycling (Certini, 2005). Although savannas are resilient to fire (Frost et al., 1986), frequency and unplanned fires may have long-term consequences for belowground sustainability. Depending on  fire regime, changes in belowground components can be either beneficial or deleterious to the entire ecosystem. Soil is an important component of ecosystems which is liable to be affected by fire (Badia et al., 2014), particularly the superficial horizon. Fire may produce a wide array of changes in other ecosystem components owing to their interdependency. The effect of fire on belowground systems appears to be related to burning with greater losses of nutrient occurring in hotter fires (Whelan, 1995). Immediately after the fire, soil desiccation and the change in albedo of the surface may result in high soil temperatures, which further influence the effect of fire on the soil (DeBano et al., 1999). Frequent fires may have a detrimental effect on soil properties by destroying the organic matter contained in the aboveground vegetation and litter (Stoof et al., 2010). The effect of fire on the physical soil properties is primarily due to the indirect effect of biomass removal (Verma, 2012), which is particularly related to the period and intensity of fire exposure, and parameters such as soil structure and texture.

The effect of fire on chemical soil properties is complex and variable, but generally, the changes are most pronounced in the topsoil (Snyman, 2003). Recent information has indicated that the use of prescribed fire for land-management purposes and the incidence and severity of wildfires have increased substantially worldwide (Pricope and Binford, 2012). In the Sudanian savanna-woodland alone, Delmas et al. (1991) have estimated that between 25 and 50% of the total area burns annually primarily due to anthropogenic causes. In most protected areas of Burkina Faso, officially approved early fire (burning taking place between October and December-January depending on the locality, region) has been adopted as a management tool to minimize the risk of severe late fire (occurring from February to May), to improve pasture production for wildlife and to maintain species’ composition and richness (Bellefontaine et al., 2000). Much of the previous research on the effects of burning in West African savanna-woodland has focused on vegetation dynamics (Savadogo et al., 2008, 2009; N’Dri et al., 2014) and fire behavior (Dayamba et al., 2010; Savadogo et al.,  2007b; Sow et al., 2013; N’Dri et al., 2018), soil macro-fauna (Doamba et al., 2014) particularly the effects of burning during the dry season on soil temperature, root biomass, soil surface CO₂ efflux and soil water properties (Savadogo et al., 2007a, 2012; Van Straaten et al., 2019). Consequently, relatively little is known about the long-term legacy of annual burning on soil properties. There is also a lack of information about the effects of fuel load and fire seasonality on soil properties. More specifically, our aim was to determine the modifications induced in soil physicochemical properties in plots subjected to repeated fire and with varying fuel load treatments.

The study addressed the following hypotheses: (1) The fuel load through increased fire severity induces alterations   in   soil   properties,   and  nutrient  pools  will decrease from low to high fuel load treatments; (2) Fuel load treatment will affect vertical nutrient partitioning of the burned soil, that is, available nutrients will decrease from the topsoil to deeper soil levels.

Study site description

The experiment was conducted in the State forest reserve of Dindéresso (11.225°N, 4.447°W), located at an altitude of 359 m above sea level in Burkina Faso, West Africa. The climate at this location is South Sudanian, with two main seasons; the rainy season occurs from May to October and the dry season from October to April (Fontès and Guinko, 1995). The mean annual rainfall for the study period 2009-2011 was 1010 ± 145 mm (mean ± standard error). Soils at the forest reserve are derived from sedimentary elements and classified as lixisols (Driessen et al., 2001). Soils are mainly deep (> 85 cm) and have relatively high silt fractions and are classified as lixisols (FAO classification system) (van Straaten et al., 2019). Phyto-geographically, the study site is located in the Sudanian Regional Centre of Endemism in the South Sudanian Zone (Fontès and Guinko, 1995). The vegetation type at the site is tree/bush savanna with a grass layer dominated by the annual grasses Andropogon pseudapricus Stapf. and Loudetia simplex (Pilger) C.E. Hubbard as well as the perennial grasses Andropogon gayanus Kunth. and Andropogon ascinodis C.B.Cl.

Burning experiment and fuel load treatment

The investigation of soil physical and chemical properties was performed on a permanent experimental site established in August 2009 (Figure 1). The plots were located on flat ground to eliminate the influence of slope on fire behavior (Trollope et al., 2002). The total experimental area (11.16 ha) comprised three non-contiguous blocks (each 3.72 ha) (Doamba et al., 2014; van Straaten et al, 2019). The blocks were located as to minimize variations in aspect, slope and soil type after a preliminary survey. Each block was further divided into seven plots of 0.24 ha (80×30 m). The plots were separated by 10m fire-breaks and each block was surrounded by a 20 m wide fire-break. All the blocks were similar and subject to moderate grazing, mainly by livestock (a mixed herd of cattle, sheep and goats), all year round. The woody vegetation covers an average of 44% in each plot.

The fuel load treatment procedure was as follow: first, except on the control plot, the herbaceous vegetation was harvested manually by cutting at the base, approximately 10 cm above the ground. The harvested biomass and available plant litter were weighed in situ before taking the material to the laboratory, where biomass was determined by again weighing. Then it was placed in an oven for 48 h at 105 °C, and finally reweighed. Fire is applied on the same day when the cutting occurred. Based on the dry weight of the material from each plot, three fuel load treatments were considered prior to early and late burning:

(i) Normal fuel load (NF), the initial available fuel in the plot was cut and left having been spread uniformly across the plot to ensure 100% fuel bed continuity by covering any areas of bare soil;

(ii) Reduced fuel load treatment (RF): half of the initial fuel load was left on the plot and the rest removed;

(iii) Increased fuel load (IF): the second half of the fuel removed from RF was added to the initial fuel collected on the IF plots, all of it was spread out evenly.

Further, the following fire treatments, each  with  3  replicates,  were randomly assigned to the 21 plots starting in 2009: no fire: control plots; early fire: fire set at the beginning of the dry season in December  and late fire: fire set at the beginning of the dry season and just before the next rainy season (Doamba et al., 2014). However, in this study, the activities concerned only early fire and control plots. In each season, fire was initiated early in the morning (6 am to 9 am) when the wind speed and air temperature were lowest. Data were collected at regular intervals, before and immediately after fire event, in three consecutive years. At the same time, sampling was conducted in no fire (control plots) and early fire.

Fire behavior data collection

The flame temperature of fire was recorded using a universal professional infrared pyrometer with a temperature range of -50 to +1000°C and a high optical resolution of 30:1 (model TP 6 TROTEC, MIS instrumentation, France). The rate of spread was determined by recording the time the fire front took to arrive at a selected distance on either side of the burning plot by using stopwatch. Fire intensity was estimated using Byram´s (1959) Equation 1:

I = H w r                                                                                   (1)

where I = fire intensity (kJ s^-1 m^-1), H = heat yield of the fuel (kJ kg^-1), w = weight of fuel consumed per unit area (kg m^-2) and r = rate of spread (m s^-1). The heat value (H) developed for grass fuel head fires (17 781 kJ kg^-1) (Trollope and Potgieter, 1985) was used to calculate fire intensity.

Soil sampling and analysis

Soil samples were collected before and immediately after the fire event on burnt plots. On control plots, only one sample was collected, on the assumption that the study period was short enough that no substantial change could have occurred. Litter and ash were removed prior to sampling. Soil samples were taken on the diagonal of the plot in order to obtain a good representation of the sample plot. On the two diagonal of each plot, soil samples were collected from five holes by using an auger. In each hole, three depth layers were considered (0-5, 5-10 and 10-15 cm). Finally, three composites samples were taken in each plot and were bagged and labeled. The following physical and chemical parameters were determined according to standard procedures for soil analysis:

(i) Soil particle size distribution (percent sand, silt and clay) was measured according to the procedure described by Feller (1979).

(ii) Soil pH was measured with a pH meter using soil: soil water ratio of 1/2.5 according to the Afnor (1981) procedure.

(iii) Soil organic carbon was determined after oxidation of organic matter following the procedure described by Walkley and Black (1934).

(iv) For Nt, and Pt, samples were first digested in a mixture of H₂SO₄–Se–H₂O₂ at 450°C for 4h following the method of Walinga et al. (1989). We used aspectrophotometer (CECIL instrument, CE 3020, Serial N’126-288, Cambridge, U.K.) to determine the Nt and Pt contents in the digested solution.

(v) Available P was measured using a Bray-1 extract as described by Olsen (1965).

(vi) Cation exchange capacity (CEC) was determined according to Metson (1956).

Statistical analysis

An analysis of variance (ANOVA) was performed to study the  effect of fuel load treatment on the rate of spread, fire temperature and fire line intensity. Fire behavior variables were tested for normality prior to statistical analysis. For soil physical and chemical parameters, data were collected based on pairs of observations before and after fire treatments (N pairs of samples) in burned plots and one set of samples for the control plot. Therefore, a Before-After Control-Impact (BACI)-ANOVA approach (Hewitt et al., 2001) was used to examine the modifications induced in soil properties. This approach is designed to detect situations where an environmental disturbance, in this case fire occurrence in a given season, causes a pattern of change in a measured parameter. The data were analyzed using an ANOVA generalized linear model or repeated measures methods with special structural parameters in the covariance matrices to take into account the carry-over effect. The set of data analyses was used to assess the behavior of the two groups of samples from the impact area (before and after) and the reference area (control) through the variable responses observed through time. The model (1) was used for fire affected plots and model (2) for control plots:

Y_ijk = μ + β_i + λ_j + (β λ)_ij + ε_j(i) + ε_j(k)                                                                  (1)

Y_ij = μ + Depth_i+ λ_j + Depth_i× λ_j + ε_ij                                                       (2)

where Y_ijk and Y_ij represent the response variables (values recorded before and after the fire) for soil parameters, μ is the overall mean, β_i is the effect of the between-subject factors, i (fuel load treatment, depth of soil sampling, the pre- and post-fire test and their interaction), λ_j is the effect of the within-subject factor, j, year, and (βλ)_ij is the interaction of the between- and within-subject factors. The parameters ε_j(i) and ε_j(k) are the random unobservable errors of the between-subject and the within-subject factors, respectively and are assumed to be normally distributed. For the soil particle size distribution, data were only available for the first year and were analyzed using the following linear model (3):

Y_ijk = μ + Treatment_i + Depth_j + Test_k + (Treatment_i × Depth_j) + (Depth_j × Test_k) + Treatment_i × Depth_j ×Test_k + ε_ijk      (3)

When the homogeneity of variance assumption was violated in the time series analysis according to Mauchly’s test of sphericity, the degrees of freedom for testing the significance of the within-subject factors were adjusted using a Huynh-Feldt correction factor, which is less biased than other correction factors (Davis, 2002). When a significant difference was detected, a pairwise comparison was made using Tukey’s test at the 5% level of significance. All statistical analyses were performed using SPSS 19.0 for Windows (IBM Corporation, USA). Graphs were plotted using Origin 7.5 software (Origin Lab Corporation, Northampton, MA, USA).

Fire behavior

There was an inter-annual variation in the pre-burn fuel load, which could be  explained  by  the variability in the amount and distribution of rainfall observed during the course of the study. In the NF, without any manipulation of biomass, the total fuel load varied to 0.48 ± 0.23; 1.40 ± 0.70; 0.4±0.12 t/ha for 2009; 2010; 2011 respectively (Figure 3). Therefore, the wettest year seems to induce an important production of fuel (Figure 2). This trend concurs with Govender et al. (2006) that observed increase in fuel load increased with rainfall in a range of only two consecutive years.

Despite the variation in fuel load, only the first year    of    study   (2009)   presented   significant variations in fire intensity. This parameter increased significantly in increased fuel load plot. The biomasses obtained from the different plots were low compared to other studies. We obtained biomasses ranging from 0.75 to 1.56 t/ha, whereas on sites protected from fires and animal grazing. Previous studies reported phytomass ranging from 2.5 to 4 t/ha for Andropogon gayanus fallow and Andropogon ascinodis/Loudetia togoensis savanna, respectively (Sow et al., 2013). As the protected forest of Dindéresso lies close to an urban area, it  is  subject  to frequent grass biomass removal, namely for animal feeding and for handicraft. This may explain the low biomass obtained. The highest measured temperature reached 657±36^oC. Similar temperatures were also obtained at 0 cm on the soil surface by previous studies (Savadogo et al., 2012). The temperature decreased between 24°C and 40°C at 10 cm from soil surface (Dayamba et al., 2010; Savadogo et al., 2012). In this study, the temperature was measured only in the flaming front. This temperature emitted from the combustion of biomass is transferred to the mineral soil layers (DeBano et al., 1999). Many studies have reported that overall fire temperature increases with fuel load (Hély and Alleaume, 2006) and that fire behavior is determined by fuel physical properties and weather parameters. For instance, the positive effect of wind speed has been attributed to an enhanced supply of oxygen to the fire (Trollope et al., 2004), which stimulates heat transfer by conduction or radiation, resulting in pre-heating of the fuel ahead of the fire front (Savadogo et al., 2007b). Wind direction and speed are known to affect the rate of fire spread due to tilting of the flames towards the fuel ahead of the fire (Sow et al., 2013). In the present study, fires were burnt under a relatively narrow range of weather conditions. Therefore, no such effect was observed. However, an interannual variation in fire intensity was observed due to the annual variation of rainfall that affects the amount of fuel. The cutting of the fuel and its homogeneous distribution on the ground before setting the fire could explain the observed lack of difference in rate of spread. Indeed the main characteristic of fuel that influences the spread of fire is the continuity of the fuel bed, that is, how unbroken the fuel bed is (Cheney and Sullivan, 2008). Indeed, this practice results in better distribution and less spatial heterogeneity in the fuel bed on the ground surface, promoting combustion that is more continuous and progresses with constant speed.

Textural and soil chemical properties changes induced by fires

The results show that, generally, burning has no effect on the particle size distribution, which is supported by Oswald et al. (1999), who reported that burning does not cause dramatic changes in soil texture. The main components contributing to soil texture (sand, silt, and clay) have high temperature thresholds and are not usually affected by fire (DeBano and Neary 2005), because most fires do not cause sufficient soil heating to produce significant changes in soil physical properties. This is particularly true for low intensity prescribed fires in savanna because elevated soil temperatures are usually brief and not depth penetrating (Savadogo et al., 2007a).

Contrary to expectations, the fuel load treatment during early fire did not significantly modify the chemical characteristics of the burnt soil, such as pH, Soil organic C, C/N, total nitrogen, total  phosphorus  and  CEC.  This, can primarily be explained by the fact that, despite the variation in the quantity of fuel (normal, increased or reduced), the measured temperatures were statistically similar. The values of soil pH, Soil organic C, Nt, CEC were significantly dependent on the depth of soil sampling. The increase in chemical elements on the soil superficial horizon can be explained by nutrient enrichment due to ash deposition. Indeed, in the post-fire environment, some of the nutrients accumulated in the aboveground biomass and litter are deposited as ash, which contains various amounts of available nutrients depending on the fire severity (Gimeno-Garcia et al., 2000). Black ashes are known to be rich in organic matter, containing more than 90%, whereas grey ashes contain about 12 to 55% (Gimeno-Garcia et al., 2000). Moreover, dead roots and twigs could provide an additional source of carbon to the soil (Knicker et al., 2005). However, the fire did not induce significant changes in the amount of total carbon. These results are in agreement with the previous studies (Jensen, 2001). Further, analysis of the results showed that the values of pH, total nitrogen and total phosphorus decreased significantly after the passage of fire, unlike in previous studies (Aref et al., 2011). An increase in pH has been commonly explained by the presence of ash and accumulation of potassium, sodium hydroxide, magnesium, calcium carbonate, as well as the destruction of certain groups of acids in the organic matter (Schafer and Mack, 2010). However, Giovannini et al. (1990) consider that the pH value is related to the combustion temperature; higher temperatures (˃ 460°C) during the fire induce increase in pH whereas lower temperatures (<220°C) result in decrease of pH. The amounts of nitrogen were also reduced after fire, most likely due to rapid loss by volatilization. Indeed, nitrogen is readily lost by volatilization at 200°C in the form of gas and ash (Christensen, 1994), while at 500°C, more than half of the nitrogen is lost to the atmosphere (Knicker et al., 2005). Phosphorus may also be lost by volatilization (Prieto-Fernandez, 2004). Further, the results indicated a significant inter-annual variation of pH, CEC and Pt. Since the impact of fire depends on the environmental conditions during burning, the inter-annual variation in fuel load and the rainfall variation over the three years of study may explain the observed variation in the effects of fire on the soil parameters.

Fire is a widespread process in the earth system and plays a key role in ecosystem composition distribution and function. The aim of this study is to evaluate the modification induced in soil physicochemical properties by repeated and prescribed burning for different fuel load treatment. The result of this study showed some variations   with   depth  of  soil  sampling  and  superficial layers have got the highest nutritive elements. This could be attributed to the release of nutrients accumulated in the aboveground biomass and litter after deposition as ashes, which contain different amounts of nutrients depending on the severity of fires. Furthermore, there was no effect of fuel load treatment on the soil properties. The biomass is not abundant in these savannah regions. Therefore, fire apply early does not affect directly and in the short term, the chemical and physical properties of the soil. However, a long-term study should be carried out in order to be able to identify the possible long-term impact on soil physicochemical properties.

The findings from the present study have an important implication for current savanna–woodland management: early fire seems to be a good management option from a soil perspective, striking a balance between total fire protection (which is utopian) and the occurrence of intense late fire. However, studies on the late season burns need to be thorough and globally on environment could be more damaging. Nevertheless, the application of annual early fire should continue with due consideration of the timing of burning, weather conditions and other possible factors related to fuel properties that influence fire intensity.

The authors have not declared any conflict of interests.

Funding for this study was provided by International Foundation for Science (Grant Agreement No. C/4816-1). The author appreciates Amoro Ouattara and Pascal T. Simporé for their help during the soil sampling and analysis. Thanks are also due to Dr Zacharia Gnankambary and Dr Jacques Gignoux for reviewing the early version of the manuscript. They are grateful to Dr John Blackwell and Sees–editing Ltd for linguistic improvements.