Assessment of the cover changes and the soil loss potential in European forestland: First approach to derive indicators to capture the ecological impacts on soil-related forest ecosystems

The Member States of the European Union have committed to the maintenance and protection of forest lands. More precisely, the Member States aim to ensure the sustainable development and management of the EU’s forests. For 2013, Eurostat’s statistics about primary and secondary wood products in the European forest land (65% thereof privately owned) estimate a roundwood production of 435 million m3 in total. Harmonised information, i.e., spatially and temporarily differentiated, on forestry and wood harvesting activities in the European forests are missing however. This lack of information impedes the scientific assessment of the impacts that forest management practices have on the soil-related forest ecosystems (e.g., accelerated water soil erosion, delivery of inert sediments and pollutants within the drainage network, pauperization of aquatic ecosystems). It also prevents national and European institutions from taking measures aimed at an effective mitigation of the rapidly advancing land degradation. This study provides a first pan-European analysis that delineates the spatial patterns of forest cover changes in 36 countries. The first dynamic assessment of the soil loss potential in the EU-28 forests is reported. The recently published High-resolution Global Forest Cover Loss map (2000–2012) was reprocessed and validated. Results show that the map is a powerful tool to spatiotemporally indicate the forest sectors that are exposed to cover change risks. The accuracy assessment performed by using a confusion matrix based on 2300 reference forest disturbances distributed across Europe shows values of 55.1% (producer accuracy) for the algorithm-derived forest cover change areas with a Kappa Index of Agreement (KIA) of 0.672. New insights into the distribution of the forest disturbance in Europe and the resulting soil loss potential were obtained. The presented maps provide spatially explicit indicators to assess the human-induced impacts of land cover changes and soil losses on the European soil-related forest ecosystems. These insights are relevant (i) to support policy making and land management decisions to ensure a sustainable forest management strategy and (ii) to provide a solid basis for further spatiotemporal investigations of the forestry practices’ impacts on the European forest ecosystems. ublis © 2015 The Authors. P


Introduction
The modern demographic growth and the socio-economic expansion come along with an increasing worldwide demand for forest resources (Foley et al., 2005;Eggers et al., 2008). The world's forestlands have been cleared, degraded and fragmented by timber harvesting, human-made fires and land-use conversion (Hansen et al., 2013). It is estimated that about 13 million ha of forestlands are converted to other land-uses every year (FAO, 2010). The way of forest management practices have a high impact on soil-related forest ecosystems (Lal, 1996) especially with regard to its biodiversity (Torras and Saura, 2008), water quality (Stott et al., 2001) and the related ecosystem services (Chazdon, 2008).
For the European Union, forests are an important ecosystem in terms of recreation, biodiversity, timber and carbon storage (Edwards et al., 2011;Martín-Martín et al., 2013;Hansen et al., 2013). They cover about 177 million ha (42.3% of the total land area) of the EU27 territory and provide living space for ca. 4 million people (forestry and forest-based industries) (FOREST EUROPE UNECE and FAO, 2011;Eurostat, 2011).
The Member States of the European Union have committed to the maintenance and protection of their forests (Forestry Strategy, 1998;EU Forest Action Plan, 2006). Their aim is to develop and manage their forestlands sustainably. The findings of the 5th Ministerial Conference on the Protection of Forests in Europe (MCPFE, 2007) attest a satisfactory condition and sustainable management to the European forests in general with human-induced forest damages being less than 1% of the total forestland (MCPFE, 2007). Nevertheless, the MCPFE (2007 report also indicates that harvesting and forest operation damages cause severe economic losses and deteriorate the ecosystems' health and vitality in specific areas (e.g., decrease in timber quality, rot, decay, destruction of natural regeneration). The general view appears to be misleading. Eurostat (2011) reports that wood is still the main source of income for the European forest owners and that about 65% of the forestlands in the EU are privately owned. The EU-28 is the secondlargest producer of industrial round timber after the United States and it produces approximately 80% of the world's cork (Eurostat, 2011). There has been a steady rise of roundwood production in the European Union 27 between 1995 and 2007, both for coniferous (softwood) and non-coniferous (broadleaved or hardwood) species (Eurostat, 2011). The recent financial and economic crisis led to a decreasing level of roundwood production during 2008 and 2009. Regaining strength in 2010, Europe's roundwood production returned to its pre-crisis growth trend with a 9.5% year on year growth rate and a total production of 420 million m 3 . In 2013 the roundwood production totalled 435 million m 3 (3.5% growth rate, Eurostat, 2014a,b). The Member States' Land-Use, Land-Use Change and Forestry activities (UNFCCC, 2014) project an increase of harvest rates by around 30% by 2020 as compared to 2010.
The challenge for the immediate future is to assure that the resources of the forests can be used for the humans' demands while minimizing the damages caused by the forest operations. Such ambitious undertaking, however, requires in-depth knowledge about the status of the human interference on the European forests. Despite the ongoing intensive exploitation of the European forest resources and documented impacts on the European soils and related functions (Cerdà et al., 2010;Borrelli et al., 2013a), today researchers still lack a well-grounded knowledge about the impacts that the forest management activities have on the soil functions within European forests (e.g., accelerated water soil erosion, delivery of inert sediments and pollutants within the drainage network, reduction of the rivers' retention capacity, pauperization of aquatic ecosystems, increased withdrawal of nutrients). Among others, the primary limitation is the lack of freely accessible cartography of the European forests that are object to wood supply. Not only does this conflict with the framework of the Forest Europe objectives (MCPFE, 2009) but it is also in strong contrast with the European Commission's thematic strategy of Soil Protection (European Commission, 2006) and the EU Water Framework Directive (European Commission, 2000). Hansen et al.'s (2013) recent study is a first step to fill the gap of knowledge in order to comprehensively monitor and analyze soil erosion processes in the European forests.
Building on their work, this study provides a pan-European analysis (36 European counties) of forest cover change across both space and time (i.e., intended as decline of the wooded cover). It lays the ground for the first dynamic assessment of the soil loss potential in the EU-28 forests. Forest cover change (or forest loss) is defined as a stand-replacement disturbance or the complete removal of the tree cover canopy at the Landsat pixel scale (Hansen et al., 2013). The High-resolution Global Forest Cover Loss map (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012) of Hansen et al. (2013) was reprocessed and validated to assess the forest cover change. The 1 arc-second per pixel data (approximately 30 m per pixel) within the three main forest units of the CORINE land cover map 2006 were used. After correcting the forest fire incidents (JRC European Fire Database -EFFIS, 2014;San-Miguel-Ayanz et al., 2013), the remaining forest cover changes were assumed to be primarily related to tree logging activities. This assumption seems to be justified as the permanent land-use conversion actions in European forestlands are at a rather low level limited and other sources of forest cover change such as windthrow, insect and pathogen outbreaks appear spatially considerably less compared to logging (Eurostat, 2011).

Study area
The geographical extent of this study included the 28 Member States of the European Union (EU-28), three European Union candidate countries (i.e., Montenegro, Serbia, former Yugoslav Republic of Macedonia), three potential European Union candidate countries (i.e., Albania, Bosnia and Herzegovina, Kosovo), Norway and Switzerland (Fig. 1). The forestland sector under analysis consisted of the three main forest units of the CORINE land cover 2006 (CORINE 2000 for Greece) with a total area of 150.2 million ha: (1) broad-leaved forests (50.7 million ha, 33.8%), (2) coniferous forests (69 million ha, 45.9%) and (3) mixed forests (30.5 million ha, 20.3%).

Input data
To fully cover the study area, 24 individual 10 × 10 degree granules of the High-resolution Global Forest Change map (Hansen et al., 2013) were downloaded from the online database of the University of Maryland (http://www.earthenginepartners.appspot.com).
The High-resolution Global Forest Change map resulted from a time series analysis of 654,178 Landsat images (period between 2000 and 2012). To cover the European region, Hansen et al. (2013) employed a dataset of estimate 12,000-15,000 images. Trees were defined as all vegetation taller than 5 m in height. The annual forest loss was defined as a stand-replacement disturbance or a change from a forest to non-forest state. The images captured during the growing season were preferred to the ones acquired during the senescence or dormant seasonal periods. The Google Earth Engine's computing facility was used to perform the global Landsat analysis. It is a cloud platform for earth observation data analysis that combines a public data catalogue with a large-scale computational facility optimized for the parallel processing of geospatial data (https://earthengine.google.org/#intro). As a final outcome, Hansen et al. (2013) provide global maps of forest cover change in a 10 × 10 degree raster pixel (cell size ca. 30 m × 30 m).

Data processing
After the acquisition, the 24 raster of the High-resolution Global Forest Cover Loss map were processed using the ArcGIS 10.2 model builder (i) to convert the data from a raster to a shapefile format, (ii) select only the forest cover changes grouped as European Land Cover classes 311, 312 and 313, (iii) reproject the European forest cover change shapefiles into metric coordinates (ETRS89-LAEA), (iv) subdivide the dataset into eleven annual shapefiles (2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012) and (v) remove the major forest fires reported in the JRC European Fire Database (European Forest Fire Information System -EFFIS, 2014).

Forest change density map
A forest change density map was created as indicator to spatially describe the forest cover change dynamics and highlight the forest sectors that were primarily involved. The map is based on the Kernel Density algorithm (Silverman, 1986) included in the Spatial Analyst toolbox of ArcGIS 10.2 (ESRI, 2006). To enable this, the feature of the shapefile was converted from polygon to point (creating a centroid point each 30 m × 30 m of forest cover change).
After the data processing, the remaining forest cover changes were considered to be primarily related to the tree logging activities. Accordingly, the forest change density map highlights the forest sectors mostly involved in the wood supply.

Accuracy assessment of forest change
The accuracy of the forest cover changes detected by Hansen et al. (2013) rested on confusion matrix by a per-pixel analysis (Aronoff, 1982) (geometric accuracy) and a linear correlation analysis manipulating the shapefile in a GIS environment (thematic accuracy). A set of forest cover changes (clear-cut areas in both coppices and high forests) were mapped via onscreen visual interpretation on Google Earth images. These changes were used as 'reference data' while the forest change data resulting from the Hansen et al. (2013) Landsat image classification represented the 'classified data'.
For the mapping of the 'reference data', the forest sectors rated as high forest change according to the forest change density map were selected. Within a first step, the forest change density map was separated into four classes using the quartile classification method. The data of the highest quartile (75-100%) formed the reference group. Afterwards, at least 50 random points were generated within the subset forest area and one or more clear-cut areas were mapped (generally the nearest to the generated random point) for each EU-28 Member States. Finally, 2300 clear-cut areas were mapped and employed as 'reference data'.

Soil erosion potential
Forest harvesting and fires are often held responsible for the very high levels of nonpoint source pollution (Stott et al., 2001;Hood et al., 2002). This study used the universal soil loss equation (USLE), as modified for forest land by Dissmeyer and Foster (1984), to model the average soil erosion potential due to logging activities (assuming all the forest cover changes described by Hansen et al. (2013) as tree harvesting) and forest fires (employing the JRC forest fire data) in the EU-28 the medium term (11-year). USLE uses a number of factors to estimate soil loss: where, A = soil loss (Mg ha −1 yr −1 ), R = rainfall and runoff factor (MJ mm h −1 ha −1 yr −1 ), K = soil erodibility factor (Mg h MJ −1 mm −1 ), LS = slope length and the slope steepness factor (dimensionless), and CP = cover management practice factor (dimensionless). The Kand R-factors were expressed spatially using the latest maps created by the soil research team of the Joint Research Centre of the European Commission (Panagos et al., , 2015. The originally proposed topographic LS-factor scheme of Wischmeier and Smith (1978) was replaced by the one of Desmet and Govers (1996) in order to incorporate the impact of flow convergence in the estimation of the slope-length factor. In the original equation, the cover and management factor (C) was developed for agricultural croplands. It has to be adjusted to the forestland characteristics to be able to spatially differentiate the various forest cover species, canopy cover densities and floor vegetations (Dissmeyer and Foster, 1984). Considering the size of the study area, the C-factor to predict the soil loss potential for the European forestlands was estimated in a slightly simplified way. The influence of the vegetation density was quantified by manipulating a biophysical parameter (FSoil) as derived by Poilve, 2010. This enabled to estimate the fraction of soil that is visible in the vertical which, in turn, allowed to assess whether the vegetation corresponded to bare soil patches or holes in developed canopies (gap fraction), sunlit or shaded from the canopy (values ranging from 0 to 1). Thus, the cover and management factor for the undisturbed forest (C UF ) is calculated as: where the Min C and Max C values were set according to Wischmeier and Smith (1978) as 0.0001 (100% canopy cover) and 0.009 (20% canopy cover). With regard to the disturbed forests, several authors stated that the restoration of the vegetation on bare ground which had been affected by harvest operations or fires should be expressed as a time sequence . Kitahara et al. (2000) suggested that the C-factor should be expressed either as a function of time or category for each year after the disturbance. Hence, to adapt the C-factor to the characteristics of disturbed deciduous coppice forests, a modified version of the C-factor that verified through field work activities in Italy was used (Garfì et al., 2006;Borrelli et al., 2013a) (Table 1). This method takes the functional efficiency of soil protection into account: Under normal circumstances, a harvested deciduous coppice forest regains the function of coverage of a moderately dense forest (i.e., a C of 0.009) four years after the clear-cut. Ten years after the harvesting, the Cfactor returns to the mature forest value (i.e., C of 0.003). Thereafter, it converges towards its pre-disturbance C UF value. Although coniferous forests may experience slightly different growth dynamics, similar dynamics were assumed taking into account the canalization dynamics of the field-layer vegetation (Bergstedt and Milberg, 2001). This assumption is in line with experimental measurements, which show that the soil loss in harvested coniferous forests dramatically decreases starting from the third year following the cut (Kitahara et al., 2000;Hood et al., 2002). Between the fifth to the tenth year after the wood harvesting, the C-factor decreases from 0.009 (corresponding to forest cover of ca. 20% or rangelands 60% covered by grass; Wischmeier and Smith, 1978) to 0.003 (corresponding to forest cover of ca. 40%, rangelands 95% covered by grass; Wischmeier and Smith, 1978). With respect to forest fires, the percentage of soil exposed to erosion and the vegetation recovery after the fire are closely related to the severity of the fire (Larsen and MacDonald, 2007). In literature, the post fire forest C-factors applied ranges from ca. 0.02 (low severity) to 0.3 (high severity), with a mean value of about 0.2 (Murai, 1972 cited in Kitahara et al., 2000;Wischmeier and Smith, 1978;Larsen and MacDonald, 2007). While the Joint Research Center of the European Commission's forest fire data (EFFIS, 2014) do not provide information about the severity of the burned area, literature based on the employed remote sensing data (Moderate Resolution Imaging Spectroradiometer -MODIS) suggests prevalent detection of moderate to severe fire incidents (Roy et al., 2006). Hence, for the first year after the fire, a C value of 0.2 was assumed (Table 1). For the following years, the C-factor values decrease annually towards a value of 0.009 in the fourth year after the event (Robichaud and Brown, 1999;Larsen and MacDonald, 2007). From the fifth year onwards, the same dynamics described for the harvested forest were assumed. The sub-factor P was assumed to be constant (equal to 1, i.e., absence of erosion conservation practices).

Forest change dynamics
The total EU forest area subject to analysis is estimated to be 1652.2 million ha (150.2 million ha for 11 years between 2002 and 2012). Temperate broadleaf and mixed forests (WWF, 2014) have the largest share of the biome. Sweden has the most extensive forest cover, followed by Spain, Finland, France and Germany. These five countries account for more than half of the total forest cover analyzed ( Table 2). The Netherlands, Ireland and the United Kingdom show the lowest relative forest cover (10.8%, 11.5% and 11.9%, respectively).
The forest cover change during the period from 2002 to 2012 is described spatially in Fig. 2a. During this period, the estimated absolute area affected by forest cover change totalled 7,022,423.1 ha, which equals 4.7% of the Corine primary forest area. The annual rate of forest cover changes ranged from 0.17% (2003) to 0.63% (2010) (x 0.52 ha; 1.67 ha). The greatest absolute woodland changes occurred during 2010, involving a forest area of 953,541 ha (Table 3). The Scandinavian countries show the largest absolute change of forestlands: With 1.82 million ha (Sweden) and 1.4 million ha (Finland), these countries account for 45.9% of the total forest cover change mapped by Hansen et al. (2013). For France, Poland and Germany the mapping indicated 0.52, 0.50 and 0.3 million ha of forest change, respectively. The highest relative forest cover changes were observed in Latvia (8.2%), Estonia (6.7%), Lithuania (6.2%), Finland (6%) and Sweden (5.8%), while the lowest rates occurred in Cyprus and the region of former Yugoslavia.
Regarding the type of forest (Corine classification), coniferous forests were the predominantly affected forest type (4.7 million ha; 67.6%), followed by mixed forests (1.3 million ha; 18.9%) and broad-leaved forests (0.9 million ha; 13.5%). Considering the bio-geographical regions, the results reveal that the Boreal Bio-geographical region showed the highest forest cover change (Table 4) with a cover change rate of 3.8% of the total biome area (absolute estimated forest cover change of 3.4 million ha). Fig. 2b shows the area involved in fires during the period from 2002 to 2012. Accordingly, there are five countries that have predominately been affected by forest fires (Portugal, Spain, Albania, Italy, Bosnia-Herzegovina) ( Table 2). In about one decade, fire incidents in these five countries burned a total area of 0.5 million ha. This equals 80.6% of the total detected forest fires in the study area. On average, the area annually burned is equal to 52,319.7 ha. The greatest fires in the European woodland occurred during 2012, affecting a forest area of 143,803.4 ha.

Accuracy assessment
Both thematic and geometric accuracy assessments were performed, comparing the forest cover change areas observed by the Landsat imagery (Hansen et al., 2013) with the clear-cut areas identified by means of an onscreen visual interpretation of aerial orthophotos.
The thematic accuracy analysis shows that 81.4% (n 1873; x 3.2 ha; 6.8 ha) of the visually identified clear-cut areas were also highlighted in Hansen et al.'s (2013) algorithm-derived forest cover change database. Two hundred fifty-three clear-cut areas were only partially detected by Hansen's research group (the forest cover change surface detected by Hansen et al. (2013) was less than 30% of the one visually mapped). By contrast, 427 clear-cut areas (x 2.6 ha; 7.5 ha) detected by visual interpretation were not detected at all by the algorithmic study.
The geometric accuracy was carried out using a per-pixel analysis (confusion matrix, Aronoff, 1982) also considering the 2300 clear-cut areas that were detected via an onscreen visual interpretation of aerial orthophotos. The producer accuracy of the algorithm-derived forest cover change areas was 55.1%, with a Kappa Index of Agreement (KIA) of 0.672. The overall classification accuracy totalled 94.3%.
The average annual soil loss in forests that remained undisturbed during the modelled period is equal to 0.086 Mg ha −1 yr −1 ( 0.172 Mg ha −1 yr −1 ). The areas of forest cover change mapped by Hansen et al. (2013), here assumed to be due to tree harvesting, accounts for 15.6% of the predicted soil loss (quantitatively equal to 2.92 × 10 6 Mg yr −1 and 0.45 Mg ha −1 yr −1 , with a of 1.05 Mg ha −1 yr −1 ). The soil loss potential predicted for the forest fire areas mapped by the Joint Research Centre of the European Commission (EFFIS, 2014) shows an average area-specific soil loss of 2.06 Mg ha −1 yr −1 (996,167.6 Mg yr −1 ), with a standard deviation of 2.2 Mg ha −1 yr −1 . Accordingly, about 73.4% of the total long-term soil loss was predicted to occur in the undisturbed forest. Notably, 26.6% of the soil loss was predicted to occur in the disturbed forest areas although these areas covered only ca. 7.1% of the EU-28 forestland area.
The average soil loss for the first 4 years after the vegetation disturbance shows rates of 2.94 Mg ha −1 yr −1 for the clear-cut areas and 13.43 Mg ha −1 yr −1 for the areas disturbed by forest fires. The soil loss in the disturbed forest accounts for the vast majority of the soil mobilized during the first four years after the vegetation disturbance. It also forms a high share of the longterm soil mobilization (ca. 20.1% of the total soil loss modelled for a 30-year period). An additional run of the USLE model for the disturbed forestlands under the assumption of the absence of forest harvesting and fires resulted in an average soil erosion of 0.038 Mg ha −1 yr −1 (246,511.6 Mg ha −1 yr −1 ). Compared to the forest-harvesting and forest fire conditions, the non-disturbedforest scenario generated only one twelfth (less than 8.5%) of the average erosion rate of the disturbed areas. In the comparison, the average erosion rate was 77 times lower focussing on the first four years of erosion after the vegetation disturbance. In this simulation, very severe erosion rates are only observable in some impluvi with a slope gradient greater than 50%. In reality forestland soil loss already starts to be severe at slope gradients around 15% in the forest-harvested areas of the EU-28.
The results of a cross-country comparison of the annual average soil loss values are reported in Table 5. Considering the predicted gross erosion values in undisturbed forests, the annual rates are driven by the density of the canopy cover, the rainfall erosivity, the soil erodibility and the topography. The forestlands that are naturally more exposed to the soil erosion processes are located in Slovenia, Italy and to a lesser extent also in the high mountain areas of Switzerland, Austria, Cyprus and Spain. Accelerated soil erosion rates caused by forest disturbance occur in Slovenia, Italy and Austria with values that are about three times above the European average (0.96 Mg ha −1 yr −1 ). The average soil erosion rates in a 48 months period following the wood harvesting event in these three countries were 26.1, 19.7 and 18.8 Mg ha −1 yr −1 , respectively. At NUTS-2 level, 10 of the 20 administrative regions with the highest soil loss were found in Italy. This is because a great share of the country's area is located in mountainous areas characterized by heavy bursts of intensive and erosive rainfalls that hit the steep slopes and these locations are subject to extensive wood extraction activities Panagos et al., 2014Panagos et al., , 2015.
With regard to forest fires, the highest average soil erosion rates were found in Italy, Slovenia, Croatia and France and to a lesser extent also in Spain, Greece and Romania. Portugal which experienced about 40% of the total fire events in Europe, shows an average soil erosion rate of 1.37 Mg ha −1 yr −1 (lower that the European average -1.45 Mg ha −1 yr −1 ). The soil erosion rates for the first 48 months after the fires were 27.5, 23.9, 18.8 and 17.2 Mg ha −1 yr −1 for Italy, Slovenia, Croatia and France, respectively.

Forest disturbance and soil loss potential in the Natura 2000 network
Notably, this study also detected forest cover changes in areas declared as Sites of Community Importance (SCI) and Special Protection Areas (SPAs) in the Natura 2000 network by the EU (Bastian, 2013). At a European-scale, about 931,880.7 ha of wooded areas suffered from some form of forest cover change (equal to 2.5% of the forested Natura 2000 area) while 209,172.6 ha were affected by forest fires (equal to 0.56% of the forested Natura 2000 area) ( Table 6). In across-country comparison among the EU-27 countries, Ireland (8.1%), Denmark (7.4%), Portugal (6.8) and Lithuania (6.1%) had the highest forest change rates (European average of 2.4%). With regard to fires, about 209,172.6 ha of the wooded areas suffered from forest fires (equal to 0.56% of the protected forest area of the EU-27) ( Table 6). The countries with the highest forest change rates were Spain (51.3%) and to a lesser extent also Portugal (26.3%) and Italy (12.3%).
The annual average gross soil loss predicted in the Natura 2000 areas was 4.15 × 10 6 Mg yr −1 (0.15 Mg ha −1 yr −1 ). This corresponds to an accelerated soil erosion rate in the 3.1% disturbed forest of 21% soil loss per year. The countries with the highest increase were Portugal (+244%) and Ireland (+116%).

Discussions
The European forests must satisfy a wide array of human demands (Nabuurs et al., 2007) which are projected to increase in the near future (UN-ECE, 2005) driven by market forces and supported by the targets of national and European energy policies (EEA, 2007). Changes in the European forest cover related to logging activities, fires and windthrow affect the delivery of vital ecosystem services such as water supplies (Ojea et al., 2012), soil-related functions (Wall et al., 2013), carbon storage (Van Oost et al., 2005), regulating floods (Robinson et al., 2003) and biodiversity richness (Gamfeldt et al., 2013). Governments and European institutions have been working on the development of reliable and current information to develop future forest management plans, policies and strategies (UN-ECE, 2005). Eurostat provides valuable statistics about the primary and secondary wood products in the European area. However, harmonised information, spatially and temporarily explicit, on the forest cover changes in the European forests are not available. Today, a need for reliable data to improve the knowledge about the conditions of European areas that are involved in wood supply is greater than ever especially considering an annual roundwood production that hit the 435.0 million m 3 yr −1 threshold in 2013 (EU-28) (Eurostat, 2014a,b). Given these developments, the resulting environmental threats and their external costs (Pimentel et al., 1995;Chiabai et al., 2011) must be assessed and effective management strategies must be designed (Lynch et al., 1985) in order to control the land degradation processes (Cerdà et al., 2010). The present study describes the methodology employed to reprocess the Global Forest Change map (Hansen et al., 2013) to derived harmonised data about the forest cover change in Europe. The analysis provides a detailed picture of the temporal and spatial patterns of the forest cover changes in Europe. It (i) provides a spatially explicit map of forest cover changes, (ii) highlights the forest sectors with a high probability to be involved in the wood supply activities in Europe (Fig. 4) and (iii) develops the first comprehensive and dynamic assessment of the soil loss potential for the European forests.
The annual forest cover changes in 36 European countries for the period from 2002 to 2012 were analyzed and subject to a validation procedure. A total of 7.022 million ha of disturbed forest were mapped for the study period. An increase in the annual forest cover loss was observed in Europe (Fig. 5). This is in accordance with the statistics reported by Eurostat (2014a,b) about the roundwood production in Europe. Moreover, these findings emphasize the consistency of the Hansen et al. (2013) outcomes, which shows a great local relevance and a high utility. The omission error of 44.9% can be partially attributed to the thorough validation procedure applied. It represented a sort of 'stress test' that was performed including among the data of the 'reference dataset' clear-cut areas that due to their small size are difficult to accurately detect clear-cut areas using Landsat images (Borrelli et al., 2013b). In fact, the accuracy assessment of forest loss between 2000 and 2012 carried out at global scale by Hansen et al. (2013) shows better results (n = 1500; producer accuracy = 87.8%; overall accuracy = 99.6%). While the inclusion of the small size clear-cut areas into the reference dataset considerably lowered the performance of the accuracy tests, it also ensured a more realistic and representative picture of the capability of the Global Forest Change map to represent the disturbances occurring in the European forestlands.
The statistical significance of the detected forest cover changes are also confirmed by the regression results comparing the 2002-2012 forest cover change values to the 2002-2012 roundwood production reported by Eurostat (2014a,b) (r 2 = 0.87, ˛ < 0.01) (Fig. 6).
The goal of large-area land cover change mapping is to identify characteristics and factors that have a causal relationship to forestland changes on a local scale and bundle and generalize these information in order to derive reliable and relevant information across scales. According to Hansen et al. (2013), the Global Forest Change map meets these requirements at least at a national scale. The present study broadened the scale and provides evidence for the applicability of this technique on a European scale (Fig. 7).
Although  recent study shows that better results can be detected for forest cover change at a national scale using Landsat imagery, Hansen et al.'s Global Forest Change map (2013) is a free and ready-to-use product that represents a powerful tool for forest management. To ensure sustainable forest management practices has been a central topic of the forest management in Europe (MCPFE, 1993(MCPFE, , 2007(MCPFE, , 2011. Based on this common policy framework, national and international forest management decision-makers have been working for more than 20 years to promote, improve and implement sustainable a forest management in Europe. At a pan-European scale, however, the impacts of forestry activities have so far been studied mainly based on statistical information provided to Eurostat by the member states. In the absence of spatially and temporarily explicit indications about the forest sectors that undergo logging, fires and windthrow, such information have little explanatory power and the environmental impacts behind these phenomena remain in the dark. For instance, 1000 m 3 of roundwood collected through short rotation forestry activities in a flat Scandinavian area could trigger completely different qualitative and quantitative effects on the soil-related forest ecosystems, compared to the extraction of the same amount of roundwood in semi-natural mountain forests of the Italian Apennine region (Sorriso-Valvo et al., 1995;Porto et al., 2014). This because Italy is repeatedly subject to heavy bursts of intensive and erosive rainfalls (five times stronger than in Scandinavia) falling on steep slopes (USLE topographic factor: Italy (3.6), Scandinavia (0.7)). The same would be true for a roundwood extraction within the same country but in two very dissimilar locations that differ in their rainfall aggressiveness, topographic conditions, susceptibility of soil to erosion, techniques of wood collection and application of soil conservation practices (Hood et al., 2002;Dissmeyer and Foster, 1984). The guiding principle of the EU is 'to use the forests and forest lands in a way, and at a rate, that maintains their biodiversity, productivity, regeneration capacity, vitality and their potential to fulfil, now and in the future, relevant ecological, economic and social functions, at local, national, and global levels, and that does not cause damage to other ecosystems' (Helsinki, 1993). To comply with this guideline, more effort must be put on the monitoring and assessment of the impacts of forest disturbances in Europe. This study indicates that about 931,880.7 ha and 209,172.6 ha of the wooded areas of the Natura 2000 network underwent significant rates of forest cover changes (most likely due to management/harvesting practices) and forest fires, respectively. Several areas declared as Sites of Community Importance (SCI) and Special Protection Areas (SPAs) in the Natura 2000 network suffered from the effects of forest disturbances. The cross-country comparison of the EU-28 provided in this study highlights the countries most exposed to this phenomenon. The study also shows that a disturbed forest area equal to 3.1% of the total protected forest could experience a nonlinear increase of the total soil erosion up to 21%. This is an expressive example of what experts regularly call severely accelerated soil erosion rates. This non-linear relation between the disturbed forested area and the acceleration of the soil erosion provides further evidence for the importance of the forest site characteristics in the soil erosion process.
To move a further step towards the assessment of the impacts of forest cover changes on the soil-related forest ecosystems, this study spatially defines the forest disturbances within the 36 European countries. It also provides the first dynamic modelling of the soil loss potential in the EU-28 forestlands. Previous studies of van der Knijff et al. (2000), Grimm et al. (2003), Kirkby et al. (2003) and Cerdan et al. (2010) conducted modelling exercises to assess the spatial distribution of water erosion in Europe, also including forestland. Still, these pioneering studies did not take care of the importance the forests changes across time and space. This study obtains spatially distributed information about the changes on the forest canopy density for the period from 2002 to 2012 by means of a GIS-based application of the USLE model. Variations of the forestland canopy cover were reconsidered annually across the study period while the climate, soil characteristics, topographic and management practices were assumed to remain constant. The processed forest cover data of Hansen et al. (2013) were considered as logging activities since (i) wood harvesting is the primary cause of forest cover change in Europe (Eurostat, 2014a,b), (ii) the annual fires reported in the European Fire Database of the Joint Research Centre of the European Commission (EFFIS, 2014) were removed by the Global Forest Change map and (iii) windthrow events that are spatially limited. With an annual mean of soil loss of 0.11 Mg ha −1 yr −1 , the modelling results closely conform to the average value measured through plot experiments in Europe (0.14 Mg ha −1 yr −1 measured in 612 plot-months; Cerdan et al., 2010). The close conformity with the measured data highlights the quality of the proposed modelling exercise which effectively represented the heterogeneous environmental characteristics of European forestlands by means of implementing high spatial resolution input data. In addition, the modelling results shed new light on the impacts of forest disturbances on the soil erosion processes at a pan-European level. A recorded forest disturbance involving about 7.1% of the EU-28 forestland area shows a total predicted soil loss of 26.6%. These numbers reflect the acceleration of the erosion rates due to forest disturbance reported in literature (Lowrance et al., 1988;Cerdà and Lasanta, 2005). According to Borrelli et al. (2013a) these accelerated erosion rates become even more severe when net erosion rates are modelled. The forest areas that exceed the annual acceptable average soil loss threshold of 10 Mg ha −1 yr −1 proposed by Morgan (2009) are spatially limited (0.01%). Solely considering the average gross erosion values predicted for the first four years after the disturbance, the areas with soil loss rates above the threshold increase up to 0.56% (equal to 10.5% of the total disturbed forest). Soil loss was found to be higher in the forests affected by fires. This is because forest fires predominantly occur in the Mediterranean region that is particularly prone to erosion as it is subject to long dry periods, followed by heavy bursts of erosive rain, falling on fragile soils on steep slopes (van der Knijff et al., 2000). About three quarters of the forest cover change occur in the North and Middle European Plains where the soil erosion rates remain lower due to the more smooth topography and the less erosive precipitations (Panagos et al., 2015). The presence of a south-to-north gradient of soil loss was confirmed observing both, the values in the disturbed and undisturbed forests.
An additional analysis of the influences of the different factors triggering soil loss, which are the age of the cut, rainfall, slope gradient, soils, allowed a more detailed assessment of each factor's contribution to the overall sediment mobilization. In this specific case, the age of the cut was obviously the primary factor influencing the predicted soil erosion values. Soil erosion rates in the forestland remain at a low level due to the dense tree cover (0.086 Mg ha −1 yr −1 ) but rise to a mean value of 5.4 and 27.3 Mg ha −1 yr −1 during the first twelve months after logging or fires, respectively. With regard to rainfall, the annual average erosivity factor computed for the EU-28 totals about 697.6 MJ mm ha −1 h −1 yr −1 . The spatial distribution of the annual average rainfall erosivity varies highly within the observed area (between 51.4 and 6228.7 MJ mm ha −1 h −1 yr −1 ). Besides the vegetation cover and rainfall, the slope gradient is a very important triggering factor for the erosive processes. In areas without forest harvesting critical values of soil erosion are absent or remain below 3 Mg ha −1 yr −1 corresponding to the class of low erosion (In the disturbed forest areas, on the other hand, the soil loss significantly increases (>5 Mg ha −1 yr −1 ) at slope gradients between 15 and 25%, becoming even severe on slopes steeper than 35%. This is because such topography encourages both inter-rill and rill denudational processes which tend to increase as the slope gradient increases (Bradford and Foster, 1996). The USLE soil erodibility factor (K) of the European forest reveals an average value of 0.029 Mg h −1 MJ −1 mm −1 and by this is 9.4% lower than the general European condition reported by Panagos et al. (2014). A gradient of the soil erodibility factor cannot be observed across Europe forests.

Conclusions
The results of this study show that the Global Forest Change map is a valid database that can be used to observe forest dynamics in Europe. Observations made during the accuracy assessment procedure and comparisons with the work of  indicate that better results can be achieved on a national-and European-scale. Still, the Global Forest Change map provides valuable and accessible information about the European forest sectors involved in the European wood supply. These data were employed to describe the forest cover change in 36 European countries and to estimate the soil erosion potential in the forestland of the EU-28 region. New insights into the distribution of the forest disturbance in Europe and the resulting soil loss potential were obtained. The presented maps provide spatially explicit indicators to assess the human-induced impacts of land cover changes and soil losses on the European soil-related forest ecosystems. These insights are relevant (i) to support policy making and land management decisions to ensure a sustainable forest management strategy and (ii) to provide a solid basis for further spatiotemporal investigations of the forestry practices' impacts on the European forest ecosystems.
The application of soil erosion models such as USLE shows that it is a suitable tool to assess accelerated soil erosion in forest environments. Moreover, this model can also be applied using scenarios that integrate pre-forest-harvesting and post-forest-harvesting soil conservation techniques. Consequently, future research should (i) focus on improving the spatiotemporal information about the forest cover change in Europe and (ii) work on the comparison between harvested forest areas where clear-cut activities take into account soil conservation practices and areas where these conservation techniques are neglected. Once the impacts of forest conservation techniques are quantified by field observations they can be parameterized and integrated in the modelling operations.

Data availability
The European map of soil loss potential, as all the maps presented in this study, is available on the European Soil Data Centre (ESDAC) web platform (http://esdac.jrc.ec.europa.eu/).