Global assessment of rural-urban interface in Portugal related to land cover changes

The rural-urban interface (RUI), known as the area were structures and other human development meet or intermingle 10 with wildland and rural area, is at present a central focus of wildfire policy and its mapping is crucial for wildfire management. In the Mediterranean basin, humans cause the vast majority of fires and fire risk is particularly high in the proximity of infrastructures and of rural/wildland areas. RUI’s extension changes under the pressure of environmental and anthropogenic factors, such as urban growth, fragmentation of rural areas, deforestation and, more in general, land use/land cover changes (LULCC). As other Mediterranean countries, Portugal experienced significant LULCC in the last decades in response to migration, rural abandonment, 15 ageing of population and trends associated to the high socioeconomic development. In the present study, we analysed the LULCC occurred in this country in the 1990 – 2012 period with the main objective of investigating how these changes affected RUI’s evolution. Moreover, we performed a qualitative and quantitative characterization of burnt areas within the RUI in relation the observed changes. Obtained results disclose important LULCC and reveal their spatial distribution, which is far from uniform within the territory. A significant increase in artificial surfaces was registered nearby the main metropolitan communities of the 20 northwest, littoral-central and southern regions, whilst the abandonment of agricultural land nearby the inland urban areas led to an increase of uncultivated semi-natural and forest areas. Within agricultural areas, heterogeneous patches suffered the greatest changes and were the main contributors to the increase of urban areas; moreover, this land cover class, together with forests, were highly affected by wildfires in terms of burnt area. Finally, from this analysis and during the investigated period, it appears that RUI increased in Portugal more than two thirds, while the total burnt area decreased one third; nevertheless, burnt area within RUI 25 doubled, which emphasizes the significance of RUI monitoring for land and fire managers.


Introduction
Mediterranean area is particularly affected by wildfires, mainly as consequence of its type of climate and vegetation cover fire proneness (Pellizzaro et al., 2012;Amraoui et al., 2015).In the European Mediterranean countries, fire incidence has dramatically increased in the last decades and the average total burnt area (hereafter, BA) has quadrupled since the 60's (San- Miguel-Ayanz et al., 2012), mainly due to changes in climate and land use (Moreira et al., 2011;Ferreira-Leite et al., 2016).
Portugal stands out from this group of countries since it counts the highest number of wildfires and has been the third most affected country in terms of BA in the last three decades (Pereira et al., 2014;San-Miguel-Ayanz et al., 2016).On average, about 95% of wildfires with known causes in Europe during the period 1995 to 2010 (corresponding to about 70% of the total number of recorded events) were associated to human activities (Ganteaume et al., 2013), while only a small percentage of fires (e.g.1% in Portugal, 5% in Spain) were naturally caused by lightings (Mateus and Fernandes, 2014;Vilar et al., 2016).Wildfires have long been considered a dynamic ecological factor and an efficient agricultural and landscape management tool, but more recently they are increasingly considered a hazard (Fernandes and Botelho, 2003;Bond and Keeley, 2005;Hardy, 2005;Van Wagtendonk, 2007;Pyke et al., 2010;Moreno and Oechel, 2012), which has motivated governments to implement measures for fires prevention, monitoring and mapping.
A key factor in this regards is to study the spatial and temporal distribution of wildfires into a given area, which, being influenced by the surrounding socio-economic and environmental factors, is far to be homogeneously distributed.For example a clustered pattern was discovered for Portugal, with fires' hot spots concentrated in the north-west and center regions, while the southern area presented lower densities of wildfires (Pereira et al., 2015;Tonini et al., 2017).Urban sprawl also affect the spatiotemporal pattern of these hazardous events and makes it difficult to define a boundary between human developments and rural areas, in order to better protect this interface where wildfires are more likely to occur.The modern urban landscape in Europe has a typical star-shaped spatial pattern (Antrop, 2000) with wedge of unchanged countryside persisting between lobes of urban development (Antrop, 2004).In this context populations and activities described either as "rural" or "urban" are closely linked and their distinction is often arbitrary (Tacoli, 1998).In the Iberian Peninsula, the diffuse urban expansion along radiating access roads connecting the center to commercial/industrial zones and isolated housing, as well as the growth of peri-urban centers in previously rural areas, leaves gaps in suburban/exurban space of urban agglomerations (Trigal, 2010).Here the expansion and reconfiguration of urban and metropolitan areas comprise the processes of sub-urbanisation and peri-urbanisation, mainly of coastal regions and, in the interior, of agricultural areas (Trigal, 2010).Land use changes are at the origin of landscape patterns and dynamics and have a strong influence on forest fires (Silva et al., 2011).On the one hand, each vegetated land cover type, such as agricultural, natural and semi-natural vegetation cover, has a specific fire proneness depending on the differences in vegetation structure, moisture content and fuel load composition (Barros and Pereira, 2014;Oliveira et al., 2014;Pereira et al., 2014).Further, fire occurrence affects landscape dynamics by changing the vegetation structure and the soil processes according to the fire adaptation of each ecosystem (Viedma, 2008;Pausas et al., 2009).
Population and urban areas significantly increased in Europe during the late XX century, which helps to understand the rapid land cover and land use changes (LULCC) (Noronha Vaz et al., 2012).In the European Mediterranean countries, LULCC are mainly caused by the increasing migration to urban centers at cost of the abandonment of rural areas, and by the expansion of costal tourism (Alodos et al., 2004;Tedim et al., 2016).One consequence is the urbanization, defined as the process involving the transformation of rural and natural landscape into urban and industrial areas, caused by the interaction of very different factors and largely influenced by communication, accessibility and mobility (Antrop, 2000).The dynamic conversion among rural and urban spaces can become extremely complex (Strubelt and Deutschland, 2001): urbanization generates the centralization of certain area by changing the land use, population density, economical activities and transportation network.This complex process is at the origin of the abandonment of remote rural areas with poor accessibility, which allows the expansion of wild low vegetation and forest (Antrop, 2004).Specifically, the abandonment of low-intensity agricultural lands and grazing practices caused the increase of forest cover and scrubland vegetation (Poyatos et al., 2003;Millington et al., 2007).
Significant LULCC occurred in Portugal in the recent period.The urbanization of coastal areas in the country occurred in concomitance with the abandonment of agricultural land in marginal areas and seemed to prevail between 1990 and 2000 while, in the later period (2000)(2001)(2002)(2003)(2004)(2005)(2006), the intensification of agriculture in areas where irrigation was available was a predominant process (Van Doorn and Bakker, 2007;Diogo and Koomen, 2012).Pereira et al. (2014) observed that among the southern European countries, in the period [2000][2001][2002][2003][2004][2005][2006] Portugal registered the highest rates of land use change marked by a significant increase in artificial surfaces and sclerophyllous vegetation and a decrease in forest area and natural grasslands, because of rural abandonment, urbanization and wildfires.
The interface between the wildland and the urban space, called Wildland-Urban Interface (WUI), has been deeply investigated by researchers and fire managers in the last decades.The United States Department of Agriculture (USDA), defined the WUI as the area "where humans and their development meet or intermix with wildland fuels" (Stewart et al., 2007); here fires can spread readily among vegetation fuels and urban structures.Anthropogenic features, such as the distance to roads and houses, negatively influence the probability of forest fire occurrence, while the population density positively affects it (Haight et al., 2004;Radeloff et al., 2005;Stewart et al., 2007;Hammer et al., 2009;Lampin-Maillet et al., 2010;Conedera et al., 2015), and these factors are broadly considered to elaborate WUI maps.Urbanization and the consequent abandonment of rural areas caused the expansion of this interface, increasing the probability that wildfires affect houses and infrastructures (Theobald and Romme, 2007;Zhang et al., 2008).There are strong evidences that the expansion of the urban and WUI area increased the fire density and related risk (Lampin-Maillet et al., 2011;Fox et al., 2015;Gallardo et al., 2015;Viedma et al., 2015), the cost of houses protection from fire (Pellizzaro et al., 2012) and have an impact on biodiversity and ecosystems (Radeloff et al., 2005).Researchers developed several geospatial models for defining and mapping the WUI, taking into account all the above-mentioned factors.In Europe, Lampin-Maillet et al. (2010) proposed an approach tested in southern France and based on the combination of four types of buildings configuration and three classes of vegetation structure.Following this model, Bouillon et al. (2012) developed WUImap, a software tool for mapping WUI in the Mediterranean region successfully applied in southeastern France, eastern Spain and Sardinia in Italy.In the Alpine context, geospatial approaches to map the WUI were developed in Switzerland by Conedera et al. (2015) and in France by Fox et al. (2015).Pellizzaro et al. (2012) characterized and mapped the WUI in Sardinia, Italy, using temporal steps of about 10 years from 1954 and 2008, and found an increase of the WUI's extension.Other studies focused on additional aspects related to the WUI, namely hazard/risk, vulnerability and fire risk management in Spain (Badia et al., 2011;Galiana-Martin et al., 2011;Herrero-Corral et al., 2012), fuel and fire modelling in France (Cohen et al., 2003;Pugnet et al., 2013).The majority of these researches were developed locally and performed at house-spatial-scale or for small regions within each country.
The active rural-urban conversion processes and the associated landscape changes, largely documented in the European context, induced to reconsider the WUI concept.In this respect, recent studies defined the Rural-Urban Interface (RUI) as an alternative to the WUI, to highlight the importance of including the rural areas, and identified the RUI as the most fire prone areas in Mediterranean countries (Badia-Perpinyà and Pallares-Barbera, 2006;Catry et al., 2010;Moreira et al., 2009).In the present study, authors investigated the RUI in Portugal: the main objective was to analyze changes in land use/land cover occurred in this country in the period 1990-2012 and to assess their impact on RUI's evolution.Moreover, a qualitative and quantitative characterization of the burnt areas within the RUI in relation to the LULCC was performed.Finally, this research provides a first attempt to map the RUI's extension and evolution at national level for the entire continental Portugal.

Study area
Continental Portugal (c.a.90,000 km 2 ) is located in the southwest of Iberian Peninsula, bathed by the Atlantic Ocean on the south and west coast and bounded by Spain at north and east.According to the census data, the population was about 10.6 million in 2011 and decreased to 10.3 million in 2017; its distribution displays a much higher density in the northwestern and southern coastal areas as well around the major cities (Fig. 1).Tagus river divides the country into two regions of approximately the same area but very different in terms of several biophysical and human drivers (Fig. 2).The north region is characterized by a temperate climate, with dry and warm summer, altitude ranging from sea level to about 2000 m and mean watercourse density of about 0.65 km/km 2 .According to CORINE Land Cover (hereafter, CLC) inventory 2012 (EEA, 2016), 54% of north's area is covered by forests and scrublands, 40% is used for agriculture and about 5% is occupied by artificial surfaces (Fig. 2).The 5 region south of Tagus river is characterized by temperate climate with dry and hot summer, low altitude range (between sea level and about 1000 m), and mean watercourse density of about 0.58 km/km 2 .According to CLC 2012 inventory, 56% of this territory is occupied by agricultural areas, 40% by forest and semi-natural areas, and only 2% by artificial surfaces (Fig. 2).covers the 1975-2015 period and comprises about 49,000 fire events for a total BA of 4,430,000 ha (Oliveira et al., 2012;Barros and Pereira, 2014).The annual fire perimeter maps resulted from semi-automatic supervised image classification (Gorte, 1999) performed with classification and regression trees algorithm of Breiman et al., (1984) on late summer-autumn Thematic Mapper/Enhanced Landsat satellite imagery.Technological improvements of satellite sensors allowed acquiring and processing over time higher-resolution images with an increasing accuracy from the initial 30 hectares to 5 hectares after 1984, and even higher after 2005.To ensure the accuracy of the data, results were compared against field statistics gathered on the ground by the National Forest Authority and by the National Civil Protection Authority.For each fire record, the dataset comprises the BA (perimeter map) and the year of occurrence.
Land use/land cover's information came from the CLC inventory provided by European Environment Agency (EEA).CLC is delivered as cartographic product, both in raster (i.e. a regular grid of cells) and in vector-shapefile (as point, line and polygons) format.The minimum cartographic unit is of 25 ha (500 by 500 m) with a geometric accuracy of 100 m minimum and a thematic accuracy over 85% (EEA, 1994).CLC nomenclature is a three-level hierarchical classification system with 44 classes at the third and most detailed level (Table 1).The five more general classes for the first level are the following: Artificial Surfaces (AS), Agricultural Areas (AA), Forest and Semi-Natural Area (FSNA), Wetlands, Water bodies.CLC inventories are currently available for four periods (1990,2000,2006,2012) with a minimum time consistency of plus/minus one year.CLC was already used for land-use change and urban dynamics studies in Portugal (Noronha Vaz et al., 2012;Pereira et al., 2014).To identify and detail the major habitats/plant communities/vegetation types corresponding to each CLC class in Portugal, we employed the Soil In the present study, the four CLC inventories were employed to analyze LULCC at different levels and to map RUI at different periods, according to the methodology described below.
The first analysis consisted in investigating LULCC within the entire the study period (i.e. from 1990 to 2012).This allowed elaborating the map of changes showing the transitions among the five more general classes (CLC first level hierarchy) and, more in detail, to quantify gained and lost areas with reference to both the first and the second level hierarchy of CLC.Moreover, the difference between gains and losses within each class divided by the total area covered by the specific class in the later period (i.e. the net change) was computed and the result expressed in percentage.
RUI was then mapped for each period (1990,2000,2006 ant 2012) using a geospatial approach designed to extract the area of intersection between a buffer around the AS and the area resulting from the sum of the FSNA plus the Heterogeneous Agricultural Areas (HAA, a sub-level of AA).Different buffer width from 100 m to 2 km were tested: finally, we adopted a buffer width of 1 km, corresponding to two times the spatial resolution of CLC inventories.This value is in line with values applied in other countries for WUI mapping (Radeloff et al., 2005;Vilar et al., 2016) and, in the same time, is enough large to avoid bias in the results.The others agricultural areas (i.e.arable lands, permanents crops and pastures) were not included in the RUI definition since these vegetated land covers are usually well managed, mostly irrigated and frequently constitute an obstruction to fire spread.Similarly, San-Miguel-Ayanz et al., (2012) suggested that HAA have to be considered in the definition and quantification of the RUI in Portugal, together with FSNA.
The geocomputation which allowed producing the RUI's maps was performed under ArcGIS TM software environment.Namely, the geoprocessing workflows was implemented into a Model Builder (Fig. 3), a specific application used to create, edit and manage models, meant as workflows that string together sequences of geoprocessing tools (e.g.selection, buffer, intersect), feeding the output of one tool into another tool as input (i.e. the raster or vector digital data).Finally, we analyzed how each land cover class (in respect of the third level hierarchy of CLC) was affected by wildfires in terms of BA for each investigated period within the RUI.To this end, polygons defining the BA registered at each CLC_year plus/minus one year (1989-1991, 1999-2001, 2005-2007, 2011-2013) were merged together and the resulting BA polygons were clipped over the corresponding RUI map.The resulting outputs, representing the BA within the RUI cumulated over three years around each investigated period, were finally overlapped with the CLC source map.  4 Results

LULC change analysis
The analysis of main changes in the area occupied by the first level hierarchy classes between CLC1990 and CLC2012 allowed to visualize the main transitions occurred within the entire investigated period (Fig. 4) and thus to have an overview of the LULCC occurred in Portugal.It resulted that main transitions occurred between vegetated areas (i.e.AA and/or FSNA) to artificial surfaces (AS) and between FSNA and AA in both directions.AS increased mainly nearby the main metropolitan communities of the northwest and littoral central and southern regions.A transition from vegetated areas (AA and FSNA) to AS is also visible in centre-north and is probably due to the intensification of the main road network to connect the emergent inhabited rural-area.The conversion from FSNA to AA and vice-versa appeared to be an active and dynamic process prevailing in the southern half of the country, but it was revealed also in the inner northern region.Figure 5 shows the areas gained and lost for each CLC first-level class and the net percentage of changes, computed relatively to the total area of each class in the later land cover.The main changes in terms of surface were registered by AS, which increased 165×10 3 ha, and AA, which decreased 184×10 3 ha, but in terms of net percentage of change the increase of AS was about 50%, while AA decreased only 4.4%.The two classes which manly contributed to the increase in AS were AA, with 110×10 3 ha, and FSNA, with 50×10 3 ha.second level hierarchy.Figure 6 shows that the majority of the CLC classes (level 2, Table 1) displayed important net changes in terms of relative gains and losses compared with values for the same classes in the later period.Scrub and/or herbaceous vegetation associations (code 32) registered a net gain of about 520 ×10 3 ha (+24%), while the Forest area (code 31) decreased about 460 ×10 3 ha (-23%).Arable land (code 21) was the only AA registering an important negative net change of -225 ×10 3 ha (-20%).Among AS, Urban fabric (code 11) significantly increased 110 ×10 3 ha (45%), and, in terms of net percentage of change 10 by class, all the other three AS sub-levels, including Industrial/commercial and transport unit (code 12), Mine/dump and construction sites (code 13), Artificial/non-agricultural vegetated areas (code 14), increased more than half.

Spatial distribution and characterization of burned areas
Almost all the CLC third-level classes belonging to FSNA (code 3) were affected by wildfires in terms of burned area (Table 2), with the Transitional woodland-shrub (code 324) and Mixed forest (code 313) as the first and second more damaged classes.
This trend was similar in all the four investigated frame-periods, as highlighted in Fig. 8, where the same results are expressed in percentage for each CLC classes considering only the areas within the RUI, as the ratio of BA over the total BA for the entire frame period.

RUI map 5
The classes FSNA and HAA were considered in the present study to describe the flammable rural area which, intermingling with the urban area, defines the RUI.Thus, RUI maps arose from the zone of intersection between the sum of HAA plus FSNA and an enhanced area around AS (Fig. 3).The result was a zone of interface evolving in space and in time due to LULCC (Fig. 9).
Analyzing the period 1990-2012, the increase of RUI was more active in the north-west and along the coast, where the transition from HAA to Urban fabric was particularly intense.This evolution was mainly due to the urban growth and to the intensification 10 of the road network.The total size of the RUI, the fraction of BA within the RUI (BAR) and the total BA (TBA) were computed (Fig. 10).It resulted that RUI increased from about 780×10 3 ha in 1990 up to about 1310×10 3 ha in 2012 following a power-law (RUI=776310.year0.1686, 2 =0.99).Moreover, we computed the contribution both to the RUI and to the BA within the RUI (BAR) of each CLC class that make up the RUI, namely HAA (code 24), Forests (code 31), Scrub and/or herbaceous vegetation associations (code 32), and Open spaces with little or no vegetation (code 33).Results can be summarized as follows (Fig. 10):

Discussion
The LULCC analysis performed in the present study indicates that from 1990 to 2012 AS (code 1) globally increased in Portugal about 50% (Fig. 5).This growing process is in good agreement with previous findings of other authors (Tavares et al., 2012;Marques et al.,2014;Meneses et al., 2014;Oliveira et al., 2017).Moreover, the present study confirms that the urban growth process in Portugal (quantified as changes in AS) was principally caused by the transition from HAA (code 24) and secondly from FSNA (code 3) (Fig. 4 and Fig. 7).The urban development mainly affected the south coastal regions, mostly in the area between Portimão and Faro, and was particularly strong nearby to the main metropolitan communities of the northwest and littoral centre, namely Porto and Lisbon (Fig. 4 and Fig. 1).Silva and Clarke (2002) described the characteristics and the recent intense urban growth of the metropolitan area of Porto (MAP) and Lisbon (MAL) associated with the economic growth in the end of the XX century (Fernandez-Villaverde et al., 2013).More in details, MAL urban pattern is characterized by a mixture of urban surfaces and large farmlands, an intense urbanization along with train lines and main roads, and the emergence of tertiary centres.On the other hand, MAP is described by scattered urbanization and dispersed settlements, towns and rural villages surrounded by mountains, within small patches of intensive agriculture and pine forests in a steep slope topography.The decline of the rental market in the country lead to the degradation of old urban areas and the increase of constructions in the immediate periphery of Lisbon, while in the north new houses were built by the owners in their small plots of land, promoting a more dispersed urban pattern and an irregular spatial growth (Silva and Clarke, 2002).The dispersion of the population and of its activities in MAP is also explained and reinforced by the absence of a regional territory planning and the adoption of polycentric models of urban growth by the national authorities (Cardoso, 1996;Silva and Clarke, 2002).
Another active process identified by the performed change analysis is the abandonment of agricultural lands nearby the inland urban areas, which leads to an increase of uncultivated semi-natural and forest areas (Fig. 4) causing an increase of the urban/rural interface.
As regards the RUI definition and mapping model developed in the present study, we tested different buffer width from 100 m to 2 km, which led to different RUI's size but, in essence, to approximately equivalent results relatively to the RUI's dynamic.In literature, Vilar et al. (2016) applied a buffer width of 100 m, corresponding to the median of the distances defined in each country's national legislation (Portugal, Spain, South-France and Italy) for protection against wildfires, which makes brush clearing obligatory within a certain radius around each house located close to forests or scrublands.In US, interface WUI was defined as developed areas in the vicinity of wildland vegetation and mapped considering census blocks above 6.17 housing units/km 2 that are within a distance of 2.4 km from wildland vegetation (Radeloff et al., 2005).Finally, we decided to show results obtained applying a buffer width of 1 km because smaller values, even if more in line with the Portuguese national indications, could bias the results, given that the spatial resolution of the CLC inventory was of 500 by 500 meters.
RUI definition aimed to map the developed area located in close proximity of wild vegetation, where wildfires can cause deaths, injured, damages to human structures, and finally where human-caused wildfires are more likely to occur.Nevertheless RUI map was not based on fire incidence measures, thus it not aimed to assess fire risk or fire regimes, which depend on other factors such as topography, weather, vegetation characteristics (Radeloff et al., 2005;Parente and Pereira, 2016).Most of RUI's area detected in Portugal (Fig. 9) was located in regions of high population density and surrounding major cities, while RUI's growths mainly occurred in the transition zones from vegetated lands (AA and FSNA) to AS (Fig. 4).Urbanization and the consequent reconfiguration of Portuguese cities caused new urban problems and challenges associated to the increased fragmentation of the cities and different rural-urban relationships, as also reported for Portugal and Spain (Trigal, 2010).It is important to underline that the impressive increase of the RUI and of BA within the RUI, detected in just a little more than two decades, is not exclusive of Portugal.In Continental US, WUI increased 52% from 1970 to 2000 and 90% of this area included high and highly variable severity fire regimes (Theobald and Romme, 2007).In Europe, Fox et al. (2015) found a progressive increase in fire risk in an increase in the number of houses and dwellers, which tripled during the study period, as well as a sharp grow in summer population; rose of road network length and; finally, increase of the WUI's extension.
The inspection of the accurate Soil Use and Occupancy Chart national map (COS2007 v2.0) allowed us to identify the vegetation types and major habitats/plant communities corresponding to each one of the CLC classes for Portugal.Table 3 of the annex/supplementary material, shows the composition of the CLC classes in terms of COS classes.For simplicity, results are only presented for COS classes with more than 1% of total CLC area.Nevertheless, it is important to underline that these 19 COS classes account for 98.3% of total CLC area.Despite some expected differences between classifications, essentially noted in CLC classes less affected by wildfires, results for all the other classes can be summarized as follows: (i) Temporary dryland crops is the larger COS class (22% of total CLC area) and accounts for higher area fraction in CLC Agricultural area (code 2) and Scrub and/or herbaceous vegetation associations (code 32); this is particularly consistent with the agricultural practices, It is important to underline that, from 1990 to 2012, the increase in the BA within the RUI is higher (100%) than the increase in the RUI area (70%).This result suggest that other factors, besides the increase of the RUI area, are responsible of the increase of the burnable area within the RUI.In this regard, it is important to take into account some of the specific characteristics of the country, well described in terms of demographic, territory and forest statistics compiled in Feliciano et al. (2015), which can help to understand the most important factors affecting the forest management in Portugal.Forest is nowadays the dominant land use in the country (with more than 35% of total area), followed by bushes and grasslands (>29%), and agricultural areas (>24%) (Feliciano et al., 2015;FAO, 2018).According to the National Forest Inventory (IFN, 2010), in the 1995-2010 period, four tree species occupied about 85% of total forest area: Eucalyptus (22%-27%), Cork oak (23%-24%), Maritime pine (30%-23%) and, Holm oak (10%-11%).The first three species have the ability to generate land and business income exceeding 50 euros/ha/year (CM, 2015).Most of the forests and wooded lands (>93% in 1995) have non-industrial private owners and there is a high fragmentation of the forest property, particularly evident in the private sector (Mendes et al., 2004).Management practices are also very different and changed significantly in the last years, especially in non-industrial private forest (Novais and Canadas, 2014).According to Feliciano et al. (2015), 1/3 of Eucalyptus area is well managed by the industrial pulp and paper companies, with their own forest management and wildfire prevention/fighting resources, while the remaining area is managed by nonindustrial private owners, characterized by different objectives and economic logics.In addition, there is a significant heterogeneity in the spatial distribution of all these characteristics/factors (Baptista and Santos, 2005).Small forest holdings (<10 ha), mainly composed by pine and eucalyptus with low profitability, are much more frequent in the northern and central Portugal, while large properties (>100 ha), essentially of cork oak or a complex and unique agroforestry system of cork oak savanna ("montado"), are predominant in the southern regions of the country.Table 4 of the annex/supplementary material, provides a general description of the main characteristics of forest holdings and forest owners and summarizes the interrelationship between these factors.
Other aspects related to LULCC, such as climate change and biodiversity, are somewhat outside the scope of the present study.
However, the abandonment of rural and forest areas, traditional agricultural practices, and the lack of forestry management practices lead to an increase of biomass and fire risk, which can be empowered in a warmer and drier future climate and may have profound impact on ecosystems and biodiversity.For example, the montado, which is composed by sparse cork oak trees and a diversity of understory vegetation (e.g., shrub formations, grasslands), supports higher levels of biodiversity.The decrease in demand and price of cork has led to a reduction in management practices and to the abandonment of these lends, leading to the invasion of shrubs, reducing the biodiversity and degrading the services provided by these ecosystems (Bugalho et al., 2011).
Results from our analyses shows that in the second half of the investigated period (2000-2006 and 2006-2012) the growth rate of RUI was lower than in the first decade (1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000), probably due to a lower growth rate in the process of urbanization of rural areas.Moreover, the decrease of relative BA within the RUI from 2006 to 2012 could be associated with the relative decrease of BA in the last investigated period, as a consequence of recent plans for territorial spatial planning and protection of forest against forest fires (Mateus and Fernandes, 2014;Parente et al., 2016).At European level, urban sprawl's complexity and magnitude motivated the European Commission to recommend actions and to coordinate land use policies, within the European Cohesion Policy 2007-2013 period (CEC, 2006;EEA, 2006).In Portugal these efforts were complemented with national programs and regional plans such as the National Policy and Territorial Management (Programa Nacional da Política de Ordenamento do Território, PNPOT) and the Regional Plan for Territorial Planning (Plano Regional de Ordenamento de Território, PROT), supporting the sustainable development and the environmental landscape quality of NUTS-III areas (Noronha Vaz et al., 2012).
However, as far as we know, there is no a specific or general Portuguese legislation about WUI or RUI.It only exists one general mention about RUI in the National Plan to Protect the Forests against Wildfires (CM, 2009).In this Plan it is suggested that to protect urban-forest interface is necessary to create and maintain an external buffer strips (10-100 m) around population clusters, especially in those with the highest fire vulnerability, as well as around parks, industrial polygons, landfills, housing, shipyards, warehouses, and other buildings.
Finally, we firmly believe that the results of this study are sufficiently motivating to promote the development of specific policies and legislation, as well as changes in forest and fire management.The increase of the RUI area and particularly of the BA within the RUI clearly suggests the need of improving fire prevention measures and preparedness policies for this interface region.In fact, as indicated by the Portuguese National Fire Plan 2006 (Oliveira, 2005), the increase of the urban/rural interface, as a consequence of the above-mentioned LULCC, causes this area to be under the use of people not educated for fire and unware of possible source of ignition.In particular, portuguese forestry/forest managers must prioritize sustainable forest management practices and make brush clearing obligatory.These paradigm shifts make even more sense if one takes into account that the risk of fire is likely to increase in a future climate with a higher frequency of longer and more intense extreme events, such as drought and heat waves.

Conclusions
Continental Portugal registered important land cover changes (about 9% of the whole area) and an increase of the rural-urban interface in the investigated period .Most significant changes were associated to transitions from the following CORINE Land Cover classes: Agricultural areas (35.1%) and Forest and semi-natural areas (15.2%) to Artificial surfaces (including Urban areas); Agricultural areas to Forest and semi-natural areas (7.3%) and vice-versa (6.3%).However, relative net changes are appreciable only for Artificial surfaces, which registered a substantial increase of about 50%, while Forest and seminatural areas remains almost constant (0.3%) and Agricultural areas slightly decreased (-4.4%).The spatial distribution of these changes was far from uniform within the territory.Urban sprawl was concentrated in the metropolitan areas of Lisbon and Porto, as well as in central-north and south coastal areas (region of Algarve).A promoted socioeconomic development within the country, the intense rural abandonment and the development of mass tourist industry could act as main drivers for the expansion and reconfiguration of urban areas.On the other hand, in the south and interior north regions we assisted to a transition to vegetated land use/land cover types, probably caused by deforestation/afforestation and rural abandonment.The CLC classes mainly affected by these changes where Scrub and/or herbaceous vegetation associations, Forests and Heterogeneous agricultural areas: the increase in Artificial surfaces was precisely due to transitions from these type of land cover.Vegetated classes with higher burnt area within the RUI in the study period were: Transitional woodland-shrub, the three types of Forests considered in the CLC inventories and, the three sub-levels of Heterogeneous agricultural areas.These findings suggest the needs of extending the notion of wildland-urban interface for Portugal to rural-urban interface, defined as Forest semi-natural plus Heterogeneous agricultural areas adjacent to Artificial surfaces.
Results of the analyses of RUI's size, burnt areas and burnt areas within the RUI in the fours investigated periods (1990,2000,2006,2012) allow to conclude that, from 1990 to 2012, RUI increased about 70%, burnt area decreased 35% but, nonetheless, burnt area within the RUI increased 100%.These findings underline the need of frequent monitoring and assessment of land use changes and RUI evolution in Portugal, and reinforce the need to focus the attention of forest and fire managers on this highly fire prone region.
The conclusions of this study suggest and encourage more accurate analyses for characterizing and mapping RUI, using high resolution and precise data (e.g.true houses footprints, road network, census data) to provide practical indications in term of land and fire management.Nevertheless, our study provides precious suggestions as for what is the global distribution and evolution of RUI in Portugal, identifying which regions need to be prioritized in term of RUI monitoring.

Figure 1 -
Figure 1 -Population density at parish level (INE, 2012) in the mainland Portugal, with the location of the of the major cities.NUTS refers to regions according to Nomenclature of Territorial Units for Statistics level II (Eurostat, 2017; Santos, 2014).

Figure 2 -
Figure 2 -Land cover of mainland Portugal based on the second level of CORINE land cover inventory 2012 (EEA, 2016) Use and Occupancy Chart (Carta de Uso e Ocupação do Solo, COS) provided by the Portuguese Directorate-General of the Territory (Direção-Geral do Território, DGT).DGT is the national public body responsible for pursuing public policies for land use and town planning.We compared CLC2006 with COS2007v2.0because these are the closest inventories (in time) within the study period.In addition, COS2007v2.0presents improvement from the thematic and geometric point of view(DGT, 2016;Sarmento et al., 2016): it includes 225 classes (32 more than the initial version) at the most detailed level, distributed over 5 hierarchical levels.
Figure 3 -Framework implemented in the Model Builder (ArcGIS TM ) to map the rural-urban interface.CLC=CORINE land cover; AS = artificial surfaces; FSNA = forest and semi natural areas; HAA = Heterogeneous agricultural areas.

Figure 4 -Figure 5 -
Figure 4 -Map of land cover/land use transition from 1990 and 2012, evaluated considering the first level hierarchy of CLC 1990 and CLC 2012

Figure 6 -
Figure 6 -Area lost and gained from 1990 to 2012 for each CORINE land cover classes, considering the second level hierarchy.Net percentage changes were computed relatively to the total area of each class in the later land cover.The bar graph of the contributions to net changes in the AS sub-levels (Fig.7) shows that Urban fabric (orange bars), which includes buildings, roads and artificially surfaced areas, grew at the expense almost exclusive of HAA (code 24).On the other 5

Figure 7 -
Figure 7 -Contribution to the net changes from 1990 to 2012 of "Urban fabric" (orange bars) and "Industrial, commercial and transport" (blue bars) from the other CLC sub-levels

Figure 8 -
Figure 8 -Classes of land use, as defined by the third level hierarchy of the CORINE Land Cover, affected by wildfires expressed in percentage as the ratio of BA affecting each CLC classes over the total BA for each frame period.

Figure 9 -
Figure 9 -Maps of the rural-urban interface (RUI) in Portugal estimated for the different periods of investigation, and corresponding to the available CORINE Land Cover inventories (1990, 2000, 2006, 2012) (a) the relative contribution of the those four CLC classes to the RUI increases in time at approximately the same rate; (b) HAA (code 24) is the CLC class with the largest area within the RUI (~50%); (c) in terms of relative BA within the RUI (BAR), the most affected class is the Scrub and/or herbaceous vegetation associations (code 32), followed by Forest (code 31), HAA (code 24) and then Open spaces with little or no vegetation (code 33).The total extent of the BA (TBA) fluctuated, with a maximum in 1990 (about 500×10 3 ha) followed by 2006 (~ 460×10 3 ha), while in in 2000 and 2012 its value was lower and equal to about 310×10 3 ha.The portion of BA included within the RUI (RBA), expressed as percentage over the total BA, tends to increase in time, passing from4% and 7% in 1990 and 2000 to 15% and 14% in 2006 and 2012, respectively.

French
Maritime Alps in the period1960 -2009. Badia et al. (2011) ) noticed that two representative Mediterranean WUI areas in Catalonia were more prone to wildfires in the most recent decade of 2000 than in the 1990s.Pellizzaro et al. (2012)  analyzed WUI's dynamics and landscape changes in a tourist area of North-East Sardinia (Italy) from 1954 to 2008 and discovered that LULCC was largely associated to a transition from an agro-pastoral economy to one based on tourism.Moreover they showed: especially in southern Portugal; (ii) COS class of Temporary irrigated crops is well distributed by CLC classes of Rice fields, Permanently irrigated land, Pastures, as well as Beaches, dunes, sands; (iii) COS Shrub classes are particularly present in CLC classes of Shrubs, Pastures as well as in Open spaces with little or no vegetation; (iv) Tree vegetation in COS is also well related to the correspondent CLC classes and allows us to better understand the composition of Forests and Agricultural areas; for example, pure or mixt forests of Cork and Holm oak trees are particularly evident in the CLC classes of Agro-forestry areas, Broad-leaved forests, and Non-irrigated arable lands; on the other hand, Eucalyptus (pure or mixed forests) in COS are important components of the CLC classes of Mixed forests, Complex cultivation patterns, Annual crops associated with permanent crops, and Broad-leaved forests; finally, pure pinus pinaster forests in COS are comprised in the CLC classes of Coniferous forests, Transitional woodland-shrub, and are especially important in Beaches, dunes, sands, where account for 30% of total area; this finding is in good agreement with the presence of pinus forests in the entire central western coast.

Table 2 -Classes of land use, as defined by the third level hierarchy of the CORINE Land Cover (CLC), affected by forest fires expressed in terms of Burned Area (BA) within the Rural Urban Area (RUI) during three investigated frame periods.
The peak of BA in Transitional woodland-shrub (code 324) equals to about 43 ×10 3 ha in the period 2005-2007, compared with about 15×10 3 ha in 2011-2013, 14×10 3 ha in 1999-2001 and 6×10 3 ha in 1989-1991.It also emerges that the three over four sub-levels of HAA (code 243, 241, 242) are highly affected by wildfires, thus confirming the need of including HAA in the RUI's definition.