A method to produce a flexible and customized fuel models dataset

Abstract

Simulation of vegetation fires very often resorts to fire-behavior models that need fuel models as input. The lack of fuel models is a common problem for researchers and fire managers because its quality depends on the quality/availability of data. In this study we present a method that combines expert- and research-based knowledge with several sources of data (e.g. satellite and fieldwork) to produce customized fuel models maps. Fuel model classes are assigned to land cover types to produce a basemap, which is then updated using empirical and user-defined rules. This method produces a map of surface fuel models as detailed as possible. It is reproducible, and its flexibility relies on juxtaposing independent spatial datasets, depending on their quality or availability. This method is developed in a ModelBuilder/ArcGis toolbox named FUMOD that integrates ten sub-models. FUMOD has been used to map the Portuguese annual fuel models grids since 2019, supporting regional fire risk assessments and suppression decisions. Datasets, models and supplementary files are available in a repository (https://github.com/anasa30/PT_FuelModels). •FUMOD is a flexible toolbox with ten sub-models included that maps updated Portuguese fuel models.

Keywords: Automatic updates; Burned areas; Expert-based knowledge; FUMOD: updated fuel models gridded dataset; Fire-atlas; Flexible approach; Fuel models; Land cover; Satellite data; Spectral vegetation indexes; Time since last fire.

Graphical abstract

Fig. 1

Example extracted from the ArcGIS Survey 123 fuel models application, which includes the list of the total number of samples collected (331) with the identification of data and time, and the corresponding surface fuel model code (a); the total area (ha) of each fuel model class collected (b); a picture taken to illustrate a pine forest area with dense natural regeneration (c) where the NFFL-4 fuel model was assigned as shown in the yellow polygon overlaid in the Google Earth image (d).

Fig. 2

Frequency distributions of fuel model classes (a); percentage of vegetation cover types (b); and color heat (legend with colors distributed in deciles) table of the distribution in % of the fuel models by time since last fire (TSLF), considering the 331 records (for a description of fuel model codes (FM Code) see Fig.S1). For example, 59 % of the areas that burned less than 3 years ago had the V-MH fuel model class assigned.

Fig. 3

Flowchart representing the proposed FUMOD method, composed by a sequence of the 10 sub-models (SM) and eight datasets (labeled from A to H) to automatically produce the final updated fuel models map (FMM). See Table 1 for the description of each dataset, and the next section for the description of each SM.

Fig. 4

Comparative box-plots of NDVI vegetation index as a function of burned area age (TSLF). NDVI values were extracted from mean monthly composites of Sentinel-2 reflectance data for the month before the date of fire occurrence. Outliers were removed from the plot. Red dots and horizontal lines inside the box represent the mean and median, respectively. The upper and lower box limits represent the 25^th and 75^th percentiles and the whiskers extend to 10^th and 90^th percentiles.

Fig. 5

Comparison between the distribution of Sentinel-2 derived SAVI vegetation index in shrubs and sparse vegetation land cover types. SAVI values were extracted from monthly composites of Sentinel-2 mean surface reflectance for the month before fire occurrence. Red dots and horizontal lines inside the box represent the mean and median, respectively. The upper and lower box limits represent the 25^th and 75^th percentiles and the whiskers extend to 10^th and 90^th percentiles.

Fig. 6

Example of a fire that burned in shrubland (COS2018) in July 2020 in Portugal. The fuel model classes assigned depend on TSLF (shrub age) with rules set from combining authors and fieldwork experience (a); the updated SM5 for this area results from using the rules based on thresholds from the SAVI composite for June of 2021 (see text for details) (b). Different levels of vegetation cover inside the burned area perimeter are shown in the basemap true-color composite (Fig. 6c). Fuel model classes correspond to: 98=unburnable; 233=V-MAa; 234=V-MAb; 235=V-MH; 236=V-MMa; and 237=V-MMb (see Table S1 for descriptions).

Fig. 7

Two examples of surface vegetation in eucalypt plantations five years after the fire. The understory is mainly composed of tall shrubs and woody debris. Photos are from fieldwork done in the Central region of Portugal.

Fig. 8

Spatial distribution of the Atlantic versus Mediterranean forests and shrublands communities, based on the combination of bioclimatic and lithologic data (BIOLIT). See Fig. S1 for the matrix resultant from all classes combination of both variables.

Fig. 9

Fuel models maps (FMM) produced after applying the FUMOD model to the national land cover map (2018).

Fig. 10

Distribution of fieldwork fuel model classes in each of the main fire-prone land cover types. It is shown four rings, from the inner to the outermost: cork oak, eucalypt, pine forests, and shrublands.

Fig. 11

Distribution of fieldwork fuel model classes per fuel model class mapped, in each of the main fire-prone land cover types. Values inside bars represent the number of inventories per fuel model class.

Fig. 12

Example of a visited area where land cover type In COS2018 is shrubs (then fuel model assigned in V-MMa (results from SM5), but in the recently developed land cover map of COSc2021 is a mixture of shrubs and grasses (a), which was confirmed in the field (b,c).