Content uploaded by George Eftichidis
Author content
All content in this area was uploaded by George Eftichidis on Jan 29, 2015
Content may be subject to copyright.
VII International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2014
1
Validation of burn scar mapping: Pilot case in Peloponnesus, Greece
G.Eftychidis1, G.Leventakis1, B.Hirn2, F.Ferrucci3 and G.Laneve4
1Center for Security Studies -KEMEA, Athens, Greece (g.eftychidis@kemea-research.gr; gleventakis@kemea.gr )
2 IESConsulting, Roma, Italy; b.hirn@iesconsulting.net
3University of Calabria, Rende, Italy; f.ferrucci@gmail.com
4 University of Rome 'La Sapienza', Dipartimento di Ingegneria Astronautica, Elettrica e Energetica, Roma, Italy;
laneve@psm.uniroma1.it
Keywords: PREFER, burn scar, burned area mapping, validation, Greece, fire management, forest
management
1. Introduction
Every year Southern Europe is affected by numerous uncontrolled forest fires having large
impact on the natural environment and the regional economies. PREFER project of the FP7 Space
Theme aims to support forest fire preparedness and prevention as well as post-fire management for
vegetation recovery. The project has the objective to develop an on line service for delivering a
series of map products, that are based on the use of Earth Observation data from space-borne
sensors, which may support the improvement of fire management in EU. Among the products are
maps of scars caused by forest fires that are produced using complex remote sensing techniques and
image processing algorithms.
PREFER is designed to improve systematically the current capacity of mapping burnt areas
both in terms of area threshold (EFFIS’ current thresholds are 40 ha in Rapid Fire and ca.10 ha in Fire
Damage assessment modes) as well as in terms of spatial resolution (currently 250m in Rapid Fire
and 30-50m in Fire Damage assessment modes, respectively). The need of systematic mapping at
much higher resolution combined with the task of managing large number of fires in real time
require improved process automation and robust solutions for operational burn scar mapping at
national or regional level. The high accuracy of mapping burn scars allows assessment of fire
affected areas to become operational.
2. The problem targeted
In recent decades, a higher frequency of wildfire in mid- and high latitudes has been observed, partly
as a result of climate warming. Burn scar refers to areas that are destroyed by forest fire, grass fire
and controlled burning and have not yet recovered (Liu et al, 2014). To assess and estimate the
impact of the fire on forest ecosystems, the area of the burn scars, as the most significant parameter
for running post-fire models, has to be acquired (Vafeidis and Drake 2005). At this stage, the task of
burn scar identification and mapping is undertaken mainly through analysis of remotely sensed data.
Multi-temporal satellite data provide several interpretative advantages over single date data for
mapping burned areas (Pereira et al. 1997, 1999, Eva and Lambin 1998). These include a reduction in
the likelihood of spectral confusion with spectrally similar static land cover types, the option to use
relative rather than absolute changes in spectral values to account for spectral differences between
pixels and dates, and the opportunity to define the date of burning more precisely.
The underlying problem for most of the burn scar mapping methods is that each of them can only
capture limited aspects of burned vegetation, but very few of the methods can comprehensively
examine as many aspects as possible, through which a more reliable result should be achieved.
Another problem with existing burn scar mapping methods is the difficulty in choosing an optimal
threshold. However, it has been noted that fixed threshold methods performed poorly for varying
atmospheric effects and different land covers (Barbosa et al., 1999).
VII International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2014
2
The Burn Scar mapping capacity is particularly challenging in areas with limited visibility because of
persistent cloud cover, haze and/or smoke (G.Laneve and G.Cadau, 2006). Such cases require the
development of robust mapping techniques based on X-band (Cosmo-SkyMED, TerraSAR-X) very
high spatial resolution data and C-band (Radarsat-2 and the forthcoming ESA Sentinel-1) Synthetic
Aperture Radar data. Three different products namely Burn Scar Map HR Optical at scale 1:25,000-
1:50,000, Burn Scar Map HR SAR at scale 1:10,000-1:50,000 and Burn Scar Map VHR at scale 1:2,000-
1:5,000 are produced using the multi-patented (B.Hirn and F.Ferrucci 2003 and 2006) procedure
called MYME2 (B.Hirn and F.Ferrucci, 2005), based on Pseudo Invariant Targets
(PCT/IT04/000376,WO2005/005926A1,EP/1642087).
The accuracy and quality of the burn scar layers, which are produced following the above
mentioned automatic procedures, have to be tested and validated in order to verify the
performance of the algorithms and the accuracy of the approach adopted for the development of
the relative products.
3. Methodology
The objective of the validation is to identify any systematic deviation between burn scar
mapping and real data based on observations as well as to identify eventual false detected fire spots
in order to investigate the source of the errors and potentially improve or document the
performance of the PREFER algorithms and the data processing techniques.
Fig. 1 Pilot area of PREFER project in Greece (Messinia Region)
VII International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2014
3
The validation of the burn scar product includes comparison of the data layers produced by
the automated PREFER techniques and the MYME2 procedures against field observed or ancillary
and archived fire data in the pilot areas of PREFER established in all the Mediterranean EU countries.
The Greek pilot area (Figure 1) for the PREFER project, where the PREFER burn scar mapping
products are validated, is the region of Messinia in Peloponnesus (Greece). Furthermore PREFER
results are compared with data delivered by other automated burned area mapping services.
Mountainous areas cover the northern and eastern part of the Greek pilot area although low
forest vegetation is spread in the entire region. Forest fire problem is quite significant in the region,
which suffered severe damages during the mega-fires of 2007 in Peloponnesus.
For the map validation, fire data and burned scar maps provided by IESC were considered.
The assessment was done during ground survey conducted the autumn of 2013. The survey referred
to the burned areas of the year 2013 while for the previous years (2009-2012), fire data from the
files of the Forest Service, the Hellenic Fire Corps and the Messinia Regional Administration services
were used.
Fig. 2 Missed fire (left), false fire (center) and properly detected fire (right) in the Greek pilot area of PREFER
The validation methodology of the burn scar mapping includes both qualitative and
quantitative aspects. Missed fires (not detected) and false fires (detected without being verified in
the field or the fire records) define the qualitative aspects of the validation methodology (Fig.2).
Assessment of the accuracy of the burned area and calculation of the fire perimeter are considered
as quantitative aspects of the metthodology (Fig.3). Burn scar mapping data can be compared with
GPS measurements on the ground and field collected data in case of recent fire (before the rain
period start) as well as with fire records in case of past fires, in which case fire trace are lost.
Fig. 3 Greek pilot area of PREFER (left) and comparison of PREFER Burn Scar Map (right) and Diachronic BSM of NOA (center)
Vector data of the burned areas within the reference period (2009-2013) were collected from
the Forest Service and compared with the produced burn scar maps of PREFER.
Interviews with potential stakeholders of the specific PREFER product are organized in order
to ensure the user acceptance concerning the accuracy of the product as well as for collecting
VII International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2014
4
feedback concerning the final design of the burn scar maps that will be delivered by the on-line
service of PREFER.
4. Main results
Although the validation task shall continue for the next years, preliminary results show that
the PREFER algorithms perform quite well and the results of qualitative validation against the 2013
fires are very accurate. Some minor issues that identified concerning past fires need to be checked
further since they may be influenced by the reliability of the fire records kept by the public services.
The length of time that the spectral signature of the burned area is detectable after the fire
depends on the physical evolution of the post burn surface (vegetation regrowth, dissipation of ash
and charcoal by wind and rain). Thus the dates of the satellite images can be important in particular
towards the beginning and the end of the fire season. Similarly distinction among scars due to
vegetation loss caused by forest fires and vegetation loss caused by other factors (eg forest area
clearing) shall be further investigated.
A first quantitative comparison of the results against records of the forest service show that
the accuracy of areal extent ranges between 73% and 92%. In any case the estimation of the
relevant PREFER service is more accurate than other automated products tested in the same region
in the past.
Furthermore the burn scar mapping product of PREFER need to be designed properly in order
to fulfill the operational requirements of the potential end users (Forest Service, Fire Service,
Environmental organizations, Local and Regional Authorities) distinguishing among wildfires and
prescribed fires in agricultural areas. Finally product availability, continuity and cost issues shall be
considered in context of the PREFER service exploitation plan.
5. Conclusions
The results of the validation of the burn scar mapping products of the PREFER portfolio of EO-
based maps are quite encouraging. The product specifications and its accuracy are suitable for
addressing the operational fire management needs. Further validation and redesign of the product
format in context of the PREFER service will be required. Relevant inputs from the respective
stakeholders’ groups are currently collected in order to document thoroughly the required
performance of the relative technique and procedures as well as for shaping properly the product in
order to be acceptable by competent end users.
Acknowledgements
The work described in this paper has been co-funded by the European Union’s Seventh
Framework Programme under Grant Agreement no. 312931 (Project PREFER Space-based
Information Support for Prevention and REcovery of Forest Fires Emergency in the MediteRranean
Area). The content of this publication is the sole responsibility of the authors and does not
necessarily reflect the views of the European Commission.
References
Barbosa P.M., Gregoire J.M. and Pereira J.M.C., 1999. An algorithm for extracting burned areas from
time series of AVHRR GAC data applied at a continental scale.Remote Sensing of Environment, 69,
pp. 253–263.
VII International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2014
5
Eva H. and Lambin E.F., 1998. Remote sensing of biomass burning in tropical regions: Sampling issues
and multisensor approach. Remote Sensing of Environment, 64, pp.292–315.
Hirn B. and Ferrucci F. "Automatic method for detectingand mapping burnt areas without
vegetation", Patent 1642087, 2006
Hirn B. and Ferrucci F. "Metodo Automatico Di Rilevazionee Mappatura, in Particolare Di Aree
Bruciate E Prive Di Vegetazione, e RelativoApparato", PatentRM2003A000336, 2003
Hirn B. and Ferrucci F. 2005. MYME2: A multi-payload integratedprocedure for the automated, high-
resolution remote sensing of burn scars, Proc. IEEE IGARSS, 2005
Laneve G. and Cadau E. G. 2006. Assessment of the fire detectionlimit using SEVIRI/MSG sensor,
Proc. IEEE Geoscienceand Remote Sensing Symp., IGARSS06, pp.4157 -4160 2006
Liu Y, Dai Q, Liu J, Liu S, Yang J, 2014. Study of Burn Scar Extraction Automatically Based on Level Set
Method using Remote Sensing Data. PLoS ONE 9(2): e87480. doi:10.1371/journal.pone.0087480
Vafeidis A.T. and Drake N.A., 2005. A two-step method for estimating the extent of burnt areas with
the use of coarse-resolution data. International Journal of Remote Sensing,26, pp. 2441–2459.