Supplementary Online Material

Bottom-up Influences Control Large-Fire Size in the Western Mediterranean Basin

Table S1. Descriptive Statistics for the Complete set of Variables Explored as Fire-size Drivers

Variable / Median / Range / Interquartile range
Climate and weather
*Fire circularity (CIR) / 0.65 / 0.28-0.90 / 0.58-0.72
a*Initial spread index (ISI) / 10 / 12 / 2-52 / 2–52 / 8-14 / 8–16
a*Buildup index (BUI) / 167 / 170 / 40-508 / 40-511 / 132-221 / 132-221
aFire weather index (FWI) / 39 / 40 / 10-109 / 10-109 / 30-46 / 31-50
*Autumn-winter rain (Rain_Oct-Mar, mm) / 638 / 108-2546 / 446-1012
*Autumn-spring rain (Rain_Oct-Jun, mm) / 799 / 124-2805 / 574-1222
*Winter rain (Rain_Jan-Mar, mm) / 209 / 23-2024 / 119-368
*Winter-spring rain (Rain_Jan-Jun, mm) / 412 / 39-2283 / 299-654
Fuel composition and connectivity
*Forest cover (%) / 22.5 / 0-98.8 / 9.2-45.1
*Deciduous forest cover (%) / 0.0 / 0-65.8 / 0.5-15.9
*Eucalypt forest cover (%) / 0.0 / 0-96.2 / 0-0.1
*Conifer forest cover (%) / 4.6 / 0-96.8 / 0-8.0
*Shrubland-grassland cover (Cov_Shrub-Grass, %) / 72.1 / 0.2-100 / 47.0-85.5
*Forest effective mesh size (eMSfor, ha) / 2.7 / 0-6,232 / 0.4-28.3
*Shrubland effect. mesh size (eMSshrub, ha) / 140.3 / 0.0-10,905 / 41.9-304.3
Agriculture effective mesh size (eMSagr, ha) / 0.1 / 0.0 –75.8 / 0.0 – 0.5
Pyrodiversity
Mean fire recurrence (MFR) / 0.46 / 0-3.16 / 0.13-0.88
*Largest patch index (LPI) / 64.3 / 10.0-100 / 45.6-87.7
Patch density (PD) / 3.57 / 0.02-39.07 / 1.50-7.06
*Mean patch size (MPS, ha) / 28.0 / 2.6-3952.1 / 14.1-66.6
Median patch size (MedPS, ha) / 1.8 / 0.1-1194.6 / 0.9-5.4
*Edge density (ED, m ha-1) / 96.1 / 0.03-326.92 / 66.85-130.86
*Mean shape index (MSI) / 1.78 / 1.46-3.88 / 1.67-1.96
Area-weighted mean shape index (AWMSI) / 2.53 / 1.55-13.43 / 2.18-3.07
Landscape Shape Index (LSI) / 451.5 / 146.0-1631.5 / 340.3-612.8
*Area-weighted mean patch fractal dimension (AWMPFD) / 1.32 / 1.22 – 1.40 / 1.30 - 1.33
Patch richness density (PRD, no. 100 ha-1) / 0.86 / 0.01-4.12 / 0.37-1.53
Shannon’s diversity index (SHDI) / 0.66 / 0-1.54 / 0.35-0.91
Shannon’s evenness index (SHEI) / 0.65 / 0-1.00 / 0.39-0.79
Topography
*Elevation (m) / 588 / 38-1,455 / 384-791
Slope (%) / 9.3 / 1.3-27.0 / 6.5-12.4
Vector ruggedness measure (VRM) / 0.0030 / 0.0003-0.0167 / 0.0020-0.0044
*Terrain ruggedness index (TRI) / 3.6 / 1.6-6.3 / 3.1-4.2
*Elevation-relief ratio (DISS) / 0.52 / 0.40-0.61 / 0.50-0.53
Surface relief ratio (SRR) / 0.51 / 0.44-0.57 / 0.50-0.52
Surface area ratio (SAR) / 920 / 901-1047 / 911-933
Topographic position index (TPI) / 0.42 / -1.61 – 3.18 / 0.13-0.90
Compound topographic index (CTI) / 6.74 / 5.69-8.12 / 6.49-7.03
*Heat load index (HLI) / 4230 / 3123-5769 / 4050-4452
*Topographic solar-radiation aspect index (TRASP) / 0.503 / 0.075-0.891 / 0.406-0.599
Fire suppression-related metrics
*Building density (Hdisp, ha ha-1) / 0.002 / 0-0.226 / 0-0.009
Building dispersion metrics:
Girth (dGirth) / 0.002 / 0-0.409 / 0.000-0.024
Depth (dDepth) / 0.012 / 0-0.453 / 0.002-0.052
*Perimeter (dPer) / 0.017 / 0-0.472 / 0-0.060
*Road density (RDens, km km-2) / 0.47 / 0-6.25 / 0.09-1.65

a Mean / Maximum values

Variables considered for the BRT analysis are signaled with *.

Figure S1. Partial dependence plots (means and 95% confidence intervals) for each explanatory variable of fire size through BRT modeling (an ensemble of 100 runs), representing estimated marginal effects when the remaining variables are held at their average. Ticks along the inside top of the plots show deciles (10% quantiles) of the observed distribution across the explanatory variables. The abbreviation of variables is defined in Table 1.

Table S2. Size of the Ten Highest Interactions Between Independent Variables Included in the BRT Model

Variables / Interaction size / 106
AWMPFD - ED / 14.31
BUI - ED / 9.07
eMSfor - ED / 7.87
ISI - ED / 2.80
eMSshrub - ED / 2.67
eMSshrub - ISI / 2.63
AWMPFD - eMSshrub / 2.09
AWMPFD - eMSfor / 1.46
BUI - eMSshrub / 1.33
AWMPFD - ISI / 1.26

Interactions reflect a tree complexity of 5, that is, the model was fitted with up to 5-way interactions. The remaining interactions were lower by one order of magnitude and less.