Sunteți pe pagina 1din 12

10.

1071/WF00015

International Journal of Wildland Fire 9(4): 235246, 1999 IAWF

Comparative Study of Various Methods of Fire Danger


Evaluation in Southern Europe
D. Xavier Viegas

Department of Mechanical Engineering, University of Coimbra


Polo II, Pinhal de Marrocos, 3030 Coimbra, Portugal
Phone:+351 239 790732; Fax: +351 239 790771; email: xavier.viegas@dem.uc.pt

Giovanni Bovio

Dipartimento di Agronomia, Selvicoltura e Gestione Del Territorio, Universit di Torino


Via Leonardo da Vinci, 44, 10095 Grugliasco, Torino, Italy
Phone: +39 011 670 8645; Fax: +39 011 670 8734 ; email: Bovio@agraria.unito.it

Almerindo Ferreira

Department of Mechanical Engineering, University of Coimbra


Polo II, Pinhal de Marrocos, 3030 Coimbra, Portugal
Phone:+351 239 790732; Fax: +351 239 790771; email: almerindo.ferreira@dem.uc.pt

Antonio Nosenzo

Dipartimento di Agronomia, Selvicoltura e Gestione Del Territorio, Universit di Torino


Via Leonardo da Vinci, 44, 10095 Grugliasco, Torino, Italy
Phone: +39 011 670 8648; Fax: +39 011 670 8734

Bernard Sol

Meteo France
2, Bvd. Chateau Double, F-13098 Aix-en-Provence Cedex 02, France
Phone: +33 442 95 9071; Fax: +33 442 95 90 19; email: Bernard.Sol@meteo.fr

Abstract

Introduction

Five methods for the evaluation of fire danger, based on


meteorological factors, were tested using statistical data for
daily number of fires and burned area from six different
regions in France, Italy and Portugal, for a period of 39
years, both for Winter and Summer fires. A normalised scale
for the output of the various methods was adopted to
compare the variation of number of fires and burned area
predicted by each method. Based on a statistical test, the relative performance of the various methods is assessed. From
the five methods that were analysed in this research, the
Canadian and the modified Nesterov methods showed the
best overall performance.

The importance of meteorological factors in forest fire activity


is well recognised and the ability to anticipate their impact on
daily fire occurrence and behaviour has been a major goal for
researchers since the start of fire science. As a result, a large
number of fire danger evaluation methods, based mainly on
current meteorological properties, have been developed.
Description of the existing methods is beyond the scope of this
paper, and can be found in Reyfsnyder (1978), Turner et al.
(1961) and in McArthur (1977). Some of these methods rely
partially on a theoretical background but, since there is no
general theory to explain the complex interaction between the
various factors involved, most methods are based on empirical
approaches. Given the large number of existing methods, when
choosing or proposing a particular one to be used in a given
geographical area, one asks oneself if there is an optimal system
or if it is worthwhile to develop a new one for that purpose. As
a first step, it seems reasonable to study the known methods in
order to have at least a partial answer to those questions.

Keywords:
Fire danger methods
Fire meteorology
Southern Europe
Mediterranean forest fires.
235

Xavier Viegas, D., Bovio, G., Ferreira, A., Nosenzo, A., and Sol, B.

The task of comparing the various methods is rendered


quite difficult when we take into account not only the empiricaland sometimes subjectivenature of their formulation,
but also due to the different levels of complexity of the existing methods, with various sets of input and output parameters.
In the opinion of the authors the best test is that of submitting
each method to the same set of meteorological and fire occurrence data, as was done by Haines et al. (1985) for some particular models of the American system in north-eastern USA.
In order to render these tests meaningful a large number of
events must be considered in these studies. To obtain this a
large spatial and temporal extension of the data base must be
used, but one has to be careful in assuring that the data base is
consistent and homogeneous in each study case.
In the countries of southern Europe that belong to the
European Union, there is also a multiplicity of methods in
use for the evaluation of the fire danger. The objective of this
work was to compare the relative performance of the various
methods by testing them with the same sets of real data from
the three countries involved in this research. The authors do
not aim to establish an absolute calibration of each of the
methods that were considered, as this would require a much
larger set of data in order to cover temporal climatic changes
in each region. This is not practicable, due to changes that
usually occur in the span of time in the data collecting
methods and in the fire prevention and suppression facilities.
Besides the methods adopted in France, Italy, Portugal
and Spain, we decided to also include the Canadian Fire
Weather Index (FWI). The authors considered also the inclusion of the National Fire-Danger Rating System (NFDRS),
documented by Deeming et al. (1977) and by Bradshaw et al.
(1983). Given the complexity of this method, which includes
other parameters besides the meteorological ones, its validation in the scope of the present study would bring great difficulties, as is shown in the work of Haines et al. (1983) and
Haines et al. (1985) and as foreseen in an earlier paper by
Van Wagner (1975). In spite of these difficulties the authors
do not exclude the possibility of including the NFDRS in a
later stage of this study. The authors were informed in a
written communication that is kept on file, that was made by
the Greek forest fire expert George Eftichidis, that there was
no meteorological danger method generally adopted in
Greece in 1994. Therefore all the main methods presently in
use in the southern European Countries are considered in this
study.
In a preliminary study, Bovio and Nosenzo (1994a) evaluated the response of each method against the same set of
varying meteorological parameters, in order to test their relative sensitivity. In a second step each research team applied
the various methods, using the same software, to analyse statistical data of selected regions of France, Italy and Portugal
in a period of 39 years in each case. As is shown below, this
study covers a wide range of geographical, meteorological,
social and vegetation conditions, with fires occurring in

WinterSpring or SummerAutumn seasons. To our knowledge there is no similar or so extensive study on this subject
in Europe. A previous version of this article was presented by
Viegas et al. (1994).

Methodology
Fire Danger Methods
A brief description of the various fire danger methods that
were tested in our study is given below as well as a bibliography for a detailed reference to each method.
Ic The Canadian Fire Weather Index, also known as
FWI. This method is described in Van Wagner (1987). It
requires daily values (at 12.00 h) of wind speed, air temperature, relative humidity and 24 h precipitation, as well as
cumulative values from previous days. The computational
program prepared by Van Wagner and Pickett (1985) was
used in this study. One of the outputs of this method is the
Initial Spread Index (ISI) that was also considered in some
parts of this research.
If The French Method known as Numerical Risk, proposed by Sol (1989) and Drouet and Sol (1990). It requires
daily values of air temperature, relative humidity, cloud
cover, wind velocity and an initial value of the water content
of the soil.
Ii The Italian Method, IREPI, proposed by Bovio et al.
(1984). This index estimates the loss of water in the soil due
to actual evapotranspiration and combines it with the potential value of evapotranspiration in order to compute the
danger index. It requires the daily average values of air temperature, relative humidity, wind and insolation, as well as
precipitation.
Ip The Portuguese Method, which is a modified
version of the Nesterov Index, currently used by the
Portuguese Meteorological Institute. A detailed description
of this index can be found in INMG (1988). This method is
composed of a daily index and a cumulative index, the
latter being a weighted sum of the daily indexes of the previous days, the weighting factor being a function of precipitation. For the calculation of the daily value of the index it
is necessary to know the air temperature and relative
humidity at 12.00 h UTC. Wind speed and direction is taken
into account in the final classification, according to local
conditions. For example, in Portugal it is considered that
wind aggravates the danger index if it comes from the
Eastern sector, from 0 to 180. In some cases the daily component of this methodwithout the cumulative effect
was also tested.
Is The Spanish Method that is used by ICONA (1988). It
evaluates a probability of fire start, based on air temperature
and relative humidity measured at 12.00 h UTC. This is the
only non-cumulative method among those that were tested.
All methods provide a numerical index that grows with
the danger conditions. In most cases this number is translated
236

Fire Danger Evaluation in Southern Europe

in France and Portugal. Due to lack of space, only some references shall be made to them here; more results can be
found in Sol (1994) and in Viegas et al. (1996a).

into a fire danger scale with three to five levels, for practical
purposes. In our study we shall retain the original numerical
output of each method in order to have a better discrimination of its performance.
The numerical output of a fire danger index usually gives
an indication of the possibility of occurrence of forest fires in
a day, based on the meteorological conditions. In some
measure it gives also an indication of the degree of danger
associated with each fire start, i.e. how fast it will spread if
not attacked. Some methods, like the Canadian, provide a
separate output, the so-called Initial Spread Index (ISI), for
this purpose. In this work we shall try to compare the same
output from each method both to fire occurrence, characterised by the number of fires started in each day, and to fire
danger due to meteorological conditions, evaluated by the
area burned in each day. Fires that burned for several days
were considered as a single event in the counting of the
number of fires, but the area burned in each day was considered separately in the analysis of daily area.
Although some methods, like the Canadian and the
Italian, were developed primarily for seasonal fires, respectively for Summer and Winter fires, we shall apply all of
them both to WinterSpring or SummerAutumn fires, in
order to assess their overall performance.
In order to make this study more complete, seven other
methods, and partial versions of some of the above mentioned systems were also tested, at least with part of the data,

Study Areas and Data


Each research group worked with statistical and meteorological data from its own country, restricting its attention to
some relevant areas that are briefly described below (see
Figure 1). For each region the numbers of fires and burned
area for each day were available.
As the large majority of forest fires in the study areas are
caused by human action, the number of daily fires in each
region depends on other factors besides the meteorological
ones. Assuming that these factors do not change drastically
in each region in the period of analysis, the number of daily
fires can be retained as an indication of the stress that is put
daily on the fire prevention and suppression systems.
The burned area is also dependent on many factors that
vary widely from one region to another. For example, a very
efficient suppression system will result in only very small or
very large fires appearing in the statistics. Therefore we must
also assume that either these factors remain practically constant or that their mutual influence is cancelled in the averaging process that is made in this analysis.
It must be noted that in France a forest fire is, by definition, one that has put in danger at least 1 ha of forest, shrub
or pasture land, regardless of its actual size. Although this

Figure 1. Study areas in Southern Europe


237

Xavier Viegas, D., Bovio, G., Ferreira, A., Nosenzo, A., and Sol, B.

definition is not entirely the same as that adopted in Italy and


in Portugal, the authors believe that in practical terms the statistical data do not change from one country to another in any
way that would affect the present results.
In order to compromise between the need to have a large
amount of data for analysis and to assure its homogeneity,
some care was taken in the selection of the temporal and
spatial extension of the data base for each region. Due to the
modifications introduced in most countries in the processes
of gathering data and also in the fire prevention and suppression methods during the past years, we decided that, as a rule,
the time span should not extend for more than 56 years.
Concerning spatial extension, it was left to each research
team to propose a geographical area with sufficient homogeneity in terms of forest cover and fire occurrence.
It is well known that the meteorological conditions can
change significantly from one place to another even over a
short distance. This variation has direct implications on the
daily fire danger situation for a given area, as was shown
by Simard (1994). In this work it was decided that meteorological data from a single weather station should be
taken for each region as representative of the weather conditions in the area. This procedure was adopted in order to
have a sufficiently large number of fires and burned area in
the study area to render the statistical analysis meaningful.
It is assumed that the non-uniformity of the meteorological
conditions in the region affects all methods in a similar
way.
For France and Italy, the study areas were sufficiently
small to make this assumption easily attainable, although
care was taken to choose in each case a weather station sufficiently exposed to wind to provide an adequate description
of the meteorological conditions in that area. In the case of
the Central Region of Portugal, as it covers a wide gradient
of geographical conditions, it was necessary to verify that the
chosen meteorological station was representative of the
study area, as is described below.

The other three were coastal departments, with fire incidence from July to September. In the period of test,
19861990, there were three bad years, 1986, 1989 and
1990, the other two being relatively mild.
(ii) The department of Bouches-du-Rhne (B.R.),
around Marseille, with 508 700 ha of surface, of which
around 200 000 ha can be endangered by fire, although only
72 869 ha are covered with woodland. The terrain is very
rough, with peaks above 1000 m. In Summer there are
around 60% of days with fire, with a maximum of 12 fires in
the same day, in the period of analysis.
(iii) The department of Var (Var), in the Cte dAzur,
with a surface of 597 300 ha, of which 314 610 ha are
forested. The terrain is also very rough, with peaks above
1500 m in the interior of the department. There are around
73% of days with fires, with up to 20 fires in the same day.
(iv) The department of Eastern-Pyrenes (E. P.), near the
Spanish border. Its surface is of 411 600 ha, of which
108 757 ha are forested. The terrain is very rough, particularly in the South of the department. There are around 70%
of days with fire, in average, with a maximum of 7 fires in
the same day.
A summary of the global statistical data for the French
regions is given in Table 1. More details can be found in
Viegas et al. (1996b).
Daily values of the indexes for each of the four departments were evaluated from meteorological data provided by
the weather stations of Saint-Auban, Marignane, Le Luc and
Perpignan, respectively. These weather stations have performed a complete set of meteorological measurements for
many years and are well representative of their regions.
Analysing a large amount of cases it was found that the
average of the maximum wind velocity recorded over 3 hour
periods in a day was approximately 60% of the maximum daily
value, with an error of less than 1 m/s. Therefore a reduction
factor of 0.6 was multiplied by the maximum instantaneous
daily value, in order to characterise the daily wind speed in the
French regions. It is quite possible that the wind blowing in a
particular fire in a department, on a given day, might be quite
different from the estimated value; however, all the indexes are
calculated with the same wind. This procedure should not
change the relative performance of the various methods in a
certain region, but it modifies the absolute values of the indexes
between the various regions as can be seen in Table 4.
For France the daily extreme values of air temperature and
relative humidity were used in the evaluation of the indexes,
based on the fact that they occur in a period of the day in
which most fires start.

(a) France
In France the tests were carried out in four Mediterranean
zones:
(i) The department of Alps of Haute Provence (A.H.P.), a
mountainous interior region, with a surface of 692 500 ha, of
which 284 504 ha are forested. This department is situated
more than 100 km from the sea, having various plateaux with
more than 500 m of altitude and peaks above 2000 m. Fire
incidence is mostly in the WinterSpring period, from January
to April. In the period of test, 19811990, there were 948 days
without fires and 134 days with at least one fire, up to five fires
in the same day. A larger period was taken for this area because
in Winter conditions, given the reduced number of fires, the
fire prevention and suppression systems are quite reduced in
this region, in comparison to the coastal areas during Summer;
therefore our analysis is not disturbed by this factor.

(b) Italy
In Italy two regions were considered in the study:
(i) The Veneto Region, situated in the north-east of Italy,
with a total area of 183 6775 ha, of which around 350 824 ha
(19.1%) is forested. The main vegetation cover is high forest,
238

Fire Danger Evaluation in Southern Europe

140 100 ha (38.2% of the forested area), and coppice,


123 000 ha (35.1%). Fires occur mainly in the Winter-Spring
season. Statistical data are taken from 1 January to 30 April,
from 1988 to 1990.
(ii) The Savona Province, situated in the north-west of
Italy, with a total area of 154 455 ha, of which around 45% is
covered by forest. The main vegetation cover in this region
is coppice and shrubland. There are two fire seasons in this
region, one in Winter-Spring, like in Veneto, and the other in
Summer, from 1 June to 30 September. Statistical data are
from the years 19871989.
Meteorological data from the weather stations of Cengio
and of Foen Case Rosse (Verona), were taken respectively
for Savona and Veneto regions. These are standard synoptical weather stations placed at chosen sites, to be representative of the surrounding area. A summary of the data is found

in Table 2. More details can be found in Bovio and Nosenzo


(1994a, 1994b).
(c) Portugal
The Central Region of Portugal, roughly limited by the rivers
Douro and Tejo, was considered in this study, given its
important forest cover and fire incidence (cf. Viegas et al.
1992). The total area of this Region is 172 1650 ha, of which
around 39.3% is covered by forest, and 17.5% by wildland.
The main vegetation cover is by Pinus pinaster (29.3%) and
Eucalyptus globulus (610%). This region is limited on the
West by the Atlantic Ocean and on East by the Spanish
border. Along this span of around 200 km we have some
influence of the ocean, near the coast, a transition zone and
an interior zone with several ranges of mountains with altitudes up to 2000 m.

Table 1. Summary of data of the French regions. A.H.R, Alps of Haute Provence. B.R.,
Bouches de Rhne. Var, Var (Cte dAzur). E.P., Eastern Pyrenes.
A. H. P.
B. R.
19811990

Period of test
Fire season

Var
E. P.
19861990

Jan./ Apr.

Days with/without fires

Jul./Sept.

134/948

279/181

332/128

164/296

Number of fires

Total
Maximum daily
Average daily

191
5
0.18

675
12
1.47

954
20
2.07

292
7
0.63

Burned
surface
(ha)

Total
Maximum Daily
Average Daily

1920
278
1.77

23434
6430
50.94

48939
14564
106.39

7098
1523
15.43

Table 2. Summary of data of the Italian regions.


Veneto
19881990

Savona
19871989

Fire season

Jan./Apr.

Jan./Apr.

Days with/without fires

171/190

145/216

Total
Maximum daily
Average daily

515
19
1.43

284
7
0.79

282
7
0.77

Burned surface (ha) Total


Maximum daily
Average daily

5244
600
14.53

2329
378.2
6.45

2017
402
5.51

Period of test

Number of fires

239

Xavier Viegas, D., Bovio, G., Ferreira, A., Nosenzo, A., and Sol, B.

Data from the years 19881992 on forest fires during the


Summer season (from 1 June to 30 September) were used in
this study. This period covers a wide range of fire occurrence. As a matter of fact, during these 5 years the absolute
minimum and maximum values of total burned area in the
past 15 years were experienced in Portugal. The data are
summarised in Table 3.
Meteorological data from the Observatory of the
University of Coimbra were used in the present study.
Comparing the available climatological data from 1931 to
1960, for Coimbra with three other meteorological stations
in the region, Cabo Carvoeiro (near the Atlantic Ocean),
Penhas Douradas (at 2000 m altitude), and Castelo Branco
(near the Spanish border) it was found that the average
values of the 15.00 h measurements from June to September
of air temperature, relative humidity, cloud cover and wind
velocity, and daily precipitation of Coimbra were near the
average of the other stations. Daily evolution of the same
parameters in the period 19881990 also showed the same
trend, with the values of Coimbra covering the entire range
of the other three stations but showing a wider variation than
each of the others. In previous studies (Viegas et al. 1992,
Viegas and Viegas 1994) it was shown that meteorological
and fuel moisture data from the area of Coimbra could be
correlated successfully with number of fires per day and
burned area in the same Region.

maximum values of this index in a certain region during the


period of analysis. Given the climatological differences
between the various regions, the range of variation of each
index varied from one country to another. Actually for the
same region it was found that the range varied from one year
to another. In order to simplify our analysis only the absolute
extreme values for the entire period were considered. These
are given in Table 4 for each region.
Using equation (1), the values of the modified or standard
danger index were grouped in 10 classes limited respectively
by 010, 1020,, 90100, as well as all the other related
parameters. Average values of the normalised index for each
class were then computed and analysed. For simplicity, the
average value of the standard index in each class was taken
as 5, 15,, 95, in all cases, while their actual values could
have been 5.4, 14.8, etc. in a given situation, and similar
values in another. The slight graphical error introduced by
this simplification in the figures is certainly not relevant for
the present analysis.
Eventually a much more important error is introduced by
the partition of the scales in equally spaced class limits. For
some methods this procedure may distort the range of variation of the index. The alternative was to create class limits
adapted to each method, but this could introduce more
factors in the comparative method, so we preferred the
simpler process described above.

Normalisation of Scales

Fire Statistical Parameters

In order to render the results obtained with the several methods


in each region comparable, a common scale was used for the
danger indexes. This scale with a range from 0 to 100 was
adopted for all methods using the following equation:

Using the statistical data of a given period of one region, for


each class of fire danger, as defined above, i. e. in the set of
days in which a certain fire danger method produced standard values in the same range, the following independent
parameters could be obtained:
y1 Total number of days of occurrence (of a class of danger
given by a certain method);
y2 Number of days with fires;
y3 Total number of fires (in the days of the same class);
y4 Total area burned (in the days of the same class).

Ix = 100 (I x I min) / (I max I min).

(1)

In this equation, Ix is the normalised value of the index Ix,


and Imin and Imax are, respectively, the minimum and

Table 3. Summary of data of the Central Region of Portugal.


Period of test

It happened in some cases that there were no days of occurrence in a certain class of danger. In these cases we considered that all the corresponding variables were not defined,
except y1 (and y5, defined below), that were equal to zero.
From these, other parameters can be easily derived, as for
example:
y5 Percentage of occurrence of days in a certain class;
y6 Percentage of days with fires in a certain class;
y7 Average number of fires per day;
y8 Average area burned per day;
y9 Average area burned per day with fires;
y10 Average area burned per fire.

19881992

Fire season

Jun./Sept.

Days with/without fires

587/23

Number of fires

Total
Maximum Daily
Average Daily

29080
229
23

Burned surface (ha)

Total
Maximum Daily
Average Daily

159373
8083
261.3

Although these parameters are not independent they put in evidence different aspects of the performance of each method, so
240

Fire Danger Evaluation in Southern Europe

Table 4. Extreme values of the Danger Indexes


Max

Min

If
Max

0
0
0
0

100
74
100
184

0
0
0
6

20
15
20
19

0
0
0
0

77
77
75
84

0
0
0
1*

10
11
11
8*

10
30
10
20

80
70
80
100

Summer

228

20

86

23

20

100

Summer

72.5

20

75

14

10

100

Region

Season

Min

Veneto
Savona
Savona
A. H. P.
B. R.
Var
E. Pyr.
C. Port.

Winter
Winter
Summer
Winter

Ic

Ii
Min Max

Min

Ip

Max

Is
Min Max

* These values correspond to the Portuguese daily index, without the cumulative effect.
they should be all taken into account in a comparative study. For
the evaluation of y5 it is of course necessary to know the total
number of days in the period of analysis. Using other total or
global parameters, like the total area or forested area of the
region, new variables could be defined that are relevant for the
comparison of the fire regimes of the various regions. As such a
study is out of the scope of this paper we shall not proceed
further in that direction.
In each country it was found that the evolution of the
above mentioned parameters changed from one year to
another. Sometimes the range of variation of the indexes
itself varied, thus reflecting more or less severe conditions in
the various years. It was found also that, in different years,
the same value of fire danger given by some method
reflected different fire occurrence conditions. The causes for
this can be many: modifications in the fire detection and suppression operations, or some other changes of natural conditions that are not accounted for in the fire danger methods. A
similar behaviour was found in Viegas et al. (1992), and the
possible long-term effect of precipitation was suggested as
an explanation for this in Viegas and Viegas (1994).
Although it would be interesting to compare the interannual performance of each method, we shall restrict ourselves to the analysis of the average values for the period of
years considered in each region. As a matter of fact it was
noticed that these average values tend to be more representative of the normal year in a region and the deviations from
them can then be explained by other sorts of arguments.

at least one fire in a given day. For SummerAutumn fires,


we shall present y7, the average number of fires per day, as a
measure of the daily fire activity. This parameter was chosen
instead of y6 because in many situations the percentage of
days with fires was 100% and no interesting conclusions
could be taken from the graphs.
Winter Fires
In Figure 2 the percentage y5 of occurrence of days in a certain
class for each method is shown, for the A.H.P. (a) and Veneto
(b) regions. The Portuguese daily index was used in the French
region. As could be expected, most methods predict a large
percentage of days with low danger in this season. For some
methods, there is a second peak for normalised index values
around 60 and 30, respectively, for the A.H.P. and Veneto
regions. In the latter case it can also be observed that the Italian
and the Portuguese methods predict a relatively large percentage of days with extreme danger, which is not realistic.
In Figure 3 the percentage y6 of days with fires is shown
for the same regions. It can be observed that practically all
methods present an increasing percentage of days with fires
with increasing danger, a singular exception occurs with the
Italian method in the upper range of the scale for Veneto.
In Figure 4 the average area y8 burned daily for each class
is shown. As there are several factors affecting this parameter, besides meteorological conditions, a considerable scatter
is observed in both cases, but especially for Veneto. For AHP
the various methods predict a general increase of fire size
with fire danger, in the case of Veneto this is not so evident.

Results

Summer Fires

Since conditions for SpringWinter and for Summer


Autumn fires are quite different, we shall present the respective results separately. In both cases we shall present results for
parameters y5 and y10. For SpringWinter fires we shall present
y6, the percentage of days with fires, as for these conditions it
is more important to be able to predict the possibility of having

Results for fires in the SummerAutumn season shall be presented for Var, Savona and Central Portugal. The department
of Var was chosen to illustrate the situation in Southern France
given its larger fire incidence (cf. Table 1), but the results for
the other departments were not essentially different.
241

Xavier Viegas, D., Bovio, G., Ferreira, A., Nosenzo, A., and Sol, B.

Figure 2. Percentage y5 of days of occurrence for each


class. (a) A.H.P.; (b) Veneto.

Figure 3. Percentage y6 of days with fires: (a) A. H.


Provence; (b) Veneto.

Figure 4. Average area y8 (ha) burned daily: (a) A. H.


Provence; (b) Veneto.

In Figure 5 the percentage y5 of days of occurrence for the


three regions are shown. It is interesting to note that the low
danger peak is not so well defined in this case. On the contrary, the second peak, around 3050, is quite clear for Var
and Savona. In all regions the French and the Italian methods
predict a large percentage of days with extreme danger. This
is not convenient for practical application.
In Figure 6 the average number y7 of daily fires is shown.
Given the relatively large number of total fires in Portugal,
the curves for all methods are quite regular and show a
marked increase in the number of fires with the danger index.
In spite of some discrepancies, the same general behaviour
can be observed in the other two regions. In this figure one
point corresponding to Is (5; 166) was omitted for the sake
of clarity of the figure.
In Figure 7 the average area y8 burned daily for a given class
of danger is shown. Once again we find some scatter in the
data. In France, given the powerful fire suppression organisation and methods, the expected increase of fire size with fire
danger is not apparent, except for the Canadian method. But
this can be due to the very limited number of cases of occurrence in the upper range of this method (cf. Figure 5a). In
Central Portugal the increase of fire size is very regular, except
for the initial points of the French and Spanish methods.
242

Fire Danger Evaluation in Southern Europe

Figure 5. Percentage y5 of days of occurrence for each


class: (a) Var; (b) Savona; (c) Portugal.

Figure 6. Average number y7 of daily fires: (a) Var;


(b) Savona; (c) Portugal.

Discussion

called Mahalanobis Distance (Md), (cf. Der Megreditchian


1981), which discriminates among two samples in a given
population. Let us consider a random variable X that is
defined unequivocally for each member of the population.
The average value of this variable is X0 for the entire population and its standard deviation is . If we take two samples
S1 and S2, with average values X1 and X2, respectively, the
Mahalanobis Distance is defined by:

The visual observation of the figures like those shown in the


previous section does not provide a definitive indication
about the superior performance of any of the methods that
were tested. Although some opinions have already been
given above, about the positive and negative behaviour of the
methods in some aspects, we shall now attempt a more
objective appreciation of their performance, based on statistical analysis.
Assuming that for practical purposes a danger method
must provide a clear discrimination at least among those
days with very low or very high fire risk, we shall use the so-

Md = [(X1X2)/]2.

(2)

In this definition, has a role of normalising the


Mahalanobis distance, rendering its value independent of the
243

Xavier Viegas, D., Bovio, G., Ferreira, A., Nosenzo, A., and Sol, B.

them, and their reunion made the total population. Taking


into account the different characteristics of the statistical data
from the various regions and seasons, several criteria for
sampling were tested in our study. We give now the adopted
criteria, in each case, for selecting sample S1; the criteria for
S2 are obviously the negation of the previous one.
Winter- Spring Fires:
(a) Days with fires.
SummerAutumn Fires:
France : same as (a) and
(b) Days with more than four fires or with more than
100 ha of burned surface.
Italy: same as (a).
Portugal:
(c) 10% of the days with the minimum number of fires
(d) 10% of the days with the minimum burned area
(e) 10% of the days with the maximum number of fires
(f) 10% of the days with the maximum burned area
The results that were obtained for the various cases are given
in Table 5.
In each row of Table 5, the indicated values of Md assess the
relative performance of each method for a given region and fire
season. We included in this table a relative classification of
each method, based on this parameter. Bold type is used to
identify the first places in the relative classification, but these
must be taken with caution. Assuming that the sampling criteria described above are good and also that Md is a good statistical discriminator, some conclusions can be drawn from Table 5.
It seems that the Canadian method has a superior performance in Summer conditions, with the single exception of
the Savona region. Although the Spanish index has a better
performance in tracking both the number of fires and the
burned area (see Figures 6b and 7b) it comes second in the
discrimination between fire and no-fire days. The Portuguese
method works well in Winter and reasonably well in Summer
conditions in the whole set of regions. In the discrimination
between the days of extreme danger (sampling criteria (e)
and (f)), the French and Italian methods give better results,
apparently, but it must be remembered that those methods
both have a saturation effect, with a large proportion of days
with very high values of fire danger. A test made with 20%
of the days with extreme danger, instead of 10%, gave the
same relative classification as for (c). Therefore, in a certain
sense we could attribute a better overall performance to the
Portuguese method, if we had to choose a single method for
the entire year. It is interesting to notice that the Spanish
method gives relatively good results in Winter conditions,
but performs very poorly in Summer. The Italian method,
although it was developed for Winter conditions, showed the
best results only in Savona and performed modestly in the
other cases. The French method gave the worst results in
Winter conditions and had only a moderate behaviour in
Summer conditions. Overall this and the Spanish methods
seem to be the least effective in relation to the others.

Figure 7. Average area y8 (ha) burned daily: (a) Var; (b)


Savona; (c) Portugal.
scale or range used for variable X. In this study the variable
X will be the daily value of a given meteorological index Ix.
Given the linear character of equations (1) and (2), in the
evaluation of Md it is not relevant if we use the original scale
of values for Ix or its normalised scale value Ix.
As can be seen the Mahalanobis distance is non-dimensional and always positive. The larger its value, the larger
will be the difference between the two samples in terms of
the random variable X.
In this study the criteria for choosing the samples S1 and
S2 were always such that there were no common elements in
244

Fire Danger Evaluation in Southern Europe

Table 5. The Mahalanobis Distance for the various cases. Crit., criterion for selection of S1. Cl., relative classification for each method.
Season

Region

Crit.

Md

Winter

Veneto
Savona
A. H. P.

(a)
(a)
(a)

Summer

Savona
B.R.,Var
E. Pyr.

(a)
(a)
(b)
(c)
(d)
(e)
(f)

Central
Portugal

Ic

Cl.

Md

0.16
0.32
0.86

4
4
1

0.00
0.49
1.88
1.68
1.14
1.64
1.83

5
1
1
1
1
1
1

If

Ii

Cl.

Md

0.02
0.02
0.38

5
5
5

0.97
0.79
0.55

3
1
4

0.00
0.41
0.42
0.54
0.48
1.59
1.61

4
3
4
4
3
2
2

0.04
0.18
0.12
0.83
0.35
1.24
1.43

3
5
5
3
4
3
3

It is worth remarking that a test with the ISI component of


the Canadian index (cf. Sol 1994), made to (b) type samples
of the French regions gave the larger value of Md (2.10), thus
showing the interest of this index for the prediction of
extreme fire conditions, created by an extended drought and
strong winds, like fohn type winds.

Cl.

Ip
Md

Is

Cl.

Md

Cl.

1.20
0.56
0.74

1
3
2

1.14
0.56
0.65

2
2
3

0.23
0.47
0.55
1.42
1.01
1.13
1.23

1
2
3
2
2
4
4

0.15
0.34
0.59
0.31
0.23
0.76
0.75

2
4
2
5
5
5
5

this research work was carried out. The authors also


acknowledge the comments and suggestions made by the
reviewers of an earlier version of this paper, that strongly
contributed to its improvement.

References
Bovio G, Quaglino A, Nosenzo A (1984) Individuazione di un
indice di previsione per il Pericolo di Incendi Boschivi.
Monti e Boschi, Anno XXXV(4).
Bovio G, Nosenzo A (1994a) Comparison between methods
of forecasting danger of forest fires. Agroselviter technical report of Minerve project, Turin.
Bovio G, Nosenzo A (1994b) Confronto tra indici di pericolo di
incendi boschivi in aree campione dellItalia
Settentrionale. Agroselviter regional report of the Minerve
project, Turin.
Bradshaw LS, Deeming JE, Burgan RE, Cohen JD (1983)
The 1978 NFDRS: technical documentation. USDA
Forest Service, Gen. Tech. Rep. INT-169.
Deeming JE, Burgan RE, Cohen JD (1977) The National
Fire-Danger Rating System1978. USDA Forest
Service, Gen. Tech. Report INT-39.
Der Megreditchian, G (1981) La Prevision statistique des
phnomenes mteorologiques. Note technique N 100,
Nouvelle Srie de lEtablissement dEtudes et de Recherches
Mtorologiques, France.
Drouet JC, Sol B (1990) Mise au point dun Indice numrique
de risque mtorologique dincendie. Revue Gnerale de
Securit 92.
Haines DA, Main WA, Simard AJ (1985) Operational validation of the NFDRS in the Northeast. In Proceedings
of the Eight Conference on Fire and Forest
Meteorology. pp. 169177. (Society of American
Foresters: Detroit, 29 April2 May 1985)

Conclusion
In this paper a summary of an extensive comparative study
of various methods of fire danger evaluation, based on meteorological parameters, applied to some southern European
countries has been presented. The principal methods in use in
the Mediterranean countries of the European Union were
tested against real data from a wide range of meteorological,
social and geographical conditions. According to the present
study, from the methods that were tested some perform better
in certain conditions and some in others, leading to the conclusion that it may be interesting to continue the research
with at least those that show better overall performances. In
this situation we include the Canadian and the Portuguese
methods for Summer conditions, and the Portuguese and
Spanish methods for Winter conditions. These are not of
course definitive conclusions, as we intend to continue our
research both by including other methods in our analysis and
by extending it to other geographical areas, namely in Spain
and in Greece. In a further stage we intend to propose a calibrated method to be adopted in the southern countries of
Europe, that could be either based on some of the present
methods or include components from more than one.

Acknowledgments
The authors acknowledge the support of the DG XII of the
European Commission to the research project MINERVE, in
the framework of the Environment Program, under which
245

Xavier Viegas, D., Bovio, G., Ferreira, A., Nosenzo, A., and Sol, B.

Haines DA, Main WA, Frost JS, Simard AJ (1983) FireDanger rating and wildfire occurrence in the Northeastern
United States. Forest Science 29(4), 679696.
ICONA (1988) Experimentation de un nuevo sistema para
determinacion del peligro de incendios forestales derivado
de los combustibles: instrucciones de calculo. Instituto
Nacional para la Conservacion de la Naturaleza, Madrid.
INMG (1988) Nota explicativa sobre o ndice de Risco
Meteorolgico de Incndios Rurais. Diviso de Meteorologia Agrcola, Instituto Nacional de Meteorologia e
Geofsica (Instituto de Meteorologia at present), Lisbon.
McArthur AG (1977) Fire Danger Rating Systems.
FAO/UNESCO Technical Consultation on Forest Fires in
the Mediterranean Region, FO:FFM/77/3-01, France.
Reyfsnyder WE (1978) Systems for evaluating and predicting the effects of weather and climate on wildland fires.
World Meteorological Organisation, Special Environmental Report N 11, WMO - N 496.
Simard AJ (1994) Spatial analysis of large fire probabilities.
In Proceedings of the II International conference on
forest fire research. pp. 557566. (Coimbra)
Sol B (1989) Risque numrique mtorologique dincendies de fort en Rgion Mditerranenne: dpouillement du test de lt 1988 et propositions damliorations.
Note de Travails SMIR/SE, N1, France.
Sol B (1994) Comparaison de diverse mthodes destimation du danger dincendie sur le Sud-Est de la France: Feux
dt de la zone ctire et feux dhiver des Alpes de Haute
Provence. Technical Note of the Direction Interrgionale
Sud-Est pour Mto-France [In preparation]
Turner JA, Lillywhite JW, Pieslak Z (1961) Forecasting for
Forest Fire Services.World Meteorological Organisation,
Technical Note N 42, WMO-N 110.

Van Wagner CE (1975) A Comparison of the Canadian and


American Forest Fire Danger Rating Systems. Petawawa
Forest Experiment Station, Ontario.
Van Wagner CE (1987) Development and structure of the
Canadian Forest Fire Weather Index System. Canadian
Forestry Service, Forestry Technical Report 35, Otawa.
Van Wagner CE, Pickett TL (1985) Equations and FORTRAN
program for the Canadian Forest Fire Weather Index
System. Canadian Forestry Service, Forestry Technical
Report 33, Ottawa.
Viegas DX, Viegas MT (1994) A Relationship between rainfall and burned area for Portugal. International Journal
of Wildland Fire 4(1), 1116.
Viegas DX, Viegas MT, Ferreira AD (1992) Moisture content
of fine forest fuels and fire occurrence in Central Portugal.
International Journal of Wildland Fire 2(4), 6986.
Viegas DX, Viegas MT, Ferreira AD (1996a) Estudo comparativo da aplicao de diversos mtodos de clculo do
ndice de Risco Meteorolgico em Portugal [In
Portuguese] Internal Report, Coimbra.
Viegas DX, Bovio G, Ferreira A, Nosenzo A, Sol B (1996b)
Critical analysis of the application of meteorological fire
danger methods in Southern Europe. Internal Report of
the MINERVE Project.
Viegas DX, Sol B, Bovio G, Nosenzo A, Ferreira AD (1994)
Comparative study of various methods of fire danger
evaluation in Southern Europe. In Proceedings of the II
International conference on forest fire research,
pp. 571590, Coimbra.

246

S-ar putea să vă placă și