Scholarly article on topic 'Towards estimates of future rainfall erosivity in Europe based on REDES and WorldClim datasets'

Towards estimates of future rainfall erosivity in Europe based on REDES and WorldClim datasets Academic research paper on "Earth and related environmental sciences"

CC BY
0
0
Share paper
Academic journal
Journal of Hydrology
Keywords
{R-factor / "Climate change" / "Rainfall intensification" / Storminess / RCP4.5 / "Erosion scenario"}

Abstract of research paper on Earth and related environmental sciences, author of scientific article — Panos Panagos, Cristiano Ballabio, Katrin Meusburger, Jonathan Spinoni, Christine Alewell, et al.

Abstract The policy requests to develop trends in soil erosion changes can be responded developing modelling scenarios of the two most dynamic factors in soil erosion, i.e. rainfall erosivity and land cover change. The recently developed Rainfall Erosivity Database at European Scale (REDES) and a statistical approach used to spatially interpolate rainfall erosivity data have the potential to become useful knowledge to predict future rainfall erosivity based on climate scenarios. The use of a thorough statistical modelling approach (Gaussian Process Regression), with the selection of the most appropriate covariates (monthly precipitation, temperature datasets and bioclimatic layers), allowed to predict the rainfall erosivity based on climate change scenarios. The mean rainfall erosivity for the European Union and Switzerland is projected to be 857MJmmha−1 h−1 yr−1 till 2050 showing a relative increase of 18% compared to baseline data (2010). The changes are heterogeneous in the European continent depending on the future projections of most erosive months (hot period: April–September). The output results report a pan-European projection of future rainfall erosivity taking into account the uncertainties of the climatic models.

Academic research paper on topic "Towards estimates of future rainfall erosivity in Europe based on REDES and WorldClim datasets"

Accepted Manuscript

Research papers

Towards estimates of future rainfall erosivity in Europe based on REDES and WorldClim datasets

Panos Panagos, Cristiano Ballabio, Katrin Meusburger, Jonathan Spinoni, Christine Alewell, Pasquale Borrelli

PII: DOI:

Reference:

S0022-1694(17)30143-9 http://dx.doi.org/10.1016/johydroL2017.03.006 HYDROL 21863

To appear in:

Journal of Hydrology

Received Date: Revised Date: Accepted Date:

4 November 2016 23 January 2017 4 March 2017

Please cite this article as: Panagos, P., Ballabio, C., Meusburger, K., Spinoni, J., Alewell, C., Borrelli, P., Towards estimates of future rainfall erosivity in Europe based on REDES and WorldClim datasets, Journal of Hydrology (2017), doi: http://dx.doi.org/10.1016/johydrol.2017.03.006

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Towards estimates of future rainfall erosivity in Europe based on REDES and WorldClim datasets

Panos Panagos1*, Cristiano Ballabio1, Katrin Meusburger2, Jonathan Spinoni1 Christine Alewell2, Pasquale Borrelli12

1European Commission, Joint Research Centre, Directorate for Sustainable Resources, Via E. Fermi 2749, I-21027 Ispra (VA), Italy.

2Environmental Geosciences, University of Basel, Switzerland "•corresponding author: Tel: 0039-0332-785574, Fax: 0039-0332-786645 E-mail: panos.panagos@jrc.ec.europa.eu

Abstract

, Fax: UU3

The policy requests to develop trends in soil erosion changes can be responded developing modelling scenarios of the two most dynamic factors in soil erosion, i.e. rainfall erosivity and land cover change. The recently developed Rainfall Erosivity Database at European Scale (REDES) and a statistical approach used to spatially interpolate rainfall erosivity data have the potential to become useful knowledge to predict future rainfall erosivity based on climate scenarios. The use of a thorough statistical modelling approach (Gaussian Process Regression), with the selection of the most appropriate covariates (monthly precipitation, temperature datasets and bioclimatic layers), allowed to predict the rainfall erosivity based on climate change scenarios. The mean rainfall erosivity for the European Union and Switzerland is projected to be 857 MJ mm ha-1 h-1 yr-1 till 2050 showing a

24 relative increase of 18% compared to baseline data (2010). The changes are

25 heterogeneous in the European continent depending on the future

26 projections of most erosive months (hot period: April - September). The output

27 results report a pan-European projection of future rainfall erosivity taking into

28 account the uncertainties of the climatic models.

30 Keywords: R-factor, climate change, rainfall intensification, storminess

31 RCP4.5, erosion scenario

32 1 Introduction

ronment

33 Soil erosion is one of the main European environmental threats, particularly in

34 Southern Europe (Panagos et al., 2015a). Its prevention and mitigation is a key

35 ecosystem service to monitor and access spatially and temporally (Guerra et

36 al., 2016). Accelerated soil erosion may lead to a decrease of ecosystem

37 stability, land productivity, land degradation in general and a loss of income

38 for farmers (Salvati and Carlucci, 2013). Soil erosion and more generally land

39 degradation is driven by unsustainable land management due to increasing

40 human pressure enhanced by climate change (Helldén and Tottrup, 2008).

41 The extent, frequency and magnitude of soil erosion in Europe is expected to

42 increase due to a general increase of extreme rain fall events caused by

43 climate change (Pruski and Nearing, 2002; Deelstra et al., 2011).

45 The prediction of soil erosion changes in the future are mainly dependent on

46 modeling future rainfall erosivity, land use changes and impacts of policies on

47 soil loss. The most commonly used erosion models are the the various types of

48 the Universal Soil Loss Equation (USLE) originally developed by Wischmeier and

Smith (1978). In the proposed algorithms, soil loss by water erosion is proportional to rainfall erosivity (R-factor), which is one of five input factors. While rainfall erosivity accounts for the effect of rainfall in soil erosion, the soil erodibility (K-factor) incorporates the soil properties defining the susceptibility of a soil to erode, the cover management (C-factor) takes into account the land use and management in agricultural lands, the slope length and steepness (LS-factor) accounts for the topography and finally the support practices (P-factor) considers the effect of conservation measures. A modified version of the USLE, the Revised Universal Soil Loss Equation (RUSLE), was originally suggested by Renard et al. (1997), and has been recently applied in Europe (RUSLE2015) for the estimation of soil loss by water at 100-m resolution (Panagos et al. 2015a). Among other improvements compared to past Pan-European soil erosion assessments, RUSLE2015 incorporates the option of running climate change, land use change and policy scenarios.

Rainfall erosivity is a multi-annual average index that measures rainfall kinetic

energy and intensity describing the effect of rainfall on sheet and rill erosion

(Wischmeier and Smith, 1978). The rainfall erosivity of a given storm in RUSLE

(referred to as R-factor) is equal to the product of the total storm energy with

its maximum 30-minutes rainfall intensity. As high temporal resolution rainfall

data are commonly not available, many studies estimated rainfall erosivity

using approximation equations based on monthly or daily rainfall data (Bonilla

and Vidal, 2011; Diodato and Bellocchi, 2010). Only recently, R-factors were

directly estimated from high temporal resolution data at national/regional

scale in Europe such as the study in Slovenia (Petan et al., 2010), Switzerland

(Meusburger et al., 2012), Ebro catchment in Spain (Angulo-Martinez et al.,

2009), Czech Republic (Janecek et al. 2013), Greece (Panagos et al., 2016a) and Italy (Borrelli et al., 2016).

The occurring and projected climate change is likely to affect soil erosion due to intensification of rain, change of precipitation amounts, change of moisture and vegetation cover change (St.Clair and Lynch 2010). The most important impact of climate change on soil erosion is expected due to an increase of rainfall intensity, in particular the increase of extreme rainfall events both at global (e.g., Sillman et al., 2012) and continental scale (for Europe see, e.g., Frei et al., 2006; Westra et al., 2014). There are few studies which have addressed the risk of increasing rainfall erosivity based on past trends (Verstraeten et al., 2006 in Ukkel, Belgium; Fiener et al., 2013 in western Germany; Hanel et al., 2015 in Czech Republic). At national scale, future trends in rainfall erosivity were addressed in the USA (Nearing, 2001; Biasutti and Seager, 2015), China (Zhang et al., 2010) and Japan (Shiono et al., 2013). The studies in USA used mean annual precipitation combined with Fournier coefficient (Arnoldus, 1980) while the ones in China and Japan have downscaled the monthly precipitation spatially and temporally. The improved understanding of General Circulation Models (GCM) and the increased data availability contributed to their wider use and allow for their integration in ecological-related disciplines (e.g. soil, water). For instance, Nearing (2001) has applied the HadCM3 climate change scenario (Gordon et al., 2000) and estimated increases in rainfall erosivity between 16% and 58% in the USA.

The objective of this study is to estimate the expected change in rainfall

erosivity and its impact on soil erosion in Europe during the first half of the 21st

101 century based on the updated IPCC climate change scenarios (IPCC, 2013).

102 This study focuses on the R-factor changes without considering the impact of

103 climate change on land/ vegetation cover. Compared to previous studies

104 that used approximation equation based on annual (or monthly)

105 precipitation, this study use as input the high-temporal-resolution Rainfall

106 Erosivity Database at European Scale (REDES) (Panagos et al., 2015b) and

107 climatic data derived from the WorldClim database, which is set of global

108 climate grids with a spatial resolution of about 1 km2 (Hijma

ans, 200

110 2 Database and modelling approach for R-factor

111 prediction

escription odelling fu

112 This chapter presents: a) a brief description of REDES and its latest updates; b)

113 WorldClim datasets modelling future climatic conditions; c) the climate

114 projections for 2050 in Europe with specific focus on rainfall, and d) the

115 regression model applied for the R-factor future prediction.

117 2.1 Rainfall Erosivity Database at European Scale (REDES)

119 (20'

118 The first version of the Rainfall Erosivity Database at European Scale (REDES) 2014) included 1,541 rainfall stations within the European Union (EU) and Switzerland (Panagos et al., 2015b). In 2015, an update of REDES was

121 performed with 134 new R-factor stations, which resulted in 1,675 REDES

122 stations. The spatial distribution and the density of rainfall stations in REDES is

123 not homogeneous in all EU countries (Ballabio et al., 2017) due to availability

(or not) of high temporal resolution rainfall data. Auerswald et al. (2015) addressed 5 comments on REDES dataset and Panagos et al.(2015c) replied to this. Both studies (Auerswald et al., 2015; Panagos et al., 2015c) agree that the use of a short time series or time series from different periods is generally a problem in all large-scale studies and requires improvement in the future.

ited in REDES b min and

The R-factor as a proxy for rainfall erosivity has been calculated in REDES by using high temporal resolution data (5-min, 10-min, 15-min, 30-min and 60-min) and applying the equations proposed by Brown and Foster (1987). The R-factor is the product of the kinetic energy of a rainfall event (E) and its maximum 30-minutes intensity (I30) (Brown and Foster, 1987).

2.2 WorldCHm datasets: baseline and projected

Global precipitation and temperature (both annually and monthly) at a high spatial resolution of 1 km2 are available from the WorldClim database (Hijmans, et al., 2005). The data layers are generated through interpolation of average monthly climate data from weather stations on a 30 arc-second resolution grid (referred to as "1 km2" resolution) and include precipitation data from 47,554 locations and maximum/minimum temperature data from 14,835 locations all over the world. The density of stations in Europe is among the highest ones. Records for at least 10-years have been used to calculate the average monthly climatic grids, which represent the baseline climatic situation.

148 The future climate projection for 2050 are derived from General Circulation

149 Models such as HadGEM2 (see next section) that was used in this study (Milton

150 et al., 2011). The yielded projections are downscaled and calibrated (bias-

151 corrected) using WorldClim as the historical (1950-2000) baseline (Hijmans,

152 2005). The datasets on future projections refer to the middle century (20

153 2060) and the midpoint year 2050 will be used as reference in the following.

154 Difference maps of WorldClim datasets (future projections compared to

155 baseline ones) show the impact of climate change in precipitation and

156 temperature (Fig. 1).

157 Fig. 1: Examples of climate change predictions according to WorldClim

158 datasets: differences between 2050 projections and baseline are shown for:

159 a) the precipitation in October, b) precipitation in May, c) Maximum

160 temperature in September, d) Maximum Temperature in November).

162 2.3 Climate projections in Europe

163 The Intergovernmental Panel on Climate Change (IPCC) recently published

164 the 5th assessment report in 2013-14 (IPCC, 2013), describing the projections of

165 climate change during the 21st century. Climate projections are model-driven

166 descriptions of possible future climates under a given set of plausible sce

scenarios of climate change (Weaver et al., 2013; Rummukainen, 2010).

169 General Circulation Models (GCMs), as well as the Regional Circulation

170 Models (RCM) represent powerful tools to produce spatially explicit

171 predictions on future climate changes based on a given scenario. More than

50 General Circulation Models are currently available for environmental studies. GCMs are numerical representations of climate systems based on physical, chemical and biological properties of oceans, land and ice surface (Harris et al., 2014). Among the 50 GCMs, we have selected the HadGEM2 climate model developed by Met Office Hadley Centre in United Kingdom (Martin et al., 2011; Jones et al., 2011). HadGEM2 represents the current state of the art and it is a valuable tool for predicting future climate and understanding the climate feedbacks within the earth system (Milton et al. 2011).

system (Milte

The climate change scenarios are called Representative Concentration Pathways (RCPs) and the 3 main used RCPs are RCP2.6, RCP4.5 and RCP8.5. Among these 3 prevailing climate change scenarios, we have selected the RCP4.5 which is the most widely used and which is neither conservative (RCP2.6) nor extreme (RCP8.5). The RCP4.5 scenario forecasts an increase in greenhouse gases that is expected to peak around 2040, afterwards a smooth decline until the end of the century is assumed.

The RCP4.5 scenario applied with the General Circulation Models HadGEM2

and calibrated with WorldClim baseline data projects a global mean surface

temperature increase by 1.4 Celsius degrees (range 0.9 - 2.0) in the period

2046-2065 and by 1.8 Celsius degrees (range 1.1 - 2.6) tin the period 2081-2100

compared to the reference period of 1986-2005 (IPCC, 2013). The projected

mean global increase in extreme precipitation events by 10% and an

increase of global precipitation amount by 5% by the end of 21st century is

relevant for rainfall erosivity changes (Kharin et al., 2013).

199 2.4 Relating R-factor and WorldClim climatic data with

Gaussian process regression

201 Since intensity, duration, frequency and amount of rainfall has large

202 uncertainty in future predictions, and the General Circulation Models (GCMs)

203 lack temporally high resolved data (<1h) for a direct R-factor estimation, the

204 application of statistics and stochastic approaches represent an alternative

205 to predict the potential change in R-factor. Previous attempts to estimate

206 changes in R-factor at catchment and national scale (Nearing, 2001; Zhang

207 et al, 2005; Ito, 2007) have used relationships between rainfall erosivity and

208 monthly or annual rainfall. However, those relationships do not consider the

209 changes in rainfall intensity and the frequency of storm events (Shiono et al.,

210 2013).

212 Here we follow a different approach (Fig. 2), because we found in a previous

213 study that rainfall erosivity (R-factor) is strongly correlated with precipitation

214 dynamics (precipitation seasonality, monthly precipitation) in Europe

215 (Panagos et al, 2015b). In this study, we chose a regression approach to

216 derive the distribution of rainfall erosivity in 2050 (dependent variable) from a

217 series of related but independent WorldClim climatic variables (covariates).

218 This is done by fitting a regression model using baseline climatic conditions

219 derived from the WorldClim dataset and the rainfall erosivity as calculated

220 from field measurements.

Fitting

Prediction

16 significant grids 2010 (baseline)

Simulated Annealing

36 Baseline climate grids + 6 Bioclimatic grids (WorldClim 1950-2000)

222 Fig. 2: Procedure followed to project future (2050) rainfall erosivity for Europe.

224 The GPR regression model establishes a statistical relation between the R-

225 factor point values (calculated from REDES) and WorldClim baseline climatic

226 data acting as a set of spatially exhaustive covariates (Fitting part in Fig. 2). In

227 a second step, this GPR regression model is applied to WorldClim future

228 climatic data layers for the year 2050 (HadGem2, Scenario 4.5) in order to

229 derive the future predictions of the rainfall erosivity (R2050) (Prediction part in

230 Fig. 2)

232 The rationale behind this procedure is that rainfall intensities and as such

233 rainfall erosivity are associated with given combinations of climatic conditions

234 that occur in the present. It is assumed that in the future, similar combinations

235 of climatic conditions are related in the same way to rainfall intensities and

236 rainfall erosivity but will likely occur at different latitudes or in different periods

237 of the year. Consequently, applying the regression model fitted on current

238 climatic dataset allows to estimate future levels of rainfall erosivity when the

239 same model is applied with covariates of projected future climatic data sets.

241 The Gaussian Process Regression (GPR) was used as regression method in this

242 study. It is a regression technique generally suited for large scale applications

243 where high dimensionality (number of degrees of freedom) of data used and

244 non-existence of linear relationships between target variable and covariates

245 (Vasudevan et al., 2009; Rasmussen and Williams, 2006) subsists. The GPR

246 model was selected in this study for two reasons: a) better performance (in

247 terms of R2, RMSE, Standard error) compared to other models and b)

248 comparability of results with the existing rainfall erosivity in Europe where GPR

249 was also applied (Panagos et al., 2015b). The details on how the regression

250 model Gaussian Process Regression is applied for the rainfall erosivity

251 prediction are described in the rainfall erosivity in Europe (Panagos et al.,

252 2015b).

254 In this model application, the optimization of the GPR by feature selection

255 was performed using a Simulated Annealing (SA) approach (Kirkpatrick et al.,

256 1983). Simulated Annealing (SA) is an optimization technique processing

257 arbitrary degrees of nonlinearities (and stochasticity) and guarantees to find

fining n-( >del is th«

258 the statistically optimal solution (Ingber, 1993). Further, SA allows finding the

259 best set of covariates to be included in the GPR model by optimizing a

260 chosen model metric; in this case the metric is cross-validation Root Mean

261 Square Error (RMSE).

263 For the first fold, n-1 of the data is used in the search while the remai

264 J) is used to estimate the internal performance. The fitted model is then

265 applied to all the data in order to obtain the external performance. This

266 allows having two metrics, one used for fitting the model (internal

267 performance) and the other used to express global model performance. SA

268 also allows to estimate variable importance by ranking variable frequency

269 candidate models through the optimization process and their influence on

270 the final model. Finally, the GPR equation together with the projected

271 changes of the same covariates will be used to estimate R-factor in 2050

273 The GPR could potentially use 42 covariates from the WorldClim database.

j n-(n-

ariates w

potentially us< the 36

274 Among them, the 36 monthly layers represent the following 3 climatic ach on

275 variables (each one has 12 monthly layers):

276 - monthly total precipitation (mm)

277 - monthly average minimum temperature (degrees C * 10)

278 - monthly average maximum temperature (degrees C * 10)

279 Moreover, we have used bioclimatic variables which are derived from

280 monthly temperature and precipitation values and generate biologically

281 meaningful variables of WorldClim. Those bioclimatic variables represent

282 annual trends, seasonality and extreme or limiting environmental factors. In

283 the prediction of rainfall erosivity, we have used six bioclimatic variables a)

the Mean diurnal range (Mean of monthly difference between maximum and minimum temperature), b) isothermality c) temperature seasonality (standard deviation * 100), d) precipitation seasonality (Coefficient of variation) e) precipitation of warmest quarter (period of 3 months; % of the year) and f) precipitation of the coldest quarter. The six bioclimatic variables are pre-

the monthly precipitation and temperature values which have been already included in the model.

3 Results and Discussion

selected among the nineteen available ones as they are not collinear with

e been alread

3.1 Gaussian Process Regression fitting

The Simulated Annealing (SA) procedure has been applied over the set of 42 proposed WorldClim covariates (see 2.4 section) and a selected set of 16 covariates was used in the final best model (Table 1). The selection procedure converges at iteration 150 where the minimum RMSE of 515.78 is reached (Fig. 3) for external validation. The stability of the model output is supported by the plateau reached between the 100th and the 200th iteration (Fig. 3). This assures the good performance of the model in generalizing properties (such as future R-factor) and reducing the likelihood of runaway estimations in predicting future rainfall intensities (unless forecasted climatic variables with runaway values are provided as input to the model). The overall performance of the model is evidenced by an R2 of 0.635, while the relative error is 0.56 for the entire dataset (Fig. 3).

Fig. 3: Optimization profiles of the SA. The vertical axis expresses the average RMSE result of internal and external cross-validation.

The best model includes 16 variables ranked as shown in table 1. We observed that if more variables are included in the model, it is not implied that the performance will be improved.

will be i

Table 1: Ranking of WorldClim variables according to the Simulated Annealing (SA) optimization. Variables are ranked according to their respective selection frequency

Parame ter Covariate explanation Selection frequency Included in the model (Y)es / (N)o

Prec8 Average precipitation (mm) in August 80 Y

Prec4 Average precipitation (mm) in April 80 Y

Bio15 Precipitation Seasonality 80 Y

Tmin3 Average minimum temperature in March 70 Y

Prec9 Average precipitation (mm) in September 70 Y

Prec7 Average precipitation (mm) in July 70 Y

Prec6 Average precipitation (mm) in June 70 Y

Prec5 Average precipitation (mm) in May 70 Y

Bio3 Isothermality 70 Y

Biol 8 Precipitation (mm) of Warmest Quarter 70 Y

Tmin6 Average minimum temperature in June 60 Y

Tmin2 Average minimum temperature in February 60 Y

Tmax8 Average maximum temperature in August .o Y

Prec2 Average precipitation (mm) in February 60 Y

Precl 1 Average precipitation (mm) in November 60 Y

Bio4 Temperature Seasonality 60 Y

Tmin9 Minimum temperature in September 60 N

Tmax6 Average maximum temperature in June 50 N

Tmax5 Average maximum temperature in May 50 N

Tmax2 Average maximum temperature in February 50 N

Tmaxl 2 Average maximum temperature in December 50 N

Tmaxl 0 Average maximum temperature in October 50 N

Precl 0 Average precipitation (mm) in October 50 N

Precl Ave ra g e p re ci pitation (mm) in January 50 N

319 Among the 16 variables for the application of the future prediction R-factor

320 model at European scale, 8 monthly precipitation datasets are included. It is

321 notable that precipitation of winter months (December, January), March and

322 October are not included in the model while the warmer months during the

vegetation period (April to September) are included. Regarding the temperature effect in the future predictions of rainfall erosivity at European scale, the GPR model included 3 monthly minimum average temperatures (February, March and June) and only one monthly maximum average temperature (August). The GPR model included also four out of six bioclimatic variables in R-factor predictions: a) isothermality (Mean Diurnal Range divided by Temperature Annual Range) b) temperature seasonality (standard deviation) c) precipitation seasonality (Coefficient of Variation) and d) precipitation (mm) of warmest quarter.

3.2 Rainfall erosivity in Europe in 2050

The projected rainfall erosivity based on REDES and WorldClim datasets according to RCP 4.5 climate change scenario driven by the HadGEM2 GCM model (Fig. 4) shows an increase of the R-factor in Northern and Central European countries. The projected mean R-factor for 2050 in the European Union and Switzerland is 857 MJ mm ha-1 h-1 yr-1 showing an increase of 18%

Switzerk

compared to the current rainfall erosivity (Panagos et al., 2015b). The mean absolute error was estimated at 319MJ mm ha-1 h-1 yr-1 and the relative error 0.56 with a model R2 of 0.64. A simulation with the RCP 2.6 climate change scenario showed a smoother increase of erosivity compared to baseline (16%) while the use of the most aggressive scenario RCP 8.5 showed a notable increase of erosivity (27%).

348 Fig. 4: Rainfall erosivity projection for the year 2050 according to RCP 4.5

349 scenario driven by the HadGEM2 GCM model.

352 3.3 Estimated changes of rainfall erosivity in Europe (2050

353 compared to 2010)

354 Besides the future projections of rainfall erosivity, it is important to highlight the

355 change compared to baseline dataset of 2010 (Fig. 5). This comparison is

356 feasible because the same GPR model is used but with different climatic input

357 conditions (2010 versus 2050 climatic data). The absolute difference in R-

358 factor between the 2050 projection and 2010 baseline allows to identify areas

359 of strong erosivity decrease or increase (Fig. 5). Based on this assessment, 81%

360 of the area in Europe (around 3.5 * 106 Km2) is predicted to have an

361 increased rainfall erosivity by 2050 and only for the remaining 19% rainfall

362 erosivity is predicted to decrease (Fig. 5). In almost 25% of the study area the

363 R-factor is increasing by at least 50% by the year 2050 compared to the

364 baseline data (2010).

366 In large parts of Italy and Slovenia, Western Croatia (Adriatic sea), Scotland,

367 eastern Spain, eastern Bulgaria, eastern Romania, Western Greece and North

368 West Iberian Peninsula a pronounced decrease of the absolute rainfall

369 erosivity is expected (Fig. 5). Most of those areas (Scotland, Italy, Slovenia,

370 Western Greece, Croatia and North west Iberian Peninsula) have very high

371 mean R-factor (> 1,300 MJ mm ha-1 h-1 yr-1) in 2010 (Panagos et al., 2015b)

372 and the projected decrease is more than 200 MJ mm ha-1 h-1 yr-1 till 2050

373 mainly due to less rainfall.

376 Fig. 5: Absolute difference of R-factor between 2050 projections and 2010

379 The potentially most problematic areas are probably the ones where an

380 increase of more than 500 MJ mm ha-1 h-1 yr-1 is projected by 2050. The higher

381 rainfall erosivity in these areas is caused by more intense rainstorms and/or by

382 more frequent erosive events. The Swiss Alps, part of the French Atlantic coast,

383 East Croatia and parts of Slovakia and southern Germany are expected to

384 have such increase (rainfall intensity and/or frequency of erosive events),

385 according to the most often applied RCP4.5-based scenarios (IPCC, 2013),

386 including also the HadGEM2 GCM used in this study. In major parts of the

387 North Europe (France, Belgium, Netherlands, Germany, Denmark and Czech

388 Republic) a notable increase of rainfall events during summer period is g oing

389 to increase erosivity by 300-500 MJ mm ha-1 h-1 yr-1. According to the ratio of

390 current erosivity compared to future (till year 2050) R-factor, in those areas it is

391 expected to double. In Baltic states and Poland this increase will be lighter

392 but quite pronounced compared to the nowadays rates.

393 The highest mean relative increase (>50%) in rainfall erosivity by 2050 is

394 projected for the Netherlands, Denmark, Czech Republic, Slovakia, Germany

395 and Poland (Table 2). A decrease in mean rainfall erosivity is projected in Italy,

396 Malta and Slovenia (> 20%). In Spain and Greece, a slight increase of mean

397 rainfall erosivity is projected while in Ireland the situation remains fairly stable.

Table 2:

400 anc

<0 !: Mean R-f<

r the pro

factor values estimated for current climatic conditions (2010) rojected future scenario RCP4.5 (2050) per country.

wj* Country Mean R-factor (2010) Mean projected R-factor (2050) Change (%) 2010 - 2050

MJ mm ha-1 h-1 yr-1

AT Austria 1,075.5 1,240.8 15.4%

BE Belgium 601.5 881.9 46.6%

BG Bulgaria 695.0 838.2 20.6%

CH Switzerland 1,039.6 1,290.9 24.2%

CY Cyprus 578.1 817.0 41.3%

CZ Czech Republic 524.0 883.5 68.6%

DE Germany 511.6 849.8 66.1% A

DK Denmark 433.5 772.3 78.2%

EE Estonia 444.3 620.5 39.7%

ES Spain 928.5 1,013.4 9.1%

FI Finland 273.0 404.1 48.1%

FR France 751.7 999.1 32.9%

GR Greece 827.7 949.8 14.8%

HR Croatia 1,276.2 1,297.6 1.7%

HU Hungary 683.3 759.3 11.1%

IE Ireland 648.6 654.6 0.9%

IT Italy 1,642.0 1,249.5 -23.9%

LT Lithuania 484.2 686.5 41.8%

LU Luxembourg 674.5 945.2 40.1%

LV Latvia y^C 480.4 664.3 38.3%

MT Malta 1,672.4 1,277.3 -23.6%

NL Netherlands 473.3 841.1 77.7%

PL Poland 537.1 814.4 51.6%

PT Portugal 775.1 960.4 23.9%

RO Romania 785.0 930.2 18.5%

SE Sweden 378.1 494.6 30.8%

SI Slovenia 2,302.0 1,780.2 -22.7%

SK Slovakia 579.7 971.9 67.7%

UK United Kingdom 746.6 780.0 4.5%

402 An analysis per main climatic zones in Europe (EEA, 2011) shows that the

403 Boreal, Continental and Atlantic regions will be relatively more affected by

404 increased rainfall erosivity by 2050 (Table 3). The Alpine climatic zone will show

405 an increase of 13% of the R-factor and will be the area with highest mean

406 rainfall erosivity (approximately 1056 MJ mm ha-1 h-1 yr-1 by the year 2050). The

407 areas around the Adriatic Sea (Italian coast, Slovenia, Croatia and Western

408 Greece) show a notable decrease of rainfall erosivity. The mean R-factor in

409 the Mediterranean zone remains stable with different spatial patterns.

411 Table 3: R-factor projections estimated for current climatic conditions (2010)

412 and for the projected future scenario RCP4.5 (2050) per Biogeographical

413 region

Climatic Zone

Proportion of the study

Mean R-factor (2010)

Mean projected R-factor (2050)

MJ mm ha-1 h-1 yr1

1056.5

Change (%)

lack Sea

Boreal

Continental

Mediterranean

1050.6

1048.5

Pannonian

Steppic 0.8 729.8 686.6 -5.9%

416 3.4 Model Uncertainty

417 The uncertainty of the predictions has been quantified by modelling the

418 normalised error of the R-factor predictions (Fig. 6). The GPR model has the

419 advantage to estimate both the prediction of the mean and the prediction

420 of the mean variance. The standard error map expresses how much the

421 estimated value of R-factor might vary. We expressed this variation as a

422 proportion of the estimated R-factor value (Fig. 6). Likely, areas with a high

423 error are those where the model has to make predictions on a combination

424 of climatic factors that are not present in 2010, the baseline situation. Thus,

425 areas where largest changes are predicted by the GCM are likely to have a

426 high uncertainty. Indeed, the areas with higher uncertainty are the

427 Scandinavian countries, Baltic States, Scotland and part of Greece and

428 Spain. Medium uncertainty is noticed in Poland, parts of Germany, Czech

429 Republic, Hungary, Central France and southern Iberian Peninsula. However,

430 it should be noted that the high normalised error values in Scandinavia are

431 due to the very low absolute estimated value of R-factor in that area and

432 might thus not be of a high relevance.

435 Fig. 6: Normalised error in the R-factor prediction (2050)

437 Regarding uncertainty, we should further emphasize that our results come

438 from a methodology which incorporates statistical parameterizations (geo-

439 statistical model) over point data (REDES) and future climatic covariates

440 (monthly precipitation, monthly maximum/minimum temperature, bioclimatic

layers). Moreover, the results include high uncertainty due to the intrinsic climate model uncertainty. Consequently, the results should be regarded as an attempt to model future rainfall erosivity in Europe and identify differences in regional patterns.

3.5 Plausibility and Comparison with Local and Regional Studies

In this study, we showed that rainfall erosivity may on average increase by 18% in the European Union and Switzerland (zones which have generally similar characteristics to the ones in U.S.A) by 2050. For the U.S.A. Nearing et al. (2004) estimated a similar average increase of rainfall erosivity of around 17% in 2050. These matching results are due to very similar rainfall characteristics between the USA and Europe. Nonetheless, these changes are geographically variable.

id Euro

indicate that eriod o

Our results indicate that particularly changes of rainfall occurring during the

warmest period of the year (April-September) would have high effects and increase rainfall erosivity. Diodato et al. (2010), identified the precipitation of autumn months as the major factor for their R-factor projections in the Mediterranean basin based only on monthly rainfall data. Highest erosivity values during summer and early autumn were also observed for major parts of the European Union (Panagos et al., 2016b) and Switzerland (Meusburger et al., 2012).

Contrasting trends of future rainfall erosivity have been identified for the Mediterranean basin (Fig. 5) which has complex geographical characteristics. According to Lionello et al. (2006), the complex morphology in the Mediterranean basin with distinct basins and gulfs and many sharp orographic features influences the sea and atmospheric circulation and lead to great spatial variability for precipitation.

According to the future climatic projections, mean annual precipitation would potentially increase in large parts of Central and Northern Europe by up to about 25% and decrease in Southern Europe (Kriegsmann et al., 2014). The heavy summer precipitation events, defined as events exceeding the intensity at the 95th percentile of daily precipitation, are modelled to decrease by about 25% in parts of Iberian Peninsula and Southern France accompanied by regional increases in parts of Spain and Portugal. Further, the heavy precipitation events in winter are modelled to increase by up to 25% in Central and Eastern Europe (Kriegsmann et al., 2014). Trends of the last 35 years already showed an increase (0.5 storm events per year) of the high-intensity storm events in lowland regions of Germany (Mueller and Pfister, 2011).

Christensen et al. (2015), focused on climate change in Europe and concluded that the change in very high precipitation extremes may have higher impact than the global temperature change. They also identified that higher scatter will take place in The British Isles and Middle Europe and lower scatter in the Mediterranean and Iberian Peninsula.

491 The predicted R-factor patterns mainly depend on the spatial patterns of the

492 projected climatic covariates of the HadGEM2 model. The predicted rainfall

493 erosivity increase in Northern and Central Europe is connected to the climate

494 simulation model used in this study. Van Haren et al. (2013) predict an

495 increase of both annual cumulated precipitation (especially in Northern and

496 North-Eastern Europe) and frequency and intensity of extreme rainfall events,

497 including the summer rain-shower and thunderstorms that can remarkably

498 affect rainfall erosivity and subsequently soil erosion. On the other hand, the

499 GCM used in this study still predicts the increase of extreme events in Southern

500 Europe, but the annual cumulated rainfall is projected to significantly

501 decrease there, in particular in the summer months, which are influencing the

502 rainfall erosivity more than the winter months. We shall also highlight that the

503 patterns described here are shared by most of the climate change scenario

504 models commonly run in climate prediction experiment (for a detailed list see

505 IPCC, 2013; Rajczak et al., 2013).

507 Even though the projections of rainfall erosivity are very plausible and

508 congruent with other climate change studies, they may vary considerable

509 depending on the choice of the scenario (e.g. within the RCP4.5 an increase

510 of temperature in the range 0.9 - 2.0 Celsius degrees till 2046-2065 and

511 compared to other ones). Moreover, the erosivity predictions also include uncertainties originating from the downscaling of GCMs. Besides the climate

513 model uncertainty, the rainfall erosivity predictions embeds the natural

514 variability of climate systems (Harris et al., 2014). Moreover, the rainfall

515 projections have larger degrees of uncertainty compared to temperature

projections because of a higher number of physical models involved and the generally higher variability of rainfall in space (Harris et al., 2014).

The projected erosivity dataset is not challenging any local (or regional) erosivity map which has been developed by using a different methodology or involved local data of better quality. Our erosivity projections were comp ared with the three regional studies modelling long-term R-factor measurements (i.e., Ukkel in Belgium, western Germany and Czech Republic). We observed good agreement in both trends and comparable magnitudes. In Belgium, Verstraeten et al (2006) calculated 31% increase of erosivity during 20th

Century compared to 40% increase that we project in 2050. In western Germany, Fiener et al (2013) observed an R-factor increase of 21% per decade during the period 1973-2007 (overall about 70%). Here, we project for the next 40 years a trend consistent with the local long-term observations (+67%). Regarding the last study, Hanel et al (2015) estimated in Czech Republic an increase of R-factor by 11% per decade (1960-1990) which is also in good agreement with our projection (+68% for the next 40 years).

Along with the quantitative comparison our erosivity projections with local

studies, we performed a further qualitative comparison of our results with

regional studies which have modelled trends of future erosivity. Our results

were compared well with local studies in Sicily and Calabria (Italy), Spain and

North Ireland while the results were different in South Portugal. Similar to our

results, Grauso et al (2010) expect the higher values in the Catania plain and

eastern slope of Iblei mountains while the lowest values are projected in

542 south-east of Palermo. In Sicily, D' Asaro et al (2007) are not expecting an

543 increase of rainfall erosivity in the future. In Calabria, Capra et al (2016)

544 projected a decrease of R-factor, similar to our results (Fig. 5). In Ebro

545 catchment (Spain), Angulo-Martínez and Beguería (2012) reported a

546 decrease of very intense rainfall events but an increased frequency

547 moderate and low events which is close to our future projections (Fig. 5).

548 Similar to our results, Mullan (2013) projected an increase of erosivity in

549 western part of North Ireland (Corrard, Loughmuck) and a decrease of

550 erosivity in eastern part (Dunadry, Hillsborough, Ballywalter). Contrary to our

551 projections, Nunes et al. (2016) show a decrease of erosivity in Portugal (1950552 2008) but an increase in precipitation concentration. In line to our study,

553 Groisman et al (2004) simulated an increase of heavy precipitations in

554 Scandinavia and Northern Europe.

556 4 Conclusions and outlook

557 We modelled the rainfall erosivity in 2050 based on a moderate climate

558 change scenario (HadGEM RCP 4.5) and using as main data sources the

559 REDES based European R-factors and as covariates the WorldClim climatic

560 datasets. Although the rainfall erosivity projections are based on many

561 uncertainties, this pan-European spatial estimation highlights the areas where

562 rainfall erosivity is projected to undergo substantial changes. The prediction of

563 future erosivity in EU can contribute in policies related to soil/land and water

564 sustainable management.

The overall increase of rainfall erosivity in Europe by 18% until 2050 are in line with projected increases of 1 7% for the U.S.A. The predicted R-factor dataset can be used for applying climate change scenarios in soil erosion models. The predicted mean increase in R-factor is expected also to increase the threat of soil erosion in Europe. However, climate change might substantially affect land cover and land use, which might counterbalance or enhance some erosional trends. In order to predict soil erosion trends in the future these feedbacks between rainfall erosivity and land use / land cover need to be considered. The most prominent increases of R-factors are predicted for North-Central Europe, the English Channel, The Netherlands and Northern France. On the contrary, parts of the Mediterranean basin show a decrease of rainfall erosivity.

The Gaussian Process Regression model applied showed a relatively good performance (R2 = 0.635, Relative error = 0.56) based on most of the monthly precipitation covariates of the WorldClim dataset. Despite this study significant contribution towards better understanding of future rainfall erosivity potential in Europe, the results should be in any case handled with care, as it should be commonly done with results derived from CCM and RCM models applied to future scenarios. Future research in climate change modelling will hopefully reduce the intrinsic climate model uncertainty and provide data on better spatial and temporal predictions of rainfall intensity trends. The projected rainfall erosivity (GeoTIFF format) at ~1 km resolution will be available for free download in the European Soil Data Centre (ESDAC): http://esdac.jrc.ec.europa.eu/.

592 Acknowledgments

593 The authors would like to thank Grainne Mulhern for revision of the article from

594 a linguistic point of view.

596 Conflict of interest

597 The authors confirm and sign that there is no conflict of interest with networks

598 organisations and data centres referred to in this paper.

600 References

601 Angulo-Martinez, M., Lopez-Vicente, M., Vicente-Serrano, S.M., Begueria, S.,

602 2009. Mapping rainfall erosivity at a regional scale: a comparison of

603 interpolation methods in the Ebro Basin (NE Spain). Hydrol. Earth Syst. Sci. 13,

604 1907-1920

606 Angulo-Martínez, M.,Beguería, S., 2012. Trends in rainfall erosivity in NE Spain at

607 annual, seasonal and daily scales, 1955-2006. Hydrol. Earth Syst. Sci. 16 (10),

608 3551-3559

610 Arnoldus, H. M. J.: An approximation of the rainfall factor in the Universal Soil

611 Loss Equation, in: Assessment of Erosion, edited by: De Boodt, M. and Gabriels,

612 D., 127-132, Chichester, New York, 1980.

614 Auerswald, K., et al. 2015. Comment on "Rainfall erosivity in Europe" by

615 Panagos et al. (Sci. Total Environ., 511, 801-814, 2015). Sci. Total Environ. 532:

616 849-852.

Ballabio, C. et al. (201 7) Mapping monthly rainfall erosivity in Europe. Sci. Total Environ. 579: 1298-1315

Biasutti M., Seager R. 2015. Projected changes in US rainfall erosivity. Hydrology and Earth System Sciences, 19 (6), pp. 2945-2961.

Journal of Die

Brown, L.C., Foster, G.R., 1987. Storm erosivity using idealized intensity distributions. Transactions of the ASAE 30, 379-386.

Borrelli, P., Diodato, N., Panagos, P. 2016. Rainfall erosivity in Italy: A national scale spatiotemporal assessment. International Journal of Digital Earth, In Press. DOI: 10.1080/17538947.2016.1148203

Capra, A., Porto, P., la Spada, C. 2016. Long-term variation of rainfall erosivity in Calabria (Southern Italy). Theoretical and Applied Climatology, In Press. DOI: 10.1007/s00 704-015-169 7-2.

Christensen O.B., Yang S., Boberg F., Maule C.F., Thejll P., Olesen M., Drews M., (...), Christensen J.H. 2015. Scalability of regional climate change in Europe for high-end scenarios. Climate Research, 64 (1) , pp. 25-38.

D'Asaro, F., D'Agostino, L., Bagarello, V. 2007. Assessing changes in rainfall erosivity in Sicily during the twentieth century. Hydrological Processes, 21 (21), pp. 2862-2871.

643 Dabney, S.M., Yoder, D.C., Vieira, D.A.N. 2012. The application of the Revised

644 Universal Soil Loss Equation, Version 2, to evaluate the impacts of alternative

645 climate change scenarios on runoff and sediment yield. Journal of Soil and

646 Water Conservation, 67 (5) , pp. 343-353.

648 Deelstra J., Oygarden L., Blankenberg A.-G.B., Eggestad H.O. 2011. Climate

649 change and runoff from agricultural catchments in Norway. Internati onal

650 Journal of Climate Change Strategies and Management, 3 (4) , pp. 345-360.

652 Diodato, N., Bellocchi, G. 2010. MedREM, a rainfall erosivity model for the

653 Mediterranean region. Journal of Hydrology, 387 (1-2), pp. 119-127

655 EEA, 2011. Biogeographical regions dataset of European Environment

656 Agency. Accessed from: http://www.eea.europa.eu/data-and-

657 maps/data/biogeographical-regions-europe (June 2014).

659 Fiener, P., Neuhaus, P., Botschek, J. 2013. Long-term trends in rainfall erosivity-

660 analysis of high resolution precipitation time series (1937-2007) from Western

661 Germany. Agricultural and Forest Meteorology, 1 71-172: 115-123

663 Grauso, S., Diodato, N., Verrubbi, V. 2010. Calibrating a rainfall erosivity

664 assessment model at regional scale in Mediterranean area. Environmental

665 Earth Sciences, 60 (8), pp. 1597-1606.

Groisman, P.Ya., Knight, R.W., Easterling, D.R., Karl, T.R., Hegerl, G.C., Razuvaev, V.N. 2005. Trends in intense precipitation in the climate record. Journal of Climate, 18 (9), pp. 1326-1350.

Guerra C.A., Maes J., Geijzendorffer I., Metzger M.J. An assessment of soil erosion prevention by vegetation in Mediterranean Europe: Current trends of ecosystem service provision (2016) Ecological Indicators, 60 , art. no. 2529 , pp. 213-222.

change of precipitation extremes in Europe: Intercomparison of scenarios from regional climate models. Journal of Geophysical Research: Atmospheres,111 (D6).

Hanel, M., Pavlaskova, A., KyselyTrends J. 2015. Trends in characteristics of sub - daily heavy precipitation and rainfall erosivity in the Czech Republic. International Journal of Climatology. DOI: 10.1002/joc.4463

Harris R.M.B., Grose M.R., Lee G., Bindoff N.L., Porfirio L.L., Fox-Hughes P. 2014. Climate projections for ecologists. Wiley Interdisciplinary Reviews: Climate Change, 5 (5) , pp. 621-637.

Hijmans, R.J., Cameron, S.E., Parra, J.L., Jones, P.G., Jarvis, A., 2005. Very high resolu-tion interpolated climate surfaces for global land areas. International Journal of Climatology 25, 1965-1978.

Frei, C., Scholl, R., Fukutome, S., Schmidli, J.,

P. L. (2006). Future

Hellden, U., Tottrup, C., 2008. Regional desertification: a global synthesis. Global and Planetary Change 64, 169-176.

Janecek, M., Kveton, V., Kubatova, E., Kobzova, D., Vosmerova, M., Chlupsova, J., 2013. Values of rainfall erosivity factor for the Czech Republic c. J

Hydrol. Hydromech. 61, 97-102.

Jones, C. D., Hughes, J. K., Bellouin, N., Hardiman, S. C., Jones, G. S., Knight, J., (...), Zerroukat, M. 2011. The HadGEM2-ES implementation of CMIP5 centennial simulations, Geosci. Model Dev., 4, 543-570

Ingber L.1993. Simulated annealing: Practice versus theory. Mathematical and Computer Modelling, 18 (11) , pp. 29-57.

IPCC, 2013: Climate Change 2013: The Physical Science Basis. Contribution of Working Group I to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change [Stocker, T.F., Qin, D., Plattner, G.-K., Tignor, M. (....), Midgley, P.M. (eds.)]. Cambridge University Press, Cambridge, United Kingdom and New York, NY, USA, 1535 pp.

Ito, A. 2007. Simulated impacts of climate and land-cover change on soil erosion and implication for the carbon cycle, 1901 to 2100. Geophysical Research Letters, 34 (9), art. no. L0940

71 7 Kharin V.V., Zwiers F.W., Zhang X., Wehner M. (2013) Changes in temperature

718 and precipitation extremes in the CMIP5 ensemble. Climatic Change, 119 (2) ,

719 pp. 345-357.

721 Kirkpatrick, S., Gelatt, C. D., and Vecchi, M. P. (1983). Optimization by

722 simulated annealing. Science, 220(4598), 671.

724 Kriegsmann A., Martin E., van Meijgaard E., Moseley feifer S.,

725 Preuschmann S., Radermacher C., (...), Kotlarski S. 2014. EL CORDEX: New

726 high-resolution climate change projections for Europ n impact research.

727 Regional Environmental Change, 14(2): 563-57

729 Martin, M., Bellouin, N., Collins, W. J., Culverwell, I. D., Halloran, P. R., Hardiman,

730 S. C, (...), , Wiltshire, A. 2011. The HadGEM2 family of Met Office Unified Model

731 climate configurations, Geosci. Model Dev., 4, 723-757

733 Meusburger K teel A., Panagos P., Montanarella L., Alewell C. 2012. Spatial

734 and tempora ariability of rainfall erosivity factor for Switzerland. Hydrology

735 and Earth System Sciences, 16 (1): 167-177.

737 Milton S.F., Rae J.G.L., Ridley J.K., Sellar A., Senior C.A., Totterdell I.J., Verhoef A., (...), Manners J.C. 2011. The HadGEM2 family of Met Office Unified Model

739 climate configurations. Geoscientific Model Development, 4 (3) , pp. 723-757.

Mueller E.N., Pfister A. 2011. Increasing occurrence of high-intensity rainstorm events relevant for the generation of soil erosion in a temperate lowland region in Central Europe. Journal of Hydrology, 411 (3-4), pp. 266-278.

Mullan, D. 2013. Soil erosion under the impacts of future climate change:

Assessing the statistical significance of future changes and the potential on-

ll erosivity in ry. Journal of

.R. 2004. iew. Journal of So

Nunes, A.N., Lourenço, L., Vieira, A., Bento-Gonçalves, A. 2016. Precipitation and Erosivity in Southern Portugal: Seasonal Variability and Trends (1950-2008). Land Degradation and Development, 27 (2), pp. 211-222

site and off-site problems. Catena, 109, pp. 234-246.

Nearing M.A. 2001. Potential changes in rainfall erosivity in the U.S. with climate change during the 21 st century. Journal of Soil and Water Conservation, 56 (3), pp. 229-232.

Nearing M.A., Pruski F.F., O'Neal M.R. 2004. Expected climate change impacts on soil erosion rates: A review. Journal of Soil and Water Conservation, 59 (1) , pp. 43-50.

Panagos, P., Borrelli, P., Poesen, J., Ballabio, C., Lugato, E., Meusburger, K., Montanarella, L., Alewell, C. 2015a. The new assessment of soil loss by water erosion in Europe. Environmental Science & Policy. 54: 438-447.

Panagos, P., Ballabio, C., Borrelli, P., Meusburger, K., Klik, A., (...), Alewell, C.,

2015b. Rainfall erosivity in Europe. Sci. Total Environ. 511, 801-814.

Panagos, P., et al. (2015c) Reply to the comment on "Rainfall erosivity in Europe" by Auerswald et al. Sci. Total Environ. 532: 853-857

Panagos, P., Ballabio, C., Borrelli, P., Meusburger, K. 2016a. Spatio-temporal analysis of rainfall erosivity and erosivity density in Greece. Catena, 137, pp. 161-172

Panagos, P., Borrelli, P., Spinoni, J., Ballabio, C., Meusburger, K., Begueria, S., (...), Alewell, C. 2016b. Monthly rainfall erosivity: conversion factors for different time resolutions and regional assessments. Water (Switzerland), 8 (4), art. no. 119.

Petan S., Rusjan S., Vidmar A., Mikos M. 2010. The rainfall kinetic energy-intensity relationship for rainfall erosivity estimation in the mediterranean part of Slovenia. Journal of Hydrology, 391 (3-4) , pp. 314-321.

Pruski F.F., Nearing M.A. 2002. Climate-induced changes in erosion during the 21st century for eight L.S. locations. Water Resources Research, 38 (12) , pp. 341-3411.

Rajczak, J., Pall, P., & Schär, C. (2013). Projections of extreme precipitation events in regional climate simulations for Europe and the Alpine Region.Journal of Geophysical Research: Atmospheres, 118(9), 3610-3626.

800 801 802

810 811 812

Renard, K.G., et al., 1997. Predicting Soil Erosion by Water: A Guide to Conservation Planning with the Revised Universal Soil Loss Equation (RUSLE) (Agricultural Handbook 703). US Department of Agriculture, Washington, DC, pp. 404.

Rummukainen M. 2010. State-of-the-art with regional climate models. Wiley Interdisciplinary Reviews: Climate Change, 1 (1) , pp. 82-96.

>dels. Wile

Salvati L., Carlucci M. The impact of mediterranean land degradation on agricultural income: A short-term scenario (2013) Land Use Policy, 32 , pp. 302308.

)13) Land

Sillmann, J., Kharin, V. V., Zwiers, F. W., Zhang, X., & Bronaugh, D. (2013). Climate extremes indices in the CMIP5 multimodel ensemble: Part 2. Future climate projections. Journal of Geophysical Research: Atmospheres, 118(6), 2473-2493.

Journa

Shiono T., Ogawa S., Miyamoto T., Kameyama K. 2013. Expected impacts of climate change on rainfall erosivity of farmlands in Japan. Ecological Engineering, 61 (1 PARTC), pp. 678-689.

St.Clair S.B., Lynch J.P. 2010. The opening of Pandora's Box: Climate change impacts on soil fertility and crop nutrition in developing countries. Plant and Soil, 335 (1) , pp. 101-115.

817 Weaver C.P., Lempert R.J., Brown C., Hall J.A., Revell D., Sarewitz D. 2013.

818 Improving the contribution of climate model information to decision making:

819 The value and demands of robust decision frameworks. Wiley Interdisciplinary

820 Reviews: Climate Change, 4 (1) , pp. 39-60.

822 Westra, S., Fowler, H. J., Evans, J. P., Alexander, L. V., Berg, P., Johnson, F.....&

juency of short -

'22-555.

823 Roberts, N. M. (2014). Future changes to the intensity and frequer

824 duration extreme rainfall. Reviews of Geophysics, 52(3), 522

826 Wischmeier, W., Smith, D. 1978. Predicting rainfall erosion losses: a Guide to

827 conservation planning. Agricultural Handbook No. 537 U.S. Department of

828 Agriculture, Washington DC, USA.

830 van Haren, R., van Oldenborgh, G. J., Lenderink, G., & Hazeleger, W., 2013.

831 Evaluation of modeled changes in extreme precipitation in Europe and the

832 Rhine basin. Environmental Research Letters, 8(1), 014053

dbook N

S., Ra

834 Vasudevan S., Ramos F., Nettleton E., Durrant Whyte H. 2009. Gaussian

835 process modeling of large-scalwe terrain. Journal of Field Robotics, 26 (10) ,

836 pp. 812-840. 838

838 Verstraeten, G., Poesen, J., Demaree, G., Salles, C., 2006. Long-term (105

839 years) variability in rain erosivity as derived from 10 - min rainfall depth data

840 for Ukkel (Brussels, Belgium): implications for assessing soil erosion rates. J.

841 Geophys. Res. 111, D22.

843 Ulbrich U., Lionello P., Xoplaki E., Malanotte-Rizzoli P., Boscolo R., Alpert P.,

844 Artale V., (...), Tsimplis M. , 2006. The Mediterranean climate: An overview of

845 the main characteristics and issues. Developments in Earth and Environmental

846 Sciences, 4 (C) , pp. 1-26.

848 Zhang Y.-G., Nearing M.A., Zhang X.-C., Xie Y., Wei H., 2010. Projected rainfal

d multisce naric (1-2) , pp. 97

849 erosivity changes under climate change from multimodel and

850 projections in Northeast China. Journal of Hydrology, 384 (1-2) , pp. 97-106.

857 Graphical abstract

861 862

Highlights

Rainfall erosivity in Europe & Switzerland is estimated to increase by 18% in 2050

Rainfall erosivity will increase in 81% of the study area and decrease in the rest

R-factor projections include the uncertainty of climatic models Highest R-factor increase is projected in Northern & Central Europe Erosivity is a driver for soil erosion, floods, natural hazards & land use change