See

discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/295074517

Tree cover and seasonal precipitation drive

understorey flammability in alpine mountain
forests
ARTICLE in JOURNAL OF BIOGEOGRAPHY JANUARY 2016
Impact Factor: 4.59 DOI: 10.1111/jbi.12745

READS

29

3 AUTHORS, INCLUDING:
Thibaut Frjaville

Thomas Curt

French National Institute for Agricultural R

National Research Institute of Science and

9 PUBLICATIONS 23 CITATIONS

90 PUBLICATIONS 1,174 CITATIONS

SEE PROFILE

All in-text references underlined in blue are linked to publications on ResearchGate,

letting you access and read them immediately.

SEE PROFILE

Available from: Thibaut Frjaville

Retrieved on: 16 March 2016

Original article

LRH

RRH Flammability of alpine forests

Tree cover and seasonal precipitation drive understorey flammability in alpine

mountain forests

Thibaut Frjaville1,2*, Thomas Curt1 and Christopher Carcaillet2,3,4

Pratique des Hautes tudes, F-75014 Paris, France, 3LEHNA (UMR5023, CNRS), Universit

Lyon 1, F-69622 Villeurbanne, France, 4LTER Zone Atelier Alpes, F-38000 Grenoble,

T. Frjaville et al.

Irstea, Ecosystmes Mditerranens et Risques, F-13182 Aix-en-Provence, France, 2cole

10

France

11

*Correspondence: Thibaut Frjaville

12

Present address: UMR BIOGECO, Universit de Bordeaux, Avenue des Facults, F-33405

13

Talence, France.

14

E-mail: thibaut.frejaville@gmail.com

15

ABSTRACT

16

Aim Little is known about the understorey flammability of European mountain forests. The

17

aim of this study was to determine the relative effects of climate, vegetation structure and

18

composition on the fuel-driven variation in fire spread and intensity.

19

Location The western Alps.

20

Methods Fire spread and intensity were simulated under constant moisture and weather

21

conditions for a wide range of understorey fuel parameters measured in the litter, grass and

22

shrub layers. Simulation outputs were used to compare understorey flammability between
1

23

different forest ecosystem types (FET). The FETs were characterized by using a co-inertia

24

analysis between composition and the environment (vegetation structure and climate). The

25

relationships between these factors, fuel properties and understorey flammability were then

26

tested using partial regression analyses.

27

Results The most flammable forests displayed an open canopy (drysubalpine and open

28

mediterranean) and grew in areas with dry autumns and wet and cold springs. Fire spread and

29

intensity were controlled by the trade-off between tree cover and dead (litter) and live (grass

30

and shrub) biomass load. Fire intensity also increased as a result of seasonal precipitation

31

patterns (differential distribution between the seasons): rainy springs enhanced biomass

32

growth whereas dry climates, especially in autumn, promoted shrub biomass and stimulated

33

litter accumulation and residence (higher litterfall and lower decomposition rates).

34

Interestingly, we found a positive relationship between fire intensity and the proportion of

35

conifers that disappeared after accounting for tree cover, indicating that, in the Alps, the open

36

canopy structure of needle-leaved forests makes them potentially more flammable than

37

broad-leaved forests because of the higher load and continuity of surface fuels.

38

Main conclusions Interrelationships between tree cover, precipitation seasonality and species

39

composition govern the understorey flammability of mountain forests. We also found

40

evidence that tree cover strongly constrains fire spread by driving the amount and type of

41

surface fuel, which suggests that land-use change can have a strong influence on

42

flammability patterns.

43

Keywords

44

Fire intensity, flammability traits, FlamMap model, mountain forest fuel,

45

pyrogeography, redundancy analysis.

46

47

INTRODUCTION

48

Plant flammability is an important driver of fire, with a fundamental role in terrestrial

49

ecosystems (Bond & Keeley, 2005), and is related to the morphological traits of plants,

50

including canopy architecture, fine fuel (i.e. particles < 6 mm thickness) biomass and surface

51

area to volume ratio (SVr) (Bowman et al., 2014). Flammability traits differ between plants

52

(Papi & Trabaud, 1991; Behm et al., 2004; Frjaville et al., 2013) and affect the spread and

53

intensity of fires (Santana et al., 2011; Schwilk & Caprio, 2011). Consequently, certain plant

54

communities are more likely to burn than others (Bond & Van Wilgen, 1996). The

55

composition of plant communities modifies the fire regime substantially (Engber & Varner,

56

2012) and, reciprocally, the fire regime alters community flammability (Curt et al., 2011;

57

Ganteaume et al., 2011). The role of variability in vegetation flammability in determining

58

past and future fire activity has recently been emphasized (Girardin et al., 2013; Paritsis et

59

al., 2013). However, predicting spatial and temporal changes in vegetation flammability still

60

requires an understanding of how flammability traits vary with environmental conditions (e.g.

61

climate). Identifying the environmental drivers of plant flammability traits, or more generally

62

modelling the biogeographical variation of plant functional traits, is thus critical in predicting

63

ecosystem responses to ongoing global changes (Lavorel & Garnier, 2002; Reichstein et al.,

64

2014).

65

Because wildfire propagates through multi-specific vegetation, the plant community is the

66

most appropriate unit for flammability analysis (Bowman et al., 2014). At the community

67

level, flammability can be assessed using the physical characteristics of fire behaviour, i.e.

68

the rate of fire spread and fire intensity (Anderson, 1970; Gill & Zylstra, 2005), which

69

enables scaling up, comparison and linkage with fire behaviour models (Fernandes & Cruz,

70

2012). Indeed, community-level differences in flammability reflect the combined effects of

71

the flammability traits of the living and dead plant material from the constituent species. The

72

flammability of a community is therefore controlled by the intrinsic fuel properties of the

73

plant assemblages, which determine the spread and intensity of fire under given conditions of

74

topography and weather. While short-term variation in temperature, humidity and wind

75

govern fuel moisture and ignitability (Flannigan et al., 2009), mid- to long-term climatic

76

parameters constrain fire spread and intensity, by controlling biomass growth and the spatial

77

arrangement of fuel (Pausas & Paula, 2012), and constrain the composition of plant

78

communities (Bradstock, 2010). Consequently, disentangling how climate acts on overall

79

community flammability is a challenging task (Parisien & Moritz, 2009).

80

In mountains, a large increase in fire activity is expected in response to climate warming

81

(Westerling et al., 2006; Moritz et al., 2012), notably in the western Alps where fires may

82

become more frequent and larger (Frjaville & Curt, 2015). In these ecosystems,

83

understanding the environmental drivers of forest flammability along broad climatic

84

gradients and within diverse plant communities will improve our ability to manage fire in

85

mountain environments. To address this, our aim with this study was to (1) compare the

86

understorey fuel properties between different forest ecosystems in the Alps and (2) determine

87

the relative influences of community composition, forest structure and climatic factors on

88

understorey flammability and the underlying fuel structural traits. We chose the western Alps

89

as our study area and applied the research framework shown in Fig. 1. First, the main

90

variability in community composition, forest structure and climate was identified by defining

91

the different forest ecosystem types (FET) present. Second, based on Rothermels equations

92

(1972), surface fuel parameters were determined using a stratified sampling scheme within

93

the FETs to simulate fire behaviour along a range of habitat conditions and elevations. Third,

94

understorey flammability was calculated using the trend of surface fuels to burn fast and with

95

high intensity under given external conditions of temperature, humidity, wind, aspect and
4

96

slope. Finally, differences in understorey flammability between forest ecosystems were

97

related to community composition, forest structure and climate using a partial regression

98

analysis. As the flammability was calculated for given weather parameters, testing the

99

influence of climatic parameters on flammability relied on mid- to long-term variation in fuel

100

characteristics, i.e. composition, load and arrangement. We focused on understorey

101

flammability, because fire regimes are typified by low-intensity surface fires in the Alps

102

(Genries et al., 2009), as in other mountain ranges (Schoennagel et al., 2004; Rother &

103

Grissino-Mayer, 2014), highlighting the dominant function of surface fuels (litter, herbs and

104

shrubs) for fire spread. By acting on both community composition and biomass growth, we

105

hypothesized that the wide variety in climatic conditions and vegetation structure in the Alps

106

controls the variability in understorey flammability within forest ecosystems.

107

METHODS

108

Data sources

109

Structure and composition parameters for mountain forests in the western Alps were obtained

110

from the data provided by seven annual fieldwork seasons (20052011) from the French

111

national forest inventory (Inventaire Forestier National, IFN). The IFN comprises a network

112

of temporary plots established on a grid of 500 500 m. We focused on a study area of

113

31,710 km2 (43304624 N, 460740 E) that extended from the northern to the southern

114

range of the French Alps (Fig. 2), within which the climate of the lowlands varies from

115

warmdry (mediterranean) to coolwet (oceanic) and continental. We excluded plots if there

116

was any evidence of recent (< 5 years), natural or anthropogenic disturbance, such as logging,

117

fire or windthrow. We also removed plots below 700 m a.s.l. and those from valley floors, to

118

focus on forests situated on slopes. The climate, forest structure and abundance of woody

119

species were therefore analysed from 2609 IFN plots distributed evenly across the French
5

120

Alps (Fig. 2). The IFN data included quantitative estimates of plant species cover using the

121

phytosociological scores of BraunBlanquet, absolute cover (% cover of the plot) of

122

vegetation layers (herb, shrub and tree) and morphological variables such as tree diameter (at

123

breast height) and height.

124

The forest composition (FC) was described using the relative cover of shrub and tree species

125

(species composition) and plant functional type (functional composition). Plant functional

126

types were distinguished based on the following dichotomies: needle-leaved versus broad-

127

leaved, deciduous versus evergreen, trees versus shrubs. Prior to analyses, BraunBlanquet

128

scores were converted to median values of the corresponding cover classes (% cover of the

129

plot). In this way we ensured compatibility with Euclidean distance-based ordination

130

methods (Podani, 2006). The forest structure (FS) component was described using canopy

131

characteristics (tree cover, tree density, basal area and mean tree height) and the absolute

132

cover of shrubs and herbs.

133

Climatic (CL) parameters were obtained from data downscaled at a 1 1-km resolution using

134

a regional regression procedure and a 500 500-m digital elevation model (Frjaville &

135

Curt, 2015). Specifically, we used long-term averages (19602009) of monthly precipitation

136

and monthly temperatures, and yearly sum of temperatures above 5 C (growing degree days;

137

GDD5). We also used long-term averages (annual mean, monthly mean and annual 95th

138

percentile) of the drought indices of the Canadian fire weather index system (Van Wagner,

139

1987): the fine fuel moisture code (FFMC), the duff moisture code (DMC) and the drought

140

code (DC). Climatic data computed over 5 years (20052009) were used preferentially for

141

analysing the flammability drivers, as these data captured climatefuel relationships better

142

than longer series.

143

Ecosystem study

144

To ensure that our flammability analysis extended to the variability in climate and vegetation

145

found in alpine forests, understorey fuel parameters were sampled after different habitats had

146

been defined. A broad classification of mountain forest ecosystems (FET) was defined using

147

a co-inertia analysis (OMI; Doldec et al., 2000) of woody species assemblages (shrubs and

148

trees) and a principal component analysis of the environment, i.e. climate and forest structure

149

(see Appendix S1 in the Supporting Information). This resulted in four FETs. The dry

150

subalpine ecosystems (FET-1), which covered the inner continental zone of the Alps (Fig. 2),

151

were dominated by Larix decidua Mill., Pinus uncinata Mill. and Pinus cembra L., forming a

152

tall and open canopy over a high cover of grass or dwarf shrubs, mainly Juniperus sibirica

153

Lodd., Vaccinium spp. and Rhododendron ferrugineum L. The moistmontane ecosystems

154

(FET-2), which covered the wet northern range (Fig. 2), were dominated by Abies alba Mill.,

155

Picea abies (L.) Karst. and Fagus sylvatica L., forming tall, closed forests with a sparse

156

understorey. At lower elevations, low-stature forests of Quercus pubescens Willd., Pinus

157

nigra Arn. and Pinus sylvestris L. characterized the warm and dry southern range under

158

mediterranean influence. The openmediterranean ecosystems (FET-3) displayed open

159

forests with moderate grass cover and a high cover of small shrubs such as Juniperus

160

communis L., Genista cinerea (Vill.) DC. and Amelanchier ovalis Medik. The closed

161

mediterranean ecosystems (FET-4) had a higher tree cover with a more or less dense

162

understorey of tall shrubs such as Buxus sempervirens L. and Acer spp.

163

Fuel structural traits

164

Surface fuel parameters were measured in 96 fuel plots (Fig. 2) based on a stratified random

165

sampling design within FET, using the proximity of FETs to the IFN plots and favouring

166

multiple locations along elevational gradients. Fuel measurement was based on standard
7

167

procedures (e.g. Bessie & Johnson, 1995) following the methods of McRae (1979) and

168

Brown et al. (1982) along a 20-m equilateral triangle. Downed woody debris was measured

169

using the line intersect technique (Van Wagner, 1968). Cover, height and load of both herb

170

and litter layers (i.e. the superficial fraction of fuel particles < 6 mm not yet decayed) were

171

measured in nine 1-m2 quadrats evenly spaced along the sides of the triangle. The SVr was

172

measured for litter and shrub leaves on the basis of the geometric dimensions of the leaf

173

middle cross-area (Hachmi et al., 2011). SVr measurements were averaged from five leaves

174

randomly sampled per forest plot for each dominant species (absolute cover > 10%). Species

175

composition and height of shrubs (crownbase and top heights) were measured every 2 m

176

along the 60-m transect (the triangle sides) for all individuals (fuel ladders) that vertically

177

intercepted the transect.

178

In the case of a multi-layered shrub stratum, all shrubs could distort the realism of fire

179

behaviour simulations because, if the fire is unable to bridge gaps between strata, then higher

180

fuels are unavailable. Therefore, we developed a procedure to detect the vertical arrangement

181

of shrubs, i.e. whether individual shrub height follows a multi-modal distribution, using

182

kernel density functions (Appendix S1). Fire simulations were conducted for each vertical

183

arrangement of shrub layers detected: the fuel bed was assessed using the critical fire

184

intensity criterion of Van Wagner (1977) and iterative fire behaviour simulations (Appendix

185

S1). Shrub load was estimated beforehand from the fine fuel fraction (leaves and twigs

186

< 6 mm) and the crown volume of dominant species:

LSx 0.25 j 1 BDj phyV j
S

187

(1)

188

where LSx is the load of shrub (multi-) layer x, BDj the bulk density and phyVj the phyto-

189

volume (assuming a rectangular prism shape) of the constituent dominant species j, and 0.25

190

a correction coefficient (Appendix S1). Community-level values of SVr were computed by

191

weighting the mean value of the constituent species j by its relative abundance, Rj:
SVr j 1 SVr j Rj
S

192

S
j 1

Rj

(2)

193

In the case of shrub leaves, Rj is the relative phyto volume of the jth shrub species in the

194

(multi-) layer; in the case of tree litter leaves, Rj is the relative basal area of the jth tree

195

species in the community.

196

Community flammability quantification

197

Andersons definition (1970) of flammability encompasses three components describing the

198

ability of fuel to ignite (ignitability), the intensity of combustion (combustibility) and its

199

sustainability. Fire-line intensity (FLI, kW m1) reflects combustibility, while the concept of

200

ignitability is reflected in rate of spread, ROS, expressed in m min1 (Gill & Zylstra, 2005),

201

with idealized successive ignition of fuel fragments from surrounding burning fuel. The heat

202

released per unit area, HPA (kJ m2), expresses both combustibility and sustainability,

203

because the amount of heat released depends on both the intensity and duration of

204

combustion.

205

Understorey flammability was quantified by the multivariate variation of ROS, FLI and HPA.

206

Fire behaviour was simulated on one virtual terrain (slope = 30%, wind blowing uphill) for

207

different weather scenarios with dead fuel moisture and wind speed using the FlamMap

208

program (Finney, 2006). Canopy characteristics (tree cover and mean height) were used to

209

take into account the buffering effect on surface wind speed (i.e. the wind adjustment factor;

210

Andrews, 2012). Live fuel moisture was set at 100%. Changes in dead fuel moisture and

211

wind speed did not significantly change either the relative differences in fire behaviour

212

between FET, or the relationships with environmental conditions (data not shown).

213

Consequently, we focused on an intermediate severity scenario, with a 30 km h1 wind speed

214

and 7% dead fuel moisture content. Fire behaviour simulations were performed

215

independently for all fuel models, i.e. for one to several homogeneous fuel complexes per

216

community (Appendix S1). When several fuel models were designed in a community, we

217

averaged fire simulation outputs according to the relative cover of the fuel model in the

218

community.

219

Variance partitioning

220

We performed ANOVA and Tukeys honestly significant difference (HSD) post-hoc tests to

221

compare understorey flammability and fuel parameters between FET. To achieve normality

222

and variance homogeneity, we applied a log-transformation of fire behaviour metrics (ROS,

223

FLI and HPA) and a fourth-root transformation of fuel parameters before comparison tests.

224

The variance in understorey flammability was partitioned into unique and common

225

contributions of the sets of predictors (FC, FS and CL) using partial redundancy analysis

226

(Peres-Neto et al., 2006). The flammability matrix included ROS, FLI and HPA community-

227

level values (multivariate response). For each set of predictors (FC, FS and CL), we applied a

228

forward selection of variables (Blanchet et al., 2008). Variance and P-values were averaged

229

from 50 partial redundancy analyses with Monte Carlo permutation tests (the residuals under

230

the full model being permuted 10,000 times; Legendre & Legendre, 2012) carried out on

231

70% of randomly selected sites (using a bootstrap procedure; Cottenie, 2005). Fuel

232

parameters were modelled in the same way, except that the procedure was performed

233

separately for each dependent variable (univariate response) and the sets of predictors were

234

limited to the most parsimonious subset of 12 variables each, using a stepwise model

235

selection (both backward and forward) with the Akaike information criterion (AIC).

10

236

From these redundancy analyses we also performed partial regressions to study to what

237

extent and how one environmental factor drove changes in flammability and underlying fuel

238

properties, i.e. by accounting for the influence of other drivers. Specifically, the effect of one

239

predictor on one response variable (FLI or fuel load) was modelled by excluding the unique

240

contributions of other confounding factors (Peres-Neto et al., 2006).

241

All analyses and calculations were carried out using R version 3.1.2 (R Core Team, 2013),

242

using the ade4 (Dray & Dufour, 2007) and vegan (Oksanen et al., 2013) packages for FET

243

classification and variance partitioning.

244

RESULTS

245

Understorey flammability

246

The comparison of surface fuel parameters indicated significant biogeographical contrasts

247

between the FETs (Table 1). Litter was the most important contributor to the fuel bed,

248

accounting for the highest proportion of total fine fuel load (Fig. S2.1). However, no

249

significant difference in fine fuel load was detected between FET for either litter or dead

250

woody debris, and thus total fuel (live and dead, P > 0.05; Table 1). In contrast, open FET

251

(FET-1and FET-3) showed higher herb loads (P < 0.05), while dense, tall forests (FET-2)

252

showed lower amounts of live fine fuels, i.e. herb and shrubs within the fuel bed (P < 0.05).

253

Dwarf shrubs of drysubalpine ecosystems had a higher bulk density, and tree litter leaves a

254

higher SVr, than other FETs (P < 0.05).

255

Considering the combined effect of these fuel structural traits, fire simulations indicated an

256

elevated flammability for the open forests (FET-1 and FET-3) that exhibited the highest

257

ROS, FLI and HPA for the given fuel moisture and wind speed (Fig. 3). Closed

258

mediterranean ecosystems (FET-4) were characterized by an intermediate flammability with

11

259

a lower mean ROS (Fig. 3a), but similar FLI and HPA values compared with open forest

260

ecosystems (Fig. 3b, c). The poor understorey moistmontane ecosystems (FET-2) were the

261

least flammable.

262

Environmental drivers of understorey flammability

263

Boosted redundancy analysis indicated that the parameters of forest composition, structure

264

and climate explained most of the flammability variance (Rmean = 0.62, Pmean < 0.001; Fig.

265

4a). The interaction between these three components explained the majority of flammability

266

variance (Rmean = 0.39) compared with the pure fractions (variance that was exclusively

267

explained by one component). Nonetheless, differences in flammability were also exclusively

268

related to forest structure (Rmean = 0.11, Pmean < 0.05; Fig. 4a) and, to a lesser extent, to forest

269

composition or climate (pure fractions, Rmean = 0.07 and 0.04, respectively, Pmean 0.05; Fig.

270

4a). The interactions between forest structure, composition and climate explained the

271

majority of fuel parameter variance (Table 2), except for fuel load (litter, herb and shrub),

272

which varied mainly with climate (half of the explained variance, P < 0.01; Table 2).

273

Community composition drove most of the variance in SVr of shrub leaves (71% of

274

explained variance, P < 0.05; Table 2).

275

Stepwise regressions revealed that tree cover, precipitation seasonality and forest functional

276

composition chiefly explained the variance of flammability between forests (adjusted

277

R = 0.56, P < 0.001; Fig. 4b). Variance was better explained by the interaction between these

278

three factors (23%) and by canopy opening (variance that was exclusively explained by tree

279

cover: adjusted R = 0.22, P < 0.001; Fig. 4b). In particular, FLI decreased non-linearly with

280

tree cover by increasing until a c. 40% of the tree cover and then decreasing with canopy

281

closure (r = 0.44, P < 0.001; Fig. 5a). A multi-linear regression indicated that FLI also

282

varied with seasonal precipitation (spring to annual precipitation ratio, t = 4.5, P < 0.001; Fig.
12

283

5b), showing that open canopy forests were the most flammable, especially in areas where

284

precipitation was abundant in spring and rare in other seasons (adjusted R = 0.54, P < 0.001;

285

Fig. 5b).

286

Among the climatic parameters, partial regressions revealed that FLI increased first with

287

increasing spring to annual precipitation ratio (P < 0.001; Fig. 6a), then decreased with

288

autumn precipitation (P < 0.001; Fig. 6b) and followed a humped relationship with the 95th

289

percentile of DC (P < 0.001; Fig. 6c). Total fine fuel load exhibited similar relationships with

290

the climatic gradients (P < 0.001; Fig. 6df). Specifically, a higher load of fine litter was

291

found in areas with drier autumns and colder springs (P < 0.01; Table S2.1). Mean winter,

292

spring and annual temperatures exhibited negative correlations with litter load, herb load,

293

shrub bulk density and SVr of leaves (P < 0.05; Table S2.1). As a consequence, the colder

294

the winters and springs, the faster the fire spread (P < 0.05; Table S2.2). However, forest

295

flammability increased significantly with increasing annual, summer or autumn drought

296

(P < 0.05; Table S2.2).

297

Partial regressions showed that FLI increased with increasing proportions of needle-leaved

298

trees in the community (r = 0.17, P < 0.001; Fig. 7a), although this effect disappeared after

299

accounting for the effect of tree cover (P = 0.12; Fig. 7b). In other words, in the Alps the

300

open canopy structure of needle-leaved forests might make them more flammable than broad-

301

leaved forests.

302

DISCUSSION

303

This study proposes a framework for testing how mid- to long-term climate changes, forest

304

structure and community composition drive the variation in mountain forest understorey

305

flammability, either exclusively or in combination. Interestingly, while precipitation and

306

temperature gradients described the main axis of variation in community composition (Fig.
13

307

S1.1), we found that tree cover also explained a substantial part of the variation by

308

contributing to the second axis. By controlling light availability to the understorey, the

309

influence of tree cover on shrub species composition has already been observed across the

310

Alps (Nieto-Lugilde et al., 2015). We found that tree cover and precipitation gradients related

311

significantly to variation in fire spread and intensity driven by fuel characteristics, thus

312

confirming our predictions. This study demonstrates that understorey flammability varies

313

between mountain ecosystems, largely as the result of the interaction between tree cover,

314

precipitation seasonality and the relative abundance of coniferous trees that jointly determine

315

the variation in fuel structural traits (i.e. fuel load, composition and arrangement).

316

Tree cover was found to control the main part of the understorey flammability variance,

317

suggesting that this structural determinant of forests is able to capture most of the fuel

318

properties (Bradstock, 2010). Fire intensity increased with canopy closure until a threshold of

319

c. 40% tree cover, above which fire intensity decreased (Fig. 5a). This threshold has also

320

been detected in relation to burned areas in southern Africa (Archibald et al., 2009), and is

321

supported by the non-linear response of grass productivity to c. 3540% tree cover (Scholes,

322

2003) because of the decreasing light intensity and increasing rainfall interception by deep

323

litter layers (Anderson et al., 1969). In the western Alps, a trade-off between the amount of

324

litter and surface fuels (herb and shrub) might maximize fire intensity at an intermediate

325

canopy cover (Fig. S2.2), i.e. a high amount of grass and shrub fuel on a continuous litter

326

bed. This supports observations in savannas, where tree cover decreases the probability of

327

burning and fire intensity by limiting the availability of fine fuels (Trauernicht et al., 2012),

328

in addition to the sheltering effect of canopy closure on microclimatic conditions (Ray et al.,

329

2005; Hoffmann et al., 2012).

330

To test the relationships between tree cover and understorey flammability found using fire

331

simulations, we analysed the distribution of lightning fire sizes for the period 19802011
14

332

according to the tree cover of surrounding unburned IFN plots (Appendix S3). Among the

333

mediterranean mountain forests of the southern Alps, we found a higher burned area ratio in

334

open (FET-3) than in closed (FET 4) ecosystems (Appendix S3), confirming the simulated

335

differences in fire behaviour between them (Fig. 3). Furthermore, we found that the largest

336

lightning fires (> 10 ha) occurred in areas characterized by more open forests (with a median

337

tree cover of 40%) than where lightning fires were small or absent (median tree cover 80%;

338

d.f. =2 = 9.37, P = 0.009; Appendix S3). These independent validation tests strengthen the

339

relationships we found between tree cover, understorey flammability and surface fire spread

340

in the Alps (Fig. 5, Fig. S2.2).

341

The fire intensity varied between forests mainly as a result of interactions between tree cover,

342

precipitation seasonality and functional composition (Fig. 4). For instance, we demonstrated

343

that coniferous (needle-leaved) forests are more flammable than broad-leaved deciduous

344

forests, a widely accepted assumption (Bond & Van Wilgen, 1996; Girardin et al., 2013), but

345

it was significantly related to differences in tree cover (Fig. 7). This observation highlights

346

the ability of broad-leaved deciduous trees to form more closed canopies and therefore less

347

flammable forests. We also found increases in the amount of fine fuels (litter, herb and shrub)

348

with precipitation seasonality, i.e. with increasing spring rainfall and decreasing autumn

349

rainfall, leading to more intense fires in climates characterized by wet springs and dry

350

autumns, where open forest ecosystems dominate (FET-1 and FET-3, with higher positions

351

on the spring to annual precipitation gradient; Fig. 6). This supports the principle that

352

climatic seasonality controls the flammability of mountain forests by promoting open

353

canopies, biomass growth under rainy springs and necromass accumulation under dry

354

autumns. Indeed, we found a negative correlation between the amount of fine litter, autumn

355

rainfall and spring temperature, suggesting a decrease in litter decomposition rate in drier or

356

colder areas (Meentemeyer et al., 1982; Aerts, 1997; Keane, 2008). To a lesser extent, litter
15

357

load varied according to tree species composition and functional traits (P < 0.05; Table 2), as

358

already found in the northern Rocky Mountains (Keane, 2008) and in the drysubalpine

359

ecosystems of the Alps (Blarquez et al., 2012). Therefore we have added strong evidence that

360

community flammability in surface-fire ecosystems is controlled by canopy cover and the

361

interplay with climate (precipitation seasonality), fuel connectivity and litter bed

362

characteristics (Bowman et al., 2014).

363

We also found that fire intensity increased with drought (independent of fuel moisture),

364

especially in autumn (Table S2.2), but followed a unimodal relationship with the most

365

extreme annual conditions (95th percentile of DC; Fig. 6c). These findings highlight dry

366

conditions as enhancing understorey flammability in the Alps by promoting shrub biomass

367

(instead of tree biomass) and litter accumulation (Table S2.1), whereas fuel limitation by

368

drought may also constrain fire in supra-mediterranean ecosystems. Under a given fuel

369

moisture content, we also found that forest ecosystems inhabiting the colder climates (low

370

annual, winter and spring temperatures) burnt more readily. These findings illustrate that,

371

although temperature is recognized as the most important variable affecting wildfire, with

372

warmer temperatures leading to increased fire activity (Flannigan et al., 2009), a warmer

373

climate could also inhibit fire by promoting less flammable broad-leaf species (Girardin et

374

al., 2013). For example, the SVr of litter leaves decreased with winter and spring

375

temperatures (Table S2.1) because composition-driven changes in this flammability trait

376

(Table 2) relate mostly to these climatic gradients.

377

By controlling fuel dryness, biomass growth, decomposition of dead plant material and the

378

community composition, climate change could strongly alter fire spread, intensity and

379

ultimately regime in mountain ecosystems by promoting synergistic (e.g. higher fuel loads

380

with drier conditions) rather than antagonist (e.g. temperature- or drought-induced fuel

381

limitation) trends. The open canopy structure of the most flammable forests probably mirrors
16

382

the transient afforestation processes of abandoned pastures and eventually ploughed areas at

383

low (openmediterranean ecosystems) and high (drysubalpine ecosystems) elevations

384

(Tasser & Tappeiner, 2002; Chauchard et al., 2007). Therefore, past land-use changes in

385

European mountain forests might have governed the current flammability patterns at the

386

landscape scale (Zumbrunnen et al., 2012). Consequently, both climatic and land-use changes

387

will probably impact the occurrence, spread and intensity of fires in mountain forests by

388

promoting drier conditions (Zumbrunnen et al., 2009; Frjaville & Curt, 2015) and by

389

inducing changes in the main environmental drivers of understorey flammability, i.e. tree

390

cover, precipitation seasonality and the relative abundance of broad-leaved versus needle-

391

leaved trees. These findings also imply that, under similar weather and fuel moisture

392

conditions (as simulated here), fire spread and intensity will differ greatly across alpine

393

forests in a predictable manner. We found higher fire susceptibility in the open forests of the

394

southern Mediterranean mountains and, more surprisingly, in the subalpine ecosystems of the

395

inner continental range, where fires are rare today. Consequently, the flammability drivers we

396

have identified should be of benefit in terms of fire prevention and forest management by

397

providing a mechanistic understanding of the areas where climatic and vegetation features

398

make forests more susceptible to fire damage. In addition, this study has highlighted

399

consideration of tree cover as a strong foundation for forest practices to reduce flammability

400

in European mountain ecosystems.

401

ACKNOWLEDGEMENTS

402

Financial support was provided by the FUME Project under the European Unions Seventh

403

Framework Programme (FP7/20072013) and by grants from the National Research Institute

404

of Science and Technology for Environment and Agriculture. We acknowledge everyone

405

who has contributed to the French national forest inventory database. We also thank Jean17

406

Philippe Vidal for providing climate data and Jean-Michel Lopez, Roland Estve, Fabien

407

Guerra, Christian Ripert and Annabelle Rivoal for their invaluable assistance in the field.

408

REFERENCES

409

Aerts, R. (1997) Climate, leaf litter chemistry and leaf litter decomposition in terrestrial

410

ecosystems: a triangular relationship. Oikos, 79, 439449.

411

Anderson, H.E. (1970) Forest fuel ignitibility. Fire Technology, 6, 312319.

412

Anderson, R.C., Loucks, O.L. & Swain, A.M. (1969) Herbaceous response to canopy cover,

413

light intensity, and throughfall precipitation in coniferous forests. Ecology, 50, 255263.

414

Andrews, P.L. (2012) Modeling wind adjustment factor and midflame wind speed for

415

Rothermels surface fire spread model. General Technical Report RMRS-GTR-266. USDA

416

Forest Service, Rocky Mountain Research Station, Fort Collins, CO.

417

Archibald, S., Roy, D.P., Van Wilgen, B.W. & Scholes, R.J. (2009) What limits fire? An

418

examination of drivers of burnt area in southern Africa. Global Change Biology, 15, 613

419

630.

420

Behm, A.L., Duryea, M.L., Long, A.J. & Zipperer, W.C. (2004) Flammability of native

421

understorey species in pine flatwood and hardwood hammock ecosystems and implications

422

for the wildlandurban interface. International Journal of Wildland Fire, 13, 355365.

423

Bessie, W.C. & Johnson, E.A. (1995) The relative importance of fuels and weather on fire

424
425
426

behaviour in subalpine forests. Ecology, 76, 747762.

Blanchet, F.G., Legendre, P. & Borcard, D. (2008) Forward selection of explanatory
variables. Ecology, 89, 26232632.
18

427

Blarquez, O., Carcaillet, C., Elzein, T.M. & Roiron, P. (2012) Needle accumulation rate

428

model-based reconstruction of palaeo-tree biomass in the western subalpine Alps. The

429

Holocene, 22, 579587.

430
431

Bond, W.J. & Keeley, J.E. (2005) Fire as a global herbivore: the ecology and evolution of
flammable ecosystems. Trends in Ecology and Evolution, 20, 387394.

432

Bond, W.J. & Van Wilgen, B.W. (1996) Fire and plants. Chapman and Hall, London.

433

Bowman, D.M.J.S., French, B.J. & Prior, L.D. (2014) Have plants evolved to self-immolate?

434
435
436
437

Frontiers in Plant Science, 5, art590.

Bradstock, R.A. (2010) A biogeographic model of fire regimes in Australia: current and
future implications. Global Ecology and Biogeography, 19, 145158.
Brown, J.K., Oberheu, R.D. & Johnston, C.M. (1982) Handbook for inventorying surface

438

fuels and biomass in the Interior West. General Technical Report INT-129. USDA Forest

439

Service, Intermountain Forest and Range Experiment Station, Ogden, UT.

440

Chauchard, S., Carcaillet, C. & Guibal, F. (2007) Patterns of land-use abandonment control

441

tree-recruitment and forest dynamics in mediterranean mountains. Ecosystems, 10, 936

442

948.

443
444

Cottenie, K. (2005) Integrating environmental and spatial processes in ecological community

dynamics. Ecology Letters, 8, 11751182.

445

Curt, T., Schaffhauser, A., Borgniet, L., Dumas, C., Esteve, R., Ganteaume, A., Jappiot, M.,

446

Martin, W., NDiaye, A. & Poilvet, B. (2011) Litter flammability in oak woodlands and

447

shrublands of southeastern France. Forest Ecology and Management, 261, 22142222.

19

448
449
450
451
452
453
454
455
456

Doldec, S., Chessel, D. & Gimaret-Carpentier, C. (2000) Niche separation in community

analysis: a new method. Ecology, 81, 29142927.
Dray, S. & Dufour, A.-B. (2007) The ade4 package: implementing the duality diagram for
ecologists. Journal of Statistical Software, 22, 120.
Engber, E.A. & Varner, J.M. (2012) Patterns of flammability of the California oaks: the role
of leaf traits. Canadian Journal of Forest Research, 42, 19651975.
Fernandes, P.M. & Cruz, M.G. (2012) Plant flammability experiments offer limited insight
into vegetationfire dynamics interactions. New Phytologist, 194, 606609.
Finney, M.A. (2006) An overview of FlamMap fire modeling capabilities. Fuels

457

managementhow to measure success: conference proceedings (ed. by P.L. Andrews and

458

B.W. Butler), pp. 213220. Proceedings RMRS-P 41. USDA Forest Service, Rocky

459

Mountain Research Station, Fort Collins.

460

Flannigan, M.D., Krawchuk, M.A., de Groot, W.J., Wotton, B.M. & Gowman, L.M. (2009)

461

Implications of changing climate for global wildland fire. International Journal of Wildland

462

Fire, 18, 483507.

463

Frjaville, T. & Curt, T. (2015) Spatiotemporal patterns of changes in fire regime and

464

climate: defining the pyroclimates of south-eastern France (Mediterranean Basin). Climatic

465

Change, in press, doi:10.1007/s10584-015-1332-3.

466
467

Frjaville, T., Curt, T. & Carcaillet, C. (2013) Bark flammability as a fire-response trait for
subalpine trees. Frontiers in Plant Science, 4, art466.

20

468

Ganteaume, A., Jappiot, M., Lampin-Maillet, C., Curt, T. & Borgniet, L. (2011) Effects of

469

vegetation type and fire regime on flammability of undisturbed litter in southeastern France.

470

Forest Ecology and Management, 261, 22232231.

471

Genries, A., Morin, X., Chauchard, S. & Carcaillet, C. (2009) The function of surface fires in

472

the dynamics and structure of a formerly grazed old subalpine forest. Journal of Ecology,

473

97, 728741.

474
475

Gill, A.M. & Zylstra, P. (2005) Flammability of Australian forests. Australian Forestry, 68,
8894.

476

Girardin, M.P., Ali, A.A., Carcaillet, C., Blarquez, O., Hly, C., Terrier, A., Genries, A. &

477

Bergeron, Y. (2013) Vegetation limits the impact of a warm climate on boreal wildfires.

478

New Phytologist, 199, 10011011.

479

Hachmi, M., Sesbou, A., Benjelloun, H. & Bouanane, F. (2011) Alternative equations to

480

estimate the surface-to-volume ratio of different forest fuel particles. International Journal

481

of Wildland Fire, 20, 648656.

482

Hoffmann, W.A., Jaconis, S.Y., Mckinley, K.L., Geiger, E.L., Gotsch, S.G. & Franco, A.C.

483

(2012) Fuels or microclimate? Understanding the drivers of fire feedbacks at savannaforest

484

boundaries. Austral Ecology, 37, 634643.

485

Keane, R.E. (2008) Biophysical controls on surface fuel litterfall and decomposition in the

486

northern Rocky Mountains, USA. Canadian Journal of Forest Research, 38, 14311445.

487

Lavorel, S. & Garnier, E. (2002) Predicting changes in community composition and

488

ecosystem functioning from plant traits: revisiting the Holy Grail. Functional Ecology, 16,

489

545556.
21

490

Legendre, P. & Legendre, L. (2012) Numerical ecology. Elsevier, Amsterdam.

491

McRae, D.J., Alexander, M.E. & Stocks, B.J. (1979) Measurement and description of fuels

492

and fire behaviour on prescribed burns: a handbook. Forest Technical Report O-X-287.

493

Canadian Forestry Service, Great Lakes Forest Research Centre, Sault Ste Marie, Ontario.

494
495

Meentemeyer, V., Box, E.O. & Thompson, R. (1982) World patterns and amounts of
terrestrial plant litter production. BioScience, 32, 125128.

496

Moritz, M.A., Parisien, M.-A., Batllori, E., Krawchuk, M.A., Van Dorn, J., Ganz, D.J. &

497

Hayhoe, K. (2012) Climate change and disruptions to global fire activity. Ecosphere, 3,

498

art49.

499

Nieto-Lugilde, D., Lenoir, J., Abdulhak, S., Aeschimann, D., Dullinger, S., Ggout, J.-C.,

500

Guisan, A., Pauli, H., Renaud, J., Theurillat, J.-P., Thuiller, W., Van Es, J., Vittoz, P.,

501

Willner, W., Wohlgemuth, T., Zimmermann,, N.E. & Svenning, J.-C. (2015) Tree cover at

502

fine and coarse spatial grains interacts with shade tolerance to shape plant species

503

distributions across the Alps. Ecography, 38, 578589.

504

Oksanen, J., Blanchet, F.G., Kindt, R., Legendre, P., OHara, R.B., Simpson, G.L., Solymos,

505

P., Stevens, M.H.H., & Wagner, H. (2013) vegan: community ecology package. R package

506

version 2.0-7. Available at: http://cran.r-project.org/web/packages/vegan/index.html.

507
508
509
510

Papi, C. & Trabaud, L. (1991) Comparative study of the aerial structure of five shrubs of
Mediterranean shrublands. Forest Science, 37, 146159.
Parisien, M.-A. & Moritz, M.A. (2009) Environmental controls on the distribution of wildfire
at multiple spatial scales. Ecological Monographs, 79, 127154.

22

511

Paritsis, J., Holz, A., Veblen, T.T. & Kitzberger, T. (2013) Habitat distribution modeling

512

reveals vegetation flammability and land use as drivers of wildfire in SW Patagonia.

513

Ecosphere, 4, art53.

514
515
516
517
518
519
520
521
522
523

Pausas, J.G. & Paula, S. (2012) Fuel shapes the fireclimate relationship: evidence from
Mediterranean ecosystems. Global Ecology and Biogeography, 21, 10741082.
Peres-Neto, P.R., Legendre, P., Dray, S. & Borcard, D. (2006) Variation partitioning of
species data matrices: estimation and comparison of fractions. Ecology, 87, 26142625.
Podani, J. (2006) BraunBlanquets legacy and data analysis in vegetation science. Journal of
Vegetation Science, 17, 113117.
Ray, D., Nepstad, D. & Moutinho, P. (2005) Micrometeorological and canopy controls of fire
susceptibility in a forested Amazon landscape. Ecological Applications, 15, 16641678.
R Core Team (2013) R: a language and environment for statistical computing. R Foundation
for Statistical Computing, Vienna.

524

Reichstein, M., Bahn, M., Mahecha, M.D., Kattge, J. & Baldocchi, D.D. (2014) Linking plant

525

and ecosystem functional biogeography. Proceedings of the National Academy of Sciences

526

USA, 111, 1369713702.

527

Rother, M.T. & Grissino-Mayer, H.D. (2014) Climatic influences on fire regimes in

528

ponderosa pine forests of the Zuni Mountains, NM, USA. Forest Ecology and Management,

529

322, 6977.

530
531

Rothermel, R.C. (1972) A mathematical model for predicting fire spread in wildland fuels.
Research Paper INT-115. USDA Forest Service, Ogden, UT.

23

532

Santana, V.M., Baeza, M.J. & Vallejo, V.R. (2011) Fuel structural traits modulating soil

533

temperatures in different species patches of Mediterranean Basin shrublands. International

534

Journal of Wildland Fire, 20, 668677.

535
536
537
538

Schoennagel, T., Veblen, T.T. & Romme, W.H. (2004) The interaction of fire, fuels, and
climate across Rocky Mountain forests. BioScience, 54, 661676.
Scholes, R.J. (2003) Convex relationships in ecosystems containing mixtures of trees and
grass. Environmental and Resource Economics, 26, 559574.

539

Schwilk, D.W. & Caprio, A.C. (2011) Scaling from leaf traits to fire behaviour: community

540

composition predicts fire severity in a temperate forest. Journal of Ecology, 99, 970980.

541
542
543

Tasser, E. & Tappeiner, U. (2002) Impact of land use changes on mountain vegetation.
Applied Vegetation Science, 5, 173184.
Trauernicht, C., Murphy, B.P., Portner, T.E. & Bowman, D.M.J.S. (2012) Tree coverfire

544

interactions promote the persistence of a fire-sensitive conifer in a highly flammable

545

savanna. Journal of Ecology, 100, 958968.

546
547
548
549
550
551

Van Wagner, C.E. (1968) The line intersect method in forest fuel sampling. Forest Science,
14, 2026.
Van Wagner, C.E. (1977) Conditions for the start and spread of crown fire. Canadian
Journal of Forest Research, 7, 2334.
Van Wagner, C.E. (1987) Development and structure of the Canadian Forest Fire Weather
Index System. Forest Technical Report 35. Canadian Forestry Service, Ottawa.

24

552
553
554
555
556

Westerling, A.L., Hidalgo, H.G., Cayan, D.R. & Swetnam, T.W. (2006) Warming and earlier
spring increase western US forest wildfire activity. Science, 313, 940943.
Zumbrunnen, T., Bugmann, H., Conedera, M. & Brgi, M. (2009) Linking forest fire regimes
and climate: a historical analysis in a dry inner Alpine valley. Ecosystems, 12, 7386.
Zumbrunnen, T., Menndez, P., Bugmann, H., Conedera, M., Gimmi, U. & Brgi, M. (2012)

557

Human impacts on fire occurrence: a case study of hundred years of forest fires in a dry

558

alpine valley in Switzerland. Regional Environmental Change, 12, 935949.

559

25

560

SUPPORTING INFORMATION

561

Additional Supporting Information may be found in the online version of this article:

562

Appendix S1 Classifying the mountain forest ecosystems of the French Alps, designing

563

homogeneous fuel models in heterogeneous forests, and quantifying flammable shrub

564

biomass from individual to community levels.

565

Appendix S2 Relative contribution of litter, herb and shrub to total fine fuel load among

566

forest ecosystem types, second-order polynomial regressions relating fuel load to tree cover,

567

and Pearsons correlation coefficients between climate and understorey fuel or flammability

568

parameters.

569

Appendix S3 Validation exercises for between-ecosystem differences in, and tree cover

570

effect on, flammability from lightning fire data (19812011).

571

26

572

BIOSKETCHES

573

Thibaut Frjaville has a PhD in Environmental Sciences from the Aix-Marseille University

574

and the Mediterranean Ecosystems and Risks Research Group of Irstea Aix-en-Provence

575

(France). His research interests centre on the vulnerability of mediterranean and mountain

576

forests to fire, including the biogeography of community flammability and fire in response to

577

regional changes in climate and human activities.

578

Thomas Curt is a researcher at the Mediterranean Ecosystems and Risks Research Group of

579

Irstea Aix-en-Provence (France). His main research interests are in fire ecology and fire risk

580

assessment in Mediterranean areas.

581

Christopher Carcaillet is professor at the Ecole Pratique des Hautes Etudes, an institute for

582

advanced studies (Paris). His main research focus is disturbance chronoecology in relation to

583

climatic change and plant diversity dynamics in mountain, boreal and Mediterranean areas,

584

testing the function of disturbance on community pattern and biogeochemical cycles, and

585

exploring the role of global change (climatic and social) on perturbation regimes.

586

Author contributions: T.F. conceived the idea with the help of T.C. and C.C.; T.F. collected

587

the data, conducted the analyses and wrote the manuscript; T.C. and C.C contributed to the

588

results, interpretation and writing.

589

Editor: Kate Parr

590

27

591

TABLES

592

Table 1 Fuel parameters, mean ( SD), of mountain forest ecosystem types (FET) of the

593

western Alps. Between-FET comparisons were performed using ANOVA and HSD post-hoc

594

tests; significant (P < 0.05) F and P-values are highlighted in bold with differences indicated

595

by different letters. Fuel parameters were fourth-root transformed before comparison tests.

596

Fuel parameters included the dead and live fine biomass available for burning by the surface

597

fire front (i.e. within the fuel bed, see the Methods). The mean annual (19602009)

598

temperatures and precipitation of FET are also presented. SVr, surface area to volume ratio.
FET-1

FET-2

FET-3

FET-4

drysubalpine

moistmontane

openmed

closedmed

Unit

F3,92 (P-

n = 27

n = 16

n = 29

n = 24

Annual temperature

(C)

70.3value)
(< 0.001)

4.5 (1.5) d

6.4 (1.2) c

8.3 (1.3) b

8.6 (1.2) a

Annual precipitation

(mm)

27.6 (< 0.001)

1193 (267) b

1551 (251) a

994 (112) d

1058 (133) c

Fine litter load

2.1 (0.104)

6.68 (3.49)

5 (1.91)

5.42 (2.84)

4.68 (2.18)

Dead woody load

1.8 (0.153)

10.46 (4.05)

11.42 (4.75)

8.75 (4.03)

9.36 (2.9)

8.2 (< 0.001)

0.52 (0.44) a

0.35 (0.89) bc

0.38 (0.43) ab

0.16 (0.25) c

4.4 (0.006)

0.56 (0.69) a

0.02 (0.07) b

0.79 (1.18) a

0.74 (1.16) ab

6.6 (< 0.001)

1.08 (0.84) a

0.37 (0.9) b

1.4 (1.61) a

0.9 (1.2) ab

1.0 (0.382)

11.54 (4.4)

11.79 (4.83)

10.15 (4.36)

10.26 (3.43)

(cm)

3.3 (0.024)

29.66 (12.01) a

17.54 (12.86) b

36.59 (28.71) a

29.52 (26.42) ab

(kg m3)

7.1 (< 0.001)

2.04 (0.96) a

0.98 (0.56) ab

0.88 (0.61) b

1.06 (0.72) b

(cm1)

7.2 (< 0.001)

77.11 (28.16) a

67.69 (12.72) ab

54.31 (14.44) b

57.42 (18.63) b

2.8 (0.051)

65.77 (7.01)

76.32 (1.75)

58.58 (10.85)

64.17 (18.03)

Herb load

(t ha1)

Shrub fine fuel load

Live fine fuel load
Total fuel load
Fuel depth
Shrub bulk density
Litter leave SVr
Shrub leave SVr

599

28

600

Table 2 Variance of forest fuel parameters (fourth root-transformed) explained by forest

601

structure, composition and climate components. These three environmental components were

602

described by the 12 most parsimonious variables for each. The total explained variance was

603

partitioned from 50 (partial) redundancy analyses into pure and interaction fractions. The

604

pure fraction describes the amount of variance that is exclusively explained by one

605

component (i.e. after accounting for other components). Significance was based on Monte

606

Carlo permutation tests. Variance and P-values were averaged. *P < 0.05, **P < 0.01,

607

***P < 0.001. SVr, surface area to volume ratio.

Explained variance (adjusted R 100)
Forest structure

Forest composition

Climate

Interaction

Total

Fine litter load

7*

27 ***

25

60 ***

Dead woody load

16 *

26

51 ***

Herb load

16 **

18 **

27

60 ***

Shrub fine fuel load

15 *

13 *

21

52 ***

Live fine fuel load

15 **

9*

45

74 ***

Total fuel load

24 **

18

50 ***

Fuel depth

11

28

50 ***

Shrub bulk density

33

40 *

Litter leave SVr

25 ***

47

78 ***

Shrub leave SVr

61 *

19

85 **

608

29

609

FIGURE CAPTIONS

610

Figure 1 Analytical framework representing the approach used to test differences in

611

understorey flammability between forest ecosystems according to plant composition and

612

environmental factors (forest structure and climate).

613

Figure 2 Map of forest ecosystem types (FET) in the western Alps: location of Inventaire

614

Forestier National (IFN) (+) and fuel sampling () sites (left), and FET contours (right).

615

Figure 3 Flammability of forest ecosystem types (FET). Weather-constant simulation outputs

616

(n = 96) of fire rate of spread (ROS; a), fire-line intensity (FLI; b) and heat released per unit

617

area (HPA; c) were compared between FET by ANOVA and HSD post-hoc tests. Differences

618

(P < 0.05) are indicated by different letters. Note the log-scale for the vertical axes.

619

Figure 4 Variance partitioning of community flammability [matrix of log-transformed rate of

620

spread (ROS), fire-line intensity (FLI) and heat released per unit area (HPA)] into unique and

621

common (inter.) contributions of (a) sets of predictors (forest structure, FS; composition, FC;

622

climate, CL) and (b) only the most parsimonious predictors. Variance was partitioned using

623

partial redundancy analyses, with a bootstrap procedure in (a) from which bootstrapped

624

values (n = 50) of adjusted R were compared between unique and common contributions by

625

ANOVA and HSD post-hoc tests, with significant differences (P < 0.05) indicated by

626

different letters.

627

Figure 5 Relationships between fire-line intensity (log-transformed) and tree cover (a) with

628

precipitation seasonality (spring to annual precipitation ratio) as a co-variable (b). Variation

629

of fire-line intensity with tree cover was regressed with a two-order polynomial function.

630

Means ( SD) of forest ecosystem types (FET) are depicted: black, drysubalpine (FET-1);

30

631

red, moistmontane (FET-2); green, openmediterranean (FET-3); blue, closed

632

mediterranean (FET-3).

633

Figure 6 Partial relationships between fire-line intensity (FLI; log-transformed) and

634

precipitation seasonality: spring to annual precipitation ratio (a), autumn precipitation (b) and

635

annual 95th percentile of daily drought code (c). Partial relationships between these climatic

636

gradients and the fine fuel load are also shown (bottom). The variance explained by the

637

environmental components X1 and X2 independently of the studied variable x (X1+X2 | x) was

638

removed from the redundancy analyses before modelling the partial relationship on the

639

unexplained variance fraction (indicated in brackets). Means ( SD) of forest ecosystem

640

types (FET) are depicted: black, drysubalpine (FET-1); red, moistmontane (FET-2); green

641

openmediterranean (FET-3); blue, closedmediterranean (FET-3). Environmental

642

components: FS, forest structure; FC, forest composition

643

Figure 7 Relationship between fire-line intensity (log-transformed) and the proportion of

644

needle-leaved versus broad-leaved trees in the community (a), and after accounting for tree

645

cover effect on fire-line intensity (partial regression) (b). Results of (partial) regression

646

analyses were significant for (a) but not for (b).

647

31

648

Figure 1

649
650

32

651

Figure 2

652
653

33

654

Figure 3

655
656

34

657

Figure 4

658
659

35

660

Figure 5

661
662

36

663

Figure 6

664
665

37

666

Figure 7

667
668

38