3D soil hydraulic database of Europe at 250 m resolution

Soil hydraulic properties are required in various modelling schemes. We propose a consistent spatial soil hydraulic database at 7 soil depths up to 2 m calculated for Europe based on SoilGrids250m and 1 km datasets and pedotransfer functions trained on the European Hydropedological Data Inventory. Saturated water content, water content at field capacity and wilting point, saturated hydraulic conductivity and Mualem‐van Genuchten parameters for the description of the moisture retention, and unsaturated hydraulic conductivity curves have been predicted. The derived 3D soil hydraulic layers (EU‐SoilHydroGrids ver1.0) can be used for environmental modelling purposes at catchment or continental scale in Europe. Currently, only EU‐SoilHydroGrids provides information on the most frequently required soil hydraulic properties with full European coverage up to 2 m depth at 250 m resolution.


| DATA
The multilayered European Soil Hydraulic Database (EU-SoilHydroGrids ver1.0) was derived with European pedotransfer functions (EU-PTFs; Tóth et al., 2015) based on the soil information of SoilGrids250m and aggregated 1 km  datasets.
EU-HYDI is a collection of data from 29 institutions in 18 European countries and contains data on taxonomical, chemical, and physical soil properties of more than 18,000 soil samples. Pedotransfer functions were calibrated using soil information of 134 to 6,074 soil samples and validated on 57 to 2,357 samples, depending on the type of soil hydraulic property (Tóth et al., 2015).
SoilGrids provides the most detailed information on soil properties with full continental coverage in Europe. It incorporates soil taxonomical, physical, and chemical data of seven soil depths at 250 m resolution . We used the following soil properties to calculate the soil hydraulic properties: clay, silt, and sand content (mass %); organic carbon content (g kg −1 ); bulk density (kg m −3 ); pH in water and depth to bedrock (cm) at 0, 5, 15, 30, 60, 100, and 200 cm depth. The first four depths, which are less than or equal to 30 cm depth, are considered as topsoil and the remaining handled as subsoil in accordance with the EU-PTFs used for calculations (Tóth et al., 2015).
In case bedrock appears within 200 cm, hydraulic properties were calculated up to the first layer underlying the top of the bedrock providing the possibility to interpolate the soil hydraulic properties through different soil depths. For modelling purposes, the predicted depth to bedrock is available from www.soilgrids.org; data are described in detail in Shangguan, Hengl, Mendes de Jesus, Yuan, and Dai (2017).

| METHODS
Soil properties included in SoilGrids database were transformed into the format needed by the EU-PTFs (Tóth et al., 2015). Sand, silt, and clay content were adjusted to sum to 100%, and USDA texture classes (Soil Survey Staff, 1975) were calculated. We selected the best performing and most reliable PTFs that were calibrated on representative data subsets to exclude too much data specific models. Sixteen soil hydraulic properties were calculated for the seven standard depths of SoilGrids: 0,5,15,30,60,100,and 200 cm at both 250 m and 1 km resolutions. Given the nonlinear relations between soil hydraulic properties and the other soil properties used as predictors, the mean of a set of predicted hydraulic properties is not equal to the prediction of the property based on the mean of the predictors. Therefore, calculations were also completed on the aggregated SoilGrids1km rather than aggregate soil hydraulic layers derived on 250 m resolution. The 1 km resolution aggregated version of the SoilGrids250m maps were generated using the "average" resampling method in the GDAL software (Mitchell and GDAL Developers, 2014). Table 1 lists the calculated soil hydraulic properties and the EU-PTFs used to predict them, indicating also the soil properties used as predictors. Saturated water content (THS) refers to the water content at 0 cm matric potential (0 MPa) (pF0). Field capacity (FC) is the water content at −330 cm matric potential (−0.03 MPa), which is the most commonly used value (pF2.5). If terminology of FC is different from the above mentioned-for example, it is assumed as water content at The van Genuchten model (van Genuchten, 1980) is used for the description of the MRC: where θ(h) is the water content of the soil (cm 3 cm −3 ) at a given matric potential value (cm of water column); θ r is the residual water content (cm 3 cm −3 ); θ s is the saturated water content (cm 3 cm −3 ); and α where K is the soil hydraulic conductivity (cm day −1 ); K 0 is the hydraulic conductivity acting as a matching point at saturation (cm day −1 ); S e is the effective saturation (−); and L is a shape parameter related to pore tortuosity (−). Further to parameters K 0 and L, parameters θ r , θ s , α, n, and m of HCC were calculated to provide a fully coupled model that describes the unsaturated hydraulic conductivity in the full matric potential range.
Calculations were executed in R (R Core Team, 2016) with the "euptf" package (Weynants & Tóth, 2014) containing the soil hydraulic prediction methods (EU-PTFs; Tóth et al., 2015). The "rgdal" (Bivand, Keitt, & Rowlingson, 2016) and "raster" (Hijmans, 2016) R packages were used to call in and perform calculations on raster files. Layers of SoilGrids250m were tiled with "GSIF" (Hengl, 2016) being able to execute the predictions. Calculations were run in parallel on SoilGrids250m with the "snowfall" (Knaus, 2015) R package. Note. BD = bulk density; clay = clay content; FC = water content at field capacity; HCC = hydraulic conductivity curve; KS = saturated hydraulic conductivity; LR = multiple linear regression; MRC = moisture retention curve; MS = mean statistics of pre-determined groups; OC = organic carbon content; pH = pH in water; PTFs = pedotransfer functions; RT = regression tree; sand = sand content; silt = silt content; T/S = topsoil and subsoil distinction; THS = saturated water content; WP = water content at wilting point.  Table 2 lists soil hydraulic parameters calculated for the seven soil depths of SoilGrids dataset at 250 m and 1 km resolution. Figure 1 shows a map of THS, FC, WP, and KS at 15 cm. In those cases where PTF is a regression tree (e.g., Figure 1  was calculated using the test sets of the EU-HYDI dataset (Tóth et al., 2015) and shown in Table 3.

| Comparison of EU-SoilHydroGrids and soil hydraulic properties available from ESDAC
We evaluated the performance of EU-SoilHydroGrids ver1.0 and ESDAC SHP to reproduce the measured soil hydraulic properties of EU-HYDI on more than 1500 samples (Table 4.). In the EU-SoilHydroGrids dataset all four soil hydraulic properties had smaller MAE and RMSE • if only and exactly THS and/or FC and/or WP and/or KS is needed from the EU-SoilHydroGrids dataset it is recommended to use THS, FC, WP, KS layers, because they were derived directly with point estimations having lower uncertainty than calculating those from MRC or HCC (Tóth et al., 2015); • in EU-SoilHydroGrids soil hydraulic data of layers deeper than the bottom of the soil are included as well, which provides the possibility to interpolate soil hydraulic properties through different soil depths; therefore depth to bedrock has to be considered from other sources e.g. www.soilgrids.org; • soil hydraulic properties are calculated for the fine earth fraction (< 2 mm) and volume of coarse fragments is not considered in the calculations; • if local soil hydraulic data or local soil hydraulic PTFs and/or local soil information are available their use is recommended because spatial accuracy of EU-SoilHydroGrids is limited, especially Note. EU-HYDI = European Hydropedological Data Inventory; FC = field capacity; HCC = hydraulic conductivity curve; KS = saturated hydraulic conductivity; MRC = moisture retention curve; PTFs = pedotransfer functions; RMSE = root mean square error; WP = wilting point.