Assessment of the recurrence interval of wildfires in mainland Portugal and the identification of affected LUC patterns

ABSTRACT Wildfires are responsible for major land use and land cover (LUC) changes. These events are frequent and catastrophic in Portugal and are responsible for great damage and loss of human life. In this study, a map to assess the probability of wildfire occurrence (PWO) in mainland Portugal was created for the period 1975-2017 (first half of 2017). The PWO was obtained by the superimposition of all layers by adding all the burned areas for the total period. It was observed that the occurrences and extent of the burned areas are highly variable from year to year. The PWO map was cross-referenced with LUC geoinformation to evaluate the LUC types that were more affected by the wildfires. The results presented and the PWO maps are important for the management and planning of forest areas and for the creation of guidelines to implement preventive and reactive actions in case of wildfires.


Introduction
The Mediterranean region has a long history of fires, and these events have been an important factor in forestry, agriculture, and grazing (Ferreira-Leite, Bento-Gonçalves, Vieira, . However, the wildfires have devastated the Portuguese forest and small villages, with large consequences at different levels: social, economic, and environmental (Pereira, Pereira, Rego, Silva, & Silva, 2006).
The land use and land cover (LUC) of Portugal suffered large changes in recent years (Meneses, Reis, Pereira, Vale, & Reis, 2017;Pôças, Cunha, & Pereira, 2011;Santos, Tenedório, Rocha, & Encarnação, 2005), and these changes are explained largely by the occurrence of wildfires (Maia, Pausas, Vasques, & Keizer, 2012;Oliveira, Pereira, & Carreiras, 2012). The climate of Portugal promotes the occurrence of wildfires, as it has a rainy season during the spring, which is favorable to the development of vegetation, followed by a very warm period that triggers the development of large wildfires (Gomes, 2006) (on the coastline of Portugal, from north to south, the average temperature varies between 20.5°C and 23.6°C, the northwest of this country and a part of center sector varies between 19°C and 21°C, and areas to the south vary between 22°C and 24°C). The different types of climate create the conditions for the existence of different fire regimes in this territory (Parente, Pereira, & Tonini, 2016).
Several causes have been identified for the occurrence and development of large wildfires in Portugal: one is the increase in area covered by eucalyptus trees (Maia et al., 2012;Oliveira et al., 2012), and others are the increase in abandoned farmlands (Ferreira-Leite et al., 2016) and the reduction in pasturing (increasing the availability of fuel materials and the presence of shrubland, grass, etc.). The human ignitions of wildfires were also noted by Portuguese authorities as the cause for large occurrences. Some were caused intentionally and others were due to carelessness in the use of fire for the preparation of agricultural fields. Parente, Pereira, Amraoui, & Tedim (2018) indicate that a high percentage of wildfires in Portugal have a human origin, whether caused by accident, negligence, or arson.
However, an in-depth assessment on wildfires in Portugal is needed, i.e. based on the knowledge of the locations where fires occur, to determine the types of LUC most affected (incidence), assess the recurrence in these locations and check whether this is consistent in consecutive periods.
The main goal of this research was to map the probability of wildfire occurrence (PWO) in mainland Portugal and the assessment of the spatial and temporal distribution of wildfires (burned areas and occurrences). The second goal was the determination of PWO by different types of LUC and to identify the LUC patterns showing greater fire incidence. The return period distribution of these events (burned areas and occurrences) was also evaluated.
The relief of this territory is quite irregular and is characterized by deeply incised valleys surrounded by mountains in the north and by lower, less rugged relief in the south (Figure 1(A)).
The climate is strongly influenced by the Atlantic Ocean and the Mediterranean Sea due to the transition between the Mediterranean and the Atlantic climatic conditions. The rainfall regime is characterized by high spatial and inter-seasonal variability. The mean annual precipitation (MAP) ranges from less than 500 mm in the south and northeast to more than 2000 mm in the northwest. The MAP tends to increase with increasing latitude, elevation, and proximity to the Atlantic Ocean. Summer months (June, July, and August) are particularly dry, and the rainfall is concentrated mainly in the period lasting from October to March (climatological normals from Portuguese Institute for Sea and Atmosphere, IPMA).
The LUC has a greatly contrasting spatial distribution ( Figure 1(B)): forests and shrubland predominate in the centre and north, with high occurrence of eucalyptus and resinous plants (mainly Pinus pinaster); the largest agricultural areas are found in the south, essentially in the Alentejo region, and are characterized by important extents of Quercus areas; and settlements predominate in the littoral areas, where the more important metropolitan areas (Lisbon and Oporto) are located.

Data
The geoinformation (GI) on burned areas (vector format, shapefile) for the study area was produced by the Institute of Agronomy (ISA-UL) (1975)(1976)(1977)(1978)(1979)(1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989), the Institute for the Conservation of Nature and Forest -ICNF (Portugal)  and the European Forest Fire Information System (EFFIS) (first half of 2017). The time series used in the wildfires map production  and first semester of 2017) is the compilation of all data available by the institutions mentioned above. The ICNF data were created from EFFIS data (obtained by remote sensing) and the visual interpretation of orthophotos. The previous process was also used by ISA-UL, which completed analyses of Landsat MSS (1975MSS ( -1983 and TM/ETM+ (1984-2009) satellite images. The minimum mapping unit for these data is 1 ha. Each polygon that delimits a burned area represents an occurrence, independently of the causes of the ignition (propagation, human or natural ignition, etc.). Therefore, the total occurrences in each year are represented by the number of polygons represented in this year independent of whether a wildfire was caused or influenced by other wildfires that occurred in the region.
The LUC GI (Carta de Ocupação do Solo -COS in Portuguese; or Portuguese Land Cover Map in English) is available in the General Directorate for Territorial Development (DGT) website for the years 1995, 2007, and 2010 and has the following characteristics: scale 1/25,000; minimum mapping unit of 1 ha; vector data model (polygons); and a 20-m minimum distance between lines. The cartographic information of 2007 was obtained from the photointerpretation of orthophotos (resolution of 0.5 × 0.5 m), a process aided by IRS (Indian remote sensing) and AWiFS (advanced wide field scanner) satellite imagery and geographic information of cadastral surveys (agriculture and forestry) performed at DGT. All results were validated (geometrically and thematically) by the DGT to obtain data with high accuracy and quality (IGP, 2010). To produce the LUC cartography of 2010, DGT used the same methodology used for the cartography of 2007. The LUC cartography of 1995, also produced by DGT, was obtained using the vectorial data of LUC boundaries of 2007 that were updated to the 1995 LUC based on orthophotos and satellite images of this year (1995). The LUC cartography for the year 1995 is only available with the nomenclature used in the Kyoto report (DGT, 2014). For this reason, and to allow LUC comparison between different periods, the legend of the LUC maps of 2007 and 2010 follows the nomenclature used in the Kyoto report.
The production of the various maps also used complementary GI from the Portuguese Environmental Agency (elevation, drainage network and water bodies) and from the Portuguese Hydrographic Institute (bathymetry and digital terrain model of West of Iberian Peninsula Oceanic regionhydrographic zero).

Methodology
Commercial GIS software was used to perform the spatial analysis of wildfire distributions and to determine the recurrence of these events (burned area). Later, all the GI integrated in the PWO model was converted to raster (25 × 25 m).
In the production of the PWO map (Main Map), each annual event (individual layers) was assigned the value '1' (burned area) in the reclassification of the GI. The PWO was obtained by the superposition of all layers by adding all the burned areas for the total period (Equation (1)). A lower PWO indicates that an area is affected by at least 1 wildfire in the time series (T ) of burned areas and increases to the highest PWO (areas most affected by wildfires).
where PWO represents the probability of wildfire occurrence, Ba the occurrences with burned area in each year and T the total period (years). PWO was represented in quintiles in order to allow the user to easily understand the visible PWO, i.e. which regions are most affected by wildfires (the places with greater PWO). The colors used in the legend for the probability vary between light yellow (1) and red (5), highlighting the places where great care should be taken (preventive and reactive intervention) concerning the occurrence of wildfires.
The PWO was crossed with LUC GI (Figure 2) to identify the affected LUC patterns. This process was complemented by crossing the LUC available (1995, 2007, and 2010) by different periods of burned areas (total number of occurrences): 1995-2006 wildfire occurrences (WO) and LUC 1995;2007-2009WO and LUC 20072010-2017WO and LUC 2010. These procedures are important to identifying the LUC patterns that are more affected by the wildfires.
The Weibull probability distribution function was applied to calculate the return period for the forest fire occurrences , the first half of 2017, in mainland Portugal) and total burned area.

Wildfires in mainland Portugal
The occurrence of wildfires in mainland Portugal varied considerably in the period 1975-2017 (first half of 2017). However, in the past few years (especially 2010-2013) (Figure 3) there was an increase in the number of fires. In general, an increase in the number of occurrences was observed, but the total burned area per year does not follow this trend. In fact, the series for the years of 2003 and 2005 shows the largest burned area (440386 and 336594 ha, respectively).
The wildfires occurred mainly in the centre and north regions of Portugal (Figure 4), where the main forest areas (pinewoods, eucalyptus, and Quercus) are present. However, the extent of burned areas is increasing in the past decades, in the center of Portugal, for example, the wildfires that occurred at 17 and 18 June 2017 (more than 40000 ha), which resulted in 64 dead people according to the Portuguese National Civil Protection Authority.
The incidence of wildfires in this territory is very high, with areas that burned 17 times in the past 41.5 years that correspond to the areas with high PWO (Main Map). The middle sector of the centre region stands out by the extent of burned areas and the high values of PWO. These areas are especially in the Zêzere watershed, where the important surface freshwater reservoirs of Cabril and Castelo de Bode dams are located. The forest of this watershed is predominantly resinous; hence the easy spread of wildfires and the difficulty of extinguishing these fires (also because of the rough relief), resulting in large extents of burned area. Currently, the forest of this watershed is composed mainly by scrub and/or herbaceous vegetation associations, pastures, and other tree species (Meneses, Reis, Vale, & Reis, 2016), being the result of not only the burned areas' extent but also of the incidence of these events and not allowing forest regeneration, particularly of coniferous species.
The location of burned areas was influenced by the location of certain types of forests. The north and centre regions present a greater tendency for the  occurrence of several fires mainly due to the presence of large forest patches (pinewoods and eucalyptus). However, the north region exhibits the greatest total area burned by a factor of at least 17 (Table 1). We should also note that these two regions (north and centre) present values very close to those observed in general (Portugal) for the burned areas registered in different times (occurrences) ( Table 1).
Approximately 27% of the area of mainland Portugal was affected by wildfires from 1975 to the first half of 2017, including 14.6% of this territory that burned only one time, i.e. 54.5% of the total burned area ( Table 2). The percentage of burned area reduces with the number of occurrences, and this percentage reduces greatly if an area is affected by fire more than five times (<1% of total area).
The burned area is also very variable when the regions are compared. The north and centre present more occurrences compared to other regions, and the burned area is very high for the first occurrences (i.e. in first years). The Algarve region presents a distinct case with few occurrences of fire in the same area, but the total burned area is high as a result of the large size of each occurrence.
The return intervals of the wildfire occurrences (Ori) and burned area (Bri) are different, especially between the first year in which occurred wildfires to the next years, with a highlight on the natural break of Ori ( Figure 5). The probability that the determined number of occurrences is exceeded is very high for short return periods ( Figure 5, graph in the right). For example, at the end of the first year, there is an approximately 50% probability of exceeding the expected number of occurrences and burned area (86 occurrences and 10573 ha, respectively). However, increases in Ori are different compared to the increase in Bri after the first years, and there is a higher Ori, resulting in large return periods.
The occurrence of extreme fires, especially considering the extent of burned area, has higher return periods, which could be favorable for the regeneration of natural vegetation in these areas. On the other hand, the higher the productivity of plants (biomass) and the greater accumulation of forest fuel facilitate an increase of fire consequences depending on the severity and intensity of these events (Fernandes & Rego, 2010).

Framework of wildfires
Wildfires are recurrent events in mainland Portugal. This fact is mentioned by several authors (Ferreira-Leite et al., 2011;Ferreira-Leite et al., 2016;Oliveira et al., 2012;Verde & Zêzere, 2010) and confirmed in this research. For annual data, the number of occurrences does not have strong correlation with the burned area. In the most recent years, more wildfires occur each year, but the total burned area is reduced. However, large wildfires have occurred, for example, the fires in 2003 that resulted in large burned areas (Montiel-Molina, 2013), and these areas are the most significant in the center of Portugal, principally in the Zêzere watershed, and have serious environment implications (Meneses, Reis, Vale, & Saraiva, 2015). The wildfires are not distributed uniformly throughout the study area. This fact was also observed by Nunes, Lourenço, & Meira (2016) both in terms of ignition density and burned area.
The incidence of wildfires may be a factor that explains the perturbation of water quality parameters (WQP), mainly for certain physicochemical elements and compounds derived from wildfires in shallow water, providing the increase in the levels of specific WQP, a fact verified by Meneses, Reis, Vale, and Saraiva (2015). The subsequent burning of the forest material in short periods does not allow for normal forest regeneration, especially for coniferous forest (Pérez-Cabello & Ibarra Benlloch, 2003). However, the occurrence of the next wildfires in the same area, due to lower available biomass, results in a smaller concentration of the elements and compounds deposited in the surface of soils (e.g. ash content post-fire), and this factor can explain the reduction in the levels of determined WQP     in the main superficial freshwater reservoirs Shin, Sidharthan, & Shin Young, 2002;Vila-Escalé, Vegas-Vilarrúbia, & Prat, 2007).
Other negative consequences were the impacts of wildfires on human life (Finlay, Moffat, Gazzard, Baker, & Murray, 2012;Fowler, 2003), and Portugal stands out in the international context for the number of deaths year after year. The year 2017 stands out in this report, with 64 deaths in a fire in the centre region. The loss of forest also influences the biotic diversity, induces ecological perturbations, reduces employment (directly or indirectly linked to the forest), and causes socio-economic difficulties for the affected populations, among other consequences. The maps of burned areas and PWO are important to evaluate these consequences and for the creation of preventive and reactive measures.

Relation between LUC and wildfires
Wildfires occur most frequently in shrubland areas, eucalyptus, and Pinus pinaster forests (Table 3), but the incidence of wildfires in the areas with these LUC types is very variable between different periods. However, the PWO is very high in shrubland areas compared to other LUC types, and this result agrees with the high frequency of burned area presented in Table 3. The results obtained reflect also the fast regeneration of this type of vegetation in the burned areas and the creation of new conditions that facilitate the next wildfires due to the biomass availability.
The eucalyptus forest is noted by several authors as being very prone to large wildfires (Catry, Moreira, Tujeira, & Silva, 2013;Ferreira-Leite et al., 2016;Oliveira et al., 2012) due to the characteristics of this vegetation (pyrophytes species) and the rapid regeneration of this species (areas more favorable to the occurrence of more fires). However, the PWO of this forest type is lower compared to that of Pinus pinaster. On the one hand, the decrease in pine forest area in Portugal is linked to wildfires (slow regeneration after the first wildfire, but if it was affected by other wildfires in a short period, the regeneration of this species does not occur because young pines do not have seeds for the regeneration process), and this trend could the result of fire severity or increasing fire recurrence (Lucas-Borja et al., 2016). On the other hand, the high incidence of the pinewood nematode vector in Portugal was also responsible for heavy losses of pine trees (Autoridade Florestal Nacional, 2012;Meneses et al., 2017). The incidence of this vector constitutes a large disincentive to replanting new pine forests, and after the wildfires, most landowners opt for other alternatives. This fact also explains the reduction of this type of forest in the study area. Arnan, Quevedo, and Rodrigo (2013) indicate that changes in fire regimes can give rise to new types of forest cover or increase the distribution range of scarce forests in a regional context. This phenomenon is indirectly evidenced in this research, and we highlight the fact that the burned areas with a high incidence of wildfires have even more difficulty in regenerating forests, especially pine forests, when those affected by fires are later occupied essentially by shrubland. This LUC (in the Portuguese territory) was quantified in recent research presented by Meneses, Reis, Vale, and Reis (2018). This fact explains the extensive burned area and high PWO observed in this LUC type at different times.
In general, the Quercus forest exhibits reduced burned area compared to shrubland, Pinus pinaster, and eucalyptus forest, and the PWO is also lower in this type of LUC ( Figure 6). This type of vegetation is more resistant to fire (Curt, Bertrand, Borgniet,  (Table 3) compared to the types of forest mentioned before. Evidently, these LUC types exhibit lower PWO, explained by a low incidence of fires in this type of forest.
Some agricultural areas also exhibit burned areas, which can be caused by the propagation of wildfires or by locals starting fires (due to uncontrolled burnings) and consequent propagation to the forest, but the fire incidence in these areas is low (anthropic management activities) (Table 3).

Conclusions
Portugal has a history of catastrophic wildfires. Those more affected are the centre and north regions, but these regions also presented the highest percentage of forested area. The incidence of wildfires is most significant in the centre (Main Map), where the largest wildfires also occurred in the past years.
The return period of wildfires with the reduced burned area is very low in mainland Portugal; however, there is a high probability of many occurrences of small-magnitude fires in relatively short periods.
Pinus pinaster and Eucalyptus globulus presented high PWO, but shrubland stands out for the highest area burned and highest PWO. This fact is explained by the regeneration of this vegetation type in burned areas and its predominance in the Portuguese territory.
The PWO map (Main Map) is important in identifying the places most affected by these events and identifying where the next wildfires can occur based on the probability calculated from the incidents recorded in the past. This map also provides support for the creation of broad guidelines for economic and territorial policies, especially for the planning and management of forests and other natural resources. This map also allows for new approaches to fire probability studies in mainland Portugal and will allow for the validation of future fire risk maps produced by different institutions.
The data and procedures used in this paper possess two characteristics that may in some way influence the results and that should be taken into account in future research. First, the relatively long periods of the 1995-2006 and 2010-2017 series do not provide any assurance that the various land cover classes present in the reference cartography are maintained throughout these series. Since it is not possible to obtain other Table 3. LUC and the number of wildfire occurrences (% burned area) in mainland Portugal. dates for land cover cartography at this stage, one possibility is to use a shorter period of burned area data after each land cover date in order to reduce this possible discrepancy. Second, in the areas burned in each year, there is a change in the land cover; so new wildfires will affect a different class compared to those present in the reference land cover cartography. One possible way to overcome this situation is to update each year with the burned areas mapped in the previous year so that each land cover class is replaced an 'area burned in the previous year' class.
Software ESRI ArcGIS 10.5 was used to evaluate the wildfire distributions, model the PWO, and design the maps.

Disclosure statement
No potential conflict of interest was reported by the authors.