Project Objective

ArcFUEL delivers a complete, up-to-date, methodology for Fuel Classification Mapping (FCM on a Web-Geodatabase) based on ‘readily available” data, harmonized, accessible & interoperable according to INSPIRE principles, for the Mediterranean Region. The methodology is demonstrated via solid pilots in: Greece, Portugal, Italy and Spain 

The problem targeted

Effective Forest Fire (FF) Management requires knowledge of Fuel Classification Maps (FCMs) that are poorly available in Mediterranean countries since:

  • They are produced only at local or regional scale, without any regular updates and no standardized methodology.
  • They are not produced via a standard structure or harmonized according to INSPIRE.
  • They are heterogeneous as produced via different methods at different dates

Therefore available FCMs cannot support the systematic use of FF modelling at operational levels (prevention, suppression planning) of FF management.

“ArcFUEL” aims to cover this gap, and:

  • Standardizes a “production flow” producing FCMs.
  • Develops an INSPIRE procedure for FCMs production.
  • Produces pilot FCMs for Greece & Portugal (full countries), Italy & Spain (pilot regions).

Why FCMs for FF Management? 

Forest vegetation is considered as a Fuel. The structure and status of the Fuel-complex govern the dynamics of a fire. This is the reason why Fuel Models and their spatial patterns (i.e. FCMs) are significant for FF Management Actions during all four phases of the FF lifecycle:

  • Awareness phase (prior to the fire): requires pre-emptive methods to assessing FF Risks to minimizing impacts (environment, people). FF Risk is related to FCM: Risk Assessment is calculated as a function of several sub-indices as Fire Ignition & Propagation, Index of Fire Intensity, and flammability, all related to Fuel Models.
  • Emergency phase (during the fire): requires for reliable real-time information, risk identification–prioritization and fast decision making. FF Simulation is related to FCM
  • Impacts phase (after the fire): FF increase Risk of Flood. Risk of Flood is related to FCM, strongly related to Fuel Models.
  • Dissemination phase (lesson learnt): Dissemination when accumulating knowledge from past FF disasters, fire statistics, and Fuel Models used (e.g., linked to Corine Land Cover).

ArcFUEL Technology

Mathematical modelling/simulation of a fire front propagation in wildland forest & woodlands, requires spatial input parameters the most essential being topography, meteorology and forest vegetation. These parameters enable GIS-based fire simulators (e.g., ArcFIRE, 2011) to assessing fire pattern, growth-rate and behaviour. In turn, this knowledge is used in fire management, risk planning, prevention and mitigation decision-making.

ArcFUEL delivers forest fuel maps for wildfire propagation modelling, and as such, it delivers a complete and up-to-date methodology for Fuel Classification & Mapping based on readily available data INSPIRE interoperable at European scale. ArcFUEL is purposeful in the Mediterranean with its acute forest fire problems, thus has been applied and validated in Greece, Portugal, Italy and Spain.
Realistic predictions of fire growth ultimately depend on the availability, consistency and accuracy of input data-layers, compulsory to execute spatially-explicit fire behaviour models (Keane et al 1998a, Finney 1998). In this case:
•Topography and meteorological data is usually available at local, national, and global EU-scale to the fire management authorities for several applications.
•Maps of forest fuel type distribution are lacking at all spatial levels, a fact also related to the lack of standardized classification of forest fuel types which would support development of relative maps.
Recently the JRC funded FUELMAP (FUELMAP project, 2011) that provides a first approach to standardized forest fuel typology and to map of the distribution of such forest types in Europe. In 2012 the JRC revised the forest fuel types of FUELMAP, thus defined a new typology of Forest fuels, which is used as background in ArcFUEL.
The lack of updated forest fuel maps is key obstacle when introducing forest fire simulation applications in operational fire management, despite the literature richness and software technologies as FireGIS (Gonçalves, 1994), G-FMIS (Eftichidis et al 1998), (Lopes et al., 2002), and ArcFIRE (Bonazountas e al, 2011). ArcFUEL addresses the aforementioned need by establishing an integrated map production chain, based on a fuel mapping approach presented herewith.
ArcFUEL delivers a consistent fuel mapping methodology based on existing spatial data sets, easily accessible earth observation data, compatible with the JRC forest fuels typology. The ArcFUEL methodology and workflow advance in six technology-steps:
1. Definition of the fuel map classification scheme
2. Use of existing datasets for the mapping of the main forest fuel classes
3. Use of multi-temporal LANDSAT Thematic Mapper (TM) images for the distinction of fuel classes with different seasonal characteristics
4. Refinement of the produced fuel classes based on canopy cover density data
5. Merging of all produced map layers in a single layer
6. Further sub-classification of the produced maps based on the ECOREGIONS data

 In more details...

Pre-processing: Landsat images topological/geometrical corrections

  • Searching, selecting, ordering / downloading Landsat scenes of interest via the official sources (USGS’ Landsat archive, GloVis and EarthExplorer web- services). As the main production work-flow involves a bi-temporal analysis of a Vegetation Index, both summer and winter scenes were considered. The selection criteria were strict in the sense to get as much as possible cloud free acquisitions which, in addition, were timely as close as possible.
  • Storing all Landsat scenes in a geospatial aware data base
  • Trimming Landsat border fringes (wherever applicable, manually using the official WRS2 shapefile)
  • Converting Digital Numbers (DNs) to Top-of-Atmosphere Reflectances (ToAR) was done by normalizing either a statistical approach for most cloud-free scenes, or the 6S algorithm for scenes that were significantly cloud-contaminated (as per the cloud cover percentage estimation given in the metadata, as well as after visual control). Were processed with the 6S algorithm (i.atcorr).
  • Detecting clouds and cloud shadows
  • Topographically correcting Imagery by using the minnaert method based on the ASTER GDEM2 dataset.
  • Relatively normalizing images (separate handling for each spectral band and season) to balance up seasonal radiometric variations based on the Histogram Matching technique.
  • Creating a large surface reflectance mosaic (for each band) for the whole study area.


Processing:Forestfuel classes map production

  • Extracting major vegetation classes from existing land data bases, i.e. (i) forested areas (Broadleaved, Coniferous and Mixed) from JRC’s Forest Type Map (2006), (ii) surface fuels from JRC’s Forest Fuel Type Map (2006), (iii) ground and azonic fuels (and non fuels) as the remaining areas from the Corine Land Cover 2000/6
  • Separating deciduous and evergreen vegetation (and, similarly, grasses and shrubs) was gained via a bi-temporal classification approach, i.e. clustering and classifying the seasonal vegetation index difference inside the two main vegetation categories (forest and surface fuels). Clustering was performed with GRASS’ i.cluster module which implements a modification of the K-Means algorithm. Note that, unlike the ISODATA algorithm, which is yet another modification of the KMeans algorithm, i.cluster expects at least two inputs to run. Therefore, except for the NDVI difference, the NIR difference was subjected to the module. The clustered data sets were then classified using a Maximum Likelihood Classifier.
  • Sub-classifying the above vegetation classes based on forest density criteria. This was done by using JRC’s up-to-date Tree Cover Map (essentially a forest density map).
  • Filtering spatially and temporally relevant burned areas extracted from JRC’s Burnt Area Perimeters data base (EFFIS 2000-2012).
  • Additionally, subclassifying the fuel type classes can be achieved by incorporating JRC’s Environmental Zones (Eco-Regions) classification.

ArcFUEL fuel types

 Field validation methodology

  1. The purpose is  to create a data set of point values for validating the ArcFUEL map
  2. We overlay LUCAS points to the ArcFUEL map and we separate LUCAS in two subsets of data, one with the LUCAS points which description coincide with ArcFUEL characterization and another with the LUCAS points which description differs from ArcFUEL. We keep the first sunset as part of the ArcFUEL validation data set
  3. We use the second subset of point 2 and we apply the following methodology per region: We select  in each region randomly LUCAS points from  the second subset and create the second part of ArcFUEL data set as follows:
    1. We define a number of sampling points per region depending on its extent
    2. We take care including all ArcFUEL types  (AFT) in the data set derived
    3. Sample points are located in a distance <300m from road and in slope <40%
  4. We select randomly 5% of the number of points defined in 3.a  in elevation >1000m (LUCAS points are up to 1000m) taking into account the coverage of the ArcFUEL types
  5. During the field work
    1. We fill the form with the data that we have defined (we have developed an android application for collecting these data)
    2. We allow taking as sample point a point close to the LUCAS point but representing better the forest fuel type in the specific location
    3. During the field work we take additional sample points of representative fuel types along the way moving from one point to another 
  6. We aggregate all data from the subsets mentioned above in point 2 (LUCAS=AFT), 3 (LUCAS≠AFT), 4 (>1000m) and 5 (additional AFT based points) for deriving the ArcFUEL validation data set
  7. Defining the above methodology:
    1. We want to create a data set of point information for validating the accuracy and quality of the ArcFUEL map. Thus the main purpose is to create such a data set with distributed points, covering the area of interest and including all the fuel types present in each of the areas
    2. We used LUCAS points in order to have information for a number of these points without new ArcFUEL survey but based on the new LUCAS survey data
    3. We have to check deviation from LUCAS characterization in the field to see if and what is the problem
    4. Since  LUCAS survey has a different purpose we have to enrich the ArcFUEL data set with additional points e.g. points at elevation greater than 1000m or points of representative fuel types
    5. We consider having more than 250 sample points distributed all over Greece and Portugal for the ArcFUEL map validation data set after the end of the field survey