Réf. Oliveri & al. 2012 - R: MANFRED

Référence bibliographique complète
OLIVERI, S., GEROSA, G., PREGNOLATO, M., 2012. Report on Extreme Fire Occurrences at Alpine Scale. MANDRED Project Report, 73 pp. [Rapport en ligne]

Preface
At alpine scale, the efficiency of fire danger forecast services, prevention activities and fire fighting actions has significantly improved in the last decades. Ongoing trends show a general decrease both in the overall frequency of forest fires and in the mean extension of burnt area per single fire occurrence. In spite of that, forest fires still represent one of the main threats impacting alpine forests. In the next future, ongoing climate changes could play a relevant role in influencing both the frequency, the geographical patterns and the regimes of fires in the alpine area. Moreover, they could play a relevant role in inducing the occurrence of big or extreme fires. As a consequence, great efforts should be made to monitor the evolution of fire patterns in the Alps; in order to understand if and how climate changes influence fire patterns and regimes in the alpine region and to use the outcomes of such analyses to steer and optimize both forecasting services and prevention/mitigation activities.

Based on these evaluations, MANFRED project directed its efforts towards that direction and carried out a set of activities specifically aimed at:
• building a pan-alpine database of fires, intended as a tool to monitor the evolution of fire patterns in the Alps;
• coming to a quantitative definition of “extreme” fire in the alpine area and making a census of such occurrences;
• investigating the geographical and frequency distribution of “extreme” fires, in order to understand if trends are coming out with regard to those kinds of occurrences and if they can be linked to climate change.

This report summarizes the main results of these activities and points out the most relevant outcomes achieved.

Mots-clés
 

Organismes / Contact
Partenaires

Stefano Oliveri (Università Cattolica del Sacro Cuore; stefano.oliveri@unicatt.it; +39 030 2406719)
Giacomo Gerosa (giacomo.gerosa@unicatt.it)
Marco Pregnolato (marco.pregnolato@odaf.mi.it)

 

Principaux rapports scientifiques sur lesquels s'est appuyé le rapport
 

(1) - Paramètre(s) atmosphérique(s) modifié(s)
(2) - Elément(s) du milieu impacté(s)
(3) - Type(s) d'aléa impacté(s)
(3) - Sous-type(s) d'aléa
       

Pays / Zone
Massif / Secteur
Site(s) d'étude
Exposition
Altitude
Période(s) d'observation
Alpine Space Liguria, Piemonte, Valle d'Aosta, Lombardia, Veneto, Friuli Venezia Giulia, Provincia di Bolzano, Provincia di Trento, Ardèche, Drôme, Gard, Alpes de Haute Provence, Hautes Alpes, Alpes Maritimes, Bouches du Rhône, Var, Vaucluse, Bas-Rhin, Haut-Rhin, Haute-Saone, Ain, Isere, Loire, Rhone, Savoie, Haute-Savoie, Slovenia, Austria, Switzerland ,Oberbayern, Schwaben     Four classes:
0 - 500 m 2.
500 - 1000 m 3.
1000 - 1500 m 4.
> 1500 m
1973-2009

(1) - Modifications des paramètres atmosphériques
Reconstitutions
 
Observations
 
Modélisations
 
Hypothèses
 

Informations complémentaires (données utilisées, méthode, scénarios, etc.)
 

(2) - Effets du changement climatique sur le milieu naturel
Reconstitutions
 
Observations
 
Modélisations
 
Hypothèses
 

Sensibilité du milieu à des paramètres climatiques
Informations complémentaires (données utilisées, méthode, scénarios, etc.)
   

(3) - Effets du changement climatique sur l'aléa
Reconstitutions
 
Observations

Historical fire occurrences

This section provides an overall picture of the “common sample dataset” (2000 - 2009) and highlights, whenever possible, ongoing trends.

Fire frequency: in the reference time span, the available dataset contains information on 26.017 fire occurrences. The highest number of fires has been registered in year 2003 (more than 4.500 occurrences). In the decade analysed the number of fires per year seems to show a progressive reduction (Figure 3). When data of year 2003 are not considered, the decreasing trend is statistically significant (p = 0.032).

When a broader time-span is considered, the tendence to a progressive reduction of the number of fires appears to be even more pronounced. Figure 4 shows the yearly frequency of fires in a set of Italian (Liguria, Piemonte, Valle d’Aosta, Lombardia, Veneto and Friuli Venezia Giulia) and French regions (Ardèche, Drôme, Gard, Alpes de Haute Provence, Hautes Alpes, Alpes Maritimes, Bouches du Rhône, Var and Vaucluse) for which comparable set of data were available since 1997.

Overall burnt area: like for frequency, also the parameter Total Burnt Area shows, at alpine scale and during the time span 2000 – 2009, a progressive decrease (statistically significative when year 2003 data ae excluded). Based on the available data, fires have interested 94.970 ha (Overall Burnt Area). Like in the former case, the most critical year was 2003 when fires burnt almost 50.000 ha of forests and agricultural lands (Figure 5). This means that almost 40% of the overall area burnt in the period 2000 – 2009 burned in 2003.

Average burnt area per fire: during the period under investigation, at alpine scale the average extension of forest fires was 5.7 ha (Figure 6). Besides representing the year with the highest overall number of fires and the highest overall burnt area in the decade under investigation, 2003 corresponds to the year with the highest value of average burnt area per fire (about 10 ha). The trend seems to be decreasind, thus demonstrating a progressive improvement in fire fighting activities at alpine level.

Biggest fire: the biggest fire registered in the period under investigation occurred in year 2003 (Figure 7). During that year, 6 out of 10 biggest fires registered in the last decade at alpine scale occurred. All these fires were registered in France and the largest one interested an area of more than 6.700 ha. When year 2003 data are not taken into consideration, the trend regarding the extension of the biggest occurrence per year is almost stable. In year 2008, the biggest fire occurred at alpine level had an extension of 415 ha, the lowest value of the parameter in the investigated period.

Seasonality: relative number of fires in summer and winter seasons has been considered (Figure 8). For the reference time span, the pan-alpine database contains data on 14.815 summer fires and 11.024 winter occurrences. There is no statistical evidence of any ongoing trend during the analysed period. In terms of relative extension of the overall burnt area in summer and winter fires, in the period 2000 – 2009, they show a stable trend (Figure 9), with no emerging tendencies from a statistical point of view.

Class of altitude: based on the available data (56% of the dataset contains information of the class of altitude of the ignition points – see Figure 2), occurrences at lower altitudes (class 0 – 500 m and class 500 – 1.000 m) show a progressive increase (Figure 10). At higher altitudes, on the contrary, the relative number of fires seems to decrease (class 1.000 – 1.500 m) or remains appreciatively stable (> 1.500 m). From a statystical perspective, the increase of the relative number of fires at lower heights (class 0) is evident (p = 0.0044; R2 = 66%). Besides class 0, also class 1 is associated with an evident increasing trend (p=0.4). The reduction of the relative number of fires at higher classes of altitude (class 2 and class 3) is not associated with any statistical evidence (p values ∼ 0.5).

Cause: based on the available data (about 47% of the overall dataset contains information on causes of fire occurrences and about 38% of the occurrences are referable to natural or anthropogenic causes), the relative number of natural fires (lightening) has increased in the reference time span. In years 2003 and 2007 the overall number of these occurrences, which could be linked to particular climatic conditions, exceeded the threshold of 7% (Figure 11). When year 2003 and 2006 data are not taken into account, the increasing trend is statistically significative: p = 0.02. It is important to highlight, however, that the increasing trend could be due to a general improvement in data acquisition and the consequent progressive reduction of fires due to “unknown” causes.

Aspect: based on the “common sample” data (due to the availability of the coordinates of the ignition points, it was possible to assign aspect values to 14.474 fires), about 50% of fires (6.814 occurrences) occurred in the aspect classes SE, S and SW (Figure 12). The less relevant classes are FLAT, NW and N (about 15% of the overall set of data). When taking into consideration seasonality (Figure 13), it comes out that winter fires are more frequent than summer occurrences in all classes of aspect (60%, 62%, 61% and 57% for E, SE, S and SW respectively) made an exception for N and FLAT. A comparison among the relative number of fires occurred in the classes N and S over the period 2000 – 2009 shows (Figure 14) that the trend is basically stable.

Most concerned forest categories: for 14.310 over 26.017 fires (about 55%) the coordinates of the ignition point were available. For those occurrences, it was possible to define the vegetation unit (from the “Map of natural vegetation of Europe”) in which the ignition point was located. When the level I of the legend (Table 2) is considered, the analysis shows (Figure 15) that almost 77% of the fires occur in the classes “Thermophilous mixed deciduous broadleaved forests” (1L, 5.561 fires) and “Mesophytic broadleaved deciduous and mixed broadleaved/conifer forests” (1E, 5.404).

With regard to level II (Table 3), the statistics show (Figure 16) that about 76% of the overall number of fires occur in the classes “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests” (2P, 4.814 fires), “Beech and mixed beech forests” (2D, 3.202), “Species-poor oak and mixed oak forests” (2M, 1.669) and “Mediterranean sclerophyllous forests and scrub” (2H, 1.131).

Analysis on level III (Table 4) highlight (Figure 17) that about 45% of the overall fire occurrences are registered in three vegetation units: “Italian-Balkan-Pannonian colline to submontane or supra-Mediterranean flowering ash-downy oak forests” (3V, 2.577 fires), “Italian-Balkan colline to montane hop hornbeam-downy oak forests and mixed hop-hornbeam forests” (3U, 2.184) and “South-west European southern temperate acidophilous mixed oak forests” (3MN, 1.629).


Trends in extreme fires occurrences

In the reference time span 2000 – 2009, 255 fires occurred with a total burnt area greater than 105 ha. In those occurrences about 95.000 ha burnt. The highest number of extreme fires was registered in 2003, with 62 occurrences and a total burnt area of more than 35.500 ha (about 40% of the total area burnt in extreme fires in the period under investigation and more than 300% of the total area burnt in 2005, the second most critical year in terms of area burnt in extreme fires).

Next graphs show trends relative to some of the variables taken into consideration in the analysis of the historical fire occurrences on the overall dataset. As in former case, the time series available for the analysis at alpine scale is too short to derive sound statistical conclusions. However, it can be pointed out that some of the evidences that rose from the previous analysis seem to be confirmed also in the deepening on extreme fires. Thus, such specific fire patterns and trends should be taken into consideration and deepened in the future monitoring activities on fires at alpine scale.

Fire frequency: as Figure 20 evidences, in the investigated period the frequency of extreme fires shows a progressive decrease. The trend is statistically significative: p = 0.094. Year 2003 is associated with the highest number of extreme occurrences (62), whilst in 2004 and 2009 it has been registered the minimum number of such occurrences (6).

Overall burnt area: like for fire frequency, also the overall burnt area in extreme occurrences shows a decreasing trend (if 2003 data are not taken into account, the trend is statistically evident: p = 0.04). In the analysed time span, year 2008 has the lowest level of the parameter (about 2.000 ha), whilst in 2003 (more than 35.000 ha) burnt about 40% of the overall hectares fires ran through during the period.

Seasonality: based on the available data, in the period 2000 – 2009 summer and winter extreme occurrences were 120 and 135, respectively. Their relative frequency (compared to the overall annual number of extreme occurrences) is approximately stable (Figure 22).

Class of altitude: on average, extreme fires at lower attitudes have been the most represented. In detail 34% of extreme fires had their ignition point at an altitude between 0 and 500 m, 35% between 500 and 1.000 m, 23% between 1.000 and 1.500 m and 9% at altitudes higher than 1.500 m, respectively. In terms of relative frequency of the occurrences, the analysis pointed out the same trend highlighted on the overall dataset (but trends are not significative from a statistical perspective): extreme fires with ignition point at lower altitudes are significantly increasing, whilst the frequency of occurrences at higher altitude is decreasing (Figure 23). In year 2009, no extreme fires occurred at altitudes higher than 1.000 m.

Cause: in the period under investigation only 4 extreme fires due to natural causes (lightning) have been registered. Two of them occurred in the region Friuli Venezia Giulia (one in 2003 and the second in 2006), one in Liguria (2003) and one in Switzerland (2003). It must be pointed out that, on the occasion of the Round Table dedicated to extreme fires in alpine forests, the delegates of Friuli Venezia Giulia highlighted a relevant increase of fires due to lightning in their area.

Aspect: more than 60% of extreme fires occurred in the time-span 2000 – 2009 had their ignition points in SE, S or SW classes (Figure 24). With regard to the classes of altitude (Figure 24), the analysis highlighted that ignition points of extreme fires in the range 0 – 500 m mainly fell in the classes of aspect NE, S and SW (24%, 21% and 17%, respectively). In the other altitude ranges 500 – 1.000 m, 1.000 – 1.500 m and > 1.500 m the most represented aspects are SE, S and SW (with average values of 25%, 23% and 17% respectively). With regard to seasonality (Figure 25), the analysis points out that winter extreme fires are more frequent in all the classes of aspect, with the only exception represented by W class (for which 55% of extreme fires occurred in summer).

Most concerned forest categories: for about 66% of the extreme fires censed (166 of 255) it was possible to define the vegetation unit (from the “Map of natural vegetation of Europe”) in which the ignition point was located. When taking into consideration the level II of the legend (Table 3), it comes out (Figure 26) that more than 65% of extreme fires occurred in “Beech and mixed beech forests” (2B, 53 fires) and in “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests (Quercus pubescens, Q. virgiliana, Q. trojana, Fraxinus ornus, Ostrya carpinifolia, Carpinus orientalis)” (2H, 57 fires). Lower number of extreme occurrences interested “Mediterranean sclerophyllous forests and scrub” (2C, 18 fires), “Species-poor oak and mixed oak forests (Quercus robur, Q. petraea, Q. pyrenaica, Pinus sylvestris, Betula pendula, B. pubescens, B. pubescens subsp. celtiberica, Castanea sativa)” (2F, 15 fires) and “Subalpine and oro-Mediterranean vegetation (forests, krummholz and dwarf shrub communities in combination with grasslands and tall herb communities)” (2G, 13 fires). In the other vegetation units occurred less than 8% of the overall number of extreme fires.
When level III of the legend is considered (Table 4), it comes out (Figure 27) that more than 75% of the overall number of extreme fires has occurred in the following units: “Italian-Balkan-Pannonian colline to submontane or supra-Mediterranean flowering ash-downy oak forests” (3G, 32 fires), “SW-Alpic-north Apennine-Corsican acidic beech and fir-beech forests” (3R, 24 fires), “Italian-Balkan colline to montane hop hornbeam-downy oak forests and mixed hop-hornbeam forests” (3F, 20 fires)”, “Mediterranean sclerophyllous forests and scrub” (3L, 18 fires) and “Illyrian-Dinaric beech and fir-beech forests” (3E, 17 fires).


Fire selectivity

Besides the analysis aimed at producing descriptive statistics of both the overall database of fires and the extreme occurrences, further investigations have been carried out with the aim to highlight specific and potentially critical patterns in forest fire occurrences at alpine scale [See Methodology]. Monte Carlo simulations are summarized in different Tables, where each vegetation unit is associated with the symbols listed in Table 9.

+++ Fire frequency (or fire extension) in the vegetation unit is significantly higher than expected.
Significance p < 0.001
++ Fire frequency (or fire extension) in the vegetation unit is higher than expected.
Significance p < 0.01

-

---

Fire frequency (or fire extension) in the vegetation unit is slightly higher than expected.
Significance p < 0.05
Fire frequency (or fire extension) in the vegetation unit is significantly lower than expected. Significance p < 0.001

--

-

n.s.

Fire frequency (or fire extension) in the vegetation unit is lower than expected.
Significance p < 0.001
Fire frequency (or fire extension) in the vegetation unit is slightly lower than expected. Significance p < 0.001
No significative difference between real and expected values
  No fires occurrences affected the vegetation unit

Table 9. Symbologies used to resume results of the Monte Carlo simulations.

The presentation of results is organized in two distinct sections: frequency and extension.

Frequency

The following paragraphs are intended to show results of the analysis performed to identify patterns in fire frequency distribution between the different vegetation units. The section is structured in three distinct levels: the first one contains the outcomes of analysis on all fires and potential extension of vegetation units (Level 1, see “Rationale and method of analysis”); the second shows the outcomes of analysis on fires with ignition points in forest areas and extension of real forest areas for the different vegetation units (Level 2, see “Rationale and method of analysis”); the last one is intended to contain results of the analysis carried out on extreme occurrences (Extreme events).

Level 1
When considering the whole dataset, frequency of fires comes out to be significantly higher than expected for four vegetation units: “Mediterranean sclerophyllous forests and scrub”, “Species-poor oak and mixed oak forests”, “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests” and “Sub-Mediterranean-subcontinental thermophilous bitter oak and Balkan oak forests, as well as mixed forests”. In some relevant vegetation units, on the contrary, the frequency of fire occurrences seems significantly lower than could be expected when considering their overall extension. Between them: “Alpine vegetation, Beech and mixed beech forests”, “Mixed oak-hornbeam forests”, “Montane to altimontane, partly submontane fir and spruce forests in the nemoral zone” (Table 10).

If the available sample is stratified based on the class of altitude of the point of ignition of fires, the analysis highlights that fires “prefer” the units (Table 11): “Beech and mixed beech forest”, > 1.500 m; “Mediterranean sclerophyllous forests and scrub” both in class of altitude 1 (0 – 500 m) and 2 (500 – 1.000 m); “Species-poor oak and mixed oak forests”, in the altitude range 0 - 1.000 m; “Subalpine and oro-Mediterranean vegetation”, both in the range 500 – 1.000 m and above 1.000 m; “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests”, at all heights; “Sub-Mediterranean-subcontinental thermophilous bitter oak and Balkan oak forests, as well as mixed forests”, at altitudes < 1.000 m; “Subnival-nival vegetation of high mountain regions of the boreal and nemoral zones”, in the range 500 – 1.000 m.

Fire ignition patterns show different behaviours at national levels (Table 12):
• in Italy frequency of fires is significantly higher then expected in the following units: “Beech and mixed beech forests”, “Mediterranean sclerophyllous forests and scrub”, “Species-poor oak and mixed oak forests”, “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests” and “Sub-Mediterranean-subcontinental thermophilous bitter oak and Balkan oak forests, as well as mixed forests”;
• in Slovenia fires “prefer” “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests”;
• in Switzerland fire ignition frequencies are significantly higher then expected in the units “Species-poor oak and mixed oak forests”, “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests” and ”Xerophytic coniferous forests, coniferous woodland and scrub”.

Table 13 highlights that, during the available time span, there were no significant changes in fire ignition patterns. At alpine level, in fact, the same four vegetation units are associated with a number of fires higher than expected during the whole decade under observation: “Mediterranean sclerophyllous forests and scrub”, “Species-poor oak and mixed oak forests”, “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests” and “Sub-Mediterranean-subcontinental thermophilous bitter oak and Balkan oak forests, as well as mixed forests”.

Level 2
When only fires with ignition points in forest areas are considered, together with the extension of real forest areas for the different vegetation units (Level 2, see “Rationale and method of analysis”), it comes out that, at alpine level, five vegetation units are characterized by frequencies of fires significantly higher then what could be expected based on the relative extension of the units (Table 14): “Mediterranean sclerophyllous forests and scrub”, “Species-poor oak and mixed oak forests”, “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests”, “Sub-Mediterranean-subcontinental thermophilous bitter oak and Balkan oak forests, as well as mixed forests” and “Vegetation of coastal sand dunes and sea shores”. Results are very similar to those obtained in Level 1 deepening.

The analysis of data stratified by class of altitude show that the following vegetation units have frequencies higher than expected (Table 15): “Beech and mixed beech forest”, for heights > 1.000 m ; “Mediterranean sclerophyllous forests and scrub” in altitude classes 1 (0 – 500 m) and 2 (500 – 1.000 m); “Mixed oak-hornbeam forests”, at altitudes > 1.500 m; “Species-poor oak and mixed oak forests”, at all heights; “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests”, at altitudes lower than 1.500 m; “Sub-Mediterranean-subcontinental thermophilous bitter oak and Balkan oak forests, as well as mixed forests”, between 500 and 1.500 m.

Frome the analysis at national level, peculiar behaviours come out both for fire patterns in Italy and Switzerland (Table 16):
• in Italy five vegetation units show frequencies of fire ignition significantly higher then expected, based on the effective extension of the vegetation units: “Mediterranean sclerophyllous forests and scrub”, “Species-poor oak and mixed oak forests”, “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests”, “Sub-Mediterranean-subcontinental thermophilous bitter oak and Balkan oak forests, as well as mixed forests”;
• in Switzerland fires “prefer” “Species-poor oak and mixed oak forests” and “Xerophytic coniferous forests, coniferous woodland and scrub”.

When the sample is stratified based on fire seasonality, the Monte Carlo analysis shows that (Table 17):
• summer fires in the “Alluvial and wet lowland forests in the nemoral zone” are more frequent then expected;
• fires “prefer” the vegetation units “Mediterranean sclerophyllous forests and scrub”, “Species-poor oak and mixed oak forests”, “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed foresti” and “Sub-Mediterranean-subcontinental thermophilous bitter oak and Balkan oak forests, as well as mixed forests” both in winter and in summer;
• summer fires in the vegetation unit “Vegetation of coastal sand dunes and sea shores, often in combination with halophytic vegetation, partly with vegetation of rocky sea shores” are more frequent then what should be expected.

When the reference sample is stratified both based on season and causes [for the analysis involving causes it was not possible to use the data from Piemonte region. As highlighted in the paragraph “Materials”, in fact, that dataset did not comprise information on causes of fires] (Table 1) of fires, the Monte Carlo simulations highlight that (Table 18):
• natural fires occur with frequency rates higher than expected only during summer, in the unit “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests”;
• fires caused by anthropogenic factors occur with frequency rates higher than expected both in summer and in winter:
o “Mediterranean sclerophyllous forests and scrub”, “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests” and “Sub-Mediterranean-subcontinental thermophilous bitter oak and Balkan oak forests, as well as mixed forests” register frequency rater higher than expected in summer and during winter;
o for “Alluvial and wet lowland forests in the nemoral zone” and “Vegetation of coastal sand dunes and sea shores, often in combination with halophytic vegetation, partly with vegetation of rocky sea shores” the number of anthropogenic fires is higher then expected in summer;
o in “Species-poor oak and mixed oak forests” anthropogenic fires have relevant frequency rates in the winter season.

A dedicated deepening has been carried out, with the aim to highlight the vegetation units for which natural fires have frequency significantly higher then expected at different altitude ranges. The analysis showed (Table 19) that frequency of natural fires is significantly higher than expected in “Beech and mixed beech forests” above 1.500 m and in the unit “Xerophytic coniferous forests, coniferous woodland and scrub” in the altitude range 500 – 1.000 m.

Extreme events

As highlighted in Table 20 and Table 21 , both with Level 1 or Level 2 approach, Monte Carlo simulations show that, at alpine scale, extreme fire occurrences selectively burn the vegetation units “Mediterranean sclerophyllous forests and scrub” and “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests”. On the contrary, extreme fires are underrepresented in some relevant vegetation units, with some differences between Level 1 and Level 2 analysis:
• “Alpine vegetation in the boreal, nemoral and Mediterranean zone”, “Beech and mixed beech forests”, “Mixed oak-hornbeam forests” and “Montane to altimontane, partly submontane fir in the nemoral zone” show frequency rates significantly lower than expected, with Level 1 approach;
• with Level 2 approach the underrepresented vegetation units are “Alpine vegetation in the boreal, nemoral and Mediterranean zone”, “Montane to altimontane, partly submontane fir in the nemoral zone” and “Subalpine and oro-Mediterranean vegetation”.

When analysis are carried out at national level (Table 22 and Table 23), Monte Carlo simulations show that evident patterns come out in Italy, where “Mediterranean sclerophyllous forests and scrub” and “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests” are overrepresented both in Level 1 and Level 2 cases. “Beech and mixed beech forests”, on the contrary, are associated with frequency rates significantly higher than expected only for Level 1 analysis.

When the seasonality in considered, Monte Carlo simulations at Level 2 show (Table 24) that extreme fires prefer “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests” during the winter season, whilst in summer the overrepresented vegetation units are “Mediterranean sclerophyllous forests and scrub” and, partially, “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests”.

Extension

The following paragraphs are intended to show results of the analysis performed to identify patterns in fires extension in the different vegetation units. The section is structured in two distinct levels: the first one contains the outcomes of analysis on all fires and potential extension of vegetation units (Level 1, see “Rationale and method of analysis”); the second shows the outcomes of analysis on fires with ignition points in forest areas and extension of real forest areas for the different vegetation units (Level 2, see “Rationale and method of analysis”).

Level 1
Monte Carlo simulations show (Table 25) that fires with ignition points in “Beech and mixed beech forests” and “Subalpine and oro-Mediterranean vegetation” have burnt area sizes higher than expected by a random null model. On the contrary, the burnt area of fires with ignition point in “Mixed oak-hornbeam forests” or“Montane to altimontane, partly submontane fir and spruce forests in the nemoral zone” is lower than the values expected by a random null model.

Analysis based on class of altitudes evidence (Table 26) that only fires with ignition point in the vegetation unit “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests” and in the altitude range 500 – 1.000 m have an average extension of burnt area significanty higher than expected.

Level 2
When only fires with ignition points in forest areas are considered, together with the extension of real forest areas for the different vegetation units (Level 2, see “Rationale and method of analysis”), Monte Carlo simulations show (Table 27) that no vegetation units have burnt area sizes significantly higher than expected from a random null model. Only fires with ignition points in “Beech and mixed beech forests” are characterized by average extensions higher then expected.

On the same sample, Monte Carlo simulations based on class of altitudes of the ignition points highlight (Table 28) that both “Mediterranean sclerophyllous forests and scrub” at heights < 500 m and “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests” in the altitude range 500 – 1.000 m present fires with average burnt area sizes significantly higher than a random null model. On the contrary, “Montane to altimontane, partly submontane fir and spruce forests in the nemoral zone” show fires with burnt area values lower than the average in all the altitude ranges.


Extreme fires and Fire Danger Indexes

The average values of percentiles, as described in the previous section, are shown in Figure 33. As reference term, the 0.5 value is also reported. In general, during the Vegetative season the Indices in the extreme fire days seem to be able to distinguish these days with respect to the general distribution (most of them with average percentiles above 0.8), and also from the fire events distribution. During the Non vegetative season the Indices showed lower predictive ability. Although the value is higher than the median of the overall distribution, in general, the Indices showed lower performances in discriminating between fire days and extreme fire days. In detail, during the Vegetative season, Indices of KBDI “family” [8], DF “family” [9] and DC (sub-index of FWI) [10] were more effective in distinguishing extreme fire events. During the Non vegetative season, Munger [11], Sharples [12] and Nesterov [13] Indices showed the highest performance.

KDBI, DF and DC Indices are mainly linked to moisture content of soil layers and depend much on drought periods (low precipitations). Even if further analyses are necessary to confirm and validate the results, it might be argued that the amount of moisture in fuel is one of the driving factors influencing the chance of extreme fire occurrences in Vegetative season. During Non vegetative season, a combination of meteorological factors seems to affect the probability to have large fires in the Alpine regions. In fact, whether Munger and Nesterov indices depend very much on a “rainfall” component (i.e. number of days since the last rainfall), Sharples only requires relative humidity and temperature as input variables [14].


Conclusions

Based on the data available at the different alpine regions, the analysis carried out considered a time-span (“common sample”: period for which almost all the regions were able to provide data on fires) ranging from 2000 to 2009. The time-span available is too short to derive definitive conclusions, from a statistical perspective. Nevertheless, the analysis of data available at alpine level highlighted some relevant evidences:

• the frequency of fires shows a progressive decrease. The measure is statistically meaningful (p = 0.032) when data of year 2003 are not taken into consideration;

• both the overall burnt area per year and the yearly average extension of fires are decreasing, even if analysis carried out do not evidence a statistically meaningful trend. This tendency can be associated with the progressive increase in the efficiency of fire prevention and fire fighting operations;

• the relative frequency of fires with ignition points at lower altitudes (0 – 500 m) is significantly increasing (p = 0.0044). Same tendency is associated with points of ignition in the altitude range 500 – 1.000 m, even if the significance of the trend is less evident (p = 0.4);

• the number of natural (lighting) fires is increasing. It must be pointed out, however, that this evidence could be due to the improvements in data acquisition and the consequent progressive reduction of fires due to “unknown” causes;

• in the reference time-span the relative frequency of winter and summer fires remained basically stable;

• more than 50% of the overall number of fires have their origins in S, SW or SE sides;

• relative frequency of winter fires is higher than the frequency of summer occurrences in all sides, exception made for fires occurring in N and FLAT areas;

• when potential distribution of vegetation units is considered (level II of the legend of the “Map of Natural Vegetation of Europe” - Table 3), the analysis carried out highlighted that highest fire frequencies are associated with “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests”, “Beech and mixed beech forests”, “Species-poor oak and mixed oak forests” and “Mediterranean sclerophyllous forests and scrub”; the Monte Carlo analysis highlighted that specific vegetation units are highly prone to fires, both in terms of frequency and average extension patterns;

• the selectivity simulations highlighted that seasonality doesn’t seem to affect the distribution of fires in the different vegetation units, whilst classes of altitude play a relevant role (high vulnerability of forest formations at the limit of their climax area);

• selectivty analysis highlighted strong differences between patterns at the national levels;

• the analysis didn’t allow to evidence any relevant ongoing trend in the time-span considered.

• although the definition of extreme fires should consider a broader set of variables (describing fire characteristics and impacts of the occurrence), the dataset available at alpine scale allowed the selection of extreme fire occurrences based on their total burnt area;

• based on the methodology developed for the aims of the project, extreme fires in the time-span 2000 – 2009 were defined as those with total burnt area ≥ 105 ha. In the reference period, 255 extreme fires occurred;

• the frequency of extreme fires is decreasing, with a significant statistical trend (p = 0.094);

• the relative frequency of winter and summer extreme fire occurrences has not significantly changed during the investigated period;

• extreme fires with ignition points in the ranges 0 – 500 m and 500 – 1.000 m represent almost 70 % of the overall set of occurrences, with relative frequencies of 34% and 35%, respectively; these relative frequencies seem to increase, but the analysis carried out did not evidence any meaningful statistical ongoing trend;

• more than 60% of the extreme fire occurrences in the investigated decade had their ignition points in S, SW or SE sides;

• when comparing, per side class, the relative frequency of fires with the relative frequency of extreme occurrences, it comes out that the relative frequencies of winter extreme fires are higher than those associated to ordinary fires for all classes. Highest values are associated with FLAT, SE, S and SW areas (60%, 27%, 24% and 19% respectively);

• taking into account the potential distribution of vegetation units (level II of the legend of the “Map of Natural Vegetation of Europe” - Table 3), the analysis carried out highlighted that more than 50% of the extreme fires had their ignition points in the units “Beech and mixed beech forests”, “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests (Quercus pubescens, Q. virgiliana, Q. trojana, Fraxinus ornus, Ostrya carpinifolia, Carpinus orientalis)” and “Mediterranean sclerophyllous forests and scrub”;

• the Monte Carlo simulations on fire selectivity highlight that, at alpine scale, extreme fire occurrences selectively burn (have ignition point in) the vegetation units “Alpine vegetation in the boreal, nemoral and Mediterranean zone”, “Beech and mixed beech forests”, “Mixed oak-hornbeam forests” and “Montane to altimontane, partly submontane fir in the nemoral zone”;

• based on the same analytic approach, extreme fires are underrepresented in some relevant vegetation units. When considering the whole set of extreme events in the Alps (both those with ignition point inside or outside forest areas), the analysis showed that those occurrences “avoid” the units “Alpine vegetation in the boreal, nemoral and Mediterranean zone”, “Beech and mixed beech forests”, “Mixed oak-hornbeam forests” and “Montane to altimontane, partly submontane fir in the nemoral zone”. When only extreme fires with ignition points inside forest areas are considered, the simulations show that “Alpine vegetation in the boreal, nemoral and Mediterranean zone”, “Montane to altimontane, partly submontane fir in the nemoral zone” and “Subalpine and oro-Mediterranean vegetation” are underrepresented;

• data at national level show significative differences if compared to the alpine “picture”. In Italy, for example, “Mediterranean sclerophyllous forests and scrub” and “Sub-Mediterranean and meso-supra-Mediterranean downy oak forests, as well as mixed forests” are overrepresented both when fires with ignition points inside and outside forest areas are considered. In Italy “Beech and mixed beech forests”, on the contrary, are associated with frequency rates significantly higher than expected when the whole set of extreme occurrences (both those with ignition point inside or outside forest areas) are considered.

• analysis carried out showed that, with regard to extreme fire events, there is a significant difference between the predictive power of fire weather indexes for the vegetative and non-vegetative seasons. During the vegetative season the KBDI “family”, the DF “family”, the FWI “family” (FFMC, DMC, DC, BUI, ISI, FWI), FFDI and Orieux indexes present high skills in predicting extreme occurrences. During the non-vegetative season the only indices quite effective in predictions are Sharples and Orieux.

Modélisations
 
Hypothèses
 

Paramètre de l'aléa
Sensibilité du paramètres de l'aléa à des paramètres climatiques
Informations complémentaires (données utilisées, méthode, scénarios, etc.)

Year / Date of the signalling

Cause:
- natural 2.
- doubt 3.
- anthropogenic

- Total Burnt Area (ha)
- Forest Burnt Area (ha)
- Non Forest Burnt Area (ha)

etc. [Cf. Table 1. Attributes of the pan-alpine database of fires]


The available dataset
First step of the set of activities carried out consisted in the generation of a pan-alpine database of fires. After a first recognition of available data, it was possible to define a reference set of attributes available at alpine scale and useful for an extended and standard description of fire occurrences. The thematic attributes used to describe fires are listed in the report [Table 1], with some relevant notes and the definition of their domains (when applicable).
As reference region for data acquisition, the alpine space area has been chosen. Data acquisition process involved a large set of data owners and data providers. It allowed to build a broad dataset, covering a very large part of the alpine area. At this regard, it must be pointed out that a fruitful cooperation has been set up with different ALP FFIRS project partners, with the aim to build a joint reference pan-alpine database of fires (e.g. Austrian data were made available by the ALP FFIRS consortium).
Availability of primary attributes data did not result uniform between the different alpine areas nor inside the single regions. For the different NUTS3 regions for which data have been acquired, the time spans of data availability of the information are shown [Table 5].
A general outline of the time series of data available is given [Table 6]. The analysis of the table highlights that the time frame for which most of the alpine regions were able to provide data (“common sample”) corresponds to the period 2000 – 2009.
With the limitations highlighted [in Table 5], the pan-alpine database contains data on 82.692 fires. In the “common sample” time frame, the dataset comprises 26.017 occurrences. The degree of completeness of this “common sample” dataset is described per single primary attribute [Figure 2].


Extreme fires selection methodology

Generally speaking, the definition of what is an extreme event is a quite abstruse plight. Like every relative concept, in fact, it has to do with one’s perception, familiarity and personal sensitivity. In the forestry sector the endeavour to give a straightforward definition of extreme event is even more difficult, especially considering the different technical backgrounds, expertises, experiences and attitudes of each forester potentially involved in the matter. Both partners who took part to the Work Package 6 meetings dedicated to reach a shared definition of extreme fire and the experts who attended the Round Table dedicated to extreme forest fires in the Alps (“Incendi estremi in foreste alpine. Imparare dal passato per meglio gestire il futuro”. Milan, 16th of February 2012) pointed out that extreme fires can be identified only in connection with an integrated evaluation of a broad set of variables.

On the occasion of the Round Table mentioned above, Giancarlo Cesti (Nucleo Anti-incendi boschivi, Valle d’Aosta Region) highlighted that extreme fires should be traced back to some general characteristics:

• inducing territorial and environmental conditions. Fire behaviour is strictly related to different kinds of factors:
o vegetational;
o orographic;
o climatic and meteorologic;

• fire propagation, which can be traced back to:
o canopy typologies;
o high fire intensity, even in case of grazing fires;
o peculiar behaviours of fire;
o also subterranean fires can induce extreme consequences, especially in case of long lasting occurrences and high sensitivity of forest formations (high rates of secondary mortality);

• overall impacts, with specific regard to:
o long lasting effects on forest stands;
o severity of impacts on forest stands;
o long lasting effects on the ecosystem;
o impacts on man made environment and anthropic communities;
o extension of the area interested by serious impacts.

From the survey on data availability carried out in the first phase of the project, it came out that the databases available at alpine scale do not allow (due to lack of numeracy) the identification of extreme fire occurrences by means of such a broad set of variables and evaluations. Moreover, the analysis at alpine scale was both intended to derive quantitative evaluations on frequency of occurrences and their geographical patterns and to make it possible to compare data of the different alpine regions.

Total Burnt Area of single occurrences came out as the only thematic attribute used to describe fires in the whole set of data made available by the alpine data providers. As a consequence, it has been chosen as reference attribute for the selection of extreme fires occurred in the alpine area in the time span 2000 – 2009.

Once that attribute has been chosen, a survey has been carried out, with the aim to evaluate what threshold of the Total Burnt Area attribute is used in the different alpine regions to identify “big” or “extreme” fire occurrences. Aim of this activity was to answer the question: does it exist, at alpine scale, a shared reference value of the total burnt area of fires which is used to identify extreme occurrences? The deepening highlighted that different regions, even inside the same nation, give different definition of extreme fires. Some examples from different Italian regions:
• Liguria: the regional Fire Fighting Plan defines as large fires those with a total burnt area > 50 ha;
• Piemonte: the regional Fire Fighting Plan sets a threshold of 10 ha;
• Lombardia: 100 ha is defined as reference threshold in the ERSAF (Regional Agency for Services to Agriculture and Forestry) publication “A fiamme spente”.

Due to the high heterogeneity of local approaches and evaluations used for the identification of “big” or “extreme” fire occurrences, a dedicated analysis has been carried out, in order to define a dedicated method suitable for a quantitative (statistical) identification of “extreme” occurrences from the MANFRED pan-alpine database of fires.

To this aim, the distribution of the data of total burnt area attribute was investigated. Data came out to be log-normally distributed (Figure 18).

Based on the result of the analysis, four methods for extreme events selection were chosen (consistent with the data distribution) and compared:

1. The first approach consisted in making a Rank of the Total Burnt Area of the available data → on the rank, 10% of the fires was selected, from the largest to the smallest (i.e. 10 fires on 100) → sum of the total burnt area and identification of the % of the overall area burnt by those fires → threshold. On the common sample dataset: 10 % of the fires = 85.23 % of the overall burnt area → threshold 8.1 ha.

2. The second method had big similarities with the previous one. The whole common sample dataset was ranked, based on the values of total burnt area data. The threshold area value has been selected as the area of the fire for which all the occurred fires greater than this area burned the 90% of the overall burnt area. On the common sample: 15 % of the fires = 90.00 % of the overall burnt area → threshold 4.5 ha.

3. As third methodology, the outliers approach was applied. Outliers fit for normally distributed series of data. On MANFRED common sample data (log normal distribution), they could be applied on the transformed data (log). Significant outliers often used for extreme occurrences selection are: Mean + 2σ and Mean + 3σ On the common sample: Mean + 2σ = 48.6 ha (97.8 percentile) Mean + 3σ = 486.9 ha (99.9 percentile)

4. Last method applied consisted in the computation of percentiles. Figure 19 shows the values of 25, 50, 75, 90, 95, 98 and 99 percentiles, estimated taking into consideration fires with Total Burnt Area ≥ 0.01 ha (23.210 occurrences).

Comments:
Both method 1 and method 2 seem not to fit with the aims of the analysis. Thresholds identified (8.1 and 4.5 ha, respectively) do not allow the selection of occurrences of “extreme” dimensions, in terms of extension.
Method 3 could allow a significative identification of ”extreme” occurrences, but it is associated with a high subjectivity in the selection of thresholds.
Like for outliers approach, also the selection of “extreme” events by means of percentiles is affected by a high degree of subjectivity. Nevertheless, percentiles are easy to compute (and easy to reply in case of database update), directly based on frequency distribution and reliable, with respect to the shape of data distribution. Effective proposal: an integration of methods 3 and 4.

Mean + 2σ 48.6 ha (97.8 percentile)
Mean + 2.5σ 153.9 ha (98.8 percentile) => 99th percentile: 105 ha
Mean + 3σ 486.9 ha (99.9 percentile)


Fire selectivity

Besides the analysis aimed at producing descriptive statistics of both the overall database of fires and the extreme occurrences, further investigations have been carried out with the aim to highlight specific and potentially critical patterns in forest fire occurrences at alpine scale. Ultimate aim of this analysis was to answer the following questions:
• do fires, in terms of frequency, burn selectively specific vegetation units? This answer implies to take into consideration both the frequency of fires in the different vegetation units and the relative extension of each unit;
• is the average extension of fires in specific vegetation units significantly different than what can be expected (fires in some specific units are significantly more or less extended than the average)?
• are extreme events characterized by any specific selectivity pattern?
• is there any evident trend in the evolution of the patterns?

Rationale and method of analysis
In forest ecosystems, fires can exert variable levels of pressure on different resources (fuel types). If several types of forest vegetation were equally fire-prone, fires would occur randomly in the forest areas with an equal proportion of available and burnt forest types. Fires are considered selective when resources are used disproportionately to their availability [1].

Fire selectivity studies represent a stream of investigations arising from the scientific debate on the relative importance of factors such as extreme weather conditions (weather hypothesis) or spatial variation in fuel (fuel hypothesis) in controlling fire behaviour [1]. In recent years, these kinds of analysis have been adopted to study selectivity patterns both in areas (e.g. North America) where most of the wildfires are nature caused [2] and in regions, like central and southern Europe, where human causes play the most relevant role in determining fire frequencies. At European level, different investigations have been carried out in the Mediterranean area. In their research on the mainland Portugal, Nunes et al. [3] showed that large and very large fires are not significantly selective for land cover (10 vegetation classes), while small fires are unequivocally selective. They suggest that under more severe fire weather conditions, usually associated with larger events, fire spread may be controlled by other factors, rather than fuels, whereas in fires burning under less severe weather conditions, fuels exert a stronger control. Bajocco et al. focused [4] on Sardinia region to investigate if land-cover types exist where fire incidence is higher (preferred) or lower (avoided) than expected from a random null model. They highlighted a close association between fire incidence and specific land covers. At Alpine level, only one investigation has been carried out on selective burning of forest vegetation. Pezzatti et al. tested [1], in the Canton Ticino area (southern Switzerland), the fire selectivity of different forest vegetation classes with higher (preferred) or lower (avoided) fire frequencies and burnt area sizes than expected by a random null model. Their work displays clear selectivity patterns with respect to forest cover.

To test whether, at alpine scale, fires burn selectively specific vegetation units, a Monte Carlo analysis has been carried out. It was applied both to understand if fires burn selectively specific vegetation units in terms of frequency and to evaluate if the extension of fires in some units is significantly higher or lower than the average.

Analysis on frequency were carried out through the following steps:
1. fires were randomly reassigned to the forest vegetation classes such that the probability of assignment of each fire to a given forest cover class was kept equal to the relative extension of that class;
2. a null hypothesis has been set, that forest fires occur randomly across the different vegetation units so that there is no significant difference between the relative abundance of fires in each forest unit and the relative extension of each vegetation unit within the analyzed area;
3. the real number of fires in each forest cover class was then compared with the results of 1.000 random simulations, each based on the number of fires in the period 2000 - 2009 for each vegetation unit considered;
4. for each vegetation unit class, p –values (two-tailed test) were computed as the proportion of Monte Carlo-derived values that were as low or lower (as high or higher) than the actual values.

With regard to fire extension, the method applied consisted in the following steps:
1. mean and median fire sizes in each vegetation unit were computed;
2. the observed values were compared with a Monte Carlo simulation for which, by keeping the number of fires in each forest cover class constant, the burnt surfaces of each fire were randomly reassigned. This way, a sample was created in which the surface burnt by each fire is distributed at random with respect to forest type;
3. the p -values (two-tailed test) were obtained as the proportion of 1.000 permutations for which the mean and median random fire sizes of each forest type are as low or lower (as high or higher) than the actual value.

Monte Carlo simulations have been carried out by means of a computation routine developed by UNICATT for the specific purposes of this analysis.

Analysis have been performed at two distinct levels:

1. Level 1:
A. all fires were considered (with ignition points in both forest and non-forest areas. Table 7);
B. the extension of the vegetation units was directly derived from the shapefiles of the “Map of Natural Vegetation of Europe” (Figure 30). This map defines the areas of potential distribution of different vegetation units and the shapes of the vegetation units do not take into account the effective distribution of forest areas. For the aims of the analysis, the level II of the legend was considered (Table 3).

2. Level 2:
A. only fires with ignition points inside forest areas were considered (Table 8);
B. the extension of the different vegetation units was estimated considering the real distribution of forest areas with respect to level II (Table 3) of the legend of the “Map of Natural Vegetation of Europe”.

Materials
The execution of fire selectivity analysis required the use of different categories of input dataset:

data on fires. When fire selectivity analysis have been performed, the pan-alpine database of fires was still under construction. Some information, in fact, still had to be provided or had been supplied in a preliminary version. As a consequence, for the aims of the investigation is was possible to make use of the following dataset:
o Italy (2000-2009): a complete set of information was available for Liguria, Lombardia, Veneto, Friuli Venezia Giulia, Trentino and Alto Adige. Partial information were available for Piemonte (data on causes were missing) and Valle d’Aosta (coordinates of the ignition points not available). Valle d’Aosta data were not used for Level 2 analysis, whilst data from Piemonte were not used in the estimates involving causes;
o Switzerland (2000-2009): complete set of information;
o Slovenia (2000-2009): coordinates of the ignition points not available. These data were not used for Level 2 analysis;
Only fires with overall burnt area ≥ 0.1 ha were considered.
Table 7 and Table 8 show, respectively, the overall number of fires used to perform Level 1 and Level 2 analysis.

data on forest distribution. Different input data for the different levels of analysis:
o Level 1: the shapefiles of the “Map of Natural Vegetation of Europe” were used, defining the overall potential extension of the vegetation units (level II of the legend. Table 3);
o Level 2: to determine the effective extension of the different vegetation units (taking into consideration the real extension of forest areas), data of the “Map of Natural Vegetation of Europe” where GIS overlayed with an alpine map of forest, produced on purpose. This map was generated by integrating CORINE Land Cover data for Italy and Slovenia (classes: 244: Agro-forestry areas; 311: Broad-leaved forest; 312: Coniferous forest; 313: Mixed forest; 321: Natural grasslands; 322: Moors and heathland; 323: Sclerophyllous vegetation; 333: Sparsely vegetated areas; 411: Inland marshes; 412: Peat bogs) and, for Switzerland (where CORINE LAND COVER data are not available), data of the Forest Cover Map (2006), produced by JRC.

Figure 29 shows, on a sample area: (a) the potential extension of vegetation units, derived from the “Map of Natural Vegetation of Europe”. Analysis at Level 1; (b) the distribution of forest areas, as derived from the map generated on purpose; (c) the extension of vegetation units, overlayed with forest areas. Analysis at Level 2.
Figure 30 and Figure 31 show, respectively, the potential extension and the forest extension of vegetation units at level II of the “Map of Natural Vegetation of Europe”.


Extreme fires and Fire Danger Indexes

The aim of this analysis, carried out by ARPA Piemonte (ALPFFIRRS Consortium) based on MANFRED data, was to analyse which fire weather indices [5] can be the most representative to describe extreme events occurred in the Alpine arc in a selected time period.

Materials

Fires data
The work has been carried out based on the data of the pan-alpine database of fires. In this work, analyses have been carried out on a set of 77 extreme occurrences registered in Liguria, Lombardia, Piemonte, Veneto and Friuli Venezia Giulia regions in the time-span 2003–2009 (period and geographical area covered by ALPFFIRS meteorological data). Table 29 lists the seasonal frequency of extreme fires in the regions and time-frame under investigation.

Meteo-climatic data
The weather data used for the calculation of the fire weather Indices were derived from the very dense non-GTS weather station network managed by the Regional Environmental Protection Agencies of the considered regions. For each station, temperature, relative humidity, wind speed and precipitation were collected. The stations data and the fires are assigned to the fire management areas as defined by the Forestal Services, and in particular the elevation of the station was carefully chosen to be representative of the fire management areas. Basically, for each fire the weather data used were the closest available in terms of elevation and distance.

Methods

19 different fire weather indices, selected among some of the most widely used in forest fire danger prediction [7], were calculated by means of the software FireCalculator (developed by the Swiss Federal Institute for Forest, Snow and Landscape Research in the framework of the ALPFFIRS project). The fire weather Indices describe the forest fire potential (ignition and spread) on the basis of measured weather parameters (precipitation, relative humidity, temperature, wind speed).

The meteorological data (2003-2009) were divided according to Vegetative (May-November) and Non vegetative seasons (December-April) and Indices were finally calculated. We based our evaluation on different fire Indices distributions: the general distribution (all events), the distribution related to fire events only and the distribution in case of extreme events only. Figure 33 depicts an example of the distributions of the DMC and DC Indices: the fire events distribution for both indices is clearly detached from the general and the no fire ones, while DC seems able to discriminate days with extreme events better than DMC.

To have a more quantitative evaluation of the skill of different indices in predicting extreme events, we calculated the percentile of the extreme fires for both “all events” and “fire events” distributions. We then averaged the obtained percentiles and used the average as a metric for the ability of a given index to evaluate extreme fires versus general distribution and fire events distribution. This method was applied for all the regions; for each index we obtained a general distribution and a fire event distribution containing the index percentile values for each extreme event recorded in the Alpine area over the common period 2003-2009.


(4) - Remarques générales
 

(5) - Préconisations et recomandations
Destinataires et portée du rapport  
Types de recommandations et / ou préconisations  

Références citées :

[1] Pezzatti et al. (2009) – “Selective burning of forest vegetation in Canton Ticino (southern Switzerland)”. Plant Biosystems, Vol. 143, No. 3, pp. 609–620

[2] Cumming (2001) – “Forest type and wildfire in the Alberta boreal mixedwood: What do fires burn?”. Ecol Appl Vol. 11, No. 1, pp. 97–110

[3] Nunes et al. (2005) - ”Land cover type and fire in Portugal: Do fires burn land cover selectively?”. Landscape Ecol Vol. 20, No. 6, pp. 661 – 673

[4] Bajocco et al (2008) – “Evidence of selective burning in Sardinia (Italy): which land-cover classes do wildfires prefer?”. Landscape Ecol Vol. 23, pp. 241–248

[5] Arpaci A., Vacik H., Formayer H., Beck A. (2010) A collection of possible Fire Weather Indices (FWI) for alpine_landscapes. ALPFFIRS project report

[6] Vacik H., Arndt N., Arpaci A., Koch V., Müller M., and Gossow H. (2011). Characterisation of forest fires in Austria. Austrian Journal of Forest Science, 128 (1), 1-32

[7] Arpaci A., Vacik H., Formayer H. and Beck A. (2010). A collection of possible Fire Weather Indices (FWI) for alpine landscapes. ALPFFIRS project report,Wien

[8] Keetch JJ., and Byram G.M. (1968). A Drought Index for Forest Fire Control. Southeastern Forest Experiment Station, Asheville, North Carolina

[9] Noble IR., Bary GAV. and Gill AM. (1980). McArthur's fire-danger meters expressed as equations. Australian Journal of Ecology, 5, 201-203

[10] Van Wagner CE. (1987). Development and structure of the canadian forest fire weather index system. Canadian Forest Service.

[11] Munger TT. (1916). Graphic method of representing and comparing drought intensities. Monthly Weather Review 44: 642-643.

[12] Sharples JJ., McRae RHD., Weber RO. and Gill AM. (2009). A simple index for assessing fire danger rating. Environmental Modelling and Software, 24(6), 764-774.

[13] Nesterov VG. (1949). Combustibility of the forest and methods for its determination (in Russian). USSR State Industry Press.

[14] Zumbrunnen T., Pezzatti GB., Menéndez P., Bugmann H., Bürgi M. and Conedera M. (2011). Weather and human impacts on forest fires: 100 years of fire history in two climatic regions of Switzerland. Forest Ecology and Management, 261, 2188-2199.