Scholarly article on topic 'Soil erodibility in Europe: A high-resolution dataset based on LUCAS'

Soil erodibility in Europe: A high-resolution dataset based on LUCAS Academic research paper on "Earth and related environmental sciences"

CC BY
0
0
Share paper
Keywords
{RUSLE / K-factor / Stoniness / Modelling / Regression / Erosion}

Abstract of research paper on Earth and related environmental sciences, author of scientific article — Panos Panagos, Katrin Meusburger, Cristiano Ballabio, Pasqualle Borrelli, Christine Alewell

Abstract The greatest obstacle to soil erosion modelling at larger spatial scales is the lack of data on soil characteristics. One key parameter for modelling soil erosion is the soil erodibility, expressed as the K-factor in the widely used soil erosion model, the Universal Soil Loss Equation (USLE) and its revised version (RUSLE). The K-factor, which expresses the susceptibility of a soil to erode, is related to soil properties such as organic matter content, soil texture, soil structure and permeability. With the Land Use/Cover Area frame Survey (LUCAS) soil survey in 2009 a pan-European soil dataset is available for the first time, consisting of around 20,000 points across 25 Member States of the European Union. The aim of this study is the generation of a harmonised high-resolution soil erodibility map (with a grid cell size of 500m) for the 25 EU Member States. Soil erodibility was calculated for the LUCAS survey points using the nomograph of Wischmeier and Smith (1978). A Cubist regression model was applied to correlate spatial data such as latitude, longitude, remotely sensed and terrain features in order to develop a high-resolution soil erodibility map. The mean K-factor for Europe was estimated at 0.032thahha−1 MJ−1 mm−1 with a standard deviation of 0.009thahha−1 MJ−1 mm−1. The yielded soil erodibility dataset compared well with the published local and regional soil erodibility data. However, the incorporation of the protective effect of surface stone cover, which is usually not considered for the soil erodibility calculations, resulted in an average 15% decrease of the K-factor. The exclusion of this effect in K-factor calculations is likely to result in an overestimation of soil erosion, particularly for the Mediterranean countries, where highest percentages of surface stone cover were observed.

Academic research paper on topic "Soil erodibility in Europe: A high-resolution dataset based on LUCAS"

Soil erodibility in Europe: A high-resolution dataset based on LUCAS ^oossMark

Panos Panagos a* Katrin Meusburger b, Cristiano Ballabio a, Pasqualle Borrellia, Christine Alewellb

a European Commission, Joint Research Centre, Institute for Environment and Sustainability, Via E. Fermi 2749,1-21027 Ispra, VA, Italy b Environmental Geosciences, University of Basel, Bernoullistrasse 30,4056 Basel, Switzerland

HIGHLIGHTS

• Soil erodibility in Europe is estimated at 0.032 t ha h ha-1 MJ-1 mm- \

• Stoniness has an important impact in Mediterranean countries.

• High resolution (500 m grid cell) dataset of K-factor is available for modelling.

• Coarse fragments, permeability and soil structure were considered in K-factor.

• K-factor map has very good correspondence with regional data in literature studies.

ARTICLE INFO

Article history:

Received 25 September 2013 Received in revised form 20 January 2014 Accepted 2 February 2014 Available online 21 February 2014

Keywords:

K-factor

Stoniness

Modelling

Regression

Erosion

ABSTRACT

The greatest obstacle to soil erosion modelling at larger spatial scales is the lack of data on soil characteristics. One key parameter for modelling soil erosion is the soil erodibility, expressed as the K-factor in the widely used soil erosion model, the Universal Soil Loss Equation (USLE) and its revised version (RUSLE). The K-factor, which expresses the susceptibility of a soil to erode, is related to soil properties such as organic matter content, soil texture, soil structure and permeability. With the Land Use/Cover Area frame Survey (LUCAS) soil survey in 2009 a pan-European soil dataset is available for the first time, consisting of around 20,000 points across 25 Member States of the European Union. The aim of this study is the generation of a harmonised high-resolution soil erodibility map (with a grid cell size of 500 m) for the 25 EU Member States. Soil erodibility was calculated for the LUCAS survey points using the nomograph of Wischmeier and Smith (1978). A Cubist regression model was applied to correlate spatial data such as latitude, longitude, remotely sensed and terrain features in order to develop a high-resolution soil erodibility map. The mean K-factor for Europe was estimated at 0.032 t ha h ha-1 MJ-1 mm-1 with a standard deviation of 0.0091 ha h ha-1 MJ-1 mm-1. The yielded soil erodibility dataset compared well with the published local and regional soil erodibility data. However, the incorporation of the protective effect of surface stone cover, which is usually not considered for the soil erodibility calculations, resulted in an average 15% decrease of the K-factor. The exclusion of this effect in K-factor calculations is likely to result in an overestimation of soil erosion, particularly for the Mediterranean countries, where highest percentages of surface stone cover were observed.

© 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license

(http://creativecommons.org/licenses/by/3.0/).

1. Introduction

Soil erosion is the most widespread form of soil degradation worldwide (Bridges and Oldeman, 1999). Since soil erosion is difficultto measure at large scales, soil erosion models are a crucial estimation tool at regional, national and European levels. The high heterogeneity of soil erosion causal factors, combined with often poor data availability is an obstacle for the application of complex soil erosion models. Thus, the empirical Revised Universal Soil Loss Equation (RUSLE) (Renard et al., 1997), which predicts the average annual soil loss resulting from raindrop splash and runoff from field slopes, is still most frequently used

* Corresponding author. Tel.: +39 0332 785574; fax: +39 0332 786394. E-mail address: panos.panagos@jrc.ec.europa.eu (P. Panagos).

at large spatial scales (Renschler and Harbor, 2002; Panagos et al., in press). The RUSLE is the simple multiplication of 5 soil erosion risk factors, of which one is the soil erodibility also called K-factor. The K-factor is a lumped parameter that represents an integrated annual value of the soil profile reaction to the process of soil detachment and transport by raindrops and surface flow (Renard et al., 1997). As such soil erodibility is best estimated by carrying out direct measurements on field plots (Kinnell, 2010). However, since field measurements are expensive and often not easily transferable in space, researchers investigated the relation between "classical" soil properties and soil erodibility.

A number of equations have been designed to predict soil erod-ibility, most famous is the soil erodibility nomograph of Wischmeier et al., 1971. Dangler and El-Swaify (1976) developed an equation for

http://dx.doLorg/10.1016/j.sritotenv.201402.010

0048-9697/© 2014 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY license (http://creativecommons.org/licenses/by/3.0/).

Hawaiian soils. Other equations, such as that of Young and Mutcher (1977), require attributes that are not widely available to predict soil erodibility (e.g. bulk density). During the 1990s, Romkens et al. (1997), Williams (1995) and Torri et al. (1997) developed simpler equations mainly based on soil texture.

At European level, Panagos et al. (2012a) estimated soil erodibility based on attributes (texture, organic carbon) which were available from the Land Use/Cover Area frame Survey (LUCAS) topsoil data (Toth et al., 2013) using the original nomograph of Wischmeier et al. (1971). Inverse distance weighting (IDW) was used to interpolate erodibility to a map with a grid-cell resolution of 10 km. The dataset attracts great interest and it is available for download from the European Soil Data Centre (ESDAC); approximately 200 users have registered and downloaded the data within two years. The great majority of these used the K-factor as an input for their USLE/RUSLE models, or for validation and comparison to their modelled or measured K-factor estimates. Past experience with the coarse-resolution soil erodibility dataset showed that it is fairly difficult for soil erosion modellers to access soil profile data in their area of interest.

However, a dataset with a resolution of 10-km grid cell can be considered too rough for most applications especially as the vast majority of users downloaded the K-factor for regional and local applications. Thus, the main objective of this paper is to produce a soil erodibility dataset with a higher spatial resolution (500-m grid cell size). In order to enable a better interpolation of the LUCAS point estimates Cubist regression-interpolation is applied. Besides the higher spatial resolution achieved through the abovementioned interpolation technique, this new soil erodibility assessment will consider soil structure and the effect of stones both on the soil permeability and the shielding of rain splash. Moreover, Malta and Cyprus have been included in the analysis. Another major improvement is that the estimated soil erodibility dataset will be verified against local, regional and national data found in the literature.

2. Materials and methods

2.1. input data

The geographical extent of this study includes 25 Member States of the European Union (EU). Bulgaria, Romania and Croatia were not included as the main input dataset (LUCAS survey 2009) does not include data for those countries.

2.12. Stone cover percentage

During the 2009 LUCAS data collection exercise, the surveyors estimated the percentage of the surface that is covered with stones. Surveyors were given a chart (LUCAS, 2009b) to help them estimate the percentage of stones present above the ground (Fig. 1). According to the instruction guide (LUCAS, 2009b), the surveyors removed the vegetation coverage and litters around the sampling point. The surveyors were trained to assign their estimation to one of the five classes (LUCAS, 2009b) based on their visual assessment and the charts provided in the instruction guide (Fig. 1). As surveyors in Cyprus and Malta did not assess the percentage of stones, class = 2 was assigned to their data as this is the predominant stone cover class in LUCAS for the southern parts of the Mediterranean countries.

2.1.3. European Soil Database

The European Soil Database (ESDB), at 1:1,000,000 resolution (King et al., 1994), is a reference dataset for assessing the state of soils in the EU. The ESDB includes, among others, attributes such as texture and soil types expressed as classes.

2.1.4. Covariates used for the Cubist regression model

Cubist (Quinlan, 1992) is a rule based model tree where the terminal leaves contain linear regression models. Prediction is obtained using the linear regression model at the terminal node of the tree and smoothed by taking into account the prediction from the linear model in the previous node of the tree. Various covariates were considered for the Cubist model, but three main types were considered to be significant:

1. Remotely sensed data derived from the Moderate Resolution Imaging Spectro-radiometer (MODIS), including vegetation indices (Normalized Difference Vegetation Index — NDVI, Enhanced Vegetation Index — EVI) and raw band data which have been re-projected using Principal Component Analysis;

2. Terrain features, derived from the Shuttle Radar Topography Mission (SRTM) Digital Elevation Model, including common geo-morphometric descriptors (elevation, slope, base level of streams, altitude above channel base level and multi-resolution index of valley bottom flatness);

3. Latitude and longitude.

2.1.1. LUCAS topsoil data

LUCAS (Land Use/Cover Area frame Survey) is an in-situ assessment, which means that the data is gathered through direct field observations. The aim of the LUCAS survey is to establish a fully harmonised database within the EU on land use/cover and to document changes over time. A soil module was included in the LUCAS dataset for the first time in 2009. Topsoil samples (0-30 cm, approximate weight of 0.5 kg) were collected from 10% of the survey points, providing 19,969 soil samples across the 25 Member States. The density of LUCAS topsoil sample points is around 1 per 199 km2, corresponding to a grid cell size of around 14 km x 14 km (Panagos et al., 2013).

The objective of the soil module in the LUCAS dataset was to improve the availability of harmonised data on soil parameters in Europe. During the period 2010-2011, the 19,969 LUCAS soil samples were analysed in a single ISO-certified laboratory to obtain a coherent pan-European dataset. The significant advantage of this method is that discrepancies arising from inter-laboratory differences (Cools et al., 2004) have been avoided. The results of the analysis are stored in the LUCAS topsoil database (Toth et al., 2013), which includes (among others) the particle size distribution expressed as percentages of clay (<0.002 mm), silt (0.0020.05 mm), sand (0.05-2.0 mm) as well as organic carbon (%) and percentage coarse material (>2.0 mm). Analysis of the soil parameters followed standard procedures (LUCAS, 2009a; ISO, 2013).

The MODIS data was acquired in 2009 during the same period as the LUCAS data, while the SRTM data refer to the year 2000.

2.2. Soil erodibility estimates for the LUCAS point dataset

As direct measurements of K-factor on field plots are not financially sustainable at the regional or national levels, the soil erodibility nomograph (Wischmeier et al., 1971) is most commonly used and cited for soil erodibility calculation. An algebraic approximation of the nomograph that includes five soil parameters (texture, organic matter, coarse fragments, structure, and permeability) is proposed by Wischmeier and Smith (1978) and Renard et al. (1997) in Eq. (1):

K = [(2.1 x 10—4 M114(12-OM) + 3.25(s—2) + 2.5(p-3))/100] * 0.1317

where:

M the textural factor with M = (msi]t + mvfs) * (100 — mj;

mc [%] clay fraction content (<0.002 mm);

msi]t [%] silt fraction content (0.002-0.05 mm);

mvfs [%] very fine sand fraction content (0.05-0.1 mm);

OM [%] the organic matter content;

Fig. 1. Methodology applied for the generation of a European K-factor (soil erodibility) map.

s the soil structure class (s = 1: very fine granular, s = 2: fine

granular, s = 3, medium or coarse granular, s = 4: blocky, platy or massive; Table 1);

p the permeability class (p = 1: very rapid.....p = 6: very

slow; Table 2).

The K-factor is expressed in the International System of units as t ha h ha-1 MJ-1 mm-1. The proposed erodibility equation (Eq. (1)) can only be recommended if organic matter content is known and silt content is below 70%. If these criteria are met, this equation is more precise than alternative equations (Declercq and Poesen, 1992). The

Table 1

Classes of soil structure derived the European Soil Database.

Structure class (s) European Soil Database

1 (very fine granular: 1-2 mm) G (good)

2 (fine granular: 2-5 mm) N (normal)

3 (medium or coarse granular: 5-10 mm) P (poor)

4 (blocky, platy or massive: > 10 mm) H (humic or peaty top soil)

methodology applied in this study (depicted in Fig. 1) was selected based on the availability of data to calculate input attributes at the European level.

The combined application of the K-factor nomograph with the LUCAS dataset required three adaptions:

• According to Wischmeier and Smith (1978), Eq. (1) is restricted to samples for which the silt fraction does not exceed 70%. A subset of 718 soil samples collected in LUCAS 2009 had silt fractions in the range of 70%-80%. As these were mainly taken from northern France, southern Belgium and central Germany, it was considered essential to be included in the calculation of the K-factor. An upper limit value of 70% silt fraction was assigned to those samples. The 212 soil samples that exceeded the 80% silt fraction were excluded from the calculation.

• In literature the sand fraction is categorised into five classes of sand: very fine, fine, medium, coarse, very coarse (Gee and Bauder, 1986; Gee and Or, 2002). The very fine sand structure (0.05-0.1 mm) as sub-factor (mvfs) in Eq. (1) is usually not subject of standard soil analysis and was therefore estimated as 20% of the sand fraction (0.05-2.0 mm) which is available in the LUCAS topsoil database.

• For soil samples with organic matter content above 4%, the upper limit of4% has been applied (Wischmeier and Smith, 1978). The application of a 4% limit to soil organic matter intends to prohibit an underestimation of soil erodibility for soils that are rich in organic matter.

22.1. Estimation of structure classes

Good soil structure and high aggregate stability are important for improving soil fertility, enhancing porosity and decreasing erodibility (Bronick and Lal, 2005). In past studies (Bonilla and Johnson, 2012; Lopez-Vicente et al., 2008; Perez-Rodriguez et al., 2007), soil structure was assigned based on soil types of the Food and Agriculture Organization (FAO). A pedotransfer rule for estimating soil structure when no direct measurements are available has been developed by Van Ranst et al. (1995). In the European Soil Database, this pedotransfer rule classifies the soil structure as humic, poor, normal or good (Table 1), using pedological inputs such as the FAO soil name and soil texture (Jones et al., 2003). The latter dataset was used to derive the structure class values needed for K-factor calculation as given in Table 1.

2.2.2. Soil permeability estimation

For the estimation of the soil permeability, classes described in the US Department of Agriculture's National Soils Handbook No. 430 (USDA, 1983) were assigned according to soil texture classes (Table 2) (Rawls et al., 1982). These soil textural classes have also been employed for the estimation of the range values of saturated hydraulic conductivity, which are explained below (Table 2).

Soil permeability is affected by the content of stones (>2 mm). The Agriculture Handbook No. 537 (Wischmeier and Smith, 1978) separates the influence of stone fragments into two components: a) surface rock fragments which can further reduce the splash detachment rate in a similar way to how vegetation protects soils from rainfall intensity; b) subsurface rock fragments that lead to increased soil loss due to reduced water infiltration.

The latter effect of coarse fragments is due to a reduction in the empty spaces (voids). As the LUCAS topsoil database includes coarse fragments (>2 mm), their effect on saturated hydraulic conductivity and soil erodibility can be calculated using the following equation (Brakensieketal., 1986):

Kb/Kf = (1-Rw) (2)

where Kb (mm day-1) is the modified saturated hydraulic conductivity after accounting for the effect of rock fragments, and Kf is the saturated hydraulic conductivity of the fine soil fraction (< 2 mm). Initial estimates for Kf were also assigned by classification of LUCAS texture information into the corresponding texture classes and associated saturated hydraulic conductivities of the US Department of Agriculture's National Soils Handbook No. 430 (USDA, 1983). Rw is the percentage of coarse fragments greater than 2 mm. Rw reduces the saturated hydraulic conductivity in the soil profile and can likely change the permeability class, as indicated in Table 2.

2.2.3. Adjustment of K-factor by inclusion of surface stone cover

Besides the percentage of coarse fragments for the 0-30 cm soil samples, LUCAS provides also a percentage estimate of the surface stone cover. Surface stone cover may have a negative effect on sediment yield and thus, can be considered as natural soil-surface stabiliser. Rubio and Recatala (2006) proposed stoniness to be included in the soil erodibility index qualitative estimation. Poesen and Ingelmo-Sanchez (1992) carried out a review of the negative relationship between stone cover and the relative interrill sedimentyield. This negative relationship is generally observed where stones are either partly embedded in the top layer or are on the surface of the soil.

Poesen et al. (1994) developed a soil erodibility reduction factor expressed as an exponential decay function based on experimental field data:

St = e-004(Rc-10) (3)

where:

St is the correction factor for the relative decrease in sediment

yield;

R is the percentage of stones cover with 10% < R < 100%.

The mean rate of decay was calculated as 0.04. Similar equations with different parameters that were proposed by other authors have given different rates of decay: 0.025 (Box, 1981), 0.044 (Simanton et al., 1986), 0.050 (Martin, 1988).

Table 2

Soil permeability classes and saturated hydraulic conductivity ranges estimated from major soil textural classes.

Table 3

Classes of percentage surface stone cover of LUCAS database.

Permeability class (p) Texture Saturated hydraulic Class Percentage of stones Value (%) used Number of St (correction factor)

conductivity, mm h-1 for the St samples and

1 (fast and very fast) Sand >61.0 calculation proportion (%)

2 (moderate fast) Loamy sand, sandy loam 20.3-61.0 0 0% 0.0% 95 (0.48%) 1

3 (moderate) Loam, silty loam 5.1-20.3 1 Stones < 10% 5.0% 14,585 (73.37%) 1

4 (moderate low) Sandy clay loam, clay loam 2.0-5.1 2 10% < Stones < 25% 17.5% 3114(15.66%) 0.740

5 (slow) Silty clay loam, sand clay 1.0-2.0 3 25% < Stones < 50% 37.5% 1442 (7.25%) 0.332

6 (very slow) Silty clay, clay <1.0 4 Stones > 50 75.0% 643 (3.23%) 0.074

Surface stone cover was estimated by the LUCAS surveyor in five classes (Table 3). The majority of the samples were found to have less or equal to 10% of stones and a correction factor cannot be applied according to Eq. (3). For classes 2, 3 and 4 (Table 3), the mean value of the percentage class (Table 3 column 3) was applied in Eq. (3) resulting in three correction factors (St).

The updated soil erodibility value (Kst) incorporating surface stone cover was calculated according to Eq. (4):

Kst = K * St. (4)

2.3. Spatial prediction of the K-factor

Given the linearity of Eq. (1), a regression approach was used to predict the K-factor in order to infer the distribution of soil erodibility from a series of related, but independent, covariates (Goovaerts, 1998). Basically, this approach aims to find a statistical relationship between the property to be predicted and a set of spatially exhaustive covariates. Once this relationship is established, the dependent property can be estimated for the area of interest.

K-factor Soil erodibility in Europe

Fig. 2. High-resolution (500 m grid cell size) map of Soil Erodibility estimated as K-factor in the European Union.

2.3.1. Spatial analysis

In this study, the K-factor value of each LUCAS point sample was interpolated using a series of spatially exhaustive environmental descriptors (covariates) in order to derive a continuous map for Europe. An alternative approach would have been to apply the equation to interpolated maps of all the soil properties needed in Eq. (1). However, the latter approach has some critical drawbacks; first of all, every predicted property has its own error and (possibly) bias. This could lead to a misestimation of the K-factor which is not constant in the geographical space. Moreover, it is inherently simpler to evaluate the effect of covariates on the value of K-factor if this is directly modelled as such, and not as the combination of the mapped variables upon which the K-factor is calculated.

The approach followed in this study made the calculation in two stages. Firstly, the regression model based on the Cubist rule (Quinlan, 1992) was used to predict the value of the K-factor using a series of covariates. Cubist is a tree model where each terminal leaf contains a linear regression model. The prediction is made using the linear regression model at the terminal nodes of the tree smoothed by taking into account the predictions from previous nodes of the tree. Cubist makes an average of the sample value over a given neighbour (Quinlan, 1993). Once the first model is fitted, the nearest neighbours of a given instance can be averaged and used as the proxy value for that instance. This procedure avoids overfitting and makes the model more robust to outliers.

In the next stage, the residuals from the Cubist model were interpolated using Multilevel B-Splines (MBS) (Lee et al., 1997). In terms of accuracy and unbiasedness, MBS performs as well as kriging but it is computationally faster and allows an easy estimation of the interpolated field (K-factor).

Model performance was tested for both the fitting and a cross-validation dataset. In the bootstrapped cross-validation the random sampling with replacement 1/10 of the original dataset (mutually exclusive with the training set) was used as a validation sample. The bootstrap procedure was repeated 100 times to produce reliable estimates of the model predictive performance over LUCAS samples.

2.3.2. Verification

The proposed high-resolution dataset was validated against local and regional studies. An extensive review of published studies that use Eq. (1) was carried out. More than 100 soil erodibility assessments were found in the literature at local, regional or even national level (Hungary, Lithuania, Czech Republic and Slovakia). The authors contacted the scientists who developed those assessments and received replies and aggregated data of 21 published studies. The authors attempted to ensure the maximum representativeness for the whole study area (with at least one study for each country).

3. Results and discussion

3.1. Soil erodibility in Europe

The mean K-factor for the 25 Member States was calculated as 0.032 t ha h ha-1 MJ-1 mm-1 with a standard deviation of 0.009 t ha h ha-1 MJ-1 mm-1 (Fig. 2). The range of values is 0.004-0.076 t ha h ha-1 MJ-1 mm-1. The map (Fig. 2) does not include lakes, bare rocks, glaciers and urban areas.

The Cubist regression model predicted the pan-European distribution of the K-factor with a good performance as R2 = 0.4 and RMSE = 0.0102 t ha h ha-1 MJ-1 mm-1 in k-fold cross validation. The interpolation using MBS further increased the prediction performance of the K-factor to an R2 of 0.94 for the fitting dataset. Cross-validation gives a less good performance (R2 of 0.74), given that part of the original LUCAS points are left out for the prediction.

The spatial pattern of areas with high soil erodibility (Fig. 2) largely follows the Loess map of Europe 1:2,500,000 according to Haase et al. (2007). The mean K-factor value for the Loess areas of Europe was

Fig. 3. K-factor compared to the European Soil Database (ESDB) soil surface texture classes.

estimated at 0.0419 t ha h ha-1 MJ-1 mm-1. A comparison between the resulting K-factors and the textural classes of the European Soil Database shows that the highest mean values of the K-factor are in the medium-fine textural class (3), followed by the fine (4) and medium (2) classes, while the lowest mean values are recorded for coarse (1) and very fine (5) classes (Fig. 3). This follows the main rules of soil science that coarse particles are relatively heavy and fine particles have, due to their relatively large surface areas, high cohesion strength and thus are less susceptible to soil detachment. Thus, the medium sized texture classes are more prone to soil erosion. The organic soils (no mineral texture) have the lowest mean K-factor value.

Most of the soil samples belong to the Normal (N) soil structure class of the ESDB corresponding to the fine granular (class: 2). The majority of the samples had a moderate permeability class (3) which was corrected to moderate low (class: 4) with the incorporation of coarse fragments. A soil sample having as attributes the mean values (Table 4) of the input parameters of Eq. (1) will result in a K-factor equal to 0.032 t ha h ha-1 MJ-1 mm-1.

The aggregated country-level statistics present an overview of the soil erodibility in Europe (Table 5). Organic matter has an important impact on the soil erodibility pattern as countries with high concentrations of organic matter have the lowest soil erodibility. Ireland, Estonia, Denmark, the Netherlands, the United Kingdom, Finland, Sweden and Latvia with high mean organic matter values (Jones et al., 2005) have mean soil erodibility values of less than 0.030 t ha h ha-1 MJ-1 mm-1. On the other hand, the highest mean values (higher than 0.035 t ha h ha-1 MJ-1 mm-1) are observed in

Table 4

Summary of input soil property values used for the estimation of the K-factor.

Eq. (1) attributes

Mean value

Standard deviation

Organic matter (OM) 0-4% 3.08% 1.05%

Structure (S) 0,1,2,3,4 2a

Permeability incorporating coarse 1,2, 3,4,5,6 4a

fragments (P)

Clay (mc) 0-100% 18.5% 13.4%

Silt corrected (<70%) (msiit) 0-70% 35.5% 19.0%

Very fine sand (mvfs) 0-20% 8.1% 5.5%

K-factor (t ha h ha-1 MJ-1 mm-1) 0.004-0.076 0.0320 0.009

a Dominant value.

Table 5

Comparison of soil erodibility with and without considering surface stone content (K-factor and K^-factor, respectively) per country.

Country K-factor equation (Eq. (1)) Kst-factor stoniness Reductio stonines

ISO Name Mean value (t hah ha-1 MJ-1 mm-1) Standard deviation (t hah ha-1 MJ-1 mm-1) Mean value (t hah ha-1 MJ-1 mm-1)

AT Austria 0.0321 0.0080 0.0291 9.5%

BE Belgium 0.0422 0.0092 0.0387 8.2%

CY Cyprus 0.0362 0.0028 0.0265 26.8%

CZ Czech Republic 0.0373 0.0076 0.0342 8.3%

DE Germany 0.0334 0.0102 0.0311 7.0%

DK Denmark 0.0246 0.0065 0.0225 8.7%

EE Estonia 0.0254 0.0074 0.0242 4.5%

EL Greece 0.0298 0.0057 0.0229 23.3%

ES Spain 0.0368 0.0058 0.0265 27.9%

FI Finland 0.0273 0.0058 0.0242 11.2%

FR France 0.0356 0.0101 0.0284 20.1%

HU Hungary 0.0349 0.0078 0.0337 3.3%

IE Ireland 0.0234 0.0047 0.0216 7.4%

IT Italy 0.0322 0.0077 0.0276 14.5%

LT Lithuania 0.0321 0.0067 0.0309 3.8%

LU Luxembourg 0.0392 0.0036 0.0345 11.9%

LV Latvia 0.0290 0.0067 0.0281 3.2%

MT Malta 0.0381 0.0022 0.0284 25.5%

NL Netherlands 0.0246 0.0084 0.0236 3.9%

PL Poland 0.0299 0.0106 0.0285 4.8%

PT Portugal 0.0333 0.0069 0.0194 41.8%

SE Sweden 0.0293 0.0068 0.0252 13.9%

SI Slovenia 0.0313 0.0052 0.0282 9.6%

SK Slovakia 0.0362 0.0074 0.0321 11.3%

UK United Kingdom 0.0271 0.0063 0.0241 11.1%

Belgium, Luxembourg, central European countries (Slovakia, Czech Republic, and Hungary), Spain and France. Those relatively high values can be attributed partly to the Loess belt and partly to relatively lower organic matter content compared to the northern countries. The smallest variations were noticed in small countries (Cyprus, Malta and Luxembourg) with more homogenous regions, while higher variations were noticed in the Loess regions (Poland, Germany and Netherlands).

3.2. The effect of surface stone cover (stoniness)

The Kst values of the LUCAS points were interpolated using the same methods and covariates as for the K-factor. The Cubist model for the Kst-factor prediction performed with R2 = 0.31 and an RMSE = 0.00811 ha h ha-1 MJ-1 mm-1 for the k-fold cross validation. The MBS was used to model the spatial distribution of the residuals. The resulting Kst-factor map (Fig. 4) is slightly different compared to the K-factor map (Fig. 2). The mean Kst-factor value is 0.0271 t ha h ha-1 MJ-1 mm-1 with a standard deviation of 0.0087 t ha h ha-1 MJ-1 mm-1. The range is 0.001-0.0737 t ha h ha-1 MJ-1 mm-1. The application of the stoniness correction factor (St) reduces the K-factor on average by 15%. The stoniness effect is much stronger in the Mediterranean Basin, as also confirmed by past studies (Danalatos et al., 1995; Poesen et al., 1998).

The considerable effect of surface stone cover (named stoniness in the following) on soil erodibility in the Mediterranean Basin has also been presented in recent studies (Zavala et al., 2010). The effect of high stoniness can be greater than the protection of vegetation in limiting soil loss. The protective effect of stoniness is strongest in Portugal, Spain, Greece and France (Table 5) where it reduces the K-factor by 20-42%. In contrast, stoniness reduces soil erodibility by less than 5% in the Baltic States, Poland, Hungary and the Netherlands (Table 5). The regional effect of stoniness, visualised as percentage reduction map (Fig. 5) is most pronounced in eastern Portugal, western Spain, southern France, the Italian islands and southern Greece (Fig. 5).

The impact of stoniness on the K-factor was included for the first time at European scale. This is a major improvement of the former K-factor map. As a future development for the next LUCAS 2015 soil survey, a larger number of stoniness classes (more than the 5 classes

in Table 3) could be made available to the surveyors and targeted training could be given on how to estimate this attribute. As past research (Poesen et al., 1994) proved that the presence of surface-level stones can lead to an exponential decrease in soil erosion, soil erosion modellers should also take the Kst-factor into account.

3.3. Mapping of soil erodibility and related uncertainties

The application of Cubist regression interpolation for the development of the high-resolution soil erodibility map facilitates the identification of the dependencies of the K-factor on other covariates such as geo-morphometric indices, hydrology, topography, elevation and land cover. As the covariates are available in high resolution (<500 m), the values can be interpolated to the pixels between the sampled points with much better accuracy than with the inverse weighted distance method and the spatial variability can be modelled.

The variable importance is defined as the relevant proportion (%) of K-factor variance which is explained by a given variable (Fig. 6). The selection of variables for the Cubist model and the variable importance was performed using Recursive Feature Elimination (RFE) (Iguyon and Elisseeff, 2003). The variables were ranked in relation to their influence on the overall performance of the model (Cross validation — RMSE) and model complexity (number of rules in the Cubist model). Variables whose removal significantly increases the RMSE of the model are retained while variables with little influence are not taken into account in the model. Latitude is the most important variable and the significance of the remaining 19 variables is relative to the latitude.

MODIS derived products (Fig. 6) are indicated by a prefix such as "red", "nir" (Near Infrared), "mir" (Medium Infrared) and "EVI" (Enhanced Vegetation Index). The suffixes "_PCAb1", "_PCAb2", etc. correspond to the 1st, 2nd, etc. axes of the Principal Component Analysis (PCA) performed on said MODIS products over a time frame of 1 year (2009). Regarding the rest of the variables (Fig. 6), 'level' is the channel network base level (Bohner and AntoniC, 2009), 'network' represents the altitude above channel network (Bohner and Antonic, 2009), 'IGBP' is the MODIS global land cover (Friedl et al., 2010), 'gradient' is the downslope distance gradient (Hjerdt et al., 2004)

Fig. 4. High-resolution (500 m grid cell size) map of Soil Erodibility estimated as K-factor in the European Union, incorporating stoniness.

and 'flatness' is the multi-resolution index of valley bottom flatness (Gallant and Dowling, 2003). The remaining variable names are self-explanatory.

Since soil erodibility is a result of complex relationships between soil properties, the authors attempted to identify the impact of a change in one input parameter on soil erodibility, keeping all the other attributes constant. The uncertainty analysis is also related to the three adaptions of the methodology (mentioned above). For example, high soil organic carbon values contribute to low K-factor values. If the input attributes

are considered to be representative of the whole dataset (Table 4), then an increase in organic matter to 4% (from 3.08%) will lead to a 9% decrease in the soil erodibility (K-factor = 0.0294). The application of a 4% limit to soil organic matter (as required by the nomograph of Wishmeier and Smith) is not taken into account in many regional assessments, which results in lower K-factor values. A possible solution would be a correction with an experimental mathematical curve of the effect of organic matter in those cases where organic matter levels are higher than 4%.

Fig. 5. Reduction in soil erodibility due to the protective effect of stones covering the soil surface (stoniness).

If the belowground coarse fragments were not taken into account, the permeability class would be much lower, with a dominant value of3 (compared to actual 4), and the average calculated soil erodibility would have been 15% lower (K-factor = 0.0279).

If the soil structure was not considered in Eq. (1), then the soil erodibility would have decreased by an average of 2.5% (K-factor = 0.0313). In most cases, soil erosion modellers ignore both permeability and soil structure due to a lack of data. In these cases, the mean decrease of soil erodibility would be around 16.3% (K-factor = 0.0276).

The very fine sand fraction is estimated to be around 20% of the total sand fraction. If the very fine sand fraction is taken to be 33.3% of the total, then the soil erodibility will increase by 11% (K-factor = 0.0361).

3.4. Comparison of K-factor estimates to local and regional assessments

Scientists of most countries provided datasets or aggregated data of K factors (Table 6: column d). However, no studies with soil erod-ibility reference data were found for the United Kingdom and Nordic countries, even though our literature review on soil erodibility was extensive.

The findings in the literature are heterogeneous in scale (from plot data to national level), nonetheless all were taken into account for the verification of K-factor dataset. None of these literature studies have included the stoniness effect. The comparison of K-factor/Kst-factor with the literature results is performed by the absolute deviation (%)

^V1 •_T_T_T__

60 70 80 90 100

Importance

Fig. 6. The twenty most important covariates and their relative importance in the application of Cubist/MBS model for K-factor interpolation.

(Table 6: columns g, h). The sign (—) is applied in case the aggregated K-factor/Kst-factor data are lower than the ones found in the literature.

The comparison of our K-factor (Table 6: column e) against the regional studies (Table 6: column d) shows a deviation of about 14.3% in absolute terms (Table 6: column g). In most of the cases, K-factor mean values are higher than those of the regional studies, with the exception of the Lithuanian study, two local Polish studies and French catchment. At national and regional levels, the correspondence of K-factor to the study area aggregated data was very positive with the exception of Brandenburg (Deumlich, 2009) where the permeability had not been taken into consideration. The 14.3% average deviation can be attributed to either the application of the 4% limit in organic matter or to the incorporation of coarse fragments in the calculation of the

permeability. However, the very good correspondence of K-factor with literature data at local and regional scales shows how the soil erodibility equation can be successfully applied at the European scale.

The relative agreement between the stoniness-corrected (Kst-factor) and the literature data is almost equally good. The Kst-factor (Table 6: column f) has an average deviation of 18.0% compared to the literature studies (Table 6: column h) and especially for the Mediterranean countries the change towards smaller K-factor values is considerable. Thus, neglecting surface stone cover will likely result in an overestimation of soil erosion risk in these countries.

4. Conclusions

The presented soil erodibility map (Fig. 2) is an important contribution to the estimation of soil erosion from local to European scales, as the K-factor is very crucial among the input factors used to estimate soil loss according to RUSLE and other models. In addition, the K-factor can usually not easily be determined or calculated by individual soil erosion modellers with no extensive data access. With the publication of this study, modellers and in general scientists will be able to download the high-resolution datasets (K-factor, Kst-factor) from the European Soil Data Centre.

Compared with past attempts to predict soil erodibility at the European level (Van der Knijff et al., 2000), the presented K-factor dataset has the advantage of pan-European harmonised soil data. In addition, topsoil data was collected within one year (2009) all across Europe and analysed by the same ISO-certified laboratory. Furthermore, the past approach to map soil erodibility at European scale (Van der Knijff et al., 2000) was based on 5 estimated textural classes of large soil mapping units of the European Soil Database while the new K-factor dataset is based on measured values.

The proposed model provides a framework for the digital soil mapping of the soil erodibility at continental scale. The Cubist regression model successfully established the relation between the K-factor and environmental features with the advantage of explaining the spatial distribution of soil erodibility. This also improves the spatial accuracy of the end product and allows establishing rules upon which the K-factor can be estimated from remotely sensed data.

Table 6

Comparison of K-factor estimates with local/regional/national studies.

Catchment/region Coverage Reference study (c) K-factor of reference K-factor Kst factor Deviation of Deviation of

(country) (a) (no of points) (b) study (d) (Fig. 2) (e) (Fig. 4) (f) K-factor vs. study (g) Kst-factor vs. study (h)

Mean value (t ha h ha- 1 MJ-1 mm- 1) (%)

Hungary (HU) National (2851) Centeri and Pataki (2000) 0.0293 0.0349 0.0337 16.0% 13.1%

Slovakia (SK) National Styket al. (2008) 0.029 0.0362 0.0321 19.9% 9.7%

Czech Republic (CZ) National Dostal et al. (2002) 0.0376 0.0373 0.0342 (-) 0.8% (-) 9.9%

Lithuania (LT) National Mazvila et al. (2010) 0.035 0.0321 0.0309 (-) 9.0% (-) 13.3%

Hessen federal state (DE) Regional Tetzlaff et al. (2013) 0.0400 0.0411 0.0382 2.6% (-) 4.8%

Bavaria federal state (DE) Regional (1051) Auerswald (1992) 0.0331 0.0367 0.0337 9.7% 1.8%

Nordrhein-Westfalen federal Regional Elhaus (2013) 0.033 0.0370 0.0337 10.7% 2.2%

state (DE)

Brandenburg federal state (DE) Regional Deumlich (2009) 0.0163 0.0232 0.0223 29.7% 27.0%

Region of Sicily (IT) Regional (1813) Bagarello et al. (2012) 0.0291 0.0300 0.0230 3.2% ( —) 26.7%

Geul catchment (Maastricht, NL) Regional de Moor and Verstraeten (2008) 0.0420 0.0449 0.0383 6.5% ( —) 9.6%

Strymonas (GR) Regional Panagos etal. (2012b) 0.0241 0.0292 0.0247 17.4% 2.3%

Andalucia (ES) Regional (8) Ruiz-Sinoga and Diaz (2010) 0.0303 0.0379 0.0245 20.1% ( —) 23.7%

Sele Catchment, Basilicata (IT) Regional Diodato etal. (2011) 0.026 0.0269 0.0230 3.5% (-) 13.2%

Lautaret, Province Alps-Cote Local Bakker et al. (2008) 0.037 0.0344 0.0254 (-) 7.6% ( —) 45.7%

d'Azur (FR)

Yialias River Catchment (CY) Local Alexakis et al. (2013) 0.0261 0.0378 0.0280 30.9% 6.6%

Gregos (PT) Local (97) Ferreira and Panagopoulos (2010) 0.0344 0.0383 0.0215 10.2% ( —) 60.2%

Pico (PT) Local (25) Ferreira and Panagopoulos (2010) 0.0290 0.0394 0.0192 26.4% ( —) 50.9%

Roncäo (PT) Local (82) Ferreira and Panagopoulos (2010) 0.0229 0.0382 0.0201 40.1% (-) 13.7%

Bogucin, Poznan (PL) Local Rejman et al. (2008) 0.0598 0.0623 0.0594 4.1% ( —) 0.7%

Lazy, Carpathian foothill (PL) Local (7 plots) Swiechowicz (2010) 0.0738 0.0588 0.0552 ( —) 25.6% ( —) 33.8%

Lublin, South Warsaw (PL) Local Wawer et al. (2005) 0.0285 0.0267 0.0261 (-) 6.6% (-) 9.2%

Overall average 21 studies 0.0344 0.0373 0.0308 14.3% 18.0.%

Another advantage is that the remote sensing products are constantly updated giving the possibility for dynamic prediction of the K-factor. On the contrary, the used remote sensing products are not tailed for the prediction of soil properties and this possibly limits the model accuracy.

The high-resolution soil erodibility map (Fig. 2) incorporates certain improvements over the coarse-resolution map (Panagos et al., 2012a):

■ Soil structure was for the first time included in the K-factor estimation.

■ Coarse fragments were taken into account for the better estimation of soil permeability.

■ Surface stone content, which acts as protection against soil erosion was for the first time included in the K-factor estimation. This correction is of great interest for the Mediterranean countries where stoniness is an important regulating parameter of soil erosion.

Soil erodibility, together with management practices (P-factor) and vegetation cover (C-factor) can be influenced by agricultural practices. Therefore, the K-factor dataset can be a guide for applying better conservation practices (e.g., increase or preserve soil organic carbon in areas prone to high levels of soil erosion risk or adaption of soil management at areas of high risk).

The K-factor dataset may also be proposed as an index for the vulnerability of ecosystems. The soil erodibility maps (Figs. 2,4) delineate areas where soil reaction to erosive rainfall events is considerably high. Areas where the stoniness effect is relatively low (< 10%) and soil erodibility is still high (Kst > 0.046 t ha h ha-1 MJ-1 mm-1) should be treated with considerable care in terms of agricultural practices and vegetation cover. For example, dependent on the force and timing of erosive rain events, local/regional policies can classify those areas as being ecologically vulnerable and propose the application of permanent crops or permanent grasslands.

The study also identified possible future improvements that can be made in the future LUCAS topsoil 2015 data collection process. Data analysis of the fraction of very fine sand and hydraulic conductivity would certainly improve the textural and permeability calculation factors, and lead to more precise estimations of soil erodibility.

Conflict of interest

The authors confirm and sign that there is no conflict of interests with networks, organisations, and data centres referred in the paper.

Acknowledgements

The authors would like to thank Grainne Mulhern for revision of the article from a linguistic point of view.

References

Alexakis DD, Hadjimitsis DG, Agapiou A. Integrated use of remote sensing, GIS and precipitation data for the assessment of soil erosion rate in the catchment area of "Yialias" in Cyprus. Atmos Res 2013;131:108-24. Auerswald K. Predicted and measured sediment Loads of large watersheds in Bavaria.

5th International Symposium of River sedimentation; 1992. Bagarello V, Di Stefano V, Ferro V, Giordano G, Iovino M, Pampalone V. Estimating the USLE soil erodibility factor in Sicily, South Italy. Appl Eng Agric 2012;28(2):199-206. Bakker MM, Govers G, van Doorn A, Quetier F, Chouvardas D, Rounsevell M. The response of soil erosion and sediment export to land-use change in four areas of Europe: the importance of landscape pattern. Geomorphology 2008;98(3-4):213-26. Bohner J, Antonic O. Chapter 8 Land-surface parameters specific to topo-climatology. In: Hengl Tomislav, Reuter Hannes I, editors. Developments in Soil Science, vol. 33. Elsevier; 2009. p. 195-226. Bonilla CA, Johnson OI. Soil erodibility mapping and its correlation with soil properties in

Central Chile. Geoderma 2012;189-190:116-23. Box J. The effects of surface slaty fragments on soil erosion by water. Soil Sci Soc Am J 1981;45:111-6.

Brakensiek DL, Rawls WJ, Stephenson GR Determining the saturated hydraulic conductivity of a soil containing rock fragments. Soil Sci Soc Am J 1986;50(3):834-5.

Bridges EM, Oldeman LR. Global assessment of human-induced soil degradation. Arid Soil Res Rehabil 1999;13(4):319-25.

Bronick CJ, Lal R Soil structure and management: a review. Geoderma 2005;124(1-2): 3-22.

Centeri Cs, Pataki R Erosion map of Hungary. Proceedings of the Conference on Environmental Management of the Rural Landscape in Central and Eastern Europe, Podbanske, Slovakia September 2-6, 2000 80-7137-940-9; 2000. p. 20-2.

Cools N, et al. Quality assurance in forest soil analyses: a comparison between European laboratories. Accred Qual Assur 2004;9:688-94.

Dangler EW, El-Swaify SA Erosion of selected Hawaii soils by simulated rainfall. Soil Sci Soc Am J 1976;40(5):769-73.

Danalatos NG, Kosmas CS, Moustakas NC, Yassoglou N. Rock fragments: II. Their impact on soil physical properties and biomass production under Mediterranean conditions. Soil Use Manage 1995;11:121-6.

de Moor JJW, Verstraeten G. Alluvial and colluvial sediment storage in the Geul River catchment (The Netherlands) — combining field and modelling data to construct a Late Holocene sediment budget. Geomorphology 2008;95(3-4):487-503.

Declercq F, Poesen J. Evaluation of two models to calculate the soil erodibility factor K. Pedologie 1992;42(2):149-69.

Deumlich D. Map of the potential erosion risk in Brandenburg using DIN19708 (ABAG) in the version of ZALF Müncheberg e.V; 2009.

Diodato N, Fagnano M, Alberico I, Chirico GB. Mapping soil erodibility from composed data set in Sele River Basin, Italy. Nat Hazards 2011;58(1):445-57.

Dostal T, Krasa J, Vaska J, Vrana K. The map of soil erosion hazard and sediment transport in scale of the Czech Republic. Vodni Hospodarstvi 2002;2:46-50.

Elhaus D. Informationssystem Bodenkarte von NRW 1: 50,000; 2013 [Personal communication].

Ferreira V, Panagopoulos T. Soil erodibility assessment at the Alqueva Dam Watershed, Portugal. Advances in climate changes, global warming, biological problems and natural hazards; 2010112-8.

Friedl MA, Sulla-Menashe D, Tan B, Schneider A, Ramankutty N, Sibley A, et al. MODIS Collection 5 global land cover: algorithm refinements and characterization of new datasets. Remote Sens Environ 2010;114:168-82.

Gallant JC, Dowling TI. A multiresolution index of valley bottom flatness for mapping depositional areas. Water Resour Res 2003;39(12).

Gee GW, Bauder JW. Particle size analysis. In: Klute A, editor. Methods of soil analysis. Part 1. Physical and mineralogical methods; 1986. p. 383-411.

Gee GW, Or D. Methods of soil analysis; 2002255-79 [Part. Accessed from http://cprl.ars. usda.gov (July 2013)].

Goovaerts P. Geostatistical tools for characterizing the spatial variability of microbiological and physico-chemical soil properties. Biol Fertil Soils 1998;27(4):315-34.

Haase D, Fink J, Haase G, Ruske R, Pecsi M, Richter H, et al. Loess in Europe—its spatial distribution based on a European Loess Map, scale 1:2,500,000. Quat Sci Rev 2007;26(9-10):1301-12.

Hjerdt KN, McDonnell JJ, Seibert J, Rodhe A. A new topographic index to quantify down-slope controls on local drainage. Water Resour Res 2004;40.

Iguyon I, Elisseeff A. An introduction to variable and feature selection. J Mach Learn Res 2003;3:1157-82.

ISO. International Organization for Standardization. ISO 10694 (1995), ISO 11277 (1998), ISO 11464 (2006). Accessed from http://www.iso.ch, 2013. [Dec 2013].

Jones RJA, Spoor G, Thomasson AJ. Vulnerability of subsoils in Europe to compaction: a preliminary analysis. Soil Tillage Res 2003;73(1-2):131-43.

Jones RJA, Hiederer R, Rusco E, Montanarella L. Estimating organic carbon in the soils of Europe for policy support. Eur J Soil Sci 2005;56(5):655-71.

King D, Daroussin J, Tavernier R Development of a soil geographic database from the Soil Map of the European Communities. Catena 1994;21(1):37-56.

Kinnell PIA. Event soil loss, runoff and the Universal Soil Loss Equation family of models: a review. J Hydrol 2010;385:384-97.

Lee S, Wolberg G, Shin SY. Scattered data interpolation with multilevel B-splines. IEEE Trans Vis Comput Graph 1997;3(3):229-44.

Lopez-Vicente M, Navas A, Machin J. Identifying erosive periods by using RUSLE factors in mountain fields of the Central Spanish Pyrenees. Hydrol Earth Syst Sci 2008;12(2): 523-35.

LUCAS. Service Provision for technical support in the LUCAS study, provision of central laboratory services for soil analysis. Accessed from http://web.jrc.ec.europa.eu/ callsfortender/index.cfm?action=app.showdoc&id=2731, 2009. [Dec 2013].

LUCAS. Technical reference document C-1. Land Use/ Cover Area Frame Survey. Instructions for surveyors. EUROSTATEUROSTAT; 2009b.

Martin W. Die Erodierbarkeit von B6den unter simulierten und natiirlichen Regen und ihre Abhaingigkeit von Bodeneigenschaften. Unpubl. Ph.D. thesis T.U. Munchen; 1988.

MazvilaJ, Staugaitis G, Kutra G, Jankauskas BZemes Okio Mokslai 2010;17(3-4):69-78.

Panagos P, Meusburger K, Alewell C, Montanarella L. Soil erodibility estimation using LUCAS point survey data of Europe. Environ Model Software 2012a;30:143-5.

Panagos P, Karydas CG, Gitas IZ, Montanarella L. Monthly soil erosion monitoring based on remotely sensed biophysical parameters: a case study in Strymonas river basin towards a functional pan-European service. Int J Digit Earth 2012b;5(6): 461-87.

Panagos P, Ballabio C, Yigini Y, Dunbar M. Estimating the soil organic carbon content for European NUTS2 regions based on LUCAS data collection. Sci Total Environ 2013;442:235-46.

Panagos P, Meusburger K, van Liedekerke M, Alewell C, Hiederer R, Montanarella L. Assessing soil erosion in Europe based on data collected through a European network. Soil Sci Plant Nutr 2014. [in press].

Perez-Rodriguez R, Marques MJ, Bienes R. Spatial variability ofthe soil erodibility parameters and their relation with the soil map at subgroup level. Sci Total Environ 2007;378(1-2):166-73.

Poesen J, Ingelmo-Sanchez F. Runoff and sediment yield from topsoils with different porosity as affected by rock fragment cover and position. Catena 1992;19(5):451-74.

Poesen JW, Torri D, Bunte K. Effects of rock fragments on soil erosion by water at different spatial scales: a review. Catena 1994;23(1-2):141-66.

Poesen JW, Van Wesemael B, Bunte K, Benet AS. Variation of rock fragment cover and size along semiarid hillslopes: a case-study from southeast Spain. Geomorphology 1998;23(2-4):323-35.

Quinlan. Learning with continuous classes. Proceedings of the 5th Australian Joint Conference On Artificial Intelligence; 1992. p. 343-8.

Quinlan. Combining instance-based and model based learning. Proceedings of the Tenth International Conference on Machine Learning; 1993. p. 236-43.

Rawls WJ, Brakensiek CL, Saxton KE, Transactions — American Society of Agricultural Engineers. Estimation of soil water properties. Trans Am Soc Agric Eng 1982;25(5): 1316-20. [1328].

Rejman J, Brodowski R, Iglik I. Annual variations of soil erodibility of silt loam developed from loess based on 10-years runoff plot studies. Annals of Warsaw University of Life ScienceAnnals of Warsaw University of Life Science; 200877-83 [Land Reclamation No 39].

Renard KG, Foster GR, Weessies GA, McCool DK. Predicting soil erosion by water: a guide to conservation planning with the Revised Universal Soil Loss Equation (RUSLE). In: Yoder DC, editor. U.S. Department of Agriculture, Agriculture Handbook 703; 1997.

Renschler CS, Harbor J. Soil erosion assessment tools from point to regional scales — the role of geomorphologists in land management research and implementation. Geomorphology 2002;47(2-4):189-209.

Romkens MJM, Young RA, Poesen JWA, McCool DK, El-Swaify SA, Bradford JM. Soil erodibility factor (K). (Compilers)In: Renard KG, Foster GR, Weesies GA, McCool DK, Yoder DC, editors. Predicting soil erosion by water: a guide to conservation planning with the Revised Universal Soil Loss Equation (RUSLE). Washington, DC, USA: Agric. HB No. 703, USDA; 1997. p. 65-99.

Rubio José L, Recatalâ L. The relevance and consequences of Mediterranean desertification including security aspects. Desertification in the Mediterranean Region. A Security Issue NATO Security Through Science Series, vol. 3; 2006133-65.

Ruiz-Sinoga JD, Diaz AR. Soil degradation factors along a Mediterranean pluviometric gradient in Southern Spain. Geomorphology 2010;118(3-4):359-68.

Simanton JR, Johnson CW, Nyhan JW, Romney EM. Rainfall simulation on rangeland erosion plots. In: Lane LJ, editor. Erosion on rangelands: emerging technology and data base. Denver, CO: Society for Range Management; 1986. p. 11-7.

StykJ, Fulajtar E, Palka B, Granec M. Updated calculation of erodibility of soil factor (k-factor) for the purpose of detailed digital layer generation. Proceedings No. 30, 2008

Soil Science and Conservation Research Institute, Bratislava (Slovak Republic) 978-80-89128-51-8; 2008. p. 139-46.

Swiechowicz J. Slopewash on agricultural foothill slopes in hydrological years 2007-2008 in Lazy. Pace I Studia Geograficzne. T, 45; 2010243-63.

Tetzlaff B, Friedrich K, Vorderbrugge T, Vereecken H, Wendland F. Distributed modelling of mean annual soil erosion and sediment delivery rates to surface waters. Catena 2013;102:13-20.

Torri D, Poesen J, Borselli L. Predictability and uncertainty of the soil erodibility factor using a global dataset. Catena 1997;31(1-2):1-22.

Toth G, Jones A, Montanarella L The LUCAS topsoil database and derived information on the regional variability of cropland topsoil properties in the European Union. Environ Monit Assess 2013;185(9):7409-25.

USDA. National Soil Survey Handbook (NSSH). Web address:http://www.nrcs.usda.gov/ wps/portal/nrcs/detail/national/soils/?cid=nrcs142p2_054242, 1983. [accessed December 2013].

Van der Knijff JM, Jones RJA, Montanarella L. Soil erosion risk assessment in Europe. European Soil Bureau. European Commission, JRC Scientific and Technical Report, EUR 19044 EN; 2000 [34 pp.].

Van Ranst E, Thomasson AJ, Daroussin J, Hollis JM, Jones RJA, Jamagne M, et al. Elaboration of an extended knowledge database to interpret the 1:1,000,000 EU Soil Map for environmental purposes. In: King D, Jones RJA, Thomasson AJ, editors. European land information systems for agro-environmental monitoring, eUr 16232 EN. Luxembourg: Office for Official Publications of the European Communities; 1995. p. 71-84.

Wawer R, Nowocien E, Podolski B. Real and calculated KUSLE erodibility factor for selected Polish soils. Pol J Environ Stud 2005;14(5):655-8.

Williams JR. Chapter 25: the EPIC model. In: Singh VP, editor. Computer models of watershed hydrology. Water Resources Publications; 1995. p. 909-1000.

Wischmeier W, Smith D. Predicting rainfall erosion losses: a guide to conservation planning. Agricultural Handbook No. 537. Washington DC, USA: U.S. Department of Agriculture; 1978.

Wischmeier W, Johnson C, Cross B. A soil erodibility nomograph for farmland and construction sites. J Soil Water Conserv 1971;26(3):189-93.

Young R, Mutchler C. Erodibility of some Minnesota soils. J. Soil Water Conserv 1977;32(3):180-2.

Zavala LM, Jordán A, Bellinfante N, Gil J. Relationships between rock fragment cover and soil hydrological response in a Mediterranean environment. Soil Sci Plant Nutr 2010;56:95-104.