Fire behaviour of the large fires of 2007 in Greece

Abstract

The large wildfires of the fire season of 2007 in Greece that left approximately 270,000 ha burned and 78 people dead, presented a clear case of extreme fire behavior. This behavior is described in brief here. Furthermore, in the context of a fire modeling study, twelve of these fires were documented in detail and their behaviour was studied. Fire behaviour observations were matched with observations and measurements of weather, topography and forest fuels that were recorded at the time of the fire. Additional complimentary information was gathered after the fires were controlled. Subsequently, detailed case studies were generated for each of the documented fires providing data that were then used for the development of a reliable database of fire behaviour observations. A subset of this database was used to test fire behaviour predictions from the BehavePlus fire behaviour prediction system versus field observations for fuel situations corresponding to a fuel model for "evergreen-schlerophyllous shrublands (maquis) (1.5 -3 m)" that had been developed, earlier for Greece. The results were positive, showing good agreement between observations and predictions, leading to the conclusion that, at least this fuel model can be used reliably for surface fire behaviour prediction for this fuel type in Greece. Additional investigations led to positive preliminary results in regard to the possibility to predict active crown fire behaviour in Pinus halepensis forests based on surface fire behaviour predictions obtained from BehavePlus and the shrub fuel model above as input.

Full text

EDITED BY
DOMINGOS XAVIER VIEGAS

VI International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2010
Fire behaviour of the large fires of 2007 in Greece
Miltiadis Athanasiou
“Environmental Impact Assessment Studies”
8 Thoma Paleologou st., 13673 Acharnes, info@m-athanasiou.gr
Gavriil Xanthopoulos
National Agricultural Research Foundation, Greece,
Institute of Mediterranean Forest Ecosystems & Forest Products Technology
Hymettus Avenue & 1 N. Chlorou st., 11528 Athens, gxnrtc@fria.gr
Abstract
The large wildfires of the fire season of 2007 in Greece that left approximately
270,000 ha burned and 78 people dead, presented a clear case of extreme fire behavior.
This behavior is described in brief here. Furthermore, in the context of a fire modeling
study, twelve of these fires were documented in detail and their behaviour was studied.
Fire behaviour observations were matched with observations and measurements of
weather, topography and forest fuels that were recorded at the time of the fire. Additional
complimentary information was gathered after the fires were controlled. Subsequently,
detailed case studies were generated for each of the documented fires providing data that
were then used for the development of a reliable database of fire behaviour observations.
A subset of this database was used to test fire behaviour predictions from the
BehavePlus fire behaviour prediction system versus field observations for fuel situations
corresponding to a fuel model for “evergreen-schlerophyllous shrublands (maquis) (1.5 - 3
m)” that had been developed, earlier for Greece. The results were positive, showing good
agreement between observations and predictions, leading to the conclusion that, at least
this fuel model can be used reliably for surface fire behaviour prediction for this fuel type
in Greece. Additional investigations led to positive preliminary results in regard to the
possibility to predict active crown fire behaviour in Pinus halepensis forests based on
surface fire behaviour predictions obtained from BehavePlus and the shrub fuel model
above as input.
Keywords: Wildfire behaviour prediction, fuel model, shrub fuels, wildfire management, Greece
1. Introduction
The fire season of 2007 in Greece was the worst in recent history. More than
270,000 ha burned, 110 villages were affected directly by the fire fronts and more than
3000 homes were totally or partially destroyed. Most important, a total of seventy eight
(78) people, mostly civilians, lost their lives in a series of fire related accidents
(Xanthopoulos et al. 2009). Whereas certain operational firefighting weaknesses may be
partially responsible for the huge disaster, it is the extreme fire behavior of many of the
fires of that summer that explains why the firefighting mechanism reached, at times, the
point of total collapse. The work presented here, provides a brief description of that fire
season and the type of fire behavior that the firefighters had to face, based to a large extent
on the fire documentation efforts of the first author in the context of a fire behavior

VI International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2010
modelling study. Part of the results of the study as related to the behavior of the fires of
2007, are presented here as well.
2. The behaviour of the large fires of 2007 in Greece
Signs about the difficulty of the fire season of 2007 were evident early on. Snowfall
in winter was deficient, even on high mountain slopes, making it impossible for many ski
areas to operate. Rainfall was also well below normal. In response to the signs, the
government increased the firefighting capacity of the country by contracting more heavy
lift helicopters.
In the last ten days of June a heat wave with temperatures exceeding 40 oC in many
parts of the country contributed to an early start of the “main” fire season. It started with
two large fires in middle Greece, one of them claiming two civilian lives in the evening of
June 27 near the village Agia on mount Ossa. At about the same time, at 19:30 on June 27,
the fire of mount Parnis, started. Mount Parnis forms the NW part of the basin in which the
city of Athens is built and is a National Park. The cause of the fire was sparks from an
overloaded power line near the village of Stefani in the prefecture of Viotia, 15 km to the
west of the mountain. Initial attack failed as aerial resources had little time to act before
sunset and ground forces proved ineffective. In the six hours between 23:30 of 27/6/2007
and 05:30 of 28/6/2007, the fire front spread over a distance of approximately 2,2 km (rate
of spread 0.36 km/h) burning mainly shrubs or Aleppo pine (Pinus halepensis) forest with
evergreen shrub understory.
In the morning of June 28, increased firefighting efforts, mainly from the air, nearly
achieved control by 10:00. However, some judgement errors in firefighting allowed the fire
to continue. On-site weather measurements around 12:30 indicated that the temperature was
34 oC, the relative humidity was 31% and the wind was blowing from various directions
with a mean wind speed of 27 km/h and gusts up to 57 km/h. In the 7 hours and 15 minutes
between 06:45 and 14:00 hours, the fire front spread over a distance of 4 km, exhibiting an
average rate of spread of 0.55 km/h. At 16:50 the fire front crossed without strong
resistance the last road and firebreak intended to protect mount Parnis from a fire
approaching from the west. After that the fire started moving into the core of the Park in
three separate uphill runs, passing from the Aleppo pine forest to the fir (Abies
cephallonica) forest at the top of the mountain. It accelerated as the slope increased and
developed a strong convection column. The main run was in a SE direction, in a long draw
between two peaks. Average fire spread during this run was estimated at 4.5 km/h with a
peak spread of 6.6 km/h. The wind was from the west, measured at 9,4 km/h at this time
with gusts up to 17 km/h. Relative humidity was at 31% and temperature was at 31 oC.
However, channelling effects in the draw may have increased the local wind speed
significantly. After reaching the rim of the mountain facing Athens at about 23:00 the fire
calmed down. It was controlled in the next morning as it was moving slowly downslope
without wind. The total burned area reached 5,600 ha, including 2,180 ha of fir forest.
A second heat wave hit the country between July 17 and 26, and with it came the
second round of disaster. Numerous fires erupted and among them many escaped initial
attack and grew to large sizes. The most notable was a fire that started on the 24th of July at
Kounina, a village at the base of the steep mountains in the area of Aigialia (near the city of
Aigio on the north coast of Peloponnese). Initial attack from the air was delayed as there
were other fires in progress. On the 25th of July, with the help of the slope and a strong

VI International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2010
wind, it accelerated and burned, practically unobstructed, through the forests, agricultural
cultivations and villages on the slopes of the mountains, until it reached the top. Within the
next three days it burned about 30,000 ha, destroyed more than 70 homes in nine villages
and killed three citizens. Most notably, and due to the drought, in addition to all the fast
spreading fires in southern Greece, a series of fires in the high elevation and low
flammability forests of northern Greece, started from various causes. Fought poorly they
became surprisingly large. They kept burning for more than ten days until, on August 5,
that part of Greece received a significant quantity of rain.
The fires in southern Greece continued in August with fire behaviour worsening as
the vegetation was becoming more water stressed. On August 16 the people of Athens
watched the fire of Penteli mountain which is opposite mount Parnis forming the NE part of
the basin of Athens. The fire started at about 10:15. The temperature at 11:11 was 30 oC,
relative humidity was 33% and the wind blew from the NE varying, according to the
measurements of meteorological stations in the area, between 13 km/h and 26 km/h. Initial
attack errors allowed the fire to grow in the young Aleppo pine forest and Mediterranean
shrub fields that had regenerated after previous fires. Soon, with the help of an unstable
atmospheric profile, a strong convection column was formed which created strong local
indrafts near the fire perimeter. The aerial resources that arrived at this “plume dominated
fire” (Rothermel 1991) by 11:00 were not able to make drops near the front due to the
strong turbulence. Later, at about 12:30 the wind increased to 35 km/h the fire became wind
driven and the strong fleet of Canadair waterbombers and heavy lift helicopters (Erickson
S-62, MIL MI-26) started drops that helped control the fire by late evening. The burned
area reached 900 ha. The maximum rate of spread of the fire, while it run unobstructed, was
measured up to 2.5 km/h in Aleppo pine stands with shrub understory. The overall spread
of the fire from 10:30 to 12:30, burning mainly Mediterranean shrub and young Aleppo
pine regeneration, was calculated at 1.2 km/h.
Intense fires continued in the following days stressing the demoralized firefighting
mechanism to the limit. Then, on 21st of August, the third heat wave of the summer arrived
bringing the temperature up to more than 39 degrees Celsius for the next three days
preheating and drying the fuels. Two fires on mount Parnon and mount Taygetus in
Peloponnese that started on Thursday, August 23, were allowed to burn overnight. On
August 24 relative humidity dropped to extremely low levels (8-20%) and wind picked-up,
in many cases, exceeding 25-30 km/h with gusts reaching 50 km/h or more, bringing fire
danger to extreme levels. When fires, starting in this explosive situation, were faced with
ineffective initial attack, the stage was set for disaster.
In the morning of Friday 24th the fires of mount Parnon and Taygetus started raging
out of control. A new fire erupted near the towns of Oitylo & Areopolis, roughly 30 km
south of the fire of Taygetos. Within a few hours it caused six deaths. It attracted
immediately the attention of the firefighters and the media until, in the afternoon, the news
about massive fatalities in Ilia (western Peloponnese) started coming. Twenty three people
near the village of Artemida were surrounded by a fire that was spreading in a mix of olive
groves and clubs of Aleppo pine trees. They were trapped between the main fire front,
which was spreading at a rate exceeding 5 km/h and spot fires that were produced
massively by the falling burning embers from the huge convection column and were drawn
back towards the column by strong indrafts (Xanthopoulos et al. 2009). Additionally, the
rate of spread of some bursts was measured from 7 km/h up to 10 km/h (Athanasiou, 2008).
Four more fires with similar extreme behavior were added within the next few hours
in Peloponnese. Furthermore, two major fires erupted on Evia island while the area around

VI International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2010
Athens also received its share of fire activity. For three days, firefighting collapsed almost
completely in spite of significant firefighting help received from abroad. Sixty seven (67)
people died due to these fires. Most of the dead were caught in the open. They were trying
to flee or were surrounded by the fire while trying to save their property. Added to the
eleven (11) people who had died earlier they add-up to seventy eight (78) dead during the
fire season, a death toll exceeding by far anything the country had experienced in the past.
Figure 1: The fires in south Greece, growing rapidly on the 25th of August (NASA image,
captured by the MODIS aqua sensor).
Figure 2: Evolution of the fires in central-west Peloponnese (August 24th to September 1st).

VI International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2010
3. Fire behaviour modelling study
Fire behaviour prediction systems play an important role in modern forest fire
management. The core of such systems is the fire behaviour model on which they are
based. During the last forty years, many such models have been developed for predicting
rate of spread, flame length, fire line intensity, etc.
Fire behaviour models can be adopted in wildfire management (forest management,
fire prevention and suppression) only if their degree of reliability, strengths and weaknesses
are well known. Preliminary evaluation can be performed in laboratory burns and in low
intensity fires in the open but full confidence can only be achieved if model predictions are
tested against observations of the behaviour of large wildfires.
Rothermel’s mathematical model (1972) for predicting fire spread in wildland fuels
is probably the most widely used model of this kind in the world. The popularity it enjoys is
due mainly to its practicality, good documentation and a significant number of studies that
tested its prediction performance for various types of fuels. Due to this extensive testing
both strengths and weaknesses are quite well known. However, much of this testing has
taken place in the laboratory, in experimental field burns or in actual wildfires of low to
medium intensity (Van Wilgen et al., 1985, Xanthopoulos, 1986, McAlpine and
Xanthopoulos, 1989). The extreme wildfires of 2007 in Greece provided an opportunity for
model testing on fires spreading under extreme conditions in some important fuel
complexes found in the country. More specifically, the current work, focused on
documenting fire behaviour in the summer of 2007, and then testing the capacity of the
BehavePlus fire behaviour prediction system (Andrews et al., 2005) which is based on
Rothermel’s (1972) model, to predict that behaviour.
4. Methodology
Twelve wildfires were documented in the regions of Attica, Viotia, Corinthia,
Achaia, Ilia, Arcadia and Messinia, from June to September 2007; overall, 12,500 km were
travelled and 750 hours of fieldwork were spent (Athanasiou, 2008).
Table 1: The twelve documented forest fires of 2007

VI International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2010
A set of specific procedures was used, for acquiring sufficient and objective data
and ensuring observer’s safety during fieldwork, as well as for the effective and efficient
utilization of the obtained data in the office. In all cases, the following data were observed,
measured and recorded:
a. Rate of spread of fire front or fire finger, in kilometers per hour (km/h),
(ROSobserved). Observer’s positions were identified by using a Garmin etrex Summit
GPS device and the geographic coordinates were recorded. Photographs of the
sequential locations of the fire front, were taken using a Canon Powershot S3is
digital camera which has a lens with a focal length ranging from 36 mm (wide
Angle) to 432 mm (telephoto). For all shots the horizontal azimuth was recorded
using a compass. Thus, later in the office, fire front locations were pinpointed on the
map using the ArcGIS 9.3 Geographic Information System (GIS) software by ESRI
corporation. Additionally, the time of every photo capture was known from the data
recorded automatically by the camera on the digital photo file that was in JPG
format. By knowing the exact time for every fire front location (Alexander and
Thomas, 2003a), ROSobserved values were calculated (figure 3(a) & (b)).
Figure 3(a) & (b): Fire spread of 68 meters in tall maquis (yellow arrow), up a 96% slope
(44ο), within a period of 1 minute and 14 seconds. ROSobserved = 3.31 km/h.
b. Flame length, in meters (m), (FLobserved) by comparing the flames with objects of
known dimensions such as trees, farmhouses, paths, etc. These data will be used in
the future, for further analysis.
c. Meteorological conditions; air temperature (T) in degrees Celsius (oC), relative
humidity (RH) in (%) and wind speed and gusts in km/h, using an electronic
weather instrument (type: Thermometer – Anemometer - Hygrometer Model No
AM4205) and, moreover, wind direction azimuth, using a compass.
d. Vegetation type in which the fire spread.
e. Topographic information such as altitude, slope and aspect of the location in which
the fire spread.
f. Forest fire description (surface, passive, active or independent crown fire, explosive
fire behaviour, spotting, fire spreading in a narrow valley, steep canyon or on a
saddle).
g. Additional general information such as presence or absence of fire fighting attempts
from the air or from the ground, volunteers and civilians participation, percentage of
fire containment, etc.
h. Additional description of the fire runs by keeping detailed notes about fire spread
direction, topography, wind direction and wind adjustment factors (Rothermel,
1983) in order to develop a reliable, errors free and easy to handle, database. For

VI International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2010
example, there were some observations for sideward or backward fire spread and
other cases in which fire spread downslope with the wind blowing downslope, also.
Observations of fire behaviour at the time of the fire, were matched with
simultaneous observations and measurements of fire weather, topography and forest fuels.
As most fieldwork was performed at a small distance from the fire fronts, continuous
observer movement was necessary. Thus, most field measurements and collected
information were being saved in a digital voice recorder.
After the fires were extinguished, additional complimentary information was
collected and assessed by interviewing local residents, forest and fire service officers and
officers from local authorities, by going over the fire area with them (Byram 1954).
5. Analysis
Fieldwork was followed by processing a huge amount of various spatial and
descriptive data by the GIS software in order to develop, in substantial detail, the overall
picture of what happened. The most important information layers were: a) fire eruption
sites, b) approximate burned areas per day (pixel size=300 m) by utilizing and combining
fieldwork, maps and low resolution satellite images of the US Maryland University FIRMS
system and the NASA MODIS Rapid Response System, c) total burned areas (pixel
size=10 m) without recording unburned islands (by digitizing maps of the “Service
Régional de Traitement d'Image et de Télédétection”, Strasbourg (Sertit) for the fires of
August http://130.79.72.201/documents/greece_2007/greece_2007.html and of the World
Wildlife Fund (WWF) of Greece for the fire of Parnis mountain
http://www.wwf.gr/storage/additional/ParnithPosterBig.jpg that was produced by the Greek
company Environet), d) observation points at the time of the fires and after the fires were
over, e) horizontal azimuths of observation line for each observation point, f) locations of
fire fronts at the time of each observation g) ROSobserved values and direction of fire fronts
and fingers, h) areas were spotting occurred, k) backfire or burnout zones, l) areas where
different fire fronts merged, etc.
Subsequently, detailed maps and case studies of these fires were generated; these
were then used for the development of the fire behaviour database. This database can be
used and is available for testing any fire behaviour model.
The database constists of fire behaviour observations that took place during surface
fires in maquis (evergreen sclerophyllus shrubs eg Pistacia lentiscus, Quercus coccifera,
Phillirea latifolia, Arbutus unedo, etc.), grass, small xeric shrubs, up to 0.5 m height
(phrygana) as well as during passive and active crown fires in fir (Abies sp.) and pine
(Pinus halepensis) forests. For many of the crown fire records in the pine forests, there was
an understory layer consisting of tall evergreen shrubs.
The analysis presented here, focuses on testing the degree of agreement of
BehavePlus predictions with field observations for a specific type of fuel that is tall
Mediterranean shrublands (maquis). This fuel corresponds to a fuel model for “evergreen-
schlerophyllous shrublands (maquis) (1.5 - 3 m)” that has been developed for Greece by
Dimitrakopoulos et al. (2001) and Dimitrakopoulos (2002). Examination of the database
showed that a subset consisting of 27 records was appropriate for the analysis. In regard to
type of fire, these records fall in one of the two following categories:
a. Surface fire in tall maquis shrubfields, and

VI International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2010
b. Passive crown fire in Pinus halepensis forest with maquis understory where the fire
rate of spread is determined by the surface rate of spread in the understory (Van
Wagner, 1977).
Figure 4: Burned areas of the twelve documented forest fires, numbered as in table 1,
including the points where the fires started.
The sample of 27 observations, all of them refering to surface fire behaviour in tall
evergreen-schlerophyllous shrubs during pure surface fires or passive crown fires, includes
a maximum ROSobserved value of 3,3 km/h, and a minimum value 0.1 km/h. The mean
ROSobserved value of the 27 data points is 1.1 km/h.
BehavePlus was used for predicting surface fire rate of spread values (ROSpredicted), using
measured slope values, on-site meteorological measurements and the above mentioned fuel
model for maquis. However, certain required adjustments were made to the fuel model
using the NEWMDL module of the original BEHAVE system (Burgan and Rotrhermel
1984), before the model could be used with BehavePlus. These adjustements were:
a. Litter weight (3.38 t/ha), which had been reported separately in the published
model, was added to the weight of fine (1-hour) dead surface fuels (14.5 t/ha), so
that the total 1-hour fuel load came up to 17.88 t/ha.
b. The overall fuel bed depth was reduced to 203.58 cm from the original value of
218.00 cm as a result of adding the litter load to the rest of the 1-hour surface fuels.
c. A value of 34% was assigned to the “dead fuel moisture of extinction”. It was
estimated using NEWMDL. Such a value is required for a fuel model to be
complete but it had not been reported for the published model.
The values of the fuel model parameters are reported in table 2.

VI International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2010
A second preliminary line of investigation concerns nine (9) cases of active crown
fire in Pinus halepensis forest with maquis understory. There, following the example of
Rothermel (1991), the observed crown ROS values (ROSobserved(crown)),, were compared with
predictions for surface fire (ROSpredicted(surface)), that were obtained utilising the same surface
fuel model for “evergreen-schlerophyllous shrublands (maquis) (1.5 - 3 m)” as above.
Table 2: The values of the parameters of the “evergreen-schlerophyllous shrublands
(maquis) (1.5-3 m)” fuel model as calculated through the NEWMDL module of
BEHAVE.
6. Results
The 27 pairs of ROS observations and BehavePlus predictions, were correlated via a
linear regression, using the SPSS (v.10.0) software, resulting in the following equation:
ROSobserved = 0.068 + 0.853 * ROSpredicted (1)
with an adjusted R2 = 0.744 and equation p-value <0.001. The p-value of the slope
coefficient is p<0.001 as well. On the other hand the p-value of the constant of the equation
is not statistically significant (p-value=0.648). In spite of this, as the value of the constant

VI International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2010
Figure 5: Correlation of surface ROSobserved and ROSpredicted values
Figure 6: Plot of the regression equation between the ROSobserved(crown) and ROSpredicted(surface)
values.

VI International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2010
(=0.068) is quite close to zero and the value of the slope coefficient is not far from unit (1)
it can be concluded that the overall agreement between observations and predictions is
good. The regression equation is plotted with the data in figure 5, confirming the good
agreement of predictions with observations.
The regression of the nine active crown fire ROS observations in Pinus halepensis
forests with evergreen schlerophyllous understory against ROS predictions for surface fire
in maquis fuels, resulted in the following linear regression equation (figure 6):
ROSobserved(crown) = 1.179 + 1.051 ROSpredicted(surface) (2)
with an adjusted R² = 0.743. The constant of the equation (1.179) is not statistically
significant (p-value = 0,118) while the coefficient of the slope of the regression line is
statistically significant (p-value = 0,002) and almost equal to the unit (1.051).
The ratio between ROSobserved(crown) and ROSpredicted(surface) varied between 0.83 and
2.95 with an average value of 1.81 and standard deviation 0.74. In general, ROSobserved(crown)
values tend to be slightly less than twice the values of ROSpredicted(surface) for this fuel model.
However, due to the small sample size, this conclusion can only be considered as
preliminary, needing further investigation.
7. Discussion
The large wildfires that were spreading for many continuous days during the fire
season of 2007 in Greece, on many occasions exhibited characteristics of extreme fire
behaviour. Additionally, the large number of simultaneous conflagrations resulted in a
collapse of firefighting during the critical days of that summer. Thus, many fires spread
unobstructed, practically without any firefighting influence, presenting an opportunity for
studying fire behaviour of a type that is not commonly observed in Greece (Xanthopoulos
et al., 2009).
The dataset of 27 fire behaviour observations presented and analyzed here mostly
includes data from intense fires spreading under a variety of meteorological and
topographic conditions. For these 27 sets of observations that concern fire spread in tall
evergreen schlerophyllous shrubs, predicted values of surface rate of spread were obtained
using the BehavePlus fire behaviour prediction system. The published fuel model for
“evergreen-schlerophyllous shrublands (maquis) (1.5 - 3m)” developed by Dimitrakopoulos
et al. (2001) and Dimitrakopoulos (2002), was used as an input in the prediction system,
after applying small required adjustments. The regression analysis that followed showed
that the predicted values of surface fire rate of spread (ROSpredicted) were correlated quite
well with the actual values that were measured in the field (ROSobserved).
The adjusted R2 = 0.744 of the regression equation indicates that the unexplained
error is relatively low. This error may be due to weaknesses in the fire behaviour model
itself or errors in the inputs on which the calculations were based. As there was a strong
emphasis in correct documentation of fire behaviour, weather and topography, and cases for
which there were doubts were not included in the database, input errors are most likely due
to weaknesses in describing the fuels. This is not surprising since a single fuel model is
used to describe a variety of fuel conditions that share as common that they have been
identified through visual assessement as consisting of tall (more than 1.5 m) Mediterranean

VI International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2010
evergreen-schlerophyllous shrubs. Obviously, the actual variation between the different
shrub situations, resulting from differences in their height, density, and even species
variation, cannot be represented by using as input just one stylized fuel model. This is
common in all applications of BehavePlus with stylized fuel models as inputs, an option
that partially sacrifices accuracy in favor of practicality. Similar deviations between
predictions and observations have been noted in practically all testing of the fire behaviour
model until now.
The analysis of active crown fires can only be considered as a preliminary effort
given the limited sample size. Nevertheless, it offers a first assessment of the possibility to
predict the rate of spread of active crown fires in Pinus halepensis forests as a function of
predicted surface fire ROS in tall “evergreen-schlerophyllous shrublands (maquis) (1.5 -
3m)” that usually form the understory layer in such forests. The relatively high adjusted R²
= 0.743 is not enough to inspire confidence in the regression equation which suffers from
the small sample size and a constant that is not statistically significant. Apart from that, it is
worth noting that the ratio of ROSobserved(crown) to ROSpredicted(surface) varied between 0.83 and
2.95 with an average value of 1.81. This value is in general agreement with the suggestion
of Cruz et al. (2005) that rates of spread in crown fires are often twice the spread of the
surface fire. On the other hand, Rothermel (1991) examined seven crown fires in conifer
forests of the northern Rocky Mountains where the surface fuels could be represented by
fuel model 10, “timber litter and understory” (Anderson 1982) and found that the average
rate of spread for the crown fires was 3.34 times faster than predicted for surface fire, with
a standard deviation of 0.59. This difference can be explained by the much lower rates of
spread exhibited by fuel model 10 type fuels, when compared with shrub fuel models
(Anderson 1982). Furthermore, this difference is an indication that a multiplication factor
for estimating crown fire ROS from surface fire ROS such as the 1.81 factor above should
be used with extreme caution. It is extremely important that the fuels beneath the canopy
are described well by the fuel model used for obtaining estimates of surface fire ROS. For
example, Kazanis et al. (2006) distinguished different types of understory fuel situations in
Pinus halepensis forests in Central Greece. The shrub understory in low-elevation drier
sites has very little biomass. For such forests it would be clearly erroneous to use a high
biomass shrub fuel model such as the “evergreen-schlerophyllous shrublands (maquis) (1.5
- 3m)” for obtaining estimates of surface fire ROS that will then be multiplied by a factor
(such as 1.81) to get an estimate of crown fire ROS.
8. Conclusions
In conclusion, based on the quite good adjusted R2, and the level of significance of
equation (1) it can be concluded that the adjusted “evergreen-schlerophyllous shrublands
(maquis) (1.5 - 3m)” model can be used with acceptable confidence with the BehavePlus
system for fire behaviour predictions in shrub fuel complexes of this type in Greece and
probably other Mediterranean countries with similar vegetation. Further examination in a
wider range of conditions is, of course, desirable in order to improve confidence and
broaden possible operational uses.
In regard to predicting active crown fire spread it can only be stated that the
preliminary results are encouraging, a correlation of surface fire behaviour prediction with
actual crown ROS seems to exist but more work is needed before a robust equation can be
developed. Such work is in progress. Data collection in crown and surface fires continues,

VI International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2010
aiming to document as many fires as possible and to develop a large database that will be
available for testing any fire behaviour prediction model in the future. The authors, in
agreement with numerous other fire experts (Byram, 1960, Turner et al., 1961, Thomas,
1994, Alexander and Thomas, 2003b), believe in the importance of developing an extensive
file of well documented fires. Such case studies can be used as teaching aids (Chandler,
1976) and can assist in improving understanding of fire behaviour, resulting in safer and
more efficient fire management (Countryman 1972).
9. Acknowledgements
The research reported here was carried out as part of the M.Sc. thesis of the first
author for the post-graduate studies program on “Prevention and Management of Natural
Disasters” of the Department of Geology and Geo-Environment of the National and
Kapodistrian University of Athens, Greece, and also in the context of a research project of
the Institute of Mediterranean Forest Ecosystems and Forest Products Technology titled
“Forest fuels standardization and management methodology in Attica”. The project was co-
funded by the European Regional Development Fund (ERDF) through the Greek General
Secretariat for Research and Technology (Project code: ATT-63).
The help of Nikolaos Zirogiannis who cooperated in post fire documentation of the
burns and preparation of burned area maps in the context of his own M.Sc. thesis, is
acknowledged and appreciated.
10. Bibliography
Alexander, M.E., Thomas D.A., 2003a. Wildland Fire Behavior Case Studies and Analyses:
Other Examples, Methods, Reporting Standards, and Some Practical Advice. Fire
Management Today, 63 (4): 4-12.
Alexander, M.E., Thomas D.A., 2003b. Wildland Fire Behavior Case Studies and
Analyses: Value, approaches, and practical uses. Fire Management Today, 63(3): 4-
8.
Anderson, H.E., 1982. Aids to determining fuel models for estimating fire behavior.
USDA For. Serv., Research Note RM - 354. 4 pp.
Andrews, P.L., Bevins, C.D., Seli, R.C., 2005. BehavePlus fire modeling system, Version
4.0: User’s Guide. General Technical Report RMRS-GTR-106WWW revised. Ogden,
UT: U.S. Department of Agriculture, Forest Service, Intermountain Forest and Range
Experiment Station. 132 p.
Athanasiou, M., 2008. Forest fire behavior in Greece: Characteristics and prediction. M.Sc.
Thesis. Inter-University Post-Graduate Studies Programme on “Prevention and
Management of Natural Disasters”, Department of Geology and Geo-Environment
of the National and Kapodistrian University of Athens, and Department of
Geoinformatics and Topography of the Technological Educational Institute of
Serres. 140 p.
Burgan, R.E., Rothermel, R.C. 1984. BEHAVE: fire behavior prediction and fuel
modelling system—FUEL subsystem. Gen. Tech. Rep. INT-167. Ogden, UT: U.S.
Department of Agriculture, Forest Service, Intermountain Forest and Range
Experiment Station. 126 p.

VI International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2010
Byram, G.M., 1954. Atmospheric conditions related to blowup fires. Stn. Pap. No. 35.
Asheville, NC: USDA Forest Service, Southeastern Forest Experiment Station.
[Reprinted as: National Fire Equipment System Publication NFES 2565 by the
National Wildfire Coordinating Group, Boise, ID.]
Byram, G.M., 1960. A problem analysis and proposed research program for the Southern
Forest Fire Laboratory. Macon, GA: USDA Forest Service, Southeastern Forest
Experiment Station, Southern Forest Fire Laboratory.
Chandler, C.C., 1976. Meteorological needs of fire danger and fire behavior. In: Baker,
D.H.; Fosberg, M.A., tech. coords. Proceedings of the Fourth National Conference
on Fire and Forest Meteorology; 1976 November 16–18; St. Louis, MO. Gen. Tech.
Rep. RM–32. Fort Collins, CO: USDA Forest Service, Rocky Mountain Forest and
Range Experiment Station: 38–41.
Countryman, C.M. 1972. The fire environment concept. Berkeley, CA: USDA Forest
Service, Pacific Southwest Forest and Range Experiment Station.
Cruz, M. G., Alexander, M. E., Wakimoto, R. H. 2005. Development and testing of models
for predicting crown fire rate of spread in conifer forest stands. Canadian Journal of
Forest Research 35, 1626-1639.
Dimitrakopoulos, A.P., Mateeva, V., Xanthopoulos, G.,. 2001. Fuel models for
Mediterranean vegetation types in Greece. Geotechnical Scientific Issues
(Geotechnical Chamber of Greece): Series VI, vol. 10(3): 192-206 (in Greek).
Dimitrakopoulos, A.P., 2002, ‘Mediterranean fuel models and potential fire behaviour in
Greece’, International Journal of Wildland Fire 11(2) 127 – 130.
Kazanis, D., Xanthopoulos, G., Arianoutsou, M., 2006. Long-term post-fire evolution of
understorey biomass in Pinus halepensis Mill. forests of Central Greece. Abstract p.
175. In “Forest Ecology and Management“, Vol. 234, Supplement 1, D. X. Viegas
(Ed.), Elsevier Publishers, Amsterdam. 289 p. Full text, in Proceedings of the 5th
International Conference on Forest Fire Research. November 27-30, 2006, Figueira
da Foz, Portugal (on CD).
McAlpine, R.S., Xanthopoulos, G., 1989. Predicted vs observed fire spread rates in
ponderosa pine fuel beds: A test of American and Canadian Systems. pp. 278-294.
In proceedings of the 10th Conference on Fire and Forest Meteorology, April 17-21,
1989, Ottawa, Canada. D.C. MacIver, H. Auld, and R. Whitewood, editors. 471 p.
Rothermel, R.C., 1972. A mathematical model for predicting fire spread in wildland fuels.
Res. Pap. INT-115. Ogden, UT: U.S. Department of Agriculture, Forest Service,
Intermountain Forest and Range Experiment Station. 40 p.
Rothermel, R.C. 1983. How to predict the spread and intensity of forest and range fires.
Gen. Tech. Rep. INT-143. Ogden, UT: U.S. Department of Agriculture, Forest
Service, Intermountain Forest and Range Experiment Station. 161 p.
Rothermel, R.C. 1991. Predicting behaviour and size of crown fires in the Northern Rocky
Mountains. Res. Pap. INT–438. Ogden, UT: USDA Forest Service, Intermountain
Research Station.
Thomas, D. 1994. A case for fire behaviour case studies. Wildfire. 3(3): 45, 47.
Turner, J.A., Lillywhite, J.W., Pieslak, Z., 1961. Forecasting for forest fire services.
Tech. Note No. 42. Geneva, Switzerland: World Meteorological Organization.
Turner, J.A.; Lillywhite, J.W.; Pieslak, Z. 1961. Forecasting for forest fire services. Tech.
Note No. 42. Geneva, Switzerland: World Meteorological Organization.
Van Wagner, C.E., 1977. Conditions for the start and spread of crown fire. Canadian
Journal of Forest Research. 7: 23–34.

VI International Conference on Forest Fire Research
D. X. Viegas (Ed.), 2010
VanWilgen, B.W., LeMaitre, D.C., Kruger, F.J., 1985. Fire behavior in South African
fynbos (macchia) vegetation and predictions from Rothermel’s fire model. J. Appl.
Ecol. 22, 207–216.
Xanthopoulos, G., 1986. A fuel model for fire behavior prediction in spotted knapweed
(Centaurea maculosa L.) grasslands in western Montana. Masters Thesis. Univ. of
Montana, MT, USA. 100 p.
Xanthopoulos G., Viegas, D.X., Caballero, D., 2009. The fatal fire entrapment of Artemida
(Greece) 2007. pp. 65-75. In “Recent Forest Fire Related Accidents in Europe”.
Domingos Xavier Viegas (Editor). European Commission, Joint Research Centre,
Institute for Environment and Sustainability. EUR 24121 EN. 75 p.