Curt et al. | Fires and management in New Caledonia

1

Understanding fire patterns and fire drivers for setting a sustainable 1

management policy of the New-Caledonian biodiversity hotspot 2

3

4

5

Thomas Curt 1, Laurent Borgniet

1, Thomas Ibanez

2,3, Vincent Moron

4,5, Christelle Hély

6 6

7

1 Irstea, UR EMAX, 3275 route Cézanne, F13182 Aix-en-Provence, France 8

2 Cerege - Centre Européen de Recherche et d‘Enseignement des Géosciences de l‘Environnement, 9

Europôle Méditerranéen de l‘Arbois, 13545 Aix en Provence, France 10

3 Current address IAC – Institut Agronomique néo-Calédonien (IAC), Diversité biologique et 11

fonctionnelle des écosystèmes terrestres, BPA5, 98848 Nouméa, New Caledonia. 12

4 Aix-Marseille Université, Aix en Provence, France 13

5 IRI, Columbia University, Palisades 10964, USA 14

6 Paléoenvironnements et Chronoécologie (PALECO- EPHE) - UMR 5059, Centre de Bio-15

Archéologie et d‘Ecologie, Institut de botanique - 163, rue Auguste Broussonet - 34090 Montpellier, 16

France 17

18

19

20

Corresponding Author: thomas.curt@irstea.fr 21

22

23

24

*ManuscriptClick here to view linked References

Curt et al. | Fires and management in New Caledonia

2

ABSTRACT 25

New Caledonia (NC) is a biodiversity hotspot sheltering terrestrial ecosystems of high 26

ecological and conservation value including tropical dry forests, rainforests, and maquis. 27

However, uncontrolled bushfires threaten this exceptional biodiversity. A science-based fire 28

management policy could reduce the impact of unwanted fires and help facing climate 29

change. However, to date, data on the location, extent, causal factors and spatial patterns of 30

fires had not been collected. We compiled a 12-year-long (1999-2011) spatially-explicit fire 31

database for NC using MODIS and Landsat data. Using boosted regression trees we 32

disentangled the role of anthropogenic factors, physiography, weather and vegetation on fire 33

activity. We also characterized the location of fires and the vegetation composition at the fire 34

edges, in order to determine which ecosystems were especially vulnerable. Fire size 35

distribution was typically asymmetric with many small fires (< 10 ha) and very few large fires 36

(> 500 ha). Ignitions were preferentially located close to villages, cities or roads, at low 37

elevation and linked to high values of fire weather index. Fires were larger at the end of the 38

dry season and during El Niño events. Most fires were bushfires burning in savannas, thickets 39

and maquis, while rainforests were rather ‗avoided‘ by fire. However, bushfires generally 40

propagated towards forests of high-conservation value, thus increasing the potential for forest 41

edge erosion. As savanna-forest and maquis-forest mosaics are dominant in the landscape, we 42

discuss the extent to which NC could become a ‗fire trap‘ where fire cannot be easily 43

extirpated. Based on our spatially-explicit information on fire activity, we make 44

recommendations for a sustainable forest and fire management policy which would balance 45

the traditional use of fire and the conservation of the most valuable ecosystems. In particular, 46

it may help by reducing the damages of large and destructive bushfires ignited during drought 47

peaks. 48

Keywords: New Caledonia, dry forest, rainforest, savanna, maquis, fire hazard, forest 49

management policy, biodiversity conservation 50

Curt et al. | Fires and management in New Caledonia

3

51

Introduction 52

53

New Caledonia (NC, southwest Pacific, -21°S, 165°E) has long been recognized as a 54

remarkable biodiversity hotspot (Myers, 1988; Myers et al., 2000; Mittermeier et al., 2004) 55

harboring terrestrial ecosystems of high ecological and conservation value (Jaffré et al., 56

1998), encompassing tropical dry forests, rainforests, and maquis (shrublands). About 75% of 57

the 3,371 plant species of NC are endemic (Morat et al., 2012). However, fires and other 58

disturbances have destroyed more than 50% of the original vegetation existing before the 59

Melanesian settlement (ca. 3000 BP). Original rainforest, dry forest and maquis have been 60

partly converted into secondary vegetation, mainly savannas, secondary thickets and maquis 61

(Jaffré et al., 1998). Nowadays, a major concern is to what extent the present fire regime may 62

threaten this exceptional flora (Jaffré et al., 1998; Pascal et al., 2008) because 25% of the 63

species are estimated to be at risk (IUCN, 2011). It is acknowledged worldwide that science-64

based fire management policies can reduce the impact of unwanted fires and help facing 65

climate change and uncertainties (e.g. FAO 2007; van Wilgen et al., 2014). However, to date, 66

no georeferenced database exists in NC and fire distribution, drivers and impacts on 67

ecosystems are poorly known. 68

69

Fire has driven vegetation dynamics for millennia in NC as in many tropical countries 70

(Cochrane 2003; Bond 2005) because it is both a natural disturbance and a management tool. 71

Palaeoecological records suggest that fires occurred long before the human settlement of NC 72

(ca. 3000 BP, Hope and Paske, 1998; Stevenson, 2004), with alternating periods of drought 73

with more frequent fires promoting the expansion of shrubland (so-called maquis) on 74

ultramafic rocks, and wetter periods with few fires characterized by rainforest expansion and 75

Curt et al. | Fires and management in New Caledonia

4

the establishment of secondary forest (McCoy et al., 1999). The increase of fires since 76

settlement of humans had major implications for the landscape of NC. First, it coincided with 77

the sudden expansion of savannas (McCoy et al., 1999; Stevenson, 2004). Savannas, probably 78

maintained by fire and grazing, currently cover 30% of NC and have replaced large areas of 79

dry forest and rainforest (Gillespie and Jaffré, 2003; Ibanez et al. 2013). Second, on the 80

ultramafic substratum, wildfires combined with forest explotation and mining has led to the 81

replacement of most original forests by secondary forests (Jaffre et al., 2010) and shrubby 82

vegetation in recent decades. Presently, fires threaten some endemic conifer species (Perry 83

and Enright, 2002; Jaffre et al., 2010) but also species typical of the dry forest (Bocquet et al., 84

2007). In this context, bushfires (i.e. fires originating in grasses, maquis, thickets or secondary 85

forests) constitute one of the main threats to NC biodiversity in the future (Jaffré et al., 1998; 86

Pascal et al. 2008). Indeed, large and uncontrolled bushfires such at the Montagne des Sources 87

(December, 2005) can burn thousands of hectares of maquis and forests. This has dramatically 88

increased the concerns of decision-makers about biodiversity conservation, as the ongoing 89

climate change could also increase fire activity (Leblon, 2005). 90

91

In this context, it was increasingly urgent to collect accurate and extensive data on 92

recent fires in NC to focus fire management efforts on the most valuable ecosystems that are 93

at risk. Until now, the exact boundaries of most fires were unknown and limited information 94

about fires has come from time-consuming studies of fire scars on trees (e.g. McCoy et al., 95

1999) and contemporary newspaper reports (e.g. Chevalier, 1996). Burned areas and fire 96

hotspot datasets now available from satellite observations (such as MODIS; Roy et al., 2008) 97

and Landsat provide information at short time intervals and offer data to characterize fire 98

patterns. Characterizing the major causes and locations of fire ignitions, as well as the extent, 99

pattern and location of burned areas are key issues for the long-term management and 100

Curt et al. | Fires and management in New Caledonia

5

conservation of many types of ecosystems throughout the world (e.g. Bergeron et al., 2002; 101

Keeley, 2002; Andersen et al., 2005; Bond, 2005). Actually, the spatial pattern of fires in the 102

landscape results from interactions between several top-down and bottom-up factors which 103

drive fire ignition and fire spread (Falk et al., 2007). They include anthropogenic drivers 104

(ignitions, fire prevention and fire suppression) and biophysical factors such as fire weather, 105

vegetation, and topography (e.g. Rothermel, 1983; Badia-Perpinya and Pallares-Barbera, 106

2006; Moreira et al., 2011). In NC fire ignition is predominant due to human activity (Perry 107

and Enright, 2002), and lightning fires are rare as in many tropical areas (Stott, 2000; 108

Cochrane, 2003). Anthropogenic fires are started for a variety of reasons including range 109

management, land clearance for cultivation, hunting and combating wild pigs, controlling 110

weeds, limiting the populations of the invasive little fire ant Wasmannia auropunctata, and as 111

a result of social conflicts (Udo, 2011; Dumas et al., 2013). It is likely that large fires may 112

result from ignitions by livestock breeders in grasslands to eliminate invading shrubs and 113

trees, while most small fires in mountainous areas may be lit for cultivation purpose. Some 114

fires may get out of control and become very large particularly when lit during prolonged or 115

intense dry spells, on steep slopes, in areas difficult to access, or when firefighting resources 116

are overwhelmed. Many studies worldwide have shown the ‗preference‘ or the ‗avoidance‘ of 117

fires for certain types of vegetation or certain topographic features (Viedma et al. 2009; 118

synthesis In Moreira et al. 2011). This selectivity results from the non-random location of 119

anthropogenic ignitions and from the local variations of physiography, fire weather, and 120

vegetation‘s composition and moisture content (Verbesselt et al. 2002; Leblon, 2005). 121

Assessing the drivers of and the spatial pattern of ignitions is crucial for fire prevention and 122

pre-positioning of firefighters, while assessing the drivers of burned areas has implications for 123

fire suppression and adapting land management practices. 124

125

Curt et al. | Fires and management in New Caledonia

6

In this study we constructed the first georeferenced fire database for NC (1999-2011) 126

using several remote sensing sources (MODIS, Landsat). We also collected georeferenced 127

information on anthropogenic and environmental drivers, and on fire weather. Using these 128

databases, we characterized the spatial and seasonal pattern of fires and we identified which 129

factors explained the ignitions versus the extent of burned areas. We finally sought to 130

determine the extent to which NC forests of high ecological and conservation value were 131

affected by bushfires ignited in flammable secondary vegetation (savannas, maquis and 132

thickets) by analyzing the contacts between those vegetation types along the fire edges. Our 133

final goal is to help improve the present fire management policy for biodiversity conservation. 134

135

136

137

Materials and Methods 138

139

Study area 140

141

New Caledonia is an archipelago of ca 18,500 km2 located in the southwest Pacific (21°30 S, 142

165°30 E); it has a subtropical oceanic climate with a mean air temperature of 23°C 143

(Caudmont and Maitrepierre, 2006). Mean annual rainfall ranges from 850 to 4500 mm, with 144

considerable spatial variation: the windward east coast is much wetter than the leeward west 145

coast. Temperatures and rainfall also vary seasonally, with a dry season from mid-August to 146

mid-December, which corresponds to the main period for bushfires (Barbero et al., 2011), and 147

a cool season from June to September. NC is under the influence of the El Niño Southern 148

Oscillation (ENSO) phenomenon (Ropelewski and Halpert, 1987; Nicet and Delcroix, 2000), 149

which has a strong impact on spatial and interannual climate variability: rainfall decreases to 150

Curt et al. | Fires and management in New Caledonia

7

less than 50% of its usual value during El Niño episodes, especially ―Central Pacific‖ events, 151

when anomalous positive sea surface temperature anomalies (SSTA) peak around the 152

dateline, while humid La Niña episodes reduce fire occurrence (Barbero et al., 2011). Trade 153

winds from south-east are predominant, and strong winds (> 8 m.s-1

) occur frequently and 154

favour the spread of fire (Caudmont and Maitrepierre, 2006). 155

156

The main island of New Caledonia (The Grande Terre) has an elongated central 157

mountain range with peaks at 700-1100 m a.s.l. (Fig. 1C) including two peaks higher than 158

1600 m, located close to the windward eastern coast. A major contrast exists with respect to 159

the soils, which are ultramafic or volcano-sedimentary. Ultramafic soils are nutrient-poor and 160

contain nickel at toxic concentration, with local iron crusts (McCoy et al., 1999) and they are 161

covered by the rainforest and the ultramafic maquis. Volcano-sedimentary soils are generally 162

richer in nutrients and more suitable for plant growth; they are covered with thickets, 163

savannas and rainforests (Fig. 1A). 164

165

Savannas, ultramafic maquis and secondary thickets represent the predominant 166

vegetation types, covering ca. 67% of the main NC island (source: Land Cover map of New 167

Caledonia, DTSI 2008; Table 1). The NC dry forests ecoregion contains only 240 small forest 168

fragments with 337 native plant species and it covers less than 10,000 ha in 2010 (ca. 1% of 169

the original forest area ; Jaffre et al., 1998; Gillespie and Jaffre, 2003). This dramatic 170

contraction has been attributed for the most part to uncontrolled fires ignited for range 171

management. Dry forest is the most vulnerable ecosystem of NC and should be the focus of 172

the greatest conservation efforts (Jaffré et al., 1998; Gillespie et al., 2014) because it is often 173

used as private rangelands. Rainforests cover about 400,000 ha and are especially species-rich 174

(2106 vascular species with 83.2% of them endemic) but they are progressively being 175

Curt et al. | Fires and management in New Caledonia

8

replaced by anthropogenic systems such as savannas at low and middle elevations. Rainforest 176

has increasingly become restricted to high elevation or inaccessible areas (Jaffré et al. 1994; 177

Jaffré et al. 1998; Ibanez et al., 2013). It is mainly located at high elevation along the central 178

main mountain ridge, and at lower elevation in wet talwegs (i.e. valley bottoms). Maquis is 179

the most common native vegetation type, covering about 600,000 ha (Jaffré et al., 1998) on 180

ultramafic substratum; the vegetation is dominated by sclerophyllous, slow-growing and 181

small-stature plants (Jaffré et al., 1998). The maquis may be replaced by a para-forested 182

vegetation dominated by the flammable species Gymnostoma deplancheanum (McCoy et al., 183

1999). Savannas and thickets constitute the other main vegetation type (about 800,000 ha) but 184

they have lower conservation value (i.e. 366 vascular species with only 15% endemic). 185

Savannas are often used for cattle, game or cultivation. This vegetation type is mostly 186

dominated by exotic grasses and shrubs (e.g. Melinis, Ocimum, Guava, Lantana, Leucaena) 187

and the native fire-resistant tree Melaleuca quinquenervia. In total, the edges (or ecotones) 188

between savannas, thickets and maquis, and the high-conservation value forests (i.e. rainforest 189

and dry forest) represent 54% of the whole area of contact between different vegetation types 190

in NC. 191

192

193

Fire data and fire contours 194

195

The contemporary fire history of NC is poorly known. In the 1870s, fire frequency increased 196

in association with vegetation clearance, prospecting, mining and logging of forest trees 197

(Chevalier, 1996). More recently, fire peaks have been reported between 1910-1960 and in 198

1971, due to wood-burning steam trains and extensive fires (Chevalier, 1996; McCoy et al., 199

1999). Because of the short duration of the period studied (12 years) imposed by the MODIS 200

Curt et al. | Fires and management in New Caledonia

9

sensor (Giglio et al., 2006) we focused here on the spatial patterns of fires rather than on the 201

temporal variations in the fire regime, which require multidecadal data (Curt et al., 2013). We 202

surveyed fires extensively from 1999 to 2011 using a set of satellite images embedded at 203

several spatial scales. Three products were available for the whole period: 204

205

(i) Active fire data based on the Terra-Aqua Moderate Resolution Imaging 206

Spectroradiometer (MODIS, resolution 250 m) (i.e. Giglio et al., 2006; Roy et al., 2008), 207

which reports active fires during the satellite overpass. MODIS achieves complete global 208

coverage twice a day. It allowed us to detect 1387 hotspots during the 1999-2011 period, this 209

number being partially biased by artifacts such as cities, mines or clouds. A filtering 210

algorithm was used to remove double counts of the same fire, defined as covering an area 211

larger than one pixel (1 km2) and also appearing on two successive images. This algorithm 212

allowed us to identify distinct fires and to determine their duration; 213

(ii) NDII (Normalized Difference Infrared Index) time series (see Dasgupta et al., 214

2007) based on MODIS data at a resolution of 250 m (the MOD13Q1 product) in which the 215

extent of burned areas was characterized using the 16-day MODIS product (250 m 216

resolution). Indeed, NDII combines near-infrared and short wave near-infrared bands and 217

their sum relates to the canopy water content (Hunt et al., 2011); this has been used to map 218

fire scars in a number of ecosystem types (e.g. Numata et al., 2011); 219

(iii) A Landsat time series was used to detect the smallest fires and to cross-check fire 220

boundaries at a 30-m spatial resolution. For this, we used 39 Landsat 7 ETM images from 221

1999 to 2010 to which geometric and radiometric corrections were applied. Among the 1387 222

active fire hotspots, 583 were confirmed as wildfires based on an analysis of the NDVI 223

extracted from Landsat. Finally, we created a map of 583 fires allowing us to determine the 224

fire size and spatial patterns such as shape. Only fires larger than ca 2 ha were detected and 225

Curt et al. | Fires and management in New Caledonia

10

mapped with accuracy, so most of the very small fires ignited for everyday purposes are not 226

considered here. We tried to assess as accurately as possible the location of the ignition point 227

for each fire by analyzing fire shape. Most small and medium fires were roughly elliptical 228

shaped (egg-shaped to fan-shaped, Finney, 2004), especially in uniform environmental 229

conditions (e.g. similar fuel and topography). In this case we determined the point of ignition 230

as the focal point of each fire contour located at the lowest elevation and upwind (Catchpole 231

et al., 1992) because most fires spread uphill in the direction of the dominant wind (Moreira 232

et al., 2011). The large fires often had an irregular shape, especially when they are located in 233

areas with heterogeneous topography. Their contour was multi-lobed because fire spreads 234

from the ignition point across the landscape along the preferential pathways for fire 235

propagation (Finney, 2004). In this case we determined the point of ignition as the lowest 236

point upwind at the opposite of the lobes. 237

238

239

Collection of georeferenced data 240

241

We collected georeferenced data for NC to try to predict the drivers of ignitions and of 242

fire size including the distance to human sources of ignition, the types of vegetation, the 243

physiographic variables, and the fire weather. All these data were gathered into the 244

Geographical Information System ArcGis 9.3.1 (ESRI, Redlands, CA, USA, 2008). 245

246

The NC population is about 258,000, with a few densely populated cities and small 247

villages scattered elsewhere. Because many fires are anthropogenic (Udo, 2010), we gathered 248

the map of the cities, villages as well as the main and secondary roads, and of the main 249

infrastructures features such as the airport (source: DTSI, Department of Technology and 250

Curt et al. | Fires and management in New Caledonia

11

Information Services, Nouméa, 2008; 1:25,000) (Fig. 1B). For vegetation and fuels, we used a 251

land cover map provided by the DTSI in 2008 (resolution 20 m), which comprised ten 252

different categories of land covers including various types of maquis and forests, but also 253

agricultural lands, savannas, mangrove and other low-flammable vegetation types. We 254

regrouped the vegetation into seven main fuel types which have different flammability and 255

combustibility (see Tinquaut, 2010), but also different dynamics and conservation value: (i) 256

savannas (SAV), (ii) maquis (MAQ, including sparse woodlands on maquis), (iii) forests on 257

ultramafic soils, (iv) forests on volcano-sedimentary soils (FVS), (v) thickets and sparse 258

secondary forest on volcano-sedimentary soils (TH), (vi) agricultural lands (AGR), and (vii) 259

other land cover types of low flammability such as mangroves (Fig. 1A). 260

261

The Digital Elevation Model (DEM, resolution 30 meters) provided information on 262

elevation, slope, aspect, slope curvature, and the location of the main and secondary streams 263

and rivers which may act as fire barriers and thus limit fire propagation (Pyne et al., 1996). 264

We computed the topographic wetness index (Jones et al., 2008) using ArcGis 9.3.1. TWI 265

indicates the spatial variations of soil moisture due to topography with high values indicating 266

wet soils which are generally located downslope or in valleys (Sorensen et al., 2006) and 267

could thus indicate lower flammability of vegetation. Finally, we finally computed a landform 268

index (LANDF) which classes the main landforms according to an increasing gradient of 269

topographical ruggedness and slope steepness from valley flats to cliffs and canyons (Riley 270

and Malecki, 2001). More complex landforms are expected to form topographic barriers to 271

fire propagation (Pyne et al., 1996). 272

273

Weather during a fire event is crucial for explaining its size and pattern. In order to 274

study this effect we used the Fire Weather Index (FWI), which is a unitless index that was 275

Curt et al. | Fires and management in New Caledonia

12

designed originally to forecast fire risk in Canada on the basis of past and current weather 276

conditions (van Wagner, 1987) including air temperature and relative humidity, surface wind 277

speed, and rainfall in the past 24 hours. It provides a uniform, numeric method of rating fire 278

danger throughout an area and has been used in the tropics (e.g. in Malaysia and Indonesia, 279

Dymond et al., 2004; Groot et al., 2007). The FWI aims to predict the probability and ease of 280

ignition and propagation of a fire on the forest floor, taking into account the weather 281

conditions. High FWI values indicate high fire danger (maximal value is ca. 100). In this 282

study FWI was computed daily for 14 weather stations in NC from which all necessary 283

weather variables were available (see Barbero et al., 2011), in order to test the effect of 284

weather on fire extent. Each fire was assigned the corresponding daily FWI value computed 285

for the nearest station. For fires that burned during several days (5% of the database) the FWI 286

value was computed as the mean of the daily values. The mean FWI value (± standard error) 287

for the days with fire activity was 13.6 ± 13. A considerable variation existed between the dry 288

western coast (20 ± 11) and the wet eastern coast and the central mountain chain (7 ± 4). 289

290

291

Analysis of fire and landscape patterning 292

293

We calculated the area and the length-perimeter ratio for each fire contour, using the Fragstats 294

software (McGarigal et al., 2002). Comparisons between fires frequency or fire size and the 295

different vegetation types were undertaken using analysis of variance (ANOVA) with the 296

Fisher Least Significant Difference (LSD) test when there was variance homogeneity, and the 297

non-parametric Kruskal-Wallis test when variances were unequal. 298

299

Curt et al. | Fires and management in New Caledonia

13

The distribution of the fire sizes was analyzed using the Lorenz curve and the Gini 300

index calculated as follows (Gini, 1936): 301

)+)((1= 1+

1=

1+ kk

nk

k

kk YYXXG ∑- eq. 1 302

where n is the number of fires, X is the cumulated fraction of the number of fires, and Y is the 303

cumulated fraction of the extent of fires. G ranges from 0 (variable distributed equally) to 1 304

(distribution of fires fully asymmetric). 305

306

To test whether, and to what degree, the fires were spatially aggregated together and to 307

potential ignitions sources (cities and villages) across NC, we used the pair correlation 308

function g(r) that provides a formal measure of the fire density at a given distance (r): 309

)2/()]([)( rrKdr

drg eq. 2 310

Where K(r) is the Ripley‘s K function (Ripley, 1981), and r is the distance to an arbitrary 311

point. g(r) > 1 indicates clustering while g(r) < 1 indicates regularity. 312

This function is related to the Ripley L function (Ripley, 1981), which has been used 313

previously for detecting spatial autocorrelation in ignition points or in burnt areas (e.g. Podur 314

et al., 2003; Vadrevu et al., 2008). The calculation of the pair correlation functions was 315

conducted using the Spatstat package with the R software (R Development Core Team, 2011), 316

which integrates edge effect corrections (Goreaud and Pelissier, 1999). We also computed the 317

distance between the ignition points for each fire and the nearest road, city, and village. 318

319

We acknowledge that our 12-years fire database was insufficient to characterize 320

temporal patterns of fires, but we calculated the fire cycle which is a simple index that 321

determines the time necessary to burn an area equivalent to the whole study area (Johnson et 322

al., 1999). Fire cycle was calculated for the whole NC and for the main vegetation types, 323

Curt et al. | Fires and management in New Caledonia

14

which allowed a rough comparison of the likelihood of reburning between all vegetation 324

types. The fire cycle is computed as: 325

SA

NFC

/ eq. 3 326

Where N is the number of years of the period studied, A is the total area burned, and S is the 327

total study area. 328

329

330

Comparing the drivers of ignition and burned areas with boosted regression trees 331

332

Boosting regression tree (BRT) analysis is a machine learning method which is well-suited for 333

exploring ecological variables and patchy data without the restrictive assumptions of 334

parametric statistics (e.g. Viedma et al., 2012; Aertsen et al., 2010), to optimize predictive 335

performance (De'ath, 2007), and it is considered to be flexible and easy to interpret (Elith et 336

al., 2008). Additionally, BRTs account for collinearity (i.e. linear regression between two 337

variables) which is likely to occur in such analyses using many anthropogenic and ecological 338

factors. It computes multiple regression trees on training data and sequentially fits the 339

residuals of the previous trees to provide a final ensemble tree – a process named ‗boosting‘ 340

(De'ath, 2007). BRT models were computed using the ‗gbm‘ R package (Ridgeway, 2013) 341

with a Bernoulli (logistic) error structure. Half of the data were used for building the model 342

(i.e. training dataset), and half to evaluate the accuracy of the classification (i.e. validation 343

dataset). Using the recommendation of Elith et al. (2008) the number of trees in each BRT 344

was set automatically to 20-folds cross-validation, we selected a bag fraction (training data 345

randomly selected for computing each tree) of 0.5, a shrinkage or learning rate of 0.005 which 346

controls the learning speed of the algorithm, and a tree complexity (or tree size) of 5. This 347

generated up to 2300 trees. We evaluated the quality of the models and their predictive 348

Curt et al. | Fires and management in New Caledonia

15

performance by using the area under the receiving operator curve (AUC; Pearce and Ferrier, 349

2000). The predictive performance was considered excellent when AUC > 0.9, and very poor 350

when AUC < 0.6. The BRT procedure comprised two main steps. In a first step, all 351

explanative variables were kept in order to get a complete model and a hierarchy of all 352

variables. The output is a figure with partial dependence functions (see Fig. 5 and 6): it 353

indicates the effect of each variable on the response after accounting for the average effect of 354

all variables in the model (Elith et al., 2008 Appendix S2). In a second step, the procedure 355

allows simplifying the model (the ‗drop-off‘ step) using methods analogous to backward 356

selection in regression (Elith et al., 2008). Only the most significant variables are thus kept 357

for the final explanation. 358

359

360

Detection and analysis of fire edges 361

362

Under a given set of weather conditions, fires may spread through a landscape and stop when 363

they reach natural obstacles such as abrupt changes in fuel type (e.g. when reaching a less 364

flammable fuel), or physiographic obstacles such as ridges or rivers (Viedma et al., 2009; 365

Moreira et al., 2011). Fires may also be stopped by fire suppression or rapid change of 366

meteorological conditions, irrespective of the natural obstacles. However, as most NC fires 367

propagate freely without active fire-fighting, we assumed that most fire edges would be 368

located at natural barriers, especially at interfaces between fuel types with different abilities to 369

ignite and to propagate fire because of large differences in fuel biomass or fuel moisture. For 370

large fires burning for more than one day, variations in meteorological conditions (e.g. 371

veering and/or speed of wind) could also affect the fire behavior but it was not possible to 372

obtain accurate information about this. Analyzing fuel discontinuities at fire edges was, 373

Curt et al. | Fires and management in New Caledonia

16

therefore, of particular importance for determining whether fires spread from savannas and 374

maquis to the endangered dry forests or rainforests. For each fire, we first defined a 30-m 375

wide buffer on the GIS inside the limit of the fire. We selected 50 random points within each 376

buffer (i.e. inside the burnt area versus outside the burnt area). Each point within the fire 377

contour was paired with the closest point outside it. For each point, we assessed the 378

predominant vegetation type. We calculated the proportion of a certain assemblage of 379

vegetation (e.g. savanna inside the fire contour and maquis outside) versus the reverse 380

(maquis inside the fire contour and savanna outside) for all the fires under study. The mean 381

difference of proportion between one case and its reverse indicated the extent to which the 382

former case was preferred to the other. We compared the values observed to those obtained 383

with 10,000 Monte Carlo random replicates to test the level of preference, which is indicated 384

by the p-value. We tested the effect of buffers of varying widths (30, 60 and 90 m) and found 385

that the 30-m buffer more accurate and stable information than wider buffers. 386

387

388

Results 389

390

During the 1999-2011 period, 583 fires recorded both on MODIS hotspots and Landsat 391

images were validated as having been wildfires (Table 1; Fig. 2). The total monitored burned 392

area was 25,511 ha (ca. 255 km2), i.e. about 2 % of the area of the main island of NC. The 393

mean fire size was 44 ± 162 ha with considerable intra- (p = 0.0065) and inter-annual 394

variations (p < 0.0001). The global fire cycle of NC was 720 years but it was much shorter for 395

ultramafic maquis, savannas, and thickets on volcano-sedimentary rocks, (i.e. 34, 41 and 60 396

years, respectively) than for forests (479 and 1346 years on ultramafic and volcano-397

sedimentary substrates, respectively). 398

Curt et al. | Fires and management in New Caledonia

17

399

The area burned per fire ranged from 2 to 3641 ha, with a highly asymmetric 400

distribution and a clear predominance of small fires (standardized coefficient of asymmetry = 401

188; Gini index = 0.74). Small fires (< 10 ha) represented 40% of the total number of fires 402

although they accounted for less than 4% of the total area burned. Large fires (> 200 ha) 403

represented only 3% of the total number of fires but they accounted for 43% of the total area 404

burned. The largest fire occurred in December 2005 at the Montagne des Sources (3641 ha) 405

and accounted for 13% of the whole burned area during the entire 1999-2011 period. The 406

mean area of fires did not differ significantly among the vegetation types (Table 1). 407

408

At interannual time scales, fires were significantly more numerous (86% of the total) 409

during the Central Pacific El Niño events associated with severe drought in September-410

December (2002, 2004, 2009), and less common (14%) during the La Niña wet years (1999, 411

2000, 2007, and 2008) or unclassified years (p < 0.0001; Fig. 3). Most fires, as would be 412

expected, occurred during the main dry season (August to December), when seasonal dry 413

conditions are combined with increasing temperatures, and then during the dry inter-season 414

(April and May). The seasonal distribution of the burned areas roughly followed the 415

distribution of the number of fires. However, the mean fire size decreased during the two peak 416

months of fire frequency (November and April). The figure 3 showed that burned areas were 417

larger when FWI was high, i.e. during the dry season and El Niño years. FWI was 418

significantly higher during El Niño years than during La Niña years (7.2 ± 0.3 versus 3.6 ± 419

0.6, respectively; F = 29.8, P < 0.0001), and during the austral summer (September to 420

February) than during the rest of the year (7.9 ± 5.0 versus 3.4 ± 3.4, respectively; F = 73.6, P 421

< 0.0001). 422

423

Curt et al. | Fires and management in New Caledonia

18

The BRTs indicated that ignitions were explained by human factors, fire weather and 424

topography while the burned areas were explained by vegetation fuels, fire weather and 425

human factors (Fig. 4). All models are considered as highly informative as AUC is 0.894 ± 426

0.005 and 0.919 ± 0.005 for ignitions and burned areas, respectively. Specifically, the BRT 427

drop-off procedure indicated the chief drivers for ignitions and burned areas. Ignitions were 428

preferentially located close to villages (< 10 km), rivers (< 5 km) or roads (< 3 km), they 429

preferentially occurred during days with a FWI > 10-15, at elevation lower than 350 m and in 430

areas with mean annual rainfall below 1200 mm (Fig. 5). Vegetation type weakly explained 431

ignitions. However, they were more likely in savannas, maquis, thickets and agricultural lands 432

whereas they were unlikely in forests and other vegetation types (Fig. 5). The burned areas 433

were mostly located preferentially in savannas, maquis and agricultural lands, and to a lesser 434

degree in thickets and in other vegetation types (Fig. 5). Forests burned much less, especially 435

those on volcano-sedimentary substrates. Burned areas were more likely at low to medium 436

rainfall (< 3000 mm/year) and unlikely for FWI < 10. Topographic variables were poor 437

predictors of the burned areas. These results were confirmed by the spatial analysis of fires. 438

They were spatially aggregated at the scale of the whole NC, with a significantly positive 439

Ripley‘s L(r) index from 1 to 20 km (Fig. 7). They were also significantly aggregated at 440

distances less than 10 km from villages. 441

442

The analysis of fire edges indicated that 60% of fires were stopped at a forest edge. 443

Bushfires ignited in savannas, maquis and thickets preferentially stopped in forests, while the 444

inverse (i.e. fires ignited in forest and stopping in the bush) was much less frequent (Table 2). 445

In example, 77% of fires that stopped at the ultramafic forest edge originated from ultramafic 446

maquis, while 46% of fires that stopped at the edge of volcano-sedimentary forests originated 447

from thickets and brushes on volcano-sedimentary substrates, and 41% from savannas. 448

Curt et al. | Fires and management in New Caledonia

19

Meanwhile, less than 4% of fires were ignited in forests and spread to other vegetation types. 449

It is noteworthy that fires ignited in savannas and stopping in maquis were much frequent than 450

the inverse. The total area of forest edges burned in NC during the 1999-2011 period 451

corresponded to 0.15% of the total area of forests on volcano-sedimentary soils, and 0.14% of 452

the total area of forest on ultramafic soils. 453

454

455

Discussion 456

457

Anthropogenic and environmental determinants of fire activity in NC 458

459

The anthropogenic influence on fire activity is acknowledged worldwide: humans 460

make landscapes more or less flammable through the modification of land covers, they ignite 461

fires, and they also extinguish them (Krawchuk et al., 2009; Bowman et al., 2011). In NC, the 462

causes of fires are poorly known and literature suggested that most are human-made (Udo, 463

2010; Dumas et al., 2013). In this study we have shown that most ignitions are likely 464

anthropogenic. Indeed, they are located near villages and roads where most population 465

inhabits, and they occurred preferentially in agricultural lands and savannas burned by 466

humans for agriculture or range management. Favorable fire weather (FWI > 10) is also 467

necessary for ignitions to occur. Natural ignitions by lightning strikes are possible but likely 468

rare (Chevalier 1996). In contrast, burned areas mainly depend on fuels and weather because 469

they preferentially occur in flammable fuel types (savannas, maquis and pastures) and in areas 470

with low to moderate rainfall and FWI > 10, which allow fire to better propagate into dry 471

fuels. 472

473

Curt et al. | Fires and management in New Caledonia

20

Climate exerts a superordinate top-down control worldwide over the resources to burn 474

(i.e. the distribution and quantity of flammable vegetation) and the fire weather (Krawchuk et 475

al., 2009; Meyn et al., 2007). Although the human role on ignition is undisputed in NC, we 476

have shown that weather and climate exert a forcing role on fire activity and fire size. First, 477

fires were more frequent or larger during dry El Niño events, which correspond to negative 478

rainfall anomalies, especially in austral spring (Barbero et al., 2011). Second, fires were more 479

frequent at the end of the dry season (September to December) and during the inter-season 480

(April-May) when high air temperatures are associated with dry topsoil and fuels (Barbero et 481

al., 2011) as in many tropical areas (Van der Werf et al., 2008; Aragão et al., 2007). Third, 482

fire activity is higher during periods with high FWI values, and in low-elevation areas with 483

low mean rainfall. Fire activity can thus be predicted from weather at two spatial and 484

temporal scales: (i) a rough estimation of the seasonal fire activity can be inferred from local 485

negative rainfall anomalies and the sea surface temperature in July over the Niño 4 box during 486

at least the previous 3 months (Moron et al., 2013); (ii) a local estimation can base on the 487

daily FWI values above 10 which is sufficient to promote ignitions and fire propagation in 488

NC. This low FWI threshold is low in comparison to Mediterranean areas (generally 20-90, 489

Moriondo et al. 2006). However, it is coherent with those of Dymond et al. (2004) or De 490

Groot et al. (2007) for Indonesia or Malaysia where fires ignite when FWI > 7. This fits with 491

the statement that fires can ignite and creep in moist litter and understory fuels in the Tropics 492

(Cochrane 2003). 493

Savannas, secondary thickets, grasslands and maquis are the most flammable and fire-494

prone vegetation types in NC. First, these grassy and shrubby fuels have fine particles, low 495

moisture content, and rapid curing during the dry season (Williams et al., 2002; de Groot et 496

al., 2005). Second, they exhibit rapid post fire fuel build-up as vegetation recovers rapidly by 497

auto-succession (Bond 2005). Indeed, grassland can reburn a single year after a previous fire, 498

Curt et al. | Fires and management in New Caledonia

21

and maquis can reburn after 50 (Jaffré 1997; McCoy et al., 1999). As a consequence, grass 499

and shrub communities (including secondary thickets) of NC have short fire cycle, similar to 500

that of fire-prone South African grasslands (2-10 yrs., Archibald et al., 2011) or fynbos (10-13 501

yrs., van Wilgen et al., 2010), shrublands of south-western Australia (47 yrs., O'Donnell et 502

al., 2011; 60 years, Bradstock et al., 2002a), and southern Californian chaparral (33-42 yrs., 503

Moritz et al., 2004). In contrast, NC forests on both substrates have long to very long fire 504

cycles typical of most rainforests such as those in Amazonia (Aragão and Shimabukuro, 505

2010) and Australia (Bowman, 2003), confirming that they are not very flammable per se. 506

507

We acknowledge that the 12-years-long fire sequence limits the scope of our results 508

because several decades are usually necessary to characterize accurately the spatial and 509

temporal patterns of fires in a region (Moritz et al. 2004). However, we found clear 510

differences of spatial patterns and drivers between ignitions and burned areas. This suggests 511

that our georeferenced database is sufficient for a first characterization of the spatial patterns 512

of fires in NC. 513

514

515

Fire-landscape interactions 516

517

Fire and vegetation strongly interact in NC. The present landscape results from strong 518

changes in vegetation since the Melanesian settlement ca. 3000 BP due to fire and other 519

disturbances (Jaffré et al., 1998). Two major vegetation mosaics cohabit nowadays: the 520

savanna-forest mosaic (Jaffré et al., 1998) which is frequent in the north-west and the maquis-521

forest mosaic which is frequent in the center and the south-east (Ibanez et al., 2013). This 522

feature influences the present fire regime but it will also likely drive the landscape changes in 523

Curt et al. | Fires and management in New Caledonia

22

the future. Indeed, we have shown that bushfires predominate. They originate from savannas, 524

thickets, grasslands and maquis, and then propagate until reaching forest edges. 525

Therefore, we suggest that bushfires constitute a high fire hazard with respect to 526

forests. Undoubtedly, the dry sclerophyllous forests of NC are top-threatened because they are 527

highly fire-sensitive and reduced to small remnants due to fires and grazing (Jaffré et al., 528

1998). These remnants should be especially protected against fires when located near 529

flammable vegetation. Rainforests are thought to be of low flammability due to the scarcity of 530

fuel and the high fuel moisture content of the understory fuel bed and the litter (Stott, 2000). 531

These characteristics typically generate slow and low-intensity surface fires creeping through 532

the leaf litter (Cochrane 2003). In addition, understory fuels and tree canopies are not 533

connected vertically, which limits the probability of fire crowning. However, concerns have 534

been raised recently for rainforests worldwide: recurrent surface fires promoted by a micro-535

climatic edge effect during the dry season (e.g. Didham and Lawton, 1999; Hennenberg et al., 536

2008) may increase the intensity of new fires and facilitate burning (Goldammer, 1999; 537

Barlow and Peres, 2008). While rainforest usually act as a barrier preventing the spread of 538

fire, unusually strong El Niño years have contributed to massive fires in forest areas of 539

Amazonia (Alencar et al., 2011) or Borneo (Wooster et al., 2012). Repeated bushfires have 540

thus significantly affected rainforest boundaries, limited the recruitment of rainforest species, 541

and finally restricted forests to ‗fire-proof‘ locations such as gullies in Amazonia (Aragão and 542

Shimabukuro, 2010), Colombia (Armenteras-Pascual et al., 2011) or Australia (Bowman, 543

2003). A similar trend has been observed using remote sensing in maquis-forests mosaics of 544

central NC (Ibanez et al., 2012). Such fires that reach forest edges may also cause serious 545

damage to the base of tree stems because most rainforest tree species lack efficient traits for 546

resisting fire (Ibanez et al., 2013). Forest contraction can be especially rapid in maquis- or 547

Curt et al. | Fires and management in New Caledonia

23

savanna-forest mosaics when massive residues from forest exploitation and logs are left on 548

the ground and burn intensively (e.g. Wooster et al., 2012 in Borneo). 549

The conservation of dry forests and rainforests are undoubtedly important issues 550

worldwide (Myers et al., 2000; Olson and Dinerstein, 2002) as the global rate of forest habitat 551

destruction is alarmingly high (FAO, 2010). A major concern is that the present vegetation 552

mosaics of NC may become ‗fire traps‘ where fire cannot be easily extirpated (see 553

Lindenmayer et al., 2011 for Australian ash wet forests). Indeed, both the savanna-forest 554

mosaic and the maquis-forest mosaic are considered as alternative stable states (Perry and 555

Enright, 2002; Ibanez et al., 2013) which associate a flammable vegetation (savanna and 556

maquis) to fire-sensitive forests with a high conservation value. Simulation exercises have 557

shown that fire could maintain for centuries in such ‗fire landscapes‘ (Perry and Enright, 558

2002) and finally degrade a part of forests. 559

560

Keys for a sustainable fire management policy: challenges and opportunities in New-561

Caledonia 562

563

We provided the first georeferenced fire database and analysis of spatial patterns of 564

ignitions and burned areas for the New-Caledonian biodiversity hotspot. A major implication 565

of this study is to improve the fire policy that was put in place a few years ago in order to 566

protect people and the ecological assets. Indeed, literature shows that a comprehensive and 567

coherent policy can greatly reduce the activity and impact of fires, while an inadequate one 568

can be totally ineffective (FAO, 2007). We argue that an efficient and sustainable fire policy 569

should be realistic (in particular from a cost-benefit point of view), accepted by the 570

population, focused on the main stakes to protect, and comprehensive (i.e. including the chain 571

Curt et al. | Fires and management in New Caledonia

24

from prevention to suppression). In addition, it should account for the possible increase of fire 572

activity due to climate change. New-Caledonia faces fire issues like many similar tropical 573

biomes worldwide, (Cary and Banks, 2000; Bradstock et al., 2002b; Hughes, 2003; Aragão et 574

al., 2007). However, it has some opportunities that can help controlling fire impacts: this 575

rather prosperous territory of France is well-regulated in contrast with many regions where 576

massive deforestation and land clearing are poorly regulated (e.g. Indonesia: Herawati and 577

Santoso, 2011). 578

579

Based on this study, we suggest two main directions for a sustainable fire policy in 580

NC. In the one hand, we hypothesize that a zero-burn policy is unrealistic because: (i) fire is 581

used for multiple purposes including land management as in many tropical areas (Bowman et 582

al., 2011) and it has cultural significance, in particular for the kanak population (Udo 2011), 583

thus total fire exclusion is unlikely to be accepted by the population; (ii) low severity 584

infrequent fires are compatible with the maintenance of maquis and savannas (Martin and 585

Sapsis, 1992); and (iii) firefighting is often hard due to a complex topography, a poor road 586

network, and the scarcity of fire suppression forces on the ground and aerial. On the other 587

hand, this study provided information useful to improve the present policy towards 588

biodiversity conservation and reduce the social and ecological damages of unplanned fires. 589

This is especially true for bushfires ignited during fire danger peaks, which often overwhelm 590

the firefighting capacity, escape and become large and devastative. In contrast, small fires 591

ignited by experimented people for cultivation or range management during periods of low 592

fire danger are tolerable. Conserving the traditional firewise use of fire is of interest 593

(Carmenta et al., 2011). 594

595

Curt et al. | Fires and management in New Caledonia

25

In order to improve the current fire policy we first strongly advocate for the gathering 596

of georeferenced fire data over the longest time period possible. This database should collect 597

and map all information on fires including the points of ignitions, the fire contours, and the 598

causes of ignition. Such a database exists for the metropolitan France 599

(http://www.promethee.com/) where it is a vital piece of information for the planning of 600

sustainable management (e.g. Williams et al., 2002), calculating fire return intervals, 601

modeling fire ignition and fire risk, and pre-position the fire suppression crews. Burned areas 602

and fire hotspots datasets now available from satellite observations (e.g. MODIS; Roy et al., 603

2008) and precise fire mapping using Landsat images would bring together important 604

information for the conservation of the NC‘s ecosystem mosaic and biodiversity. Secondly, 605

this study indicated where the areas most frequently burned were, which is important for 606

protecting the high-conservation and vulnerable assets located near these fire-prone areas. 607

Their vegetation should be managed, and fire crew should be pre-positioned to speed up the 608

intervention. We support the strategy of preferentially managing the contact areas between 609

fire-prone vegetation and forests, in order to protect and conserve the latter from fire. 610

Extending the protected areas and forest reserves with fire exclusion would also be effective 611

(Adeney et al., 2009; Armenteras et al., 2013). Thirdly, we have shown that ignitions are 612

more likely during El Niño years and especially when FWI is higher than 10. During these 613

periods, fire use should be restricted. A new fire danger scale based on FWI computation 614

(Prévifeu) has been set up by MeteoFrance in NC in order to predict daily fire risk, inform the 615

population, and prepare the adequate fire suppression crews. 616

617

In conclusion, this study provided spatially-explicit information to keep under 618

surveillance and manage the areas where most ignitions occur, and the most frequently burned 619

areas. It supports the idea that improving fire prevention targeted towards the areas at risk and 620

Curt et al. | Fires and management in New Caledonia

26

a firewise management strategy are efficient ways to prevent the postulated effects of climate 621

change and their ecological consequences. Research perspectives are numerous, including a 622

better assessment of fire impact on forest edges, and fire-vegetation simulations at the scale of 623

NC to assess which forests are especially vulnerable to future fires. 624

625

Acknowledgments 626

We acknowledge the New Caledonian Government (DTSI) for providing us with the land 627

cover maps, and MétéoFrance Nouméa (Yves Bidet) for meteorological data. This study was 628

funded by the Agence Nationale pour la Recherche (France) in the frame of the ANR-INC 629

Project ―Incendies et Biodiversité en Nouvelle Calédonie‖ (ANR-07-BDIV-008-01). 630

631

Curt et al. | Fires and management in New Caledonia

27

Figure captions 632

633

Fig. 1: Maps of New Caledonia. Left: Main vegetation types (Savannas are in yellow, 634

rainforest is in green, maquis are in red, and thickets on volcano-sedimentary substrates are in 635

grey). Note that dry sclerophyllous forests are restricted to small remnants on the western 636

coast (< 100 ha). Center: Location of the main cities, villages, and roads. Right: Elevation. 637

The stars represent the capitals of the New Caledonian provinces. 638

639

Fig. 2: Maps of the fire contours in New Caledonia (1999-2010). Fires larger than 2 hectares 640

have been drawn in red. Forests are in grey, and other vegetation types and land covers are in 641

white. The two zooms into the framed small areas illustrate the main fire patterns existing 642

(fires are hatched in red). A: Zoom into the Eastern coast with small fires. B: Zoom into the 643

large Montagne des Sources fire (3,641 ha). Both figures show that most fires start in non-644

forest vegetation (bushfires) and stop at the forest edge (in grey). 645

646

Fig. 3: Annual (A) and monthly (B) burned area (black points and full lines) and number of 647

fires (white points and dotted lines) in the New Caledonia main island between 1999 and 648

2010. Light grey areas indicate La Niña ENSO years (wet weather, presumably unfavourable 649

to fires) in A and the secondary dry inter-season in B, dark grey areas indicate El Niño ENSO 650

years (dry weather, presumably favorable to fires) in A and the main dry season in B. 651

652

Fig. 4. Relative contribution of the vegetation, weather, human and topographical variables in 653

the boosted regression trees (BRTs) predicting the fire ignition and the burned areas in New 654

Caledonia. Large bars are means and small bars are the standard deviation for an ensemble of 655

20 BRT models. All models are considered as highly informative as AUC is 0.894 ± 0.005 for 656

Curt et al. | Fires and management in New Caledonia

28

ignition AUC is 0.919 ± 0.005 for burned areas. VEG: vegetation types (AGR: agricultural 657

lands, MQ: maquis, OT: other vegetation types, SV: savanna, TH: thickets of secondary 658

forest, VF: forest on volcanosedimentary substratum, UF: forest on ultramafic substratum); 659

RAINF: mean annual rainfall; FWI: fire weather index; DISTVill: distance to the nearest 660

village (m); DISTCity: distance to the nearest large city (m); DISTRoad: distance to the 661

nearest road (m); ELEV: elevation (m); TWI: topographic wetness index (unitless); SLOPE: 662

slope angle (°); ASP: aspect (0 for North, 90 for East, 180 for South and 270 for West); 663

CURV: curvature of the slope (unitless); LANDF: type of landform (unitless). For details see 664

the Materials and Methods section 665

666

Fig. 5. Partial dependence of the vegetation, weather, human and topographical variables for 667

estimating the probability of ignitions in New Caledonia. The curve represents the mean value 668

and the grey area is the confidence interval for 20 models. Values located above the grey area 669

are statistically significantly and positively associated with ignitions, while values located 670

below the grey area are statistically significantly and negatively associated with ignitions. For 671

the meaning of variables see legend Fig. 4 672

673

Fig. 6. Partial dependence of the vegetation, weather, human and topographical variables for 674

estimating the probability of burned areas in New Caledonia. The curve represents the mean 675

value and the grey area is the confidence interval for 20 models. Values located above the 676

grey area are statistically significantly and positively associated with ignitions, while values 677

located below the grey area are statistically significantly and negativel associated with 678

ignitions. For the meaning of variables see legend Fig. 4 679

680

Curt et al. | Fires and management in New Caledonia

29

Fig. 7: Pair correlation function (Ripley L) between fires and cities (A) and fires and villages 681

(B). The full lines represent the observed patterns and the grey areas represent the complete 682

spatial randomness pattern (95 % confident intervals for 500 random simulations). Clustering 683

of fires around cities or villages is significant when the full line is above the grey area. 684

Curt et al. | Fires and management in New Caledonia

30

Fig. 1 685

686

687

Curt et al. | Fires and management in New Caledonia

31

Fig. 2 688

689

Curt et al. | Fires and management in New Caledonia

32

Fig. 3 690

691

692

693

694

695

696

697

Curt et al. | Fires and management in New Caledonia

33

Fig. 4 698

699

700

701

702

703

704

Curt et al. | Fires and management in New Caledonia

34

Fig. 5. 705

706

A B C D

E F G H

I J K L

707

708

709

710

711

Curt et al. | Fires and management in New Caledonia

35

Fig. 6. 712

713

A B C D

E F G H

I J K L

714

715

716

717

718

719

Curt et al. | Fires and management in New Caledonia

36

Fig. 7 720

721

722

723

Curt et al. | Fires and management in New Caledonia

37

Table 1: Main characteristics of New Caledonia wildfires (1999-2010) according to the main vegetation types. The total area burned and the mean fire size 724

were computed for the New Caledonia main island (so-called ‗NC‘ or ‗Grande Terre‘). For the mean fire size we computed the standard error (SE). Different 725

letters in rows indicated statistically significant difference (Duncan‘s multiple range test, at 95% confidence level. For each vegetation type, the fire cycle is 726

the time necessary to burn an area equivalent to the whole area of this vegetation in NC. 727

728

Vegetation Types

Total Area NC

(ha)

Number of Fires

(n)

Mean Fire Size

(mean ± SE, ha)

Total Area

Burned (ha)

Fire Cycle

(yrs.)

Savannas 407,054 267 38 ± 60 a 10,001 41

Ultramafic Maquis 369,283 149 72 ± 305 a 10,788 34

Ultramafic Forests 172,422 15 24 ± 38 a 360 479

Volcano-sedimentary thickets and brushes 248,231 141 29 ± 43 a 4,114 60

Volcano-sedimentary Forests 334,319 11 23 ± 25 a 248 1346

Total 1,531,308 583 44 ± 162 25,511 720

729

730

731

Curt et al. | Fires and management in New Caledonia

38

Table 2: Index of preference of certain vegetation types along the fire edges. The preference 732

is computed as the mean difference of surface between two vegetation types (left: inside the 733

fire contour, right: outside the fire contour) versus the same vegetation types but inversed 734

(left: outside the fire contour, right: inside the fire contour). A mean difference of 0.30 for 735

SAV_FUM versus FUM-SAV means that there is 30% more cases when fire propagated into 736

savanna and stopped at the edge of an ultramafic forest than cases when fire propagated into 737

ultramafic forest and stopped into the savanna. SAV: savanna; FUM: forest on ultramafics; 738

FVS: forest on volcanosedimentary substrates; MAQ: maquis; OTH: other vegetation types 739

including agricultural lands. SD is the standard deviation for 10,000 random simulations 740

741

Types of Edges Mean Difference SD p-value

SAV_FUM versus FUM-SAV 0.30 0.22 0.0020

SAV_FVS versus FVS_SAV 0.44 0.01 < 0.0001

SAV_OTH versus OTH_SAV 0.07 0.03 0.0057

SAV_MAQ versus MAQ_SAV 0.25 0.01 < 0.0001

MAQ_FUM versus

FUM_MAQ 0.24 0.02 < 0.0001

MAQ_FVS versus FVS_MAQ 0.38 0.01 < 0.0001

OTH_FUM versus FUM_OTH 0.14 0.13 NS

OTH_FVS versus FVS_OTH 0.27 0.06 < 0.0001

OTH_MAQ versus MAQ_OTH -0.07 0.03 0.0205

FVS_FUM versus FUM_FVS 0.32 0.15 0.0308

742

743

Curt et al. | Fires and management in New Caledonia

39

References 744

745

Adeney, J.M., Christensen, N.L., Pimm, S.L., 2009. Reserves Protect against Deforestation Fires in the Amazon. 746

Plos One 4. 747

Aertsen, W., Kint, V., van Orshoven, J., Özkan, K., Muys, B., 2010. Comparison and ranking of different 748

modelling techniques for prediction of site index in Mediterranean mountain forests. Ecological 749

Modelling 221, 1119-1130. 750

Alencar, A., Asner, G.P., Knapp, D., Zarin, D., 2011. Temporal variability of forest fires in eastern Amazonia. 751

Ecological Applications 21, 2397-2412. 752

Andersen, A.N., Cook, G.D., Corbett, L.K., Douglas, M.M., Eager, R.W., Russell-Smith, J., Setterfield, S.A., 753

Williams, R.J., Woinarski, J.C.Z., 2005. Fire frequency and biodiversity conservation in Australian 754

tropical savannas: implications from the Kapalga fire experiment. Austral Ecology 30, 155-167. 755

Aragão, L., Shimabukuro, Y.E., 2010. The Incidence of Fire in Amazonian Forests with Implications for REDD. 756

Science 328, 1275-1278. 757

Aragão, L.E.O.C., Malhi, Y., Roman-Cuesta, R.M., Saatchi, S., Anderson, L.O., Shimabukuro, Y.E., 2007. 758

Spatial patterns and fire response of recent Amazonian droughts. Geophys. Res. Lett. 34, L07701. 759

Archibald, S., Scholes, R., Roy, D., Roberts, G., Boschetti, L., 2011. Southern African fire regimes as revealed 760

by remote sensing International Journal of Wildland Fire 19, 861-878. 761

Armenteras-Pascual, D., Retana-Alumbreros, J., Molowny-Horas, R., Maria Roman-Cuesta, R., Gonzalez-762

Alonso, F., Morales-Rivas, M., 2011. Characterising fire spatial pattern interactions with climate and 763

vegetation in Colombia. Agricultural and Forest Meteorology 151, 279-289. 764

Badia-Perpinya, A., Pallares-Barbera, M., 2006. Spatial distribution of ignitions in Mediterranean periurban and 765

rural areas: the case of Catalonia. International Journal of Wildland Fire 15, 187-196. 766

Barbero, R., Moron, V., Mangeas, M., Despinoy, M., Hély, C., 2011. Relationships between MODIS and ATSR 767

fires and atmospheric variability in New Caledonia (SW Pacific). J. Geophysical Research 2011. 768

Barlow, J., Peres, C.A., 2008. Fire-mediated dieback and compositional cascade in an Amazonian forest. 769

Philosophical Transactions of the Royal Society B-Biological Sciences 363, 1787-1794. 770

Bergeron, Y., Leduc, A., Harvey, B.D., Gauthier, S., 2002. Natural fire regime: A guide for sustainable 771

management of the Canadian boreal forest. Silva Fennica 36, 81-95. 772

Curt et al. | Fires and management in New Caledonia

40

Bocquet, A., Chambrey, C., Derroire, G., Tassin, J., 2007. Modalities of colonization of woody savannah by dry 773

forest plants in New Caledonia. Revue D Ecologie-La Terre Et La Vie 62, 87-91. 774

Bond, W.J., 2005. Large parts of the world are brown or black: A different view on the 'Green World' 775

hypothesis. Journal of Vegetation Science 16, 261-266. 776

Bowman, D., 2003. Australian landscape burning: a continental and evolutionary perspective. Fire in 777

Ecosystems of South-West Western Australia: Impacts and Management, 107-118. 778

Bowman, D.M.J.S., Balch, J., Artaxo, P., Bond, W.J., Cochrane, M.A., D'Antonio, C.M., DeFries, R., Johnston, 779

F.H., Keeley, J.E., Krawchuk, M.A., Kull, C.A., Mack, M., Moritz, M.A., Pyne, S., Roos, C.I., Scott, 780

A.C., Sodhi, N.S., Swetnam, T.W., 2011. The human dimension of fire regimes on Earth. Journal of 781

Biogeography 38, 2223-2236. 782

Bradstock, R.A., Cohn, J.S., Bradstock, R.A., Williams, J.E., Gill, A.M., 2002a. Fire regimes and biodiversity in 783

semi-arid mallee ecosystems. Flammable Australia: the fire regimes and biodiversity of a continent., 238-784

258. 785

Bradstock, R.A., Williams, J.E., Gill, A.M., 2002b. Flammable Australia: the fire regimes and biodiversity of a 786

continent. Flammable Australia: the fire regimes and biodiversity of a continent, x + 462 pp. 787

Carmenta, R., Parry, L., Blackburn, A., Vermeylen, S., Barlow, J., 2011. Understanding human-fire interactions 788

in tropical forest regions: a case for interdisciplinary research across the natural and social sciences. 789

Ecology and Society 16, 53-63. 790

Cary, G.J., Banks, J.C.G., 2000. Fire regime sensitivity to global climate change: An Australian perspective. 791

Biomass Burning and Its Inter-Relationships with the Climate System 3, 233-246. 792

Catchpole, E.A., Alexander, M.E., Gill, A.M., 1992. Elliptical-fire perimeter- and area-intensity distributions. 793

Canadian Journal of Forest Research 22, 968-972. 794

Caudmont, S., Maitrepierre, L., 2006. Atlas climatique de la Nouvelle-Calédonie. . Direction interrégionale de 795

Nouvelle-Calédonie, 50 pp. 796

Chevalier, L., 1996. Terre de fer et de jade: de la baie de Prony à l'ile Ouen. Editions du Kagou, Nouméa (New 797

Caledonia), 149 pp. 798

Cochrane, M., 2003. Fire science for rainforests. Nature 421, 913-919. 799

Curt, T., Borgniet, L., Bouillon, C., 2013. Wildfire frequency varies with the size and shape of fuel types in 800

southeastern France: Implications for environmental management. Journal of Environmental Management 801

117, 150-161. 802

Curt et al. | Fires and management in New Caledonia

41

Dasgupta, S., Qu, J.J., Hao, X.J., Bhoi, S., 2007. Evaluating remotely sensed live fuel moisture estimations for 803

fire behavior predictions in Georgia, USA. Remote Sensing of Environment 108, 138-150. 804

De'ath, G., 2007. Boosted trees for ecological modeling and prediction. Ecology 88, 243-251. 805

de Groot, W.J., Wardati, Wang, Y.H., 2005. Calibrating the fine fuel moisture code for grass ignition potential in 806

Sumatra, Indonesia. International Journal of Wildland Fire 14, 161-168. 807

Didham, R.K., Lawton, J.H., 1999. Edge structure determines the magnitude of changes in microclimate and 808

vegetation structure in tropical forest fragments. Biotropica 31, 17-30. 809

Dumas, P., Toussaint, M., Herrenschmidt, J., Conte, A., Mangeas, M., 2013. Le risque de feux de brousse sur la 810

Grande Terre de Nouvelle-Calédonie : l‘Homme responsable, mais pas coupable. Revue Geographique de 811

l'Est (In French) 53, 1-20. 812

Dymond, C.C., Roswintiarti, O., Brady, M., 2004. Characterizing and mapping fuels for Malaysia and western 813

Indonesia. International Journal of Wildland Fire 13, 323-334. 814

Elith, J., Leathwick, J.R., Hastie, T., 2008. A working guide to boosted regression trees. Journal of Animal 815

Ecology 77, 802-813. 816

Falk, D.A., Miller, C., McKenzie, D., Black, A.E., 2007. Cross-scale analysis of fire regimes. Ecosystems 10, 817

809-823. 818

FAO, 2007. Fire management — global assessment 2006. A Thematic Study Prepared in the Framework of the 819

Global Forest Resource Assessment 2005. FAO, Rome. 820

FAO, 2010. Global forest resources assessment 2010. Food and Agriculture Organization of the United Nations 821

Rome, 80 pp. 822

Finney, M., 2004. FARSITE: Fire Area Simulator. Model Development and Evaluation. USDA Forest Service, 823

Research Paper RMRS-RP-4, 52 pp. 824

Giglio, L., Csiszar, I., Justice, C., 2006. Global distribution and seasonality of active fires as observed with the 825

Terra and Aqua Moderate Resolution Imaging Spectroradiometer (MODIS) sensors. Journal of 826

Geophysical Research 111, 1-12. 827

Gillespie, T.W., Jaffre, T., 2003. Tropical dry forests in New Caledonia. Biodiversity and Conservation 12, 828

1687-1697. 829

Gillespie, T.W., O'Neill, K., Keppel, G., Pau, S., Meyer, J.Y., Price, J.P., Jaffre, T., 2014. Prioritizing 830

conservation of tropical dry forests in the Pacific. Oryx 48, 337-344. 831

Curt et al. | Fires and management in New Caledonia

42

Gini, C., 1936. On the Measure of Concentration with Special Reference to Income and Statistics. Colorado 832

College Publication, General Series No. 208, 73-79. 833

Goldammer, J.G., 1999. Ecology - Forests on fire. Science 284, 1782-1783. 834

Goreaud, F., Pelissier, R., 1999. On explicit formulas of edge effect correction for Ripley's K-function. Journal 835

of Vegetation Science 10, 433-438. 836

Groot, W.J.d., Field, R.D., Brady, M.A., Roswintiarti, O., Mohamad, M., 2007. Development of the Indonesian 837

and Malaysian Fire Danger Rating Systems. Mitigation and Adaptation Strategies for Global Change 12, 838

165-180. 839

Hennenberg, K.J., Goetze, D., Szarzynski, J., Orthmann, B., Reineking, B., Steinke, I., Porembski, S., 2008. 840

Detection of seasonal variability in microclimatic borders and ecotones between forest and savanna. Basic 841

and Applied Ecology 9, 275-285. 842

Herawati, H., Santoso, H., 2011. Tropical forest susceptibility to and risk of fire under changing climate: A 843

review of fire nature, policy and institutions in Indonesia. Forest Policy and Economics 13, 227-233. 844

Hope, G., Paske, J., 1998. Pleistocene environmental upheaval in New Caledonia. Palaeogeogr. Palaeoclim. 845

Palaeoecol. 142, 1-21. 846

Hughes, L., 2003. Climate change and Australia: Trends, projections and impacts. Austral Ecology 28, 423-443. 847

Hunt, E.R., Li, L., Yilmaz, M.T., Jackson, T.J., 2011. Comparison of vegetation water contents derived from 848

shortwave-infrared and passive-microwave sensors over central Iowa. Remote Sensing of Environment 849

115, 2376-2383. 850

Ibanez, T., Borgniet, L., Mangeas, M., Gaucherel, C., GÉRaux, H., HÉLy, C., 2012. Rainforest and savanna 851

landscape dynamics in New Caledonia: Towards a mosaic of stable rainforest and savanna states? Austral 852

Ecology, no-no. 853

Ibanez, T., Curt, T., Hély, C., 2013. Low tolerance of New Caledonian secondary forest species to savanna fires. 854

Journal of Vegetation Science 24, 177-188. 855

IUCN, 2011. IUCN Red List of Threatened Species. International Union for Conservation of Nature and Natural 856

Resources (http://www.iucnredlist.org/). 857

Jaffré, T., Bouchet, P., Veillon, J., 1998. Threatened plants of New Caledonia: Is the system of protected areas 858

adequate? Biodiversity and Conservation 7, 109-135. 859

Jaffre, T., Bouchet, P., Veillon, J.M., 1998. Threatened plants of New Caledonia: Is the system of protected areas 860

adequate? Biodiversity and Conservation 7, 109-135. 861

Curt et al. | Fires and management in New Caledonia

43

Jaffre, T., Munzinger, J., Lowry, P.P., 2010. Threats to the conifer species found on New Caledonia's ultramafic 862

massifs and proposals for urgently needed measures to improve their protection. Biodiversity and 863

Conservation 19, 1485-1502. 864

Johnson, E.A., Miyanishi, K., O'Brien, N., 1999. Long-term reconstruction of the fire season in the mixedwood 865

boreal forest of Western Canada. Canadian Journal of Botany-Revue Canadienne De Botanique 77, 1185-866

1188. 867

Jones, G., Chew, J., Silverstein, R., Stalling, C., Sullivan, J., Troutwine, J., Weise, D., Garwood, D., 2008. 868

Spatial analysis of fuel treatment options for chaparral on the Angeles National Forest. General Technical 869

Report - Pacific Southwest Research Station, USDA Forest Service, 237-245. 870

Keeley, J.E., 2002. Fire management of California shrubland landscapes. Environmental Management 29, 395-871

408. 872

Krawchuk, M.A., Moritz, M.A., Parisien, M.-A., Van Dorn, J., Hayhoe, K., 2009. Global Pyrogeography: the 873

Current and Future Distribution of Wildfire. PLoS ONE 4, e5102. 874

Leblon, B., 2005. Monitoring forest fire danger with remote sensing. Natural Hazards 35, 343-359. 875

Lindenmayer, D.B., Hobbs, R.J., Likens, G.E., Krebs, C.J., Banks, S.C., 2011. Newly discovered landscape traps 876

produce regime shifts in wet forests. Proceedings of the National Academy of Sciences of the United 877

States of America 108, 15887-15891. 878

Martin, R., Sapsis, D., 1992. Fires as agents of biodiversity: Pyrodiversity promotes biodiversity. In: Proceedings 879

Symposium Biodiversity of Northern California, Univ. California, Berkeley, pp. 150-157. 880

McCoy, S., Jaffre, T., Rigault, F., Ash, J.E., 1999. Fire and succession in the ultramafic maquis of New 881

Caledonia. Journal of Biogeography 26, 579-594. 882

McGarigal, K., Cushman, S., Neel, M., Ene, E., 2002. FRAGSTATS: Spatial Pattern Analysis Program for 883

Categorical Maps. Computer software program produced by the authors at the University of 884

Massachusetts, Amherst. http://www.umass.edu/landeco/research/fragstats/fragstats.html. 885

Meyn, A., White, P.S., Buhk, C., Jentsch, A., 2007. Environmental drivers of large, infrequent wildfires: the 886

emerging conceptual model. Progress in Physical Geography 31, 287-312. 887

Mittermeier, R., Robles Gil, P., Hoffmann, M., Pilgrim, J., Brooks, T., Mittermeier, C., Lamoureux, J., da 888

Fonseca, G., 2004. Hospots Revisited. Chicago University Press, Chicago. 889

Curt et al. | Fires and management in New Caledonia

44

Morat, P., Jaffre, T., Tronchet, F., Munzinger, J., Pillon, Y., Veillon, J., Chalopin, M., 2012. The taxonomic 890

reference base "Florical" and characteristics of the native vascular flora of New Caledonia. In press., 243-891

254. 892

Moreira, F., Viedma, O., Arianoutsou, M., Curt, T., Koutsias, N., Rigolot, E., Barbati, A., Corona, P., Vaz, P., 893

Xanthopoulos, G., Mouillot, F., Bilgili, E., 2011. Landscape - wildfire interactions in southern Europe: 894

Implications for landscape management. Journal of Environmental Management 92, 2389-2402. 895

Moritz, M.A., Keeley, J.E., Johnson, E.A., Schaffner, A.A., 2004. Testing a basic assumption of shrubland fire 896

management: how important is fuel age? Frontiers in Ecology and the Environment 2, 67-72. 897

Moron, V., Barbero, R., Mangeas, M., Borgniet, L., Curt, T., Berti-Equille, L., 2013. Prediction of September-898

December Fire in New Caledonia (Southwestern Pacific) Using July Nino-4 Sea Surface Temperature 899

Index. Journal of Applied Meteorology and Climatology 52, 623-633. 900

Myers, N., 1988. Threatened biotas: "hot spots" in tropical forests. Environmentalist 8, 187-208. 901

Myers, N., Mittermeier, R.A., Mittermeier, C.G., da Fonseca, G.A.B., Kent, J., 2000. Biodiversity hotspots for 902

conservation priorities. Nature 403, 853-858. 903

Nicet, J., Delcroix, T., 2000. ENSO-Related Precipitation Changes in New Caledonia, Southwestern Tropical 904

Pacific: 1969–98. Mon. Weather Rev., 3001-3006. 905

Numata, I., Cochrane, M.A., Galvao, L.S., 2011. Analyzing the Impacts of Frequency and Severity of Forest Fire 906

on the Recovery of Disturbed Forest using Landsat Time Series and EO-1 Hyperion in the Southern 907

Brazilian Amazon. Earth Interactions 15. 908

O'Donnell, A.J., Boer, M.M., McCaw, W.L., Grierson, P.F., 2011. Vegetation and landscape connectivity control 909

wildfire intervals in unmanaged semi-arid shrublands and woodlands in Australia. Journal of 910

Biogeography 38, 112-124. 911

Olson, D., Dinerstein, E., 2002. The Global 200: Priority ecoregions for global conservation. Annals of the 912

Missouri Botanical Garden 89, 199-224. 913

Pascal, M., Richier de Forges, B., Le Guyader, H., Simberloff, D., 2008. Mining and other threats to the New 914

Caledonia biodiversity hotspot. Conservation Biology 22, 498-499. 915

Pearce, J., Ferrier, S., 2000. Evaluating the predictive performance of habitat models developed using logistic 916

regression. Ecological Modelling 133, 225-245. 917

Curt et al. | Fires and management in New Caledonia

45

Perry, G.L.W., Enright, N.J., 2002. Humans, fire and landscape pattern: understanding a maquis-forest complex, 918

Mont Do, New Caledonia, using a spatial 'state-and-transition' model. Journal of Biogeography 29, 1143-919

1158. 920

Podur, J., Martell, D.L., Csillag, F., 2003. Spatial patterns of lightning-caused forest fires in Ontario, 1976-1998. 921

Ecological Modelling 164, 1-20. 922

Pyne, S.J., Andrews, P.L., Laven, R.D., 1996. Introduction to wildland fire. 2nd ed. New York, NY: John Wiley 923

& Sons, 808 pp. 924

R Development Core Team, 2011. R: A language and environment for statistical computing, reference index 925

version v. 2.14.1. R Foundation for Statistical Computing, Vienna, Austria. 926

Ridgeway, G., 2013. Generalized Boosted Models: A guide to the gbm package. http://cran.r-project.org/, 34 pp. 927

Riley, S.J., Malecki, R.A., 2001. A landscape analysis of cougar distribution and abundance in Montana, USA. 928

Environmental Management 28, 317-323. 929

Ripley, B.D., 1981. Spatial Statistics. John Wiley and Sons, New York. 930

Ropelewski, C., Halpert, M., 1987. Global and regional scale precipitation patterns associated with the El 931

Niño/Southern Oscillation. Monthly Weather Review 115, 1606-1626. 932

Rothermel, R.C., 1983. How to predict the spread and intensity of forest and range fire. In. USDA Forest 933

Service, National Wildfire Coordinating Group, General Technical Report INT-143, pp. (Ogden, Utah). 934

Roy, D.P., Boschetti, L., Justice, C.O., Ju, J., 2008. The collection 5 MODIS burned area product - Global 935

evaluation by comparison with the MODIS active fire product. Remote Sensing of Environment 112, 936

3690-3707. 937

Sorensen, R., Zinko, U., Seibert, J., 2006. On the calculation of the topographic wetness index: evaluation of 938

different methods based on field observations. Hydrology and Earth System Sciences 10, 101-112. 939

Stevenson, J., 2004. A late-Holocene record of human impact from the southwest coast of New Caledonia. 940

Holocene, 888-898. 941

Stott, P., 2000. Combustion in tropical biomass fires: a critical review. Progress in Physical Geography 24, 355-942

377. 943

Tinquaut, F., 2010. Analyse des combustibles et modélisation du comportement du feu dans les principaux 944

écosystèmes de Nouvelle Calédonie. Master Thesis Univ. Clermont-Ferrand (France), 42 pp. 945

Udo, N., 2011. Fire, natural resources and territories in New Caledonia: Perception, use and management of fire. 946

ANR-INC Report, 40 pp. 947

Curt et al. | Fires and management in New Caledonia

46

Vadrevu, K.P., Badarinath, K.V.S., Eaturu, A., 2008. Spatio-temporal analysis of fire events in India: 948

implications for environmental conservation. Journal of Environmental Planning and Management 51, 949

817-832. 950

Van der Werf, G., Randerson, J., Giglio, L., Gobron, N., Dolman, A., 2008. Climate controls on the variability of 951

fires in the tropics and subtropics. Global Biogeochemical Cycles 22, 13-20. 952

van Wagner, C., 1987. Development and structure of the Canadian Forest Fire Weather Index System. Forestry 953

Technical Report - Canadian Forestry Service, viii + 37 pp. 954

van Wilgen, B.W., Forsyth, G.G., de Klerk, H., Das, S., Khuluse, S., Schmitz, P., 2010. Fire management in 955

Mediterranean-climate shrublands: a case study from the Cape fynbos, South Africa. Journal of Applied 956

Ecology 47, 631-638. 957

van Wilgen, B.W., Govender, N., Smit, I.P.J., MacFadyen, S., 2014. The ongoing development of a pragmatic 958

and adaptive fire management policy in a large African savanna protected area. Journal of Environmental 959

Management 132, 358-368. 960

Viedma, O., Angeler, D.G., Moreno, J.M., 2009. Landscape structural features control fire size in a 961

Mediterranean forested area of central Spain. International Journal of Wildland Fire 18, 575-583. 962

Viedma, O., Torres, I., Perez, B., Moreno, J.M., 2012. Modeling plant species richness using reflectance and 963

texture data derived from QuickBird in a recently burned area of Central Spain. Remote Sensing of 964

Environment 119, 208-221. 965

Williams, R.J., Griffiths, A.D., Allan, G.E., Bradstock, R.A., Williams, J.E., Gill, A.M., 2002. Fire regimes and 966

biodiversity in the savannas of northern Australia. Flammable Australia: the fire regimes and biodiversity 967

of a continent., 281-304. 968

Wooster, M.J., Perry, G.L.W., Zoumas, A., 2012. Fire, drought and El Nino relationships on Borneo (Southeast 969

Asia) in the pre-MODIS era (1980-2000). Biogeosciences 9, 317-340. 970

971

972