Forest fires are the major disturbance in the ecosystems of the Mediterranean regions of Europe, affecting their functioning and dynamic and with consequences on air pollution and human activities. Decreasing populations in rural areas and the associated abandonment of agricultural activities and animal grazing has lead to an increase in forest fuel and shrubland/woodland areas (i.e. Lloret et al., 2002; Mouillot et al., 2003; Pereira et al., 2005). In particular, 50% of forest fires in Europe are detected in the western Iberian Peninsula (Portugal and the Spanish region of Galicia). This is the highest percentage in Europe, followed by southern Italy, with only 10% (Pereira et al., 2005).
Several studies have shown that in the Iberian Peninsula wildland fires are mainly originated by accident and arsonists, instead of by natural causes (i.e. Váz quez and Moreno, 1995; Vasconcelos et al., 2001). However, although fire management and fuels have a considerable impact, present, and past meteorological conditions are widely recognized as being crucial for both the ignition and extension of wildland fires (i.e. García Díez et al., 1994; Kunkel, 2001). The meteorological and climatic conditions of each location are determined by spatial and temporal interactions of atmospheric patterns. This can be studied by combining and grouping individual atmospheric elements into classes, representative of the situations occurring at each particular moment. Procedures based on the determination of weather types and their application in environmental analyses have been used extensively. As an example, Raziei et al. (2011) used circulation weather types to study the causes of dry and wet spells in western Iran, and on the Iberian Peninsula this method has been used to study the precipitation regime (Trigo and DaCamara, 2000) and the lightning activity (Tomás et al., 2004).
Several studies relating atmospheric circulations and forest fires have been conducted in different areas. Brenner (1991) analysed the possible relationship between wildland fires and the El Niño-Southern Oscillation in the United States. Kitzberger et al. (2001) shown that Southerly US and Northern Patagonia share similar fire–climate relationship related to similar climate anomalies associated with El Niño-Southern Oscillation. Skinner et al. (2002) reported the impact of large-scale circulation patterns at the 500 hPa pressure level on wildland fire severity in Canada. Pereira et al. (2005) have described the synoptic patterns related to large forest fires in Portugal using anomaly fields of temperature and humidity.
Here, we analyze the relationship between circulation weather types and wildland fire activity in the western Iberian Peninsula. The results may help to better establish the risk of forest fires and to develop forecasting of forest fire risk based on the results from regional circulation models.
The data and methodology employed to obtain the circulation weather types (CWT) are described in Section 2. Section 3 is devoted to presenting and discussing the relationship between CWT and forest fire activity. A modelling of the daily number of forest fires is reported in Section 4. Finally, the conclusions are given in Section 5.
2. Data and CWT classification methodology
The present study covers the 6 year-period between 2000 and 2005. This time period seems representative of the circulation conditions over the area. Table I shows the long-term (1948–2005) average frequencies for the dominant circulation types, and they are similar to those of the 2000–2005 period. This is specially seen for the circulation types associated with strong fire activity [northeast (NE), east (E), and hybrid anticyclonic northeast (HANE), see Section 3]. The analysis was performed for the summer season (defined here as July, August, and September) because this is the time of the year with the highest forest fire activity in the spatial domain considered: Portugal and the Spanish region of Galicia (see Figure 1). For example Pereira et al. (2005) reported for Portugal (∼76% of the total surface considered in this study) that 83% of forest fires was registered during the summer months. Also García Díez et al. (1994) indicated that July, August, and September were the months with ‘the highest incidence of forest fires’ over Galicia.
Table 1. Total number of days and frequencies associated with each dominant circulation weather type (from July to September) in the western Iberian Peninsula
Long-term frequencies (1948–2005)
Total number of days associated with each dominant circulation weather type for the summer season (from July to September) in the western Iberian Peninsula. The 10 dominant circulation weather types (CWT) are northeast (NE), east (E), west (W), northwest (NW), north (N), cyclonic (C), anticyclonic (A), hybrid anticyclonic northeast (HANE), hybrid anticyclonic northwest (HANW), and hybrid anticyclonic north (HAN). The long-term (1948–2005) frequencies are also shown.
The meteorological data employed were the daily time-series of sea level pressure (SLP), temperature, and relative humidity at the 850 hPa pressure level (T850 and HR850, respectively), and wind intensity at a height of 10 m (V10), which was calculated through the use of the zonal and meridional wind components. They were obtained from and the NCEP/NCAR reanalyses data (Kalnay et al., 1996).
From the procedures developed by Trigo and DaCamara (2000) and Spellman (2000), the daily atmospheric circulation affecting the Iberian Peninsula can be characterized by the use of a series of indices associated with the direction and vorticity of the geostrophic flow. According to the definition of geostrophic wind, the pressure difference between two points can be used as an estimation of the geostrophic intensity. The westerly (WF) [southerly (SF)] flow index is associated with points located to the south and north (west and east) of the domain studied. The composition of WF and SF gives the total flow index (F). In a similar way, indices based on vorticity, are obtained: the horizontal variation in southerly (ZS) and westerly (ZW) shear vorticities and the total shear vorticity (Z).
These indices were calculated using SLP values at 16 grid points (p1 − p16), as shown in Figure 1. The following expressions, proposed by Trigo and DaCamara (2000) were used:
There are two main categories of CWT: (1) directional flow types (northeast, NE; east, E; southeast, SE; south, S; southwest, SW; west, W; northwest, NW; north, N), associated with marked wind direction (|Z|< F) and (2) rotational flow types [(cyclonic (C) or anticyclonic (A)], where rotation is more relevant (|Z|> 2F). There is also the possibility of hybrid categories (F < |Z|< 2F), which can be any combination of the two main categories (for example: hybrid anticyclonic easterly, HAE).
The data on wildland fires used in the present study were the daily number of forest fires (one forest fire is an individual ignition spot). They were provided by the Direcçao Geral das Florestas for Portugal and by the Dirección General del Medio Natural for Galicia (these are the Portuguese and Spanish Governmental Agencies, respectively). It is worth mentioning that most of studies on the relationship between climate and wildland fires used burnt area instead of number of fires. Delgado Martín et al. (1997) shown for the NW of Spain (an area related to that considered here) that the meteorological conditions associated with large numbers of forest fires are also those that contribute to more burned areas. However, the response of number of fires and burnt areas may be different, especially during strong wind events (Section 4).
3. Results and discussion
The total number of forest fires detected during the summertime of the period analysed was 145 566. Their interannual variation is shown in Figure 2. The maximum value was seen in the summer of 2000, with a total of 31 252 forest fires, followed by the summer of 2005, when 25 482 fires were detected. By contrast, the minimum value (16 746 forest fires) was seen during the summer of 2004. The large variation between maximum and minimum will be discussed in Section 4. Monthly variation in the number of forest fires is shown in Figure 3, where a considerable variability can be observed.
During the period analysed (a total of 552 d), 10 circulation types were seen to be dominant from 26 determined circulation types. A circulation type was considered dominant when it appeared in a number of days greater than 21 (552/26). The frequencies for each dominant type are shown in Table I. It may be observed that these 10 circulation types account for 88.0% of all days (486), although the percentage varies between a minimum of 79.3% in 2002 and a maximum of 91.3% in 2005. By contrast, only 66 d belong to the 16 non-dominant circulation types, whose frequencies are shown in Table II. The most frequent circulation during the summer in the domain studied is the NE type (22.8%), followed by the N and A types. The less frequent dominant circulations are the E and W types (4.0%). By contrast the highest non-dominant frequency was only 1.6% (HAW type). It is noteworthy that easterly circulations (NE, E, and HANE) account for 34% at all days and can therefore be taken as the dominant zonal flow direction during the summer. However, northerly circulations (NE, NW, N, HANE, HANW, and HAN) are the most frequent (62.8%).
Table 2. Total number of days and average frequencies associated with each non-dominant circulation weather type (from July to September) in the western Iberian Peninsula
Total number of days and average frequencies associated with each non-dominant circulation weather type for the summer season (from July to September) in the western Iberian Peninsula. The 16 non-dominant circulation weather types (CWT) are southeast (SE), south (S), southwest (SW), hybrid cyclonic northeast (HCNE), hybrid cyclonic east (HCE), hybrid cyclonic southeast (HCSE), hybrid cyclonic south (HCS), hybrid cyclonic southwest (HCSW), hybrid cyclonic west (HCW), hybrid cyclonic northwest (HCNW), hybrid cyclonic north (HCN), hybrid anticyclonic east (HAE), hybrid anticyclonic southeast (HASE), hybrid anticyclonic south (HAS), hybrid anticyclonic southwest (HASW), and hybrid anticyclonic west (HAW).
Table III shows the number of forest fires detected for each dominant circulation type. It is worth mentioning here that 94.5% of all forest fires (154 051) detected in the spatial domain and time period considered in this study occurred in the 10 dominant circulation types. The highest numbers of forest fires were found to be associated with the NE and N types, which is expected because they also are the most frequent circulation types. The relatively large number of forest fires in the E and HANE types should be noted, in spite of their relatively low frequencies. The possible relationship between circulation types and forest fires can be estimated by dividing the number of forest fires detected in each circulation type (numbers in Table III) by the number of days belonging to each circulation type (numbers in Table I). The result is shown in Table IV. Clearly, the most favourable circulation type for the generation of forest fires in the domain studied is the E type (548.9 fires per day on the average), followed by the NE type (357.3 fires per day), and the HANE type (309.4 fires per day). The number of forest fires per day for the other circulation types is lower than the average value (299.5). This indicates that easterly circulations are the most favourable for the generation of forest fires in the domain studied. Our results are in partial agreement with the finding reported by Pereira et al. (2005), who only analysed large summer forest fires for Portugal. They concluded that large forest fires were dominant on days associated with south-easterly conditions. It should be mentioned that there was only one day without forest fires (the 17 September 2002 with C circulation type; by contrast the maximum daily number of forest fires was 962 the 1 September 2002 with NE circulation type). Then, the probability of days with at least one forest fire occurrence can be taken as 100%, considering the whole spatial domain.
Table 3. Number of forest fires detected in each dominant circulation weather type (CWT) during the summer (from July to September) in the western Iberian Peninsula
Table 4. The average number of forest fires per day in each dominant circulation weather type (CWT; from July to September) in the western Iberian Peninsula. The highest figures are in bold
The meteorological variables with the greatest impact on the generation of forest fires are temperature and humidity (i.e. Delgado Martín et al., 1997). High temperature and strong drying affect the fuel moisture, facilitating the ignition of wildland fires. The role of wind is also important because it promotes convective heat transfer, affecting the development of convective columns on fires. Thus, in order to explain why easterly circulations favour forest fires in the western Iberian Peninsula, the 850 hPa level temperature and relative humidity, and the 10-m height wind were considered. The 850 hPa level has the advantage of representing low-level conditions without undergoing local effects. The impact of easterly circulations on these variables was analysed through the use of anomaly fields. The anomalies were calculated with respect to the mean values of the time period considered (2000–2005) and for the E, NE, and HANE circulation types.
The spatial distribution of the T850 anomalies is depicted in Figure 4. For the three circulation types, low-level temperature increases in the western Iberian Peninsula, especially for the E type, which is the circulation associated with the highest number of forest fires per day. The T850 anomalies for the NE and HANE types are similar, in agreement with the numbers in Table IV. The spatial distribution of the HR850 anomalies is shown in Figure 5. The negative anomalies in the western Iberian Peninsula indicate that low-level drying is associated with easterly circulations. Figure 5 is also in agreement with the figures in Table IV because the largest decrease in HR850 is clearly associated with the E circulation type, and the NE and HANE types show similar negative anomalies. Thus, the increasing temperature and decreasing humidity associated with easterly circulations explain their impact on the generation of forest fires in the domain studied. It should be noted that the results shown in Figures 4 and 5 are reasonable because low-level air from the East reaches the western Iberian Peninsula after crossing the Iberian plateau, which becomes overheated and dries out during the summer.
Figure 6 shows the spatial distribution of V10. Type E circulation shows positive anomalies north of ∼41°N, corresponding to Galicia and Northern Portugal, which is the area associated with the highest values in the number of forest fires within the domain studied (Moreno, 1999; Pereira et al., 2005). In the case of the NE circulation type, positive anomalies extend southward, but the figures are similar to those found for the E type. Accordingly, increased low-level wind seems to favour the ignition of forest fires on days associated with the E and NE circulation types. By contrast, in the case of the HANE type, V10 anomalies are negative in the area located north of ∼41°N. Although, the HANE circulation type is associated with a lower value of the daily number of forest fires than the E and NE types (see Table IV), the disagreement between the E/NE and HANE circulation types does not allow the impact of the wind on the ignition of wildland fires to be established clearly. Increased wind affects on development and extent of the wildland fires, but their impact on the generation of fires is not so clear.
4. Modelling the number of forest fires
Circulation types can only account for part of the variance in the number of forest fires. This is because of the limited number of types and the relatively arbitrary division between the groups. Another possibility to study the relationship between forest fires and circulations is through a regression of the number of forest fires to the six indices employed to classify the circulation types. Since those indices are continuous, this statistical approach provides a mathematical modelling of the number of daily forest fires. It must be taken into account that part of the information present in one index is also contained in another because of atmospheric dynamics and the nature of the indices. This means that the indices are interconnected and, hence, a stepwise multiple regression is appropriate to relate number of forest fires and the indices. A similar approach has been used for modelling the effect of circulations on winter temperature in Sweden (Chen, 2000). This procedure leads to the following model:
where N is the number of daily forest fires in the domain studied. We have considered the 552 d corresponding to the time period studied here. The multiple correlation coefficient is 0.5, which is significant at the 0.01% probability level. In view of the variability present in the number of forest fires, this is a good result. The scatterplot of the observed versus the predicted number of daily forest fires is shown in Figure 7. Most points fall inside the 95% confidence interval. The points outside of 95% confidence interval tend to be associated with the highest numbers of daily forest fires (most of them with more than 600 fires per day). The current meteorological conditions only account for part of the forest fires variability because there are other factors affecting ignition. One relevant factor is the past weather (i.e. García Díez et al., 1994). For example, high (low) precipitation values in the spring are generally related with much (poor) forest fine fuel growth. This may explain the difference between the number of forest fires detected in 2000 and 2004. The summer circulations were similar (Table I), but precipitation in the spring was different. The spring 2004 was very dry over the spatial domain considered in this study [amount of precipitation lower than half the normal climatic values (1961–1990)]. By contrast, 2000 was a humid year, especially April with precipitation higher than twice the normal climatic values.
Since WF is the westerly component of geostrophic wind, the model indicates that easterly flows have a positive effect on the generation of forest fires, in agreement with the results reported in Section 3. The model also shows that the total flow component (F) has a negative impact on the number of forest fires. Wind speed is dictated by the westerly component (the meridional component is usually one order of magnitude lower than the zonal component). Thus, high wind speed tends to reduce the number of forest fires, even for easterly flows. This may be interpreted as follows: for high wind speed, fires from several close points are easily merged, and the number of forest fires detected decreases.
The aim of this study was to obtain information about the impact of low-level atmospheric circulations on the generation of summer forest fires over the western Iberian Peninsula, which is the area with the strongest wildland fire activity in Europe. For this purpose, the procedure developed by Trigo and DaCamara (2000) to obtain circulation weather types was applied. This method is based on the use of a set of six indices associated with geostrophic wind and vorticity. Six years of data were used, but circulation type frequencies were similar to long-term (1948–2005) average frequencies. There was only one day without forest fires (considering the whole spatial domain) and it was found that easterly circulations were associated with a higher number of daily forest fires than the average; specifically, the weather types identified as east, northeast and hybrid anticyclonic easterly. The strongest impact was seen for the East weather type. In order to explain this result, it is shown that the three easterly circulations are associated with positive anomalies in low-level temperature and negative anomalies in low-level relative humidity over the western Iberian Peninsula. This means that easterly flows heat and dry the domain studied at low levels, thus favouring the ignition of forest fires. The possible impact of easterly circulations on low-level wind was also considered. However, in this case the result was not clear because positive anomalies were found for the area with strong fire activity in the case of the E and NE circulation weather types, whereas negative anomalies were seen for the HANE type.
The indices used to classify the weather types were also employed to develop a statistical model for predicting the daily number of forest fires. Stepwise multiple regression showed that only two indices are necessary in the domain studied: the westerly component of geostrophic wind and the total geostrophic wind intensity. The model indicates that the number of forest fires increases as the easterly flow increases, but decreases if the total wind speed increases. This latter seems to indicate that high wind speed allows near fires to merge, thus reducing the number of fires detected.
The results obtained here show that atmospheric circulations play an important role in determining the risk wildland fires and that regional circulation models can be used to forecast the risk forest fires, although other factors, as past weather conditions, may also be relevant.