When soil becomes fuel: identifying a safe window for prescribed burning of Tasmanian vegetation growing on organic soils

Introduction

Organic soils are globally important because they contain an immense store of carbon and can play a critical role in regional hydrology, acting as both a ‘sponge’ and a filter to maintain water quality and quantity (Joosten 2009; Turetsky et al. 2015; Goldstein et al. 2020; Wilson et al. 2022). Organic soils are defined as soils containing more than 12–18% carbon depending on clay content (Isbell 2002). Organic soils are rare in mainland Australia (Whinam and Hope 2005; Joosten 2009), but in the island state of Tasmania, 15% (approximately 1 million ha) has been mapped as having organosols (Cotching et al. 2009). These occur mostly in western Tasmania, and include the most extensive blanket bogs in the southern hemisphere (Bridle et al. 2003; Whinam and Hope 2005; di Folco 2007). The accumulation of organic matter that forms organic soils and various peat landforms is recognised as part of the Outstanding Universal Values of the Tasmanian Wilderness World Heritage Area. Tasmanian organic soils present a challenge for fire managers because of their embodied conservation values, the sharply contrasting fire ecology of vegetation types that grow on them and their vulnerability to fire damage; soil fire can cause significant and permanent ecosystem change in a single event (Hill 1982; Bowman et al. 2021; Prior et al. 2023). Organic soils form so slowly (0.4–4.2 mm year⁻¹) that if burnt, it can take decades to centuries for them to recover (Wood et al. 2011). In addition to the ecological and human health damage caused by soil fires (Hill 1982; Bridle et al. 2003; Johnston et al. 2012), organic soils are a troublesome fuel. Once ignited they can be extremely difficult to extinguish, and can smoulder for months, flaring up to ignite surface fuels when weather becomes more favourable for vegetation fires (Lin et al. 2019, Santoso et al. 2019). Unfortunately, extinguishing hotspots can be time consuming, expensive, and often impractical.

Understanding the drivers of combustibility is key to predicting when organic soils are vulnerable to ignition. Such information is needed for fire managers to assess risks when planning fuel reduction burns, and to triage suppression efforts when there are multiple fires. It is widely accepted that soil moisture is a major control on the combustibility of organic soils (Frandsen 1987; Garlough and Keyes 2011; Prat-Guitart et al. 2016; Prior et al. 2020). Laboratory studies have been invaluable in determining how the risk of soil ignition varies with individual factors such as soil moisture content, soil bulk density and organic matter content (Frandsen 1997; Garlough and Keyes 2011; Prat-Guitart et al. 2016). However, scaling up the results of laboratory studies to make predictions at the landscape level is fraught with difficulty, because landscapes are topographically heterogeneous, such that soils and soil moisture can vary over fine spatial scales (di Folco and Kirkpatrick 2011; Lukenbach et al. 2015; Hokanson et al. 2016). One example where scaling up was attempted was when Prior et al. (2020) derived soil moisture thresholds from their laboratory studies, and then estimated the volumetric water content of Tasmanian organic soils using a geographic information system and a high-resolution climate database. This showed that most Tasmanian organic soils are likely to be combustible at some time almost every summer, but there was no field validation of this inference.

Thus, while laboratory studies can indicate the factors controlling soil combustibility, they need to be complemented with field observations of when organic soils are likely to ignite. Direct field observations are challenging given the intractable terrains that support organic soils, the stochastic occurrence of wildfires, and practical difficulties in collecting data during fire management operations.

Based on a preliminary study of fire behaviour, the potential for peat fires in flat buttongrass moorland has been considered to be low, because the soil moisture content is typically high throughout the year (Marsden-Smedley 1993). Nonetheless, peat fires do occur in buttongrass moorland, possibly initiated at the base of large shrubs and copses of small trees (Marsden-Smedley 1993; French et al. 2016). The occurrence of organic soil fires has become an important management issue with guidelines recommending that planned burning in buttongrass occurs in spring and autumn, under mild fire weather conditions and when soils are wet, although predictions of fire risk based on soil dryness index (SDI) have been described as unreliable for buttongrass soils (Marsden-Smedley 2009). This study is motivated to advance understanding of the fire risk to organic soils. We investigated when it is safe to burn Tasmanian vegetation growing on organic soils by assembling a dataset based on field records of vegetation fires (both management fires and wildfires), some of which resulted in soil fires, while others did not. Our dataset has good spatial coverage (Fig. 1), and spans two decades (May 2001 to August 2021). We note that despite best efforts and close engagement with fire managers, practical constraints made this is a very incomplete record of fires within that period, and there are likely biases in the data that we discuss and propose ways of overcoming in the future. The incidence of soil fire is likely to be over-represented in the data, because it is easier to observe presence than absence of soil fire, and occurrence is also more likely to be reported. We discuss these problems and develop recommendations for future work.

Fig. 1.

(a) Locations of study sites (99 burnt, 43 unburnt) and extent of organic soil in Tasmania, Australia. The boundary of the Tasmanian Wilderness World Heritage Area is shown in blue. (b) Terrain map of Tasmania, with the 1000-mm isohyet shown in red (higher rainfall to the west and mountains in the north-east of the island).

When land managers are planning fuel reduction burns, they are constrained not only by the risk of soil fires occurring, but also by whether the vegetation will carry a surface fire (Fig. 2). Globally, empirical research has found that the combustibility of forest vegetation can be predicted from the moisture content of dead fine fuels, which are composed of surface litter and other fine matter, and are readily ignitable (Ellis et al. 2022). The cumulative frequency of total area burnt in relation to fine fuel moisture content has been used to derive critical thresholds indicating high or extreme fire potential (Nolan et al. 2016; Ellis et al. 2022). While these thresholds are not appropriate to use for fuel reduction burns, we applied a similar approach to determine maximum fuel moisture thresholds indicating when vegetation is able to carry surface fire using available fire history data and associated estimates of fine fuel moisture using gridded climate data for Tasmania.

To identify the window for prescribed burning on organic soils across Tasmania, we: (1) investigated the probability of a vegetation fire igniting organic soils, in relation to modelled soil dryness, focussing on moorland sites. From this relationship, we proposed a ‘safe’ threshold, below which soils are unlikely to burn; (2) identified when vegetation is likely to be dry enough to carry fire, and examined the relationship between area burnt and estimated fine fuel moisture content; (3) produced maps of Tasmania showing the estimated number of days when vegetation fires were possible, and organic soils were unlikely to burn based on our selected ‘safe’ thresholds; and (4) discuss the assumptions and limitations associated with our results, and make suggestions for future improvements. In particular, we highlight the importance of comprehensively collecting and recording data regarding organic soil fires, including non-occurrence when these soils are exposed to vegetation fire. This is a vital step for adaptive management of fire on Tasmania’s organic soils.

Fig. 2.

Conceptual model showing probability of a flame starting a fire in organic soil (red) or vegetation (green) in relation to dryness. Management fires are usually planned to minimise the risk of organic soils catching fire (represented by red left-pointing arrow), yet must also be scheduled for when vegetation can carry fire (green right-pointing arrow), and are thus restricted to the narrow window (black double-headed arrow) when these conditions intersect, assuming the indicated probabilities for vegetation and soil fire. We assume a more conservative (lower) threshold for organic soil fire because the consequences of burning an organic soil are more serious than those of a management fire failing to ignite.

Materials and methods

Results

Probability of soil fire and threshold selection

Our dataset showed organic soil fires were possible over the entire range of SDI values (Fig. 3), but soils were more likely to burn when at the drier end of their range (Fig. 4). The average SDI was 60 when soils ignited, compared to 15 when they did not, although there was substantial variation in SDI for both when organic soils ignited (1.5–141) or did not (0–88). Overall, 21% of vegetation fires on organic soil, but only 7% of soil fires, occurred on the 45% of days with SDI less than or equal to 5 (Fig. 5a). Similarly, only 12% of the soil fires and 28% of the vegetation fires occurred on the 55% of days when SDI <10. However, even at an SDI of zero, across all sites the probability of soil burning when exposed to vegetation fire was higher than our proposed 20% probability threshold.

Fig. 3.

Range in soil dryness index (SDI) at each site for all days between July 2010 and June 2015 (blue is minimum, grey maximum), compared to Day 0 of each vegetation fire where the soil (a) burned (red) or (b) did not burn (green). Data are sorted by minimum Day 0 SDI. Squares show moorland sites, circles other sites. Note that although there is a definite bias towards fire starting at high SDI, there are many exceptions. Conversely, most unburnt sites had low SDI, but there were sites with moderate SDI where soil fire did not occur.

Fig. 4.

Count of sites exposed to vegetation fire according to soil dryness index (SDI), showing the counts of when fires did and did not occur (left hand panels). The corresponding proportion of those sites in which soil fires occurred is shown in the right-hand panels, with the lines indicating model predictions. Top row of panels show all sites, second row of panels show moorland sites, third row of panels show scrub sites, and bottom row of panels show forest sites. Note the different scales on the y-axes. For presentation, SDI was binned into 10-unit classes, with the average of each class used here, but actual data were used in the modelling.

Fig. 5.

Cumulative relative frequency distribution of soil dryness index (SDI) for all days, vegetation fire days and soil fire days (a) across all sites, and (b) in moorland sites. Vegetation fire days, and especially soil fire days, are skewed towards higher SDI (drier soils).

Twelve of 35 buttongrass moorland sites had soil fires, compared with 45 of 53 scrub sites and 25 of 26 forest sites. Modelling showed that in addition to SDI, vegetation (buttongrass moorland, scrub or forest) affected the probability of soil igniting if exposed to a vegetation fire (Table 1, Fig. 4, Fig. S1). The model that included both SDI and vegetation indicated soil fire was significantly more likely in scrub or forest than in buttongrass moorland when exposed to vegetation fire (Fig. 4). For buttongrass moorland (for which our dataset is most comprehensive), the probability of soil burning if exposed to vegetation fire was 10% (95% CI of 3–29%) at SDI of zero, 13% at SDI of 5 and 17% at SDI of 10. A 20% risk of soil fire was associated with an SDI of 12.9. At buttongrass moorland sites, SDI was below the existing recommended ‘optimum’ value of 5 for 54% of days, which included 49% of vegetation fires but only 9.6% of soil fires (Fig. 5b). Similarly, 65% of all days at our moorland sites have an SDI below 10, and 55% of vegetation fires but only 14% of soil fires occurred when SDI was below 10. Using the SDI value of 10 as the threshold, 80% of soil fire outcomes (positive and negative) were successfully predicted: soil fires occurred at two sites where it was predicted not to occur, and there was no soil fire at five sites where it was predicted (Table S2). We consider these results support the existing recommended SDI range of 0–10, with an optimum of 5, for planned burning in buttongrass moorlands.

Table 1.Summary of binomial generalised linear models describing the probability of an organic soil igniting when exposed to a vegetation fire in terms of Soil Dryness Index (SDI) and vegetation (VEG; buttongrass moorland, scrub or forest).

Model	k	AICc	ΔAICc	w i	Explained deviance (%)
SDI + VEG	4	74.3	0.0	0.998	52.8
SDI	2	87.0	12.7	0.002	40.3
VEG	3	106.8	32.5	0.000	27.3
Intercept only	1	139.4	65.1	0.000	NA

k, number of parameters in the model; AIC_c, Akaike’s information criterion modified for small sample size; ΔAIC_c; difference in AIC_c relative to the best model; w_i, Akaike weight for model i in the candidate set (these sum to 1), and deviance explained is relative to the intercept-only model.

Vegetation combustibility

The vegetation combustibility analysis showed that the area burnt was negligible until FMC% fell below 50%. It then increased gradually as FMC% declined further to about 25% (Fig. 6), and more rapidly below 25%. We considered vegetation to be combustible when FMC% fell below 28.9%, which corresponded to 90% of the total area burned.

Fig. 6.

Cumulative area burnt in relation to fine fuel moisture content (FMC%) for Tasmania between 1990 and 2022. The green vertical line indicates the value of fine fuel moisture content below which 90% of the total burnt area was burned; we chose to use this threshold of 28.9% to indicate when vegetation is combustible.

Maps of ‘safe’ and ‘safe combustible’ days

There was a strong east-west division in the average number of days per year below a ‘safe’ SDI threshold, with most of the west of the state having at least 200 days per year below an SDI of 5, compared with fewer than 64 days per year in much of the east (Fig. 7, Fig. S2). There was only a modest difference between a threshold of SDI 5 and 10 on the number of ‘safe’ days per year (Fig. 7, Fig. S3). However, screening out days when vegetation was not combustible altered this spatial pattern and greatly reduced the days suitable for prescribed burning (Figs. 7b, c). With an SDI threshold of 5, only a small area had more than nine suitable days on average each year, and mountainous areas in the west of the state, with shallow soils and high runoff, were rarely suitable. Safe prescribed burning was much more feasible with a threshold SDI of 10, with much of the state having more than 36 suitable days per year (Fig. 7, Fig. S3). The window for ‘safe’ prescribed burning varied considerably between years (Fig. 8). At one of the wetter buttongrass moorland sites, with an SDI threshold of 10, there were on average 53 safe combustible days per year, but this varied from 5 to 90 during the period 1990–2018 (Fig. 8b). By comparison, at one of the drier moorland sites, there were on average 26 safe combustible days per year, ranging from 14 to 39 days per year during this period.

Fig. 7.

Number of ‘safe’ days and ‘safe combustible’ days per year corresponding to soil dryness index (SDI) thresholds of 5 (considered optimal; top panels) and 10 (upper limit; lower panels). The risk of soil burning, if exposed to vegetation fire, was predicted to be 13% for SDI of 5, and 17% for SDI of 10. The corresponding ‘safe combustible’ days are those that are below the SDI threshold but fine fuels are dry enough to combust (considered to be FMC% <28.9%). Supplementary Figs S2 and S3 show the sensitivity of counts of ‘safe days’ and ‘safe combustible days’ to the selected SDI threshold.

Fig. 8.

The number of ‘safe combustible’ days each calendar year between 1990 and 2018, for sedgeland at a dry site (Surv-1, in eastern Tasmania), and a wet site (Serp-2, in western Tasmania). The upper panel corresponds to a threshold soil dryness index (SDI) of 5, the lower panel to SDI of 10. Note the different scales on the y-axis.

Discussion

Our overall aim was to develop a method to identify the window for prescribed burning on vegetation growing organic soils across Tasmania when these organic soils are unlikely to ignite. The most important finding of this study is that the safe window for prescribed burning on organic soils is limited. Our analysis highlights that burning vegetation on organic soils involves a difficult balancing act between vegetation combustibility and protecting organic soil from combustion. As we discuss below, there are caveats related to the quantity of, and probable bias in, the soil fire data and inaccuracies in the estimates of soil dryness and fine fuel moisture content at a coarse spatial scale. Nonetheless, our inference that burning vegetation on organic soils carries substantial risk is very likely to hold. Moreover, this risk is likely to increase due to anthropogenic climate change. Broadly, our findings harmonise with those of the laboratory-based study of Tasmanian organic soil combustion conducted by Prior et al. (2020).

Our work supports using the existing recommended guidelines that prescribed burning in buttongrass moorland vegetation should occur when SDI is between 0 and 10, and ideally around 5 (Marsden-Smedley 2009). At an SDI of 5, our binomial modelling showed that the risk of organic soil igniting was 13%, but this is likely an overestimate because of the dataset bias towards positive observations. We note that the number of ‘safe’ days per year was moderately sensitive to the specified SDI threshold (Fig. S2), but the number of ‘safe, combustible’ days was very sensitive to it (Fig. S3).

In buttongrass moorland, the estimated ‘safe combustible’ window ranged spatially from 8 to 35 days per year on average for an SDI threshold of 5, compared with 25–54 days for an SDI threshold of 10. While using an upper threshold of 10 offers more flexibility for conducting prescribed burns, it is associated with a slightly higher (17% vs 13%) risk of soil fire. We stress that these suggested SDI thresholds represent provisional guidelines only; managers should use their own experience of burning in this landscape, as well as prevailing weather conditions and direct measurement of soil moisture to make the final decision as to when and where to burn.

The vegetation combustibility constraint considerably narrowed the window for prescribed burning on organic soils. To indicate vegetation combustibility we selected an upper threshold of 28.9%, which is supported by Marsden-Smedley’s (1993) work showing a large increase in rate of spread of fire in buttongrass moorlands when moisture content of dead fuel dropped from 40 to 20%. The value of 28.9% is near the upper end of the 22–28% range considered to be very difficult to burn for eucalypt litter, and above that for grass to become flammable (Sullivan et al. 2012). It is therefore unlikely that the fine fuel moisture threshold could be substantially increased to widen the window for safe prescribed burning. The effects of climate change also need also be recognised; in western Tasmania there is a pronounced warming and drying trend apparent in the instrumental record with climate projections suggesting it is likely to increase the number of days when vegetation is combustible, but reduce the period when burning is safe (Grose et al. 2010; Fox-Hughes et al. 2014; Love et al. 2016). Additionally, marked interannual variation in ‘safe, combustible days’ means that prescribed burning should be avoided completely in particularly dry years. Rather than setting annual targets, fire management should be flexible, and focus on periods when conditions are safe for prescribed burning.

We acknowledge that the fires in our dataset are a very incomplete record of fires on organic soils in the last 20 years. They are necessarily biased towards prescribed burns and more accessible areas, but this is an advantage given the data will predominantly be used for planning management fires. Of more concern is that there is a likely bias towards vegetation fires that caused soil fires because it is easier to detect a soil fire than establish none occurred, and more likely that such a detection will be recalled and recorded. We stress that it is crucial to record negative observations: without them, predictions will always be biased towards higher probability of soil fire. For this reason, we advocate recording this information for every prescribed burn on organic soil.

There are many opportunities to refine the predictions presented here. First, we need more data on when soil fires do or do not occur to establish more reliable thresholds. While it is clear that when SDI is above about 40, soil fire is very likely, there is much less certainty for moister soil, when prescribed burns are scheduled. Therefore, more observations are required from days when SDI is in the low to moderate range. Second, higher spatial resolution of fires and burned soils will provide more accurate estimates of soil moisture at a given location, and potentially lead to more reliable predictions (Table S3). Fine scale spatial variation in soil moisture is likely responsible for apparently anomalous instances of soil fire occurring when SDI is 0; SDI is a coarse scale estimate and a value of 0 does not mean that soil is saturated across the entire landscape. Third, predictions are likely to be improved using a soil moisture index that better deals with the complex influences on soil hydrology. Selecting the optimal soil depth to use for moisture estimates is also likely to substantially improve predictions. Fourth, incorporating factors such as fire weather, landform, terrain and more detailed vegetation classes into the model will almost certainly improve predictions when these data are available, as they often will be in advance of prescribed burning. Compiling and analysing data from field measurements of soil (Prior et al. 2020) should complement the broad scale approach suggested here, Finally, the choice of threshold is ultimately arbitrary, although constrained by reality. In general, a conservatively low soil moisture threshold should be adopted to minimise risk of soil fire, while the choice of the FMC% threshold could be more flexible, because the consequences of vegetation not carrying fire are less serious.

Despite the limitations of our approach, this study, based on two decades of observations, represents a vital step for adaptive management of fire on Tasmania’s organic soils. We have demonstrated how simple records of date and location of vegetation fires can be used to estimate the risk of soil fire occurring. Refining the procedure described here will allow land managers to objectively plan the timing of prescribed burns on combustible organic soils, even in remote areas where access is difficult or expensive. By being aware of the associated risks, they can take precautions such as having adequate personnel and equipment available to quickly extinguish soil fires when they start. Additional data can be incorporated into the dataset and allow predictions to be more targeted to the conditions to which they will be applied.

Supplementary material

Supplementary material is available online.