By continuing to browse this site you agree to us using cookies as described in About Cookies
Notice: Wiley Online Library will be unavailable on Saturday 7th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 08.00 EDT / 13.00 BST / 17:30 IST / 20.00 SGT and Sunday 8th Oct from 03.00 EDT / 08:00 BST / 12:30 IST / 15.00 SGT to 06.00 EDT / 11.00 BST / 15:30 IST / 18.00 SGT for essential maintenance. Apologies for the inconvenience.
 Wildfires greatly increase a landscape's vulnerability to flooding and erosion events by removing vegetation and changing soils. Fire damage to soil increases with increasing soil temperature, and, for fires where smoldering combustion is absent, the current understanding is that soil temperatures increase as fuel load and fire intensity increase. Here, however, we show that this understanding that is based on experiments under homogeneous conditions does not necessarily apply at the more relevant larger scale where soils, vegetation, and fire characteristics are heterogeneous. In a catchment-scale fire experiment, soils were surprisingly cool where fuel load was high and fire was hot and, conversely, soils were hot where expected to be cooler. This indicates that the greatest fire damage to soil can occur where fuel load and fire intensity are low rather than high, and has important implications for management of fire-prone areas prior to, during, and after fire events.
 Vegetation fires annually burn an average 3.7 million km2 worldwide [Giglio et al., 2010], affecting an area larger than the contiguous western United States. By removing vegetation and changing soils, fire greatly increases a landscape's vulnerability to flooding and erosion events [Cerdà and Robichaud, 2009; Lanini et al., 2009; Rulli and Rosso, 2005; Stoof et al., 2012] which pose a substantial threat to life, property, and drinking water supplies in and downstream of burned areas. Although fire is part of many ecosystems and recovery is often rapid [Cerdà and Doerr, 2005; Cerdà et al., 1995; Pausas et al., 2008], it can in other cases contribute to long-lasting degradation and even desertification since removal of fertile topsoil during post-fire erosion events is much faster than soil formation by weathering [Shakesby, 2011]. Because fire damage to the soil system increases with increasing soil temperature [Stoof et al., 2010], the soil temperature reached during a fire is an important factor for post-fire erosion risk and plant regeneration, and thus the recovery potential of burned areas. Yet, while fire behavior at the landscape scale is frequently studied, little attention has been given to the interaction between fire and soil heating at larger scales.
 High soil temperatures can considerably alter belowground ecosystem functioning. Effects range from seed and microbe mortality (~60°C) [Granström and Schimmel, 1993; Grasso et al., 1996] and losses of nutrients (>200°C) [Certini, 2005], to the development of soil water repellency (175°C) [DeBano, 1981], losses of soil organic matter and water retention (>200°C) [Stoof et al., 2010], and structural degradation [García-Corona et al., 2004]. These latter changes contribute to a reduction in infiltration that is frequently observed after fire [e.g., Martin and Moody, 2001], and greatly contribute to the fire-induced increase in runoff and erosion. Information on soil temperatures reached during fire is therefore valuable for field managers [Keeley, 2009] and for the planning of hillslope rehabilitation treatments aimed at reducing environmental damage and increasing resilience of burned areas [Robichaud et al., 2010]. Because of this, fire damage to soil (“soil burn severity”) is often assessed and used to map runoff and erosion risk and identify areas requiring post-fire treatments to mitigate these risks [Robichaud et al., 2010]. Unfortunately, soil burn severity cannot be mapped from satellite imagery alone; identification requires time-consuming ground-based measurements, thus making the risk mapping of post-fire land degradation tedious and complex.
 Soil temperature prediction from observation or reconstruction of fire behavior could be a solution. However, this is currently hampered by insufficient understanding of the interaction between fuel, fire, and soils. While fire response to factors like vegetation characteristics, topography, and weather conditions at the landscape scale is now quite predictable [Bowman et al., 2009], relationships between fire behavior and soil heating at larger scales are not as well studied. When organic layers protect the soil, soil heating is largely a consequence of post-frontal (non-flaming, or “smoldering”) combustion of ground fuels and soil temperature is essentially independent of fuel load and fire intensity [Hartford and Frandsen, 1992; Valette et al., 1994]. However, there is little evidence that this uncoupling of fuel load, fire intensity, and soil temperature holds in ecosystems with shallow or no litter and duff layers. Instead, there is a solid body of literature reporting positive relations between fuel load and/or fire intensity and soil temperature, with study areas ranging from Mediterranean shrublands [Molina and Llinares, 2001] to Californian chaparral [Moreno and Oechel, 1991] and ponderosa pine/oak forest [Keeley and McGinnis, 2007], and from Brazilian cerrado [Miranda et al., 1993] to Australian grassland [Morgan, 1999], desert [Wright and Clarke, 2008], and eucalypt forests [Bradstock and Auld, 1995; Granged et al., 2011]. While some of these studies only find weak [Keeley and McGinnis, 2007] or partial relationships (Bradstock and Auld  only found a fuel load–soil temperature relationship for near-surface standing fuels <0.5 m from the ground), the solid body of literature cited above contrasts considerably with the view that fuel load, fire intensity, and soil temperature are uncoupled. It is interesting to note that the studies cited above were all performed under fairly homogeneous conditions—in the laboratory or in small experimental or prescribed burns in the field. However, wildfires, and in some countries also prescribed fires, occur at much larger scales at which soils and vegetation show considerable spatial variability [e.g., Steffens et al., 2009] and fire behavior dynamics are different than in small areas [Kerby et al., 2007]. Because of this, and because of emerging discoveries about threshold behaviors [Zehe and Sivapalan, 2009], it is unclear if the fuel load–fire intensity–soil temperature relationship observed at smaller scale holds at the more complex landscape scale.
 For an interdisciplinary study into the effects of fire on land degradation, we intentionally burned a 9 ha shrubland catchment in Portugal, a country facing severe problems with fires and post-fire land degradation. This gave us the opportunity to study the effect of landscape and fire heterogeneity on the relationship between fuel load, fire behavior, and soil temperature in a large-scale field experiment. Before the fire, we mapped the catchment, calculated solar radiation, and collected fuel data, and during the fire, we monitored soil surface temperatures, flame temperatures, fire intensity, and fire spread.
2 Materials and Methods
2.1 Site Description and Characterization
 The study area is the Portuguese Valtorto catchment (40°06′21″N, 8°07′03″W, Figure 1a), where the climate is Mediterranean with some oceanic influence (Csb) [Köppen, 1923]. Stoof et al.  describe the catchment, which has shallow soils (Figure 1b) and was covered by dense heathland dominated by Ericaceae (Erica umbellata, Erica cinerea, and Calluna vulgaris) and Pterospartum tridentatum. Litter cover in this vegetation type is discontinuous and barely amounts to 0.2 kg m−2 [Fernandes et al., 2000]. We intensively surveyed the area in 2007 for vegetation height (n = 266) and soil depth (n = 283) using five replicates per sampling point. Vegetation cover and surface rock cover were visually estimated at all sites including the temperature monitoring sites (below). In November 2008, we sampled six 1 m2 plots by measuring vegetation height (n = 15/plot), and then determining fuel load per plot (ranging from 1.1 to 5.9 kg m−2) by harvesting and weighing the vegetation, and calculating its dry weight from three oven-dried subsamples (80°C, 24 h). Given the strong correlation between fuel load and vegetation height ((1)), we used vegetation height as a proxy for fuel load to create a fuel load map (Figure 2a).
where FL = fuel load (kg m−2) and h = vegetation height (m).
2.2 Experimental Fire
 The experiment took place after a 10 day dry period on the morning of 20 February 2009 using backfire and headfire techniques (Figure 1a) to maximize convection and reach the maximum fire intensity possible under the prevailing weather conditions. Mean air temperature was 14°C, relative humidity was 33%, wind direction was S-SE, wind speed was 1.7 m s−1 (Table S1), average soil moisture content (0–2.5 cm depth, n = 40; see Stoof et al. ) was 0.28 m3 m−3, and fire danger was low to moderate (Table S2). During the early stages of the fire (Video S1), we measured flame temperatures on the southern flank of the catchment (n = 226) using a handheld infrared pyrometer (Omegascope OS534E, Omega Engineering, USA). For safety reasons, flame temperatures were not measured during the final stage of the fire (Video S2). We recorded soil surface temperatures every 2 s at 52 sites using K-type thermocouples (Ø1.5 mm, TC-direct, Netherlands) connected to data loggers (EL-USB-TC; 0.5°C resolution, 1°C logger accuracy, Lascar Electronics, UK) buried with the thermocouple tip at the soil surface the day before the fire. Replicate sensors were installed at eight of these sites, and their results were taken along in the analyses. Post-fire inspection learned that actual thermocouple depth was 0.0 ± 0.4 cm, with ⅔ of thermocouples installed at 0.0 cm and no trend apparent in the temperature data. Because the resulting small error in depth/height was randomly distributed, all data were taken along for analyses after a quality check of the time series. One logger was discarded from analysis due to poor probe-logger contact. The timing of maximum temperature was used to create a map of fire spread that was interpolated using ordinary kriging. Finally, we estimated fire intensity classes using an empirical relationship derived for similar vegetation [Vega et al., 1998, Text S1] for which we estimated flame lengths in the field and from 20 photo and film snapshots taken across the catchment during the fire. This information was used to draw a fire intensity map.
2.3 Statistical Analyses
 Effects of fuel load on fire intensity and fire intensity on soil temperature were analyzed using the Kruskal-Wallis rank sum test, with Wilcoxon rank sum tests performed to assess differences between fire intensity classes. Underlying causes of soil temperature variation were moreover assessed by determining the effect of vegetation and landscape parameters (fuel load, vegetation cover, rock cover, slope, elevation, aspect, and solar radiation during the 10 dry days before the fire) using Classification and Regression Tree (CART) analysis in JMP 9 (SAS Institute Inc., USA). The latter four parameters were extracted from a DEM, with the sine of aspect taken to get a continuous variable. The CART splitting process was stopped when the lowest Akaike information criterion (corrected for finite sample size, AICc) was reached; the minimum number of records in each subgroup was set to five.
3 Results and Discussion
 Flame temperatures were 736 ± 126°C and fire intensity throughout the catchment ranged from <500 (low) to >15.000 kW m−1 (extreme) (Figure 2b). Fire spread particularly fast in the bottom of the valley (Figure 2c), where 25% of the area burned in just 10% of the time. Smoldering and post-frontal flaming combustion did not occur. Shrubs were completely consumed except in the parts of the valley and the northwest-facing slope (Figure 2b) where there was less solar radiation prior to the fire (Figure 1c) and, therefore, higher initial moisture. Although soil temperatures during the fire were locally as high as 800°C, the soil in most of the catchment remained below 100°C (Figure 2d).
 As expected, fire intensity significantly increased with fuel load (p < 0.0001, Figure 3a), because the increased amount of biomass provided fuel for the fire, but also because the ignition method caused fire fronts to merge and intensity to increase in the bottom of the valley where fuel loads were higher. In contrast, soil temperatures were inversely related to fire intensity (p = 0.024, Figures 2 and 3). CART analysis showed that vegetation cover, fuel load, and aspect affected soil temperature most (Table 1). Explanation of soil temperature variation through the regression tree was modest (R2 = 0.25), possibly because of three processes discussed further below. Nevertheless, the resulting patterns were interesting. Sparsely vegetated areas (<61.5% vegetation cover) experienced higher temperatures than more densely vegetated areas (188°C versus 87°C, respectively); within the sparsely vegetated areas, soil temperatures were higher for north(east)ern aspects (aspect <86°, Table 1). However, for the remaining majority of sampling points (80%), soil temperature variation was determined solely by fuel load, with none of the other variables (slope, elevation, aspect, rock cover, solar radiation) adding to the explanation. As for vegetation cover, substantially higher temperatures were recorded for sites with lower fuel load: 115°C versus 45°C for higher loads (Table 1 and Figure S1). Interestingly, because fuel load increased with soil depth (r = 0.58, p < 0.0001), high soil temperatures only occurred on shallow (degraded) soils (Figure S1).
Table 1. Results of the CART analysis for maximum soil temperature (Tmax; R2 = 0.25, AICc = 757), which are averages of the values in each group (n) ± SD.
Vegetation cover <61.5%
Aspect < 86°
79 ± 80
Aspect ≥ 86°
343 ± 318
Vegetation cover ≥61.5%
Fuel load < 2.9 kg m−2
115 ± 153
Fuel load ≥ 2.9 kg m−2
48 ± 30
 The inverse relationship between soil temperature and fuel load and fire intensity is likely the combined result of three processes leading to a decrease in downward heat transfer:
Where fire intensity was high, large air temperature gradients increased upward heat transfer (buoyancy). This is illustrated in Video S2 that shows the high-intensity climax caused by interacting fire fronts and the resulting strong convective smoke plume, with smoke illustrating the direction of convective heat.
Variation in fire spread rate decreased flame residence time in areas where fire spread was rapid and fire intensity was high (Figure 2d), resulting in shorter duration of soil heating.
Soils under taller vegetation had reduced exposure to heat release, because shrubland fires move primarily through the canopy, and vertical fuel continuity is likely to decrease with shrub height [e.g., Fernandes et al., 2000].
 Additionally, spatial variation in fuel moisture caused by differences in aspect, solar radiation, and vegetation characteristics played a role in soil temperature variation. Because the fire was conducted in the morning, east-facing slopes (Table 1) were drier as they were the first to be exposed to the sun. Moreover, these less vegetated east-facing slopes had in general received more incoming solar radiation during the dry days preceding the fire than elsewhere in the catchment (Figure 1c). Finally, areas with higher fuel loads (NW-facing slopes and valley floor) were moister than the sparsely vegetated and degraded areas that dried out more quickly. This protective effect of moisture has been observed previously in laboratory and field experiments with duff [Hartford and Frandsen, 1992; Hille and den Ouden, 2005] but was not expected in the absence of such ground fuels. Although this experimental fire was conducted in late winter, spatial variation in fuel moisture has been found to exist even during droughts [Fernandes et al., 2010], making this finding also potentially relevant in summer.
4 Implications for Managing Fire-Prone Areas
 Our results have important implications for understanding and managing ecosystem resilience of fire-prone areas, an issue that is particularly relevant given the current and expected increase in fire risk around the world due to climate change [Carvalho et al., 2010; Holz and Veblen, 2011]. First, because the processes governing the relationship between fuel load, fire intensity, and soil temperature are spatially heterogeneous and can be scale dependent, it is essential to study these relationships at the scale relevant to natural fires, i.e., the catchment or landscape scale. Second, and perhaps even more important, our findings show that densely vegetated areas with high fuel loads are not necessarily at greatest risk for fire-induced soil degradation, while more sparsely vegetated areas may be at higher risk than previously thought. Because high soil temperatures negatively affect post-fire recovery, already degraded areas—being more vulnerable to high soil temperatures—can be prone to further degradation. At the same time, more densely vegetated areas will be affected to a lesser extent, or not at all, when soils stay cooler during fire. The different temperature responses measured indicate that there are differences in resilience to fire across a landscape, and reverse the current understanding of where the greatest fire damage likely occurs. This study therefore highlights the importance of considering soil heating in degraded and sparsely vegetated areas when developing fire management and post-fire restoration plans. Without post-fire intervention, spatial variation in fire damage and recovery potential can not only magnify already existing differences in degradation and ecosystem resilience across landscapes but also trigger nonlinear responses, complex feedback mechanisms, and tipping-point behavior [Veraart et al., 2012; Zehe and Sivapalan, 2009].
 There are several practical applications for the findings from this study. Regarding prescribed fires, the results indicate that spatial variability of vegetation characteristics and moisture are important parameters to take into account, in terms of both timing and the size of the burn. If a uniform burn is desired, burning hillslopes of similar aspect—and thus moisture status—will better achieve soil heating or damage-related management goals than burning entire catchments, where vegetation and moisture distribution are inherently heterogeneous, leading to variable soil burn severity. Also, control of flame residence time through the ignition pattern (favoring head firing) will mitigate soil heating during prescribed burns in shrubland.
 Another practical application relates to the current difficulty in classifying soil burn severity after fire, for which alternative ways to identify fire damage to soil at larger scales are needed. The findings illustrate and expand the understanding of the range of processes governing soil heating and thus potential soil damage during forest fires in areas where litter and duff are shallow, discontinuous, or absent, such as shrublands, woodlands, and open forest types. Using these findings as a basis for increased collaboration between disciplines would allow this expanded understanding of the processes governing fire-induced soil heating to be used to develop conceptual models to predict areas at risk for soil heating and thus potential damage. It would additionally allow current spatial fire behavior models, like FARSITE [Finney, 2004], to be extended to account for soil heating during fire, and thereby enable landscape-scale hindcasting of soil temperatures. These models are well suited to predict and/or reconstruct fire behavior at the landscape scale using information on fuels, weather, and topography, but currently do not include soil heating parameters. Application of the findings of this study in this way can both greatly facilitate land managers’ predictions of areas at greatest risk of post-fire land degradation and considerably improve the effectiveness and efficiency of restoration measures.
 We thank C. Ferreira, P. Palheiro, the Sapadores Florestais of Vilarinho, Cadafaz and Aflopinhal, Góis municipality, S. Drooger, A. de Kort, E. Slingerland, J. v/d Berg, W. Mol, A. Mansholt, and P. Bingre-Amaral for assistance, and P. Vermeulen, M. Cruz, and W. v/d Putten for discussion. Editor W. Knorr, A. Cerdà, and four anonymous reviewers are thanked for valuable comments. Work was supported by an IAWF scholarship (C.R.S.) and EU-contract 037046 (C.R.S., A.J.D.F., and C.J.R).