Sensitivity of mGROWA-simulated groundwater recharge to changes in soil and land use parameters in a Mediterranean environment and conclusions in view of ensemble-based climate impact simulations.

STOTEN-17736; No of Pages 15 Science of the Total Environment xxx (2015) xxx–xxx

Contents lists available at ScienceDirect

Science of the Total Environment journal homepage: www.elsevier.com/locate/scitotenv

Sensitivity of mGROWA-simulated groundwater recharge to changes in soil and land use parameters in a Mediterranean environment and conclusions in view of ensemble-based climate impact simulations L. Ehlers a,b,⁎, F. Herrmann a, M. Blaschek b, R. Duttmann b, F. Wendland a a b

Forschungszentrum Jülich GmbH, Institut für Bio- und Geowissenschaften, Agrosphäre (IBG-3), D-52425 Jülich, Germany Christian-Albrechts-Universität zu Kiel, Geographisches Institut, D-24118 Kiel, Germany

H I G H L I G H T S • • • •

Future groundwater recharge was simulated using the water balance model mGROWA. Four different GCM–RCM combinations were used as climate forcing. Sensitivity of model output to changes in selected parameters was examined. Simulation results indicate significantly reduced groundwater recharge by 2100.

a r t i c l e

i n f o

Article history: Received 8 January 2015 Received in revised form 25 March 2015 Accepted 30 April 2015 Available online xxxx Keywords: Water balance simulations Groundwater recharge Semi-arid environment mGROWA Climate change Climate projections GCM–RCMs CLIMB project

a b s t r a c t This study examines the impact of changing climatic conditions on groundwater recharge in the Riu Mannu catchment in southern Sardinia. Based on an ensemble of four downscaled and bias corrected combinations of Global and Regional Climate Models (GCM–RCMs), the deterministic distributed water balance model mGROWA was used to simulate long-term mean annual groundwater recharge in the catchment for four 30-year periods between 1981 and 2100. The four employed GCM–RCM combinations project an adverse climatic development for the study area: by the period 2071–2100, annual rainfall will decrease considerably, while grass reference evapotranspiration will rise. Accordingly, ensemble results for our base scenario showed a climate-induced decrease in the median of annual groundwater recharge in areas covered by Macchia from 42–48 mm/a to 25–35 mm/a between the periods 1981–2010 and 2071–2100, corresponding to a reduction of 17–43%. To take into account the influence of additional plant available water storage in weathered bedrock on groundwater recharge generation, the model was extended by a regolith zone for regions covered by Mediterranean Macchia. In a set of model runs (“scenarios”), parameter values controlling the water storage capacity of this zone were increased step-wise and evaluated by comparison to the base scenario to analyze the sensitivity of the model outcome to these changes. The implementation of a regolith zone had a considerable impact on groundwater recharge and resulted in a decrease of the median in annual groundwater recharge: by 2071–2100, the 35% scenario (available water content in the regolith of 3.9 to 5.7 vol.%) showed a reduction of 67–82% as compared to the period 1981–2010 in the base scenario. In addition, we also examined the influence of changes in the crop coefficients (Kc) as well as different soil texture distributions on simulated groundwater recharge. © 2015 Elsevier B.V. All rights reserved.

1. Introduction The Mediterranean region is “among the most responsive regions to global change” (Giorgi, 2006) and expected to be severely affected by ⁎ Corresponding author at: Øster Voldgade 10, DK-1350 Copenhagen, Denmark. E-mail addresses: [email protected] (L. Ehlers), [email protected] (F. Herrmann), [email protected] (M. Blaschek), [email protected] (R. Duttmann), [email protected] (F. Wendland).

advancing water scarcity as a consequence of a projected further decrease in mean annual rainfall as well as an increasing variability of summer precipitation (IPCC, 2007). Over the past decades, an increase in sea surface temperature-linked sea evaporation as well as a drop in precipitation has been registered, which in turn exhibits considerable inter-decadal variations (Mariotti, 2010; Mariotti and Dell'aquila, 2012). Already, droughts represent a serious economic threat to the Mediterranean countries. For instance, the 2003 summer drought caused a reduction of 36% in the net primary production of maize in

http://dx.doi.org/10.1016/j.scitotenv.2015.04.122 0048-9697/© 2015 Elsevier B.V. All rights reserved.

Please cite this article as: Ehlers, L., et al., Sensitivity of mGROWA-simulated groundwater recharge to changes in soil and land use parameters in a Mediterranean environment and..., Sci Total Environ (2015), http://dx.doi.org/10.1016/j.scitotenv.2015.04.122

2

L. Ehlers et al. / Science of the Total Environment xxx (2015) xxx–xxx

the Po valley, Italy (Ciais et al., 2005). Extremely dry conditions affected the entire western Mediterranean between 2006 and 2008 (LopezBustins et al., 2013). The EU-funded research project CLIMB (“Climate-induced Changes on the Hydrology of Mediterranean Basins”, 7th Framework Program) aimed at quantifying the impact of changing climatic conditions on catchment hydrology in a set of study areas located throughout the Mediterranean (Ludwig et al. 2010). The Riu Mannu di San Sperate catchment in southern Sardinia is one of the investigated areas. The catchment is characterized by intensive agricultural activity and has experienced a successive decrease in precipitation and a period of extremely dry summers in recent years. In the frame of the CLIMB project, Mascaro et al. (2013) performed hydrological simulations for the Riu Mannu catchment using the physically based distributed model tRIBS. The authors presented an approach to compensate for scarcity of hydro-meteorological data and to generate the required forcing for their simulations. They used a subset of a historic discharge record (1930–1932) to calibrate and temporally downscale precipitation and potential evapotranspiration to an hourly resolution, obtaining good results with regard to model performance. Furthermore, Piras et al. (2014) used the tRIBS model and assessed the impact of changing climatic conditions on water resources in the Riu Mannu catchment by comparing results for a reference (1971–2000) and future period (2041–2070). Their findings indicate reductions in most water balance components, including mean annual runoff, soil water content and groundwater table. In contrast to the studies cited above, our work focused particularly on the impact of changing climatic conditions on groundwater recharge in the Riu Mannu catchment. We performed water balance simulations for four 30-year periods between 1981 and 2100, using the water balance model mGROWA (Herrmann et al., 2013) and four combinations of Global and Regional Climate Models (GCM–RCMs). To take into account additional plant available water storage in weathered bedrock, the model was extended by a regolith zone for regions covered by Mediterranean Macchia. Several studies indicated that this type of vegetation is able to withdraw water from the regolith zone through tap roots. Unfortunately, the information content of the soil maps available for the Riu Mannu catchment did not suffice to permit a robust, spatially distributed characterization of the parameters controlling the soil water storage capacity for locations covered by Macchia. Since it was assumed that the parameters of the regolith zone have a considerable influence on the water budget, the available water content θa of the regolith zone was increased step-wise in a sequence of model runs (“scenarios”). The results were then evaluated by comparison to a base scenario to analyze the sensitivity of the model outcome to these changes. Moreover, simulations were performed for two sources of pedological information: a soil texture distribution based on a soil map by Aru et al. (1990) and one derived through interpolation of sampled data using compositional kriging (cf. Section 3.2.1). Finally, further simulations were performed for different sets of crop coefficients Kc. It should be noted that the simulation results presented in this study are affected by different sources of uncertainty. There are three major sources of uncertainty (Walker et al., 2003): Firstly, there is input uncertainty which concerns the model forcing data and arises from measurement errors or data inaccuracies. Inaccuracies are for instance a result of generalized, inter- or extrapolated (e.g. precipitation) and spatially aggregated information (land use and soil maps). Secondly, as models necessarily use simplified process descriptions to reproduce physical processes, there is always structural uncertainty of a model. Thirdly, parameter uncertainty concerns the parameter values employed in the model. However, we point out that we did not explicitly quantify the various sources of uncertainties present in this work as this would have exceeded its scope. Instead, we examined the sensitivity of the model outcome to changes in the aforementioned model parameters. A

comprehensive quantitative uncertainty assessment (cf. e.g. Refsgaard et al. (2007)) remains to be conducted in a future study. 2. Study area The Riu Mannu di San Sperate catchment is located in the south of Sardinia and has a drainage area of 472.5 km2 (Fig. 1, top). The Sardinian Agency for Research in Agriculture (AGRIS) operates an experimental farm in Ussana in the south of the catchment, collecting hydrometeorological data and monitoring crop productivity. The topography of the catchment is characterized by a relatively plain area in the western and central part and rugged mountainous terrain in the eastern half of the catchment (Fig. 1, i). Elevation ranges between 69 and 961 m a.s.l., with a mean elevation of 297 m a.s.l. and a mean slope of 8.5°. The catchment has a Mediterranean-type climate (Peel et al., 2007) which is characterized by an extremely dry summer period and a wet season from September to May. The mean annual precipitation is about 500 mm (De Girolamo et al., 2010), with N90% of the annual precipitation falling in winter (Mascaro et al., 2013). The Riu Mannu di San Sperate is a 35 km long river course and has its headwaters in the north of the catchment. The flow regime is dominated by low flows throughout the year (b1 m3/s) and the irregular occurrence of high flows (and potentially overbanking) after extreme rainfall events. The river drains into the Flumini Mannu near Monastir in the south-west of the catchment. As there is currently no discharge gauge, the catchment can be classified as “ungauged” in the sense of Sivapalan (2003). The lack of a discharge record made model validation and calibration to discharge observations impossible (cf. Section 3.3). The economy in the Riu Mannu catchment largely depends on intensive and often irrigated agriculture (Fig. 1, ii). With regard to the agricultural production, durum wheat represents the most frequently cultivated crop (37%). Other land uses include olives, clover, vineyards as well as orchards, alfalfa and corn silage (De Girolamo and Lo Porto, 2012). Over the past decades, the intensification of agricultural activities in combination with ever increasing amounts of fertilizers led to a decline of water quality in the catchment due to nutrient pollution (De Girolamo et al., 2010). The prevailing geological units in the study area are post-Variscan Quaternary and Tertiary sediments, which cover most of the western half of the catchment. Towards the eastern margin of the catchment, metamorphic rocks belonging to the Sarrabus and Gerrei formation as well as a complex of volcanic intrusions can be found (Fig. 1, iii, based on Carmignani et al. (2011)). Natural vegetation is dominated by Mediterranean-type Macchia scrubland, which is a sclerophyllous shrub formation. Typical species include holm oak (Quercus ilex), cork oak (Quercus suber), tree phillyrea (Phillyrea angustifolia L.), wild asparagus (Asparagus acutifolius), lentisk (Pistacia lentiscus L.), juniper (Juniperus phoenicea L.) and wild olive (Olea silvestris) (Detto et al., 2006). Macchia vegetation is well adapted to dry conditions and has high water use efficiency (Marras et al., 2011). 3. Data and methods 3.1. Brief model description The following paragraphs are intended to outline the basic functionalities and equations of the mGROWA model. A comprehensive description of the mGROWA model is given by Herrmann et al. (2013). The water balance model mGROWA was designed for applications in medium-sized to large catchments and administrative units such as Federal States (Herrmann et al., 2013). The model design enables the simulation of groundwater recharge at a high resolution, which represents the main target variable of the mGROWA model. The model resolution is arbitrary and can be tailored to the purpose and scale of investigation. For the present study, a cell size of 50 m was deemed



3

Fig. 1. Location of study area, (i) digital elevation model DEM10, (ii) land-use categories based on CORINE, (iii) geological units.

appropriate, resulting in 188,680 grid cells in the Riu Mannu catchment. For every grid cell, the water balance was calculated. Fig. 2 provides an overview of the input data required to perform water balance simulations as well as the general mGROWA calculation procedure. In order to perform hydrological simulations, spatially distributed grids of precipitation p and grass reference evapotranspiration et0 have to be provided at a daily resolution. The water balance calculation is divided into: (i) a physically based part, resulting in daily values for soil moisture dynamics, actual evapotranspiration eta and total runoff qt, followed by (ii) an empirical part in which total runoff is split up into direct runoff qd (containing surface runoff and natural interflow) and groundwater recharge qr, based on the concept of base flow indices (BFI) (Kunkel and Wendland, 2002). (i) In the physically based part, calculations are performed for every grid cell of the respective model grid using the water balance equation which is denoted as (s storage and t time): ds ¼ p−eta −qt : dt

ð1Þ

The calculation of eta is a function of grass reference evapotranspiration et0, a land-use specific crop coefficient Kc, terrain characteristics (slope and aspect/β, γ) and a storage term s. eta ¼ et0 Kc f ðβ; γÞ f ðsÞ

ð2Þ

Grass reference evapotranspiration et0 is calculated by means of the FAO Penman–Monteith Equation (Allen et al., 1998). The crop coefficient Kc (ibid.) adjusts et0 with regard to different land-use classes, thus accounting for different types of vegetation and crops whose surface characteristics differ from those of the reference grass surface. Hence, they lead to different amounts of actual evapotranspiration. This approach also allows adjusting the Kc values according to the phenological stage of a given type of vegetation (e.g. ground cover, crop height, leaf area). Kc values used in mGROWA simulations were retrieved from the literature, e.g. ATV-DVWK (2002). Furthermore, three calculation modes are implemented in mGROWA, accounting for the characteristic properties of vegetated, sealed and open water surfaces. The water balance calculation for vegetated surfaces is performed by the sub-model BOWAB, a

Fig. 2. mGROWA data basis and modeling scheme, modified after Herrmann et al. (2013).


4


Fig. 3. mGROWA groundwater recharge scheme, modified after Herrmann et al. (2013).

one-dimensional multi-layer soil water balance model (described in Engel et al. (2012)). For these surfaces, s is equivalent to the soil moisture content θ. In BOWAB, the water storage capacity of the soil at a given grid location is dependent on the soil profile depth and the soil texture. In the model, a soil profile is represented by a sequence of model layers of a certain thickness. Each model layer is assigned a soil texture, which in turn has a fixed field capacity θfc and available water capacity θa as defined in Müller and Waldeck (2011). Water uptake by plants from a soil profile is controlled by monthly exhaustion factors, which determine from which depths (i.e. model layers) vegetation withdraws water for evapotranspiration; the depths vary among different types of vegetation and are subject to seasonal variations. The factors also determine the relative contribution of each layer to total eta, i.e. their contributions sum up to 100%. For a more detailed account of the processes which are reproduced through the submodules of the mGROWA model, cf. Herrmann et al. (2013). (ii) As indicated above, the empirically based separation of total runoff qt into direct runoff (containing surface runoff and natural interflow) and groundwater recharge is performed using calibrated base flow indices, BFI (cf. Kunkel and Wendland (2002)). The first assumption underlying the BFI approach is that long-term mean annual base flow is identical to long-term mean annual groundwater discharge which in turn represents a good estimate for groundwater recharge (cf. Risser et al. (2005)). The second assumption is that, in the long-term, the ratio of groundwater

Fig. 4. Soil texture distribution as derived based on compositional kriging of soil separates (left) based on 77 samples (black x-marks) and based on the soil texture map by Aru et al., 1990 (right). The soil texture distribution for the regionalized soil data is based on the German classification system (AD-HOC-AG BODEN, 2005), the labels for the soil texture classes relate to the closest FAO classes (FAO, 2006).


L. Ehlers et al. / Science of the Total Environment xxx (2015) xxx–xxx Table 1 Names and acronyms for Global Climate Models (GCMs) and ENSEMBLES Regional Climate Models (RCMs) as well as GCM–RCM combinations employed in this study. Model name GCMs HadCM3 (Hadley Centre for Climate Prediction, Met Office, UK) ECHAM5/MPI (Max Planck Institute for Meteorology) RCMs RCA (Swedish Meteorological and Hydrological Institute (SMHI), Sweden) REMO (Max Planck Institute for Meteorology, Germany) RACMO2 (Koninklijk Nederlands Meteorologisch Instituut (KNMI), Netherlands)

Acronym HCH ECH RCA REM RMO

GCM–RCM combinations used for water balance simulations: ECH–REM, ECH–RMO, ECH– RCA & HCH–RCA.

recharge to direct runoff remains stable, being a result of the near static nature of the geological structure and surface characteristics in the study area. Hence, BFI values are considered being constant. The equation for the separation of total runoff qt into its components is: qt ¼ B FI qt þ ð1−BFIÞ qt ¼ qr þ qd :

ð3Þ

Fig. 3 displays a scheme illustrating the procedure for the calculation of groundwater recharge in the study area. On impervious surfaces, groundwater recharge equals zero as total runoff is equivalent to direct runoff. On pervious surfaces, groundwater recharge is identical to total runoff if the surface consists of unconsolidated rock (BFI = 1), whereas groundwater recharge on solid rock is the product of the respective BFI value and total runoff. 3.2. Input data 3.2.1. Geospatial data The employed digital elevation model had a resolution of 10 m (DEM10) and was retrieved from the platform Sardegna Geoportale (Regione Autonoma della Sardegna). The terrain parameters slope and aspect required in Eq. (2) were generated based on the DEM10 using ESRI ArcGIS (Version 10.3). The land-use information is based on the CORINE data set (Coordination of Information on the Environment) and had a resolution of 1:100,000. The initial CORINE land-use classes were aggregated into a number of major land use classes (cf. Fig. 1, ii). To each class, a crop coefficient Kc was assigned. The main geological units were determined based on a digital version of a geological map of Sardinia (scale 1:25,000, corresponding to Carmignani et al. (2011)). Pedological information represented the most problematic input data and was hard to come by, which appears to be a general problem in Sardinia (Vacca et al., 2014). The soil map which was used for the simulations corresponded to a printed map created by Aru et al. (1990) (henceforth “soil texture map”) and had a very low spatial

5

resolution of 1:250,000. The map showed delineations of soil regions, each of which was further described in the corresponding map legend. Unfortunately, the description only gave ranges for soil textures typical for a respective soil region, with no indication about which texture was considered to be dominating. As a consequence, several regions with similar descriptions were grouped together and assigned the first soil texture which was stated in the legend description, resulting in four aggregated soil texture classes (Fig. 4, right). These soil texture classes are based on the FAO soil texture classification system (FAO, 2006). Moreover, a relationship between the gradient of the earth's surface and the coarse grain fraction was assumed, resulting in reduced water storage capacities in mountainous regions of the catchment: areas with a slope gradient of 0 to 2° were assigned a value corresponding to 100% of the respective available water content θa as given by Müller and Waldeck (2011), cf. Section 3.1. Slopes with a gradient between 2 and 9° received values amounting to 90% and slopes with a gradient N9° amounting to 75% of the initial values. As an additional source of pedological information, a soil texture distribution derived through the regionalization of the soil separates clay, silt and sand using compositional kriging (CK) was included into the simulations (henceforth “regionalized soil data”). The approach will be outlined in the following paragraph. Note that the boundaries between different soil textures as depicted in Fig. 4 (left) were derived based on the German classification system (AD-HOC-AG BODEN, 2005), which uses a 63 μm threshold between sand and silt. The labels for the soil textures were assigned the closest term of the FAO taxonomy (FAO, 2006). A compositional kriging approach was applied to map clay, silt and sand content of the top 30 cm soil layer. CK is a spatial interpolation technique that respects the constant sum and non-negativity constraints of compositional data such as soil textural fractions by adding these restrictions to the standard kriging system (De Gruijter et al., 1997; Walvoort and De Gruijter, 2001). A total of 77 soil samples were collected during three field campaigns since 2010 and lab-analyzed to determine their grain-size distribution using the sieve–pipette method in accordance with Köhn (Gee and Or, 2002). Efficiency of the CK-model is evaluated by root mean squared errors (RMSE) and coefficients of determination (R2). Both accuracy measures are calculated from leave-one-out cross-validation errors considering each of the 77 observed values. The multivariate prediction of soil textural fractions reached an explained variance level of 47% (RMSE = 9.9) with respect to clay and 39% (RMSE = 13.6) regarding sand. However, the prediction of silt remained rather weak with an R2 value close to 0 and an RMSE of 7.4. Note that the overall variation is considerably lower among the measured silt values as compared to the observation for clay and sand. Thus, the lower RMSE value for silt does not indicate a more accurate model. All calculations regarding CK were done using functions from the R-compositions package (Van Den Boogaart et al., 2014). As Fig. 4 illustrates, there are considerable differences in the soil texture distributions. The soil texture distribution based on the soil texture

Fig. 5. Boxplots of annual grass reference evapotranspiration for the employed GCM–RCM combinations, 30-year hydrologic periods 1981 to 2100. Outliers (circles) represent values outside 1.5 times the IQR.


6


Fig. 6. Boxplots of annual precipitation for the employed GCM–RCM combinations, 30-year hydrologic periods 1981 to 2100. Outliers (circles) represent values outside 1.5 times the IQR.

map by Aru et al. (1990) (right) clearly reflects the main geological units as well as topographic characteristics and consists of two dominating texture classes (clay loam and sandy clay loam). In contrast, the soil texture distribution obtained through the application of compositional kriging to a set of 77 samples collected in the catchment (left) is more diverse and exhibits several more texture classes ranging from sandy loam to clay. The resulting boundaries between texture classes are smoother than their soil texture map counterparts, being a result of the CK interpolation. The most relevant difference also with regard to the water balance simulation is the extensive clay patch in the northwest of the study area, which implies a considerably reduced permeability of the soil layers with regard to the soil moisture dynamics. To account for reduced permeability of clay, the BFI values for clay were lowered for the regionalized soil data. 3.2.2. Climate input In order to study future changes in groundwater recharge qr, mGROWA simulations were performed for the period 1981 to 2100. Long-term mean annual groundwater recharge qr was calculated for four 30-year periods (1981–2010, 2011–2040, 2041–2070 and 2071–2100). The climate input employed to perform the simulations (spatially distributed precipitation p and grass reference evapotranspiration et0), was generated at an earlier stage of the CLIMB project (Deidda et al., 2013). It consists of four different GCM–RCM combinations (cf. Table 1), which belong to a set of 14 RCMs originally generated in the course of the EU-FP6 ENSEMBLES project (Van Der Linden and

Mitchell, 2009). All model combinations implemented the A1B IPCC emission scenario and had a resolution of app. 25 km. The RCMs were validated based on regular grids of spatially interpolated daily precipitation and surface temperature (“E-OBS”, described in Haylock et al. (2008)), resulting in the selection of the four best-performing (with regard to the Riu Mannu catchment) GCM–RCM combinations (Deidda et al., 2013). This final set of GCM–RCMs consisted of two different Global Climate Models driving the same Regional Climate Model and one Global Climate Model forcing two different Regional Climate Models. The selected RCMs were subsequently downscaled using a multi-fractal technique (Deidda, 1999, 2000; Deidda et al., 1999) to increase their spatial resolution and facilitate climate change impact analyses on a local scale. The results were then utilized for the mGROWA simulations in the Riu Mannu study area. Each of the four employed GCM–RCM combinations draws a different picture of the projected changes regarding the tempo-spatial development as well as statistical behavior of the climatic variables grass reference evapotranspiration et0 and precipitation p. In order to illustrate both similarities and deviations between these scenarios, Figs. 5 and 6 show boxplots for annual values of these climatic elements over the four examined hydrologic periods between 1981 and 2100. Fig. 5 shows that the GCM–RCM combinations agree in projecting a steady increase of the median (black horizontal bar) of annual grass reference evapotranspiration et0 in the study area, throughout the four investigated 30-year periods between 1981 and 2100. The figure illustrates that the GCM–RCM combinations vary with regard to the magnitude of

Table 2 Parameter scenarios (ID) for which mGROWA simulations were performed. For each scenario, simulations were carried out for all four GCM–RCM combinations. n = number of simulated model layers. dmax = maximum depth for plant water uptake for eta in cm. ID

n

Denotation

dmax

Description of modification

Pedological information

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19

4 10 10 10 10 10 10 10 10 4 10 10 10 10 10 10 10 10 4

Base scenario 5% scenario 15% scenario 25% scenario 35% scenario 45% scenario 55% scenario 65% scenario 75% scenario Base scenario 5% scenario 15% scenario 25% scenario 35% scenario 45% scenario 55% scenario 65% scenario 75% scenario Kc adj

120 300 300 300 300 300 300 300 300 120 300 300 300 300 300 300 300 300 120

Soil texture map Soil texture map Soil texture map Soil texture map Soil texture map Soil texture map Soil texture map Soil texture map Soil texture map Regionalized soil data Regionalized soil data Regionalized soil data Regionalized soil data Regionalized soil data Regionalized soil data Regionalized soil data Regionalized soil data Regionalized soil data Soil texture map

20 21

4 4

Kc−0.1 Kc +0.1

120 120

No regolith zone implemented, water storage limited to soil profiles, default setup Regolith zone implemented, available water content of the regolith zone set to 5% of the topsoil Regolith zone implemented, available water content of the regolith zone set to 15% of the topsoil Regolith zone implemented, available water content of the regolith zone set to 25% of the topsoil Regolith zone implemented, available water content of the regolith zone set to 35% of the topsoil Regolith zone implemented, available water content of the regolith zone set to 45% of the topsoil Regolith zone implemented, available water content of the regolith zone set to 55% of the topsoil Regolith zone implemented, available water content of the regolith zone set to 65% of the topsoil Regolith zone implemented, available water content of the regolith zone set to 75% of the topsoil No regolith zone implemented, water storage limited to soil profiles, default setup Regolith zone implemented, available water content of the regolith zone set to 5% of the topsoil Regolith zone implemented, available water content of the regolith zone set to 15% of the topsoil Regolith zone implemented, available water content of the regolith zone set to 25% of the topsoil Regolith zone implemented, available water content of the regolith zone set to 35% of the topsoil Regolith zone implemented, available water content of the regolith zone set to 45% of the topsoil Regolith zone implemented, available water content of the regolith zone set to 55% of the topsoil Regolith zone implemented, available water content of the regolith zone set to 65% of the topsoil Regolith zone implemented, available water content of the regolith zone set to 75% of the topsoil No regolith zone implemented, initial Kc values for Macchia and Agriculture were exchanged by literature values, Table 3 The above Kc values were decreased by 0.1, Table 3 The above Kc values were increased by 0.1, Table 3

Soil texture map Soil texture map



the projected increase as well as the width of the interquartile range (IQR), which is an indicator for variability. The changes in the location of the median in Fig. 6 illustrate that all employed GCM–RCM combinations project a decrease in annual precipitation p. As observed for the grass reference evapotranspiration et0, the combinations vary with regard to the magnitude of the projected decrease and the width of the IQR. 3.3. Model calibration and validation As pointed out in Section 2, the Riu Mannu catchment can be categorized as “ungauged” (Sivapalan, 2003) due to the lack of adequate discharge observations. The only existing discharge observations were a historic discharge record covering the period 1925–1935. A comprehensive calibration of mGROWA parameters based on this record and the subsequent application of the resulting calibrated model to the 1981–2100 period was unsuitable for the following reasons. Documents from the Italian Hydrologic Survey indicate that the quality of the discharge measurements was impaired repeatedly during the years 1925–1935. Also, large parts of the originally swampy catchment were meanwhile converted into intensively irrigated agricultural land, including the installation of drainage systems (Mascaro et al. (2013)). Thus, there were significant changes in the hydrological response of the catchment due to the anthropogenic interventions during the 20th century. The lack of accurate historic data concerning land-use and water table depths made a calibration of mGROWA parameters related to actual evapotranspiration and total runoff generation pointless. Likewise, a validation of mGROWA simulation results (total runoff and groundwater recharge) based on low-quality historical discharge seemed unadvisable. As a consequence, Kc values required for the simulation of actual evapotranspiration were taken initially from literature (ATV-DVWK, 2002) and varied in the frame of the sensitivity analysis for selected vegetation types (Table 3). The only parameter set that was calibrated based on the historical discharge record were the geology-specific BFI values, as the natural

7

geological setting of the Riu Mannu catchment can be regarded as static and thus the ratio of groundwater recharge and total runoff is assumed to be constant over long periods, i.e. the last century. The calibration was conducted as follows: The initial BFI values for the geological units were assigned using empirical values for similar geological units found in comparable studies (Karpuzcu et al., 2008). These values were modified until the weighted mean BFI value for all geological units was in the range of acceptable values obtained using a base flow separation program developed by Arnold et al. (1995), Arnold and Allen (1999). By applying this program, the BFI value for the whole catchment was obtained to be ca. 0.4. As result of the calibration procedure, the hydro-geological unit of unconsolidated sediments was assigned a BFI value of 1, volcanicsedimentary successions were given a BFI of 0.4, while metamorphic rocks (0.15) and volcanic rocks (0.1) were assigned low BFI values, corresponding with their reduced permeability (compare Fig. 1, (iii)). An exemplary local validation of the simulated soil moisture was performed based on soil moisture sensor data gathered at the AGRIS experimental farm located in the south of the Riu Mannu catchment. The record covered a period of 11 months between 2011 and 2012. The model performance with regard to the agreement of simulated soil moisture and soil moisture recorded at the sensor location was assessed by computing the Nash–Sutcliffe Efficiency (Nash and Sutcliffe, 1970). According to Moriasi et al. (2007), the resulting value of 0.75 indicated a satisfying performance of the soil water balance module implemented in mGROWA. 3.4. Parameter variations We performed mGROWA simulations for a base scenario and a succession of scenarios during which we gradually varied selected mGROWA parameters. In total, simulations for 21 different scenarios were performed (cf. Table 2 for an overview). Herein, the base scenario represented the initial mGROWA parameter setup; for this scenario, the simulated soil water storage extended to

Fig. 7. Modified monthly exhaustion depths for Macchia vegetation.


8


Table 3 Initial monthly Kc values (IDs 1–18, from ATV DVWK 2002) and adjusted monthly Kc values (ID 19, Table 2) for land-use classes Macchia and Agriculture. ID

Land-use class

Kc

Jan

Feb

Mar

Apr

May

Jun

Jul

Aug

Sep

Oct

Nov

Dec

1–9, 10–18 19 1–9, 10–18 19

Macchia

init. adj. init. adj.

1.173 1.15 0.733 0.7

1.173 1.15 0.733 0.7

1.173 1.15 0.744 0.9

1.264 1.15 0.947 1.1

1.211 1.15 1.188 1.2

1.280 1.15 1.181 0.9

1.280 1.15 1.185 0.7

1.294 1.15 1.151 0.7

1.294 1.15 0.974 0.7

1.250 1.15 0.853 0.7

1.238 1.15 0.775 0.7

1.173 1.15 0.733 0.7

Agriculture

120 cm depth (four model layers of 30 cm thickness) and did not consider additional water storage in the regolith. This depth also defined the maximum depth for plant water uptake for evapotranspiration (parameter dmax, Table 2). In the scenarios that considered a regolith zone (IDs 2–9 and 11–18), the simulated soil water storage reached to a depth of 300 cm (ten model layers of 30 cm thickness). The variations of the available water content θa of the regolith were intended to illustrate the sensitivity of simulated groundwater recharge qr towards changes in the respective parameter by comparison with the base scenario. The parameters subject to variations were: i) the available water content θa of a newly implemented regolith zone beneath Macchia vegetation (along with adjusted exhaustion factors) and ii) the K c values for the land-use classes Macchia and Agriculture. We focused on Macchia and Agriculture, because they represent two major land-uses in the catchment and are hence important with regard to groundwater recharge. All scenarios were simulated for both regionalized soil data and the soil texture map. The procedure for the parameter variations will be outlined in detail in the following Sections 3.4.1 and 3.4.2.

3.4.1. Implementation of a regolith zone under Macchia vegetation By implementing a regolith zone, we took into account that weathered bedrock can contain considerable amounts of plant available water (Jones and Graham, 1993; Katsura et al., 2006). For instance, Jones and Graham (1993) report available water capacities for granite in the range of 1–7%. As Sardinian granitic rocks exhibit a profound weathered mantle (Barrocu (2007)), we assumed this to be valid for the granitic rocks present in the Riu Mannu catchment. As a result, the regolith zone was parameterized as follows: we increased the depth of the simulated soil water storage from 120 cm to 300 cm (following Witty et al. (2003)). Within this storage, the regolith zone reached from the bottom of the respective soil profile down to 300 cm depth. In a sequence of eight scenarios (ID 2–9 and ID 11–18 respectively, Table 2), the regolith was assigned a step-wise increasing percentage (from 5% to 75%, with 10% increments) of the available water content θa of the overlying soil texture. Herein, the 5% and 75% scenario defined the lower and upper bound of the range we considered plausible. Each scenario thus represented a different composition of the regolith zone: the 5% scenario stood for little weathered bedrock with few cracks, a high coarse grain fraction content and hence a low water

Fig. 8. Differences in long-term mean annual groundwater recharge on Macchia between base scenario (ID 10) and the 5 to 75% scenarios (IDs 11–18, Table 2) using regionalized soil data. Hydrologic period 1981–2010, ECH–REM combination.



9

Fig. 9. Comparison of long-term mean annual groundwater recharge for the 35% scenario using regionalized soil data (top, ID 14, Table 2) and the soil texture map (bottom, ID 5, Table 2).

storage capacity; the 75% was the most extreme scenario representing a strong degree of weathering resulting in a high water storage capacity. Moreover, we regarded the 35% scenario to be the most plausible (resulting in available water contents in the regolith between 3.9 and 5.7 vol.% depending on the soil texture and thus within the range reported by Jones and Graham (1993)) and will refer to it as such in the remainder of this paper. In addition, we changed the exhaustion factors (cf. Section 3.1) for the land-use class Macchia to allow this type of vegetation to consume water from the newly implemented regolith zone in the simulations. The exhaustion factors were modified based on indications found in the literature: David et al. (2007) report that Macchia is able to maintain a constant eta during the summer months (≥ 1 mm/d) through its drought resistance and the ability to access deep lying water resources

by tap roots. Also, Rambal (1984) reports a shift in water uptake to deeper zones for Quercus coccifera as an adaptation strategy to the drying up of upper soil layers in the course of the summer. This is reproduced in our simulations by adjusting the exhaustion depths as depicted in Fig. 7.

3.4.2. Use of adjusted crop coefficients As outlined in Section 3.1, the initially employed Kc values (IDs 1 to 18, Table 2) for the land-uses Macchia and Agriculture (cf. Table 3) were adopted from ATV-DVWK (2002). These values represent large-scale average surface characteristics and compositions of the vegetation cover, respectively. Thus, their usage in a meso-scale catchment and in a semi-arid environment required a critical review.

Table 4 Changes in long-term mean annual groundwater recharge for different land-use categories. 35% scenario using the soil texture map (ID 5) and regionalized soil data (ID 14, Table 2). ECH– REM combination. ID

Land-use category

% of total area

5 14 5 14 5 14 5 14 5 14

Agriculture

41.2

Macchia

21.7

Vineyards, olives, permanent crops

12.6

Forest

7.1

Natural grassland

6.0

1981–2010

2011–2040

2041–2070

2071–2100

mm

(%)

mm

(%)

mm

(%)

mm

(%)

72.1 56.6 26.9 26.8 62.1 58.8 47.4 47.0 58.6 57.2

(100) (100) (100) (100) (100) (100) (100) (100) (100) (100)

64.4 50.8 24.3 24.2 56.0 53.2 43.6 43.2 53.8 52.5

(89.4) (89.8) (90.3) (90.3) (90.2) (90.5) (91.9) (92.0) (91.7) (91.8)

59.0 49.9 23.1 21.8 53.9 52.0 42.0 41.3 51.8 50.2

(81.8) (88.1) (85.7) (81.3) (86.8) (88.4) (88.6) (88.0) (88.6) (88.0)

36.1 28.3 12.2 12.3 31.6 30.1 28.3 28.3 36.7 36.0

(50.0) (50.0) (45.5) (45.8) (50.9) (51.1) (59.8) (60.2) (59.8) (60.2)

∑ = 88.6%


10


Kc values for the land-use classes Macchia and Agriculture could be obtained from publications with a closer relation to the actual vegetation in the model domain: The Kc values for Macchia were taken from Goldhamer (1989), who suggests a constant value of 1.15 for evergreen shrubbery in a Mediterranean climate. The Kc value for Agriculture was

taken from Lovelli et al. (2010), who adjusted the original FAO Kc values for wheat based on observed changes in the Mediterranean climate. We regarded the exchange of the Kc values for Agriculture by values for wheat as being justified because durum wheat is the most frequently cultivated crop in the study area (De Girolamo and

Fig. 10. Temporal and spatial changes in long-term mean annual groundwater recharge for all four GCM–RCM combinations, 30-year periods, 1981–2100. 35% scenario using regionalized soil data (ID 14, Table 2).



Lo Porto, 2012). Since Bassu et al. (2009) report that the sowing date for durum wheat in a Mediterranean environment is usually at the onset of the rainy season between October and December, we shifted the Kc values obtained from Lovelli et al. (2010) back by two months: as a consequence, crop maturity is simulated in May (cf. Table 3, Kc value of 1.2) instead of July. Finally, we performed simulations for two more scenarios (Kc −0.1 and Kc +0.1, ID 20 and 21, Table 2) for which we reduced, respectively increased the Kc values by 0.1. As before, the variations were intended to give an impression of the sensitivity of simulated groundwater recharge towards changes in the Kc values. 4. Results and discussion In the following sections, the impacts of the modifications to the mGROWA parameter setup with regard to simulated groundwater recharge qr will be presented and discussed. 4.1. Impact of the implementation of a regolith zone under Macchia vegetation 4.1.1. Spatial distribution of long-term mean annual groundwater recharge Firstly, the impact of the implementation of a regolith zone under Macchia vegetation on long-term mean annual groundwater recharge qr will be illustrated. In this context, Fig. 8 visualizes the changes in long-term mean annual groundwater recharge q r between the base scenario (ID 10, Table 2) and the 5% to 75% scenarios (IDs 11–18, Table 2). The underlying pedological information is regionalized soil data and the model was forced by the ECH–REM climate model combination. The first observation is that the 5% scenario yields a slight increase of groundwater recharge under Macchia vegetation. This is somewhat unexpected and appears to be a technical issue: although the water storage capacity was extended by a regolith zone, characterized by 5% of the available water content θa of the overlying soil, the simulations resulted in decreased water consumption for Macchia vegetation as compared to the base scenario (ID 10) and hence an (marginal) increase in groundwater recharge. The reduced consumption is a result of the changes to the exhaustion factors which control from which model layer and how much water is withdrawn for evapotranspiration. In relation to the small total water storage capacity in the regolith layers, the assigned exhaustion factors were assumed too high in this scenario

11

(e.g. in August, compare Fig. 7). In the case of comparatively wet summers, this led to unconsumed water in the upper layers and explains the observed slight increase. In contrast, the 15 to 75% scenarios (IDs 11–18, Table 2) lead to reduced values for long-term mean annual groundwater recharge (Fig. 8). This observation is in line with our expectations: the incorporation of a regolith zone creates additional water storage volume. This additional storage has to be filled up with water before it can become groundwater by exiting the lowermost (i.e. 10th) model layer. As the modification of Macchia exhaustion factors enables water uptake from the entire range of model layers, more water is consumed by the evapotranspiration as compared to the base scenario and hence groundwater recharge is reduced. The modeling results show considerable sensitivity towards changes in the available water content of the regolith zone. In view of the considerations presented in Section 3.1, the 35% scenario is considered the most plausible scenario. For the 35% scenario, long-term mean annual qr with regard to the base scenario is reduced by 10–20 mm/a in large parts of the areas covered by Macchia. In some patches, the reduction of groundwater recharge amounts to up to 40 mm/a. Given the comparatively low groundwater recharge under Macchia (see for instance Fig. 9), these reductions are indeed considerable. Furthermore, we exemplarily compared the effect of different pedological input data on groundwater recharge in the study area for the 35% scenario. This comparison is presented in Fig. 9 for the ECH–RMO climate model combination and all 30-year hydrologic periods from 1981 to 2100. Since both scenarios have the same climate forcing, the results agree in projecting an overall decline in long-term mean annual qr by the period 2071–2100. However, there are several regions which exhibit clear differences in long-term mean annual groundwater recharge qr. These deviations obviously reflect the respective underlying soil texture distribution. It appears that the impact of the different sources of pedological information on qr in the eastern half of the catchment is rather marginal, as these areas show almost identical values for qr. In western half of the catchment, the usage of the soil texture map (bottom) results in considerably higher qr towards the northern fringe of the study area as compared to those obtained using regionalized soil data (top). A look at the respective soil texture distributions (Fig. 4) reveals that the regionalized soil data exhibits a clay dominated area in the north of the catchment. The limits of this region can be clearly distinguished by the abrupt change in qr with regard to the adjacent regions.

Fig. 11. Monthly sums of groundwater recharge on Macchia covered areas for base, 5%, 35% and 75% scenarios using regionalized soil data (IDs 10, 11, 14, 18, cf. Table 2), ECH–RMO combination, years 1951 to 2100.


12


In comparison, the soil texture based on the soil texture map is characterized by a lower clay content in this area. As a consequence of the low BFI values in the clayey area, qr obtained using regionalized soil data is considerably lower in this region than qr obtained based on the soil texture map: qr in the clay dominated area amounts to values in the range 10–60 mm/a, which is similar to values obtained for mountainous regions in the eastern part of the catchment. Values exceeding 80 mm are limited to a few patches located in the central and southern part of the study area as well as a region at the very top of the catchment. In contrast, qr obtained based on the soil texture map exceeds 80 mm/a in a large part of the north. In order to get a better grasp of how different sources of soil information affect long-term mean annual groundwater recharge in the catchment, Table 4 compares qr for the five most frequent land-use categories in the study area. In agreement with the observations from Fig. 9, it can be noted that the deviation in qr caused by different sources of soil information is relatively small. For example, differences in long-term mean annual qr range from 0 to 1.5 mm/a in the period 2071–2100 for most land-use categories. An exception represents Agriculture, for which the difference amounts to 7.8 mm/a in the period 2071–2100. A look at the land-use in Fig. 1 and the soil texture

distribution in Fig. 4 suggests that this difference is, too, a result of the adjustment of BFI values in the clay dominated area of the regionalized soil data. In conclusion, it appears that the impact of using different sources of pedological information on groundwater recharge is moderate to low. Yet, given the highly generalized information derived from the soil texture map, we consider the regionalized soil data to be preferable, as its derivation is based on a set of 77 soil samples and the application of a geostatistical approach (CK). Furthermore, Fig. 10 displays the spatial distributions of long-term mean annual qr in the catchment for the 35% scenario and regionalized soil data (Table 2, ID 14) as well as all four GCM–RCM combinations. When examining the results for the period 2071–2100, the most distinct differences between the GCM–RCM combinations are found in the western part of the catchment which is characterized by low lying, plain terrain, deep soils and intensive agricultural activity. Also, it can be observed that the HCH–RCA combination is the moistest scenario, with qr mainly being in the range of 40–60 mm/a in the southwest of the study area. Regions with zero recharge are limited to areas with shallow soils and impermeable bedrock In contrast, ECH–RCA represents the driest scenario, with qr values between 0 and 20 mm/a occurring in

Fig. 12. Box plots of annual groundwater recharge on Macchia in the base, 5%, 35% and 75% scenarios using regionalized soil data (IDs 10, 11, 14, 18, Table 2), all GCM–RCM combinations, four 30-year periods 1981–2100. Outliers (circles) represent values outside 1.5 times IQR.



almost the entire catchment and in regions with intensive agricultural production. The ECH–REM and ECH–RMO scenarios yield qr values which lie in between the two more extreme scenarios. Despite local differences, the results for qr for all GCM–RCM combinations collectively point towards an overall decrease in long-term mean annual qr by the period 2071–2100. 4.1.2. Monthly sums of groundwater recharge Another possibility to display how groundwater recharge qr under Macchia is affected in the different scenarios for water storage in the regolith zone is presented by Fig. 11. The representation of groundwater recharge in a raster hydrograph allows examining both inter-annual and intra-annual variability of the data and follows an idea of Strandhagen et al. (2006) and Koehler (2004). Displayed are monthly sums of simulated groundwater recharge under Macchia vegetation for each month of the water year (x-axis) over the years 1951–2100 (y-axis) and based on the ECH–RMO combination. In the base scenario (Fig. 11, left), the onset of groundwater recharge is between September and October, for almost the entire period between 1951 and 2100. November, December and January are the months during which groundwater recharge occurs most intensively, reaching and sometimes exceeding values of 40 mm per month. With the beginning of the summer period in May, groundwater recharge ceases and eventually stops. When comparing both the number of months with recharge N0 mm and the monthly recharge amounts over the entire period, it can be noted that there is an evident decrease in groundwater recharge, particularly towards the year 2100. In comparison to the base scenario, results obtained for groundwater recharge in the 5% scenario (Fig. 11, center-left) show only small differences. However, it appears that the modifications have resulted in a slight increase in the number of months with recharge N0 mm/a. As discussed before, the parameter modifications did not have the anticipated effect in this scenario (cf. Section 4.1.1). In line with our expectation, considerable differences become apparent when looking at groundwater recharge obtained for the 35% scenario (Fig. 11, centerright). This scenario leads to a distinct decrease in qr both with regard to the number of months with recharge N 0 mm as well as the absolute amounts per month and causes a shift of the onset of groundwater recharge towards December. Finally, the 75% scenario (Fig. 11, right) displays dramatically reduced qr, in terms of monthly totals as well as the number of months and years with zero qr. In this scenario, qr rarely exceeds or even reaches 20 mm. However, the 75% scenario represents

13

the most extreme of all simulated scenarios and is likely to be unrealistic.

4.1.3. Boxplots of annual groundwater recharge While the previous paragraphs demonstrated how variations of the water storage capacity under Macchia affected the spatial distribution of simulated groundwater recharge in the catchment, Fig. 12 displays boxplots for annual groundwater recharge in areas covered by Macchia for four hydrologic periods between 1981 and 2100 and each of the employed climate scenarios. As before, the regionalized soil data served as pedological input. Two general observations can be made: medians of annual qr decrease both i) over time, i.e. from the first period 1981–2010 to the last period 2071–2100 and ii) from the base scenario to the 75% scenario. When looking closer at the results obtained for the base scenario (Fig. 12, left), one can observe that the medians for annual groundwater recharge under Macchia lie between 42 mm/a (HCH–RCA) and 48 mm/a (ECH–RCA) in the period 1981–2010. In comparison, the corresponding values for the period 2071–2100 are markedly lower: the medians lie between 25 and 35 mm/a, which is equivalent to a reduction in annual qr by 17–43%, depending on the respective forcing. This reduction is solely due to the employed GCM–RCM combinations and thus an illustration of the impact of changing climatic conditions on groundwater recharge for locations covered by Macchia. In order to determine the impact of the implementation of a regolith zone and the step-wise variation of the available water content θa on groundwater recharge, a comparison of the medians between the base scenario and the 35% scenario is worthwhile. In the period 1981–2010, the 35% scenario yields qr in the range of 22–30 mm/a, which corresponds to a distinct reduction of 38–48%. Moreover, by the period 2071–2100, medians for qr based on the 35% scenario lie between 8 and 14 mm/a, equivalent to reductions between 67 and 82% as compared to the 1981–2010 base scenario. These results emphasize the following: the implementation of a regolith zone significantly affects simulated groundwater recharge; provided that the 35% scenario represents a plausible parameterization of this zone, the reduction of qr as compared to the base scenario (38–48%) is larger than the reduction resulting from changing climatic conditions until the period 2071–2100 (17–43%), thus illustrating the sensitivity of the model outcome towards these changes. This, in turn, underlines the importance of acquiring supplementary field data in order to support a realistic parameterization of the regolith zone for future simulations.

Fig. 13. Box-plots of annual groundwater recharge on Macchia and agricultural land (wheat) using different Kc value configurations, based on the soil texture map (IDs 1, 19, 20, 21, Table 2). All GCM–RCM combinations, hydrologic period 1981–2010. Outliers (circles) represent values outside 1.5 times IQR.


14


4.2. Impact of using adjusted crop coefficients Finally, Fig. 13 exemplarily compares median and IQR of annual groundwater recharge for Macchia and Agriculture based on different crop coefficient scenarios (IDs 1, 19, 20 and 21, Table 2). Pedological information was the soil texture map, the hydrologic period presented is 1981–2010. An examination of the resulting changes in groundwater recharge under Macchia (Fig. 13, upper plot) reveals that the impact of these modifications for this type of vegetation is rather small. In the base scenario (ID 1, Table 2), the median for annual qr under Macchia lies between 43 and 49 mm. In comparison, the corresponding medians in the scenario with modified Kc values (ID 19, Table 2) indicate a slight increase (smaller than 3 mm) of qr, corresponding to an increase of annual qr in the range of 2–7%. A reduction of these Kc values by 0.1 (ID 20, Table 2) results in a more notable increase in qr between 7 and 13%. In contrast, when employing Kc values that are increased by 0.1 (ID 21, Table 2), qr is only very slightly affected, showing a reduction from 0 to 2%. This is an indication that water availability for consumption by vegetation is limited, thus preventing a more pronounced decrease of qr. When looking at the effect of the modifications to the Kc values for Agriculture (Fig. 13, bottom), one can observe a stronger impact on annual groundwater recharge qr. From the base scenario (ID 1) to the model run using adjusted Kc values (ID 19), the medians for annual groundwater recharge increase by 6–10 mm for the different GCM– RCM combinations, resulting in an increase of annual qr in the range of 7–14%. Moreover, when looking at the −0.1 scenario, a very distinct increase in qr of 23–33% can be observed. As seen before for areas with Macchia vegetation, qr in the scenarios varied by 0.1 (IDs 20 and 21) is not as much affected, showing a reduction of annual qr by 1–9%. In summary, it can be noted that the impact on groundwater recharge when using adjusted Kc values for Macchia and Agriculture is relatively small. This is mainly a result of the fact that the initial Kc values for Macchia and Agriculture (ATV-DVWK, 2002) and the adjusted Kc values showed a strong resemblance (cf. Table 3) and is hence no surprise. As the adjusted Kc values represent values which were recommended for an application in Mediterranean environments, this observation adds support to the assumption that the initially employed ATV-DVWK (2002) values may also be appropriate for an application in the Riu Mannu catchment. From the observations made in the −0.1 and +0.1 scenario, it becomes clear that the Kc values do indeed represent a parameter that has a significant impact on simulated groundwater recharge.

configuration for the regolith zone and hence recommend its use if actual field data is not available. With regard to the exchange of monthly Kc values for the land-use classes Agriculture and Macchia, we observed that the changes in groundwater recharge when using Kc values based on ATV-DVWK (2002) and those obtained from two other sources were rather small but more significant for agricultural areas. In conclusion however, the comparatively small differences in the medians of simulated annual groundwater recharge suggest that the initially employed ATV-DVWK (2002) values represent a reasonable choice also for an application in a Mediterranean environment. However, another interesting observation was that the increase of Kc values only slightly affected groundwater recharge. In contrast, the scenarios employing decreased Kc values resulted in distinctly higher groundwater recharge, especially for agricultural areas. This demonstrates that the Kc values do in fact have a considerable impact on simulated groundwater recharge. Furthermore, the use of two different sources of soil texture distributions (soil texture map based on Aru et al. (1990)/regionalized soil data) had a rather low impact on long-term mean annual groundwater recharge spatially averaged for several land-use classes. Given the high level of generalization and simplification with regard to the soil texture map (Aru et al., 1990), we note that we consider the regionalized soil data to be preferable, as its derivation is based on a set of 77 soil samples and the application of a state-of-the-art interpolation technique (compositional kriging). Finally and from a more general perspective, it can be summarized that the adverse climatic development (decrease in annual rainfall in the order of 100 mm, increase in et0 larger than 50 mm by the period 2071–2100) as projected by the employed GCM–RCM combinations results in a considerable decrease in simulated long-term mean annual groundwater recharge in the Riu Mannu catchment by the period 2071–2100. This development is likely to have implications for water security in the catchment. The results presented in this study can provide valuable orientation for decision makers in the local water management. Acknowledgments This study received funding in the context of the EU-FP 7 CLIMB project (http://www.climb-fp7.eu), grant number 244151. The authors express their gratitude towards AGRIS (Sardinian Agency for Research in Agriculture), especially Dr. Antonino Soddu. Finally, we thank our two anonymous reviewers for their constructive comments that helped to further improve the quality of this manuscript.

5. Summary and conclusion References The distributed water balance model mGROWA was used to simulate groundwater recharge in the Riu Mannu catchment under changing climatic conditions. Simulations were carried out based on four different GCM–RCM combinations, representing possible pathways for a future climate development. We examined the sensitivity of the mGROWA simulation results to changes in the water storage capacity of a newly implemented regolith zone and to changes in the Kc values for the land-use classes Agriculture and Macchia. In addition, we used two different sources of soil texture distributions. The implementation of a regolith zone extended the soil water storage capacity considered by the model. This zone is commonly not described and parameterized in soil maps available for large and meso-scale water balance modeling. Several studies indicated that Mediterranean scrubland vegetation is able to access this water storage. We simulated qr for a number of scenarios while step-wise increasing water storage capacities, leading to a reduction in qr. The resulting bandwidth of reduction emphasizes the necessity to take this regolith zone into account in groundwater recharge modeling. We regard the 35% scenario (with available water contents in the regolith in the range of 3.9–5.7 vol.%) as representing a plausible parameter

AD-HOC-AG BODEN, 2005. Bodenkundliche Kartieranleitung (KA5). Bundesanstalt für Geowissenschaften und Rohstoffe, Hannover. Allen, R.G., Pereira, L.S., Raes, D., Smith, M., 1998. Crop evapotranspiration: guidelines for computing crop water requirements. Food and Agriculture Organization of the United Nations. Arnold, J.G., Allen, P.M., 1999. Automated methods for estimating baseflow and ground water recharge from streamflow records. J. Am. Water Resour. Assoc. 35, 411–424. Arnold, J.G., Allen, P.M., Muttiah, R., Bernhardt, G., 1995. Automated base-flow separation and recession analysis techniques. Ground Water 33, 1010–1018. Aru, A., Baldaccini, P., Delogu, G., Dessena, M. A., Madrau, S., Melis, R. T., Vacca, A. & Vacca, S. 1990. Carta dei suoli della Sardegna, in scala 1:250.000. Dipartimento Scienze della Terra Universitá di Cagliari, Assessorato Regionale alla Programmazione Bilancio ed Assetto del Territorio, SELCA, Firenze. ATV-DVWK, 2002. Verdunstung in Bezug zu Landnutzung, Bewuchs und Boden, Merkblatt ATV-DVWK-M 504. Deutsche Vereinigung für Wasserwirtschaft Abwasser und Abfall. GFA-Ges. zur Förderung d. Abwassertechnik e.V. Barrocu, G., 2007. Hydrogeology of granite rocks in Sardinia. Groundwater in Fractured Rocks. Taylor & Francis. Bassu, S., Asseng, S., Motzo, R., Giunta, F., 2009. Optimising sowing date of durum wheat in a variable Mediterranean environment. Field Crop Res. 111, 109–118. Carmignani, L., Oggiano, G., Funedda, A., Conti, P., Pasci, S. & Barca, S. 2011. Geological Map of Sardinia. Scale 1:25,000. Universitá degli Studi di Siena — Centro GeoTecnologie, San Giovani Valdarno. Ciais, P., Reichstein, M., Viovy, N., Granier, A., Ogee, J., Allard, V., Aubinet, M., Buchmann, N., Bernhofer, C., Carrara, A., Chevallier, F., De Noblet, N., Friend, A.D., Friedlingstein,


L. Ehlers et al. / Science of the Total Environment xxx (2015) xxx–xxx P., Grunwald, T., Heinesch, B., Keronen, P., Knohl, A., Krinner, G., Loustau, D., Manca, G., Matteucci, G., Miglietta, F., Ourcival, J.M., Papale, D., Pilegaard, K., Rambal, S., Seufert, G., Soussana, J.F., Sanz, M.J., Schulze, E.D., Vesala, T., Valentini, R., 2005. Europe-wide reduction in primary productivity caused by the heat and drought in 2003. Nature 437, 529–533. David, T.S., Henriques, M.O., Kurz-Besson, C., Nunes, J., Valente, F., Vaz, M., Pereira, J.S., Siegwolf, R., Chaves, M.M., Gazarini, L.C., David, J.S., 2007. Water-use strategies in two co-occurring Mediterranean evergreen oaks: surviving the summer drought. Tree Physiol. 27, 793–803. De Girolamo, A.M., Lo Porto, A., 2012. Land use scenario development as a tool for watershed management within the Rio Mannu Basin. Land Use Policy 29, 691–701. De Girolamo, A.M., Lo Porto, A., Oueslatioueslati, O., 2010. Modeling nutrient balances in the Rio Mannu basin ans scenario analysis of the impact of fertilizer reduction. Fresenius Environ. Bull. 1854–1861. De Gruijter, J.J., Walvoort, D.J.J., Van Gams, P.F.M., 1997. Continuous soil maps — a fuzzy set approach to bridge the gap between aggregation levels of process and distribution models. Geoderma 77, 169–195. Deidda, R., 1999. Multifractal analysis and simulation of rainfall fields in space. Phys. Chem. Earth Part B 24, 73–78. Deidda, R., 2000. Rainfall downscaling in a space–time multifractal framework. Water Resour. Res. 36, 1779–1794. Deidda, R., Benzi, R., Siccardi, F., 1999. Multifractal modeling of anomalous scaling laws in rainfall. Water Resour. Res. 35, 1853–1867. Deidda, R., Marrocu, M., Caroletti, G., Pusceddu, G., Langousis, A., Lucarini, V., Puliga, M., Speranza, A., 2013. Regional climate models' performance in representing precipitation and temperature over selected Mediterranean areas. Hydrol. Earth Syst. Sci. 17, 5041–5059. Detto, M., Montaldo, N., Albertson, J.D., Mancini, M., Katul, G., 2006. Soil moisture and vegetation controls on evapotranspiration in a heterogeneous Mediterranean ecosystem on Sardinia, Italy. Water Resour. Res. 42. Engel, N., Müller, U., Schäfer, W., 2012. BOWAB — Ein MehrschichtBodenwasserhaushaltsmodell. Geo Berichte 20. Landesamt für Bergbau, Energie und Geologie. FAO, 2006. Guidelines for Soil Description. Food and Agriculture Organization, Rome. Gee, G.W., Or, D., 2002. Particle size analysis. In: Dane, J.H., Topp, G.C. (Eds.), Methods of Soil Analysis, Part 4. Physical Methods. Soil Science Society of America, Madison, WI (ASA and SSA). Giorgi, F., 2006. Climate change hot-spots. Geophys. Res. Lett. 33. Goldhamer, D.A., Snyder, R.L., 1989. Irrigation Scheduling: A Guide for Efficent On-farm Water Management. University of California, Division of Agriculture and Natural Resources, Berkeley. Haylock, M.R., Hofstra, N., Klein Tank, A.M.G., Klok, E.J., Jones, P.D., New, M., 2008. A European daily high-resolution gridded data set of surface temperature and precipitation for 1950–2006. J. Geophys. Res.-Atmos. 113. Herrmann, F., Chen, S.N., Heidt, L., Elbracht, J., Engel, N., Kunkel, R., Muller, U., Rohm, H., Vereecken, H., Wendland, F., 2013. Highly resolved distributed water-balance modelling in Lower Saxony using mGROWA. Hydrol. Wasserbewirtsch. 57, 206–224. IPCC (Intergovernmental Panel On Climate Change), 2007. Climate change 2007: impacts, adaptation and vulnerability. Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, UK. Jones, D.P., Graham, R.C., 1993. Water-holding characteristics of weathered granitic rock in chaparral and forest ecosystems. Soil Sci. Soc. Am. J. 57, 256–261. Karpuzcu, M., Agiralioglu, N., Alpaslan, N., Engin, G., Gömann, H., Gunduz, O., Kocal, M., Oncel, S., Pekdeger, A., Voigt, H., Kuhr, P., Montzka, C., Tetzlaff, B., Vereecken, H., Wendland, F., 2008. Integrated modelling of nutrients in selected river basins of Turkey — results of a bilateral German–Turkish research project. Forschungszentrum, Zentralbibliothek Jülich. Katsura, S., Kosugi, K., Yamamoto, N., Mizuyama, T., 2006. Saturated and unsaturated hydraulic conductivities and water retention characteristics of weathered granitic bedrock. Vadose Zone J. 5, 35–47. Koehler, R., 2004. Raster Based Analysis and Visualization of Hydrologic Time Series. Department of Watershed Management, The University of Arizona, Tucson, AZ. Kunkel, R., Wendland, F., 2002. The GROWA98 model for water balance analysis in large river basins — the river Elbe case study. J. Hydrol. 259, 152–162.

15

Lopez-Bustins, J.A., Pascual, D., Pla, E., Retana, J., 2013. Future variability of droughts in three Mediterranean catchments. Nat. Hazards 69, 1405–1421. Lovelli, S., Perniola, M., Di Tommaso, T., Ventrella, D., Moriondo, M., Amato, M., 2010. Effects of rising atmospheric CO2 on crop evapotranspiration in a Mediterranean area. Agric. Water Manag. 97, 1287–1292. Ludwig, R., Soddu, A., Duttmann, R., Baghdadi, N., Benabdallah, S., Deidda, R., Marrocu, M., Strunz, G., Wendland, F., Engin, G., Paniconi, C., Prettenthaler, F., Lajeunesse, I., Affifi, S., Cassiani, G., Bellin, A., Mabrouk, B., Bach, H., Ammerl, T., 2010. Climate-induced changes on the hydrology of Mediterranean basins — a research concept to reduce uncertainty and quantify risk. Fresenius Environ. Bull. 19 (10A), 2379–2384. Mariotti, A., 2010. Recent changes in the Mediterranean water cycle: a pathway toward long-term regional hydroclimatic change? J. Clim. 23, 1513–1525. Mariotti, A., Dell'aquila, A., 2012. Decadal climate variability in the Mediterranean region: roles of large-scale forcings and regional processes. Clim. Dyn. 38, 1129–1145. Marras, S., Pyles, R.D., Sirca, C., Paw U, K.T., Snyder, R.L., Duce, P., Spano, D., 2011. Evaluation of the advanced canopy–atmosphere–soil algorithm (ACASA) model performance over Mediterranean maquis ecosystem. Agric. For. Meteorol. 151, 730–745. Mascaro, G., Piras, M., Deidda, R., Vivoni, E.R., 2013. Distributed hydrologic modeling of a sparsely monitored basin in Sardinia, Italy, through hydrometeorological downscaling. Hydrol. Earth Syst. Sci. 17, 4143–4158. Moriasi, D.N., Arnold, J.G., Van Liew, M.W., Bingner, R.L., Harmel, R.D., Veith, T.L., 2007. Model evaluation guidelines for systematic quantification of accuracy in watershed simulations. Trans. ASABE 50, 885–900. Müller, U., Waldeck, A., 2011. Auswertungsmethoden im Bodenschutz. Dokumentation zur Methodendatenbank des Niedersächsischen Bodeninformationssystems. Geo Berichte 19. Landesamt für Bergbau, Energie und Geologie. Nash, J.E., Sutcliffe, J.V., 1970. River flow forecasting through conceptual models part I — a discussion of principles. J. Hydrol. 10, 282–290. Peel, M.C., Finlayson, B.L., McMahon, T.A., 2007. Updated world map of the Koppen– Geiger climate classification. Hydrol. Earth Syst. Sci. 11, 1633–1644. Piras, M., Mascaro, G., Deidda, R., Vivoni, E.R., 2014. Quantification of hydrologic impacts of climate change in a Mediterranean basin in Sardinia, Italy, through highresolution simulations. Hydrol. Earth Syst. Sci. 18, 5201–5217. Rambal, S., 1984. Water-balance and pattern of root water-uptake by a Quercus-coccifera L evergreen scrub. Oecologia 62, 18–25. Refsgaard, J.C., Van Der Sluijs, J.P., Hojberg, A.L., Vanrolleghem, P.A., 2007. Uncertainty in the environmental modelling process — a framework and guidance. Environ. Model Softw. 22, 1543–1556. Risser, D.W., Gburek, W.J., Folmar, G.J., 2005. Comparison of Methods for Estimating Ground-water Recharge and Base Flow at a Small Watershed Underlain by Fractured Bedrock in the Eastern United States. U.S. Geological Survey Scientific Investigations Report 2005–5038, 31 p. Sivapalan, M., 2003. Prediction in ungauged basins: a grand challenge for theoretical hydrology. Hydrol. Process. 17, 3163–3170. Strandhagen, E., Marcus, W.A., Meacham, J.E., 2006. Views of the rivers: representing streamflow of the greater yellowstone ecosystem. Cartogr. Perspect. 55. Vacca, A., Loddo, S., Melis, M.T., Funedda, A., Puddu, R., Verona, M., Fanni, S., Fantola, F., Madrau, S., Marrone, V.A., Serra, G., Tore, C., Manca, D., Pasci, S., Puddu, M.R., Schirru, P., 2014. A GIS based method for soil mapping in Sardinia, Italy: a geomatic approach. J. Environ. Manag. 138, 87–96. Van Den Boogaart, K.G., Tolosana, R., Bren, M., 2014. Compositions: Compositional Data Analysis. R Package Version 1.40-1 Ed. Van Der Linden, P., Mitchell, J.F.B., 2009. ENSEMBLES: Climate Change and Its Impacts: Summary of Research and Results From the ENSEMBLES Project. Met Office Haley Centre, FitzRoy Road, Exeter EX1 3 PB, UK (160 pp.). Walker, W.E., Harremoës, P., Romans, J., Van Der Sluus, J.P., Van Asselt, M.B.A., Janssen, P., Krauss, M.K.V., Walker, W.E., Harremoës, P., Romans, J., Van Der Sluus, J.P., Van Asselt, M.B.A., Janssen, P., Krauss, M.K.V., 2003. Defining uncertainty. A conceptual basis for uncertainty management in model-based decision support. Integr. Assess. 4, 5–17. Walvoort, D.J., De Gruijter, J., 2001. Compositional kriging: a spatial interpolation method for compositional data. Math. Geol. 33, 951–966. Witty, J.H., Graham, R.C., Hubbert, K.R., Doolittle, J.A., Wald, J.A., 2003. Contributions of water supply from the weathered bedrock zone to forest soil quality. Geoderma 114, 389–400.


Enhanced groundwater recharge rates and altered recharge sensitivity to climate variability through subsurface heterogeneity.

Simulation of future groundwater recharge using a climate model ensemble and SAR-image based soil parameter distributions - A case study in an intensively-used Mediterranean catchment.

Impact of Climate Change on Irrigation Demand and Crop Growth in a Mediterranean Environment of Turkey.

Extreme climate events counteract the effects of climate and land-use changes in Alpine treelines.

Impacts of land use changes on soil properties and processes.

Depth-Dependent Mineral Soil CO2 Production Processes: Sensitivity to Harvesting-Induced Changes in Soil Climate.

Climate and land use controls on soil organic carbon in the loess plateau region of China.

Impacts of groundwater recharge from rubber dams on the hydrogeological environment in Luoyang Basin, China.

Uncertainty in soil data can outweigh climate impact signals in global crop yield simulations.

Impact of climate changes during the last 5 million years on groundwater in basement aquifers.

Impact of Spatial Soil and Climate Input Data Aggregation on Regional Yield Simulations.

Erratum to: Riparian Land-Use and Rehabilitation: Impact on Organic Matter Input and Soil Respiration.

Short-Term Behavior of Slag Concretes Exposed to a Real In Situ Mediterranean Climate Environment.

Identification of priority organic compounds in groundwater recharge of China.

Fewer clouds in the Mediterranean: consistency of observations and climate simulations.

Climate and soil properties limit the positive effects of land use reversion on carbon storage in Eastern Australia.

Changes in climate variability with reference to land quality and agriculture in Scotland.

Land use impact on soil quality in eastern Himalayan region of India.

Impact of distillery effluent application to land on soil microflora.

A Generalized Deforestation and Land-Use Change Scenario Generator for Use in Climate Modelling Studies.

Land-use and carbon cycle responses to moderate climate change: implications for land-based mitigation?

Effects of Land Use Changes from Paddy Fields on Soil Bacterial Communities in a Hilly and Mountainous Area.

Responses of Water and Salt Parameters to Groundwater Levels for Soil Columns Planted with Tamarix chinensis.

Changes in soil microbial biomass and residual indices as ecological indicators of land use change in temperate permanent grassland.