Modelling and monitoring of hydrological processes in the unsaturated zone of chalk, a porous medium with fractures, is important to optimize water resource assessment and management practices in the United Kingdom (UK). However, incorporating the processes governing water movement through a chalk unsaturated zone in a numerical model is complicated mainly due to the fractured nature of chalk that creates high-velocity preferential flow paths in the subsurface. In general, flow through a chalk unsaturated zone is simulated using the dual-porosity concept, which often involves calibration of a relatively large number of model parameters, potentially undermining applications to large regions. In this study, a simplified parameterization, namely the Bulk Conductivity (BC) model, is proposed for simulating hydrology in a chalk unsaturated zone. This new parameterization introduces only two additional parameters (namely the macroporosity factor and the soil wetness threshold parameter for fracture flow activation) and uses the saturated hydraulic conductivity from the chalk matrix. The BC model is implemented in the Joint UK Land Environment Simulator (JULES) and applied to a study area encompassing the Kennet catchment in the southern UK. This parameterization is further calibrated at the point scale using soil moisture profile observations. The performance of the calibrated BC model in JULES is assessed and compared against the performance of both the default JULES parameterization and the uncalibrated version of the BC model implemented in JULES. Finally, the model performance at the catchment scale is evaluated against independent data sets (e.g. runoff and latent heat flux). The results demonstrate that the inclusion of the BC model in JULES improves simulated land surface mass and energy fluxes over the chalk-dominated Kennet catchment. Therefore, the simple approach described in this study may be used to incorporate the flow processes through a chalk unsaturated zone in large-scale land surface modelling applications.
Chalk can be described as a fine-grained porous medium traversed by fractures (Price et al., 1993). Previous studies showed that the unsaturated zone of the chalk aquifers plays an important role in groundwater recharge in the UK (e.g. Lee et al., 2006; Ireson et al., 2009). Therefore, both monitoring (e.g. Bloomfield, 1997; Ireson et al., 2006) and modelling (e.g. Bakopoulou, 2015; Brouyère, 2006; Ireson and Butler, 2011, 2013; Sorensen et al., 2014) strategies have been adapted previously to understand the governing hydrological processes in the chalk unsaturated zone.
In chalk, the matrix provides porosity and storage capacity, while the
fractures greatly enhance permeability (Van den Daele et al., 2007). Water
movement through the chalk matrix is slow due to its relatively high
porosity (0.3–0.4) and low permeability (10
Simulating water flow through the matrix–fracture system of chalk has been the subject of research for some time. Both conceptual (e.g. Price et al., 2000; Haria et al., 2003) and physics-based (e.g. Mathias et al., 2006; Ireson et al., 2009) models have been proposed previously to describe water flow through the chalk unsaturated zone. The physics-based models mentioned above were developed based on a dual-continua approach and required relatively large numbers of parameters (i.e. of the order of 20–30 parameters) that were calibrated via inverse modelling using observed soil moisture and matric potential data (e.g. Ireson et al., 2009; Mathias et al., 2006).
In recent years, representation of chalk has gained attention in land surface modelling. For example, Gascoin et al. (2009) applied the Catchment Land Surface Model (CLSM) over the Somme River basin in northern France. A linear reservoir was included in the TOPMODEL-based runoff formulation of the CLSM to account for the contribution of chalk aquifers to river discharge. Le Vine et al. (2016) applied the Joint UK Land Environment Simulator (JULES; Best et al., 2011) over the Kennet catchment in southern England to evaluate the hydrological limitations of land surface models. In that study, two intersecting Brooks and Corey curves were proposed, which allowed a dual-curve soil moisture retention representation for the two distinct flow domains of chalk (i.e. matrix and fracture) in the model. Considering this dual Brooks and Corey curve, a three-dimensional groundwater flow model (ZOOMQ3D; Jackson and Spink, 2004) was coupled to JULES to demonstrate the strong influence of representing chalk hydrology and groundwater dynamics on simulated soil moisture and runoff.
The above-mentioned studies illustrate the importance of representing chalk in land surface modelling. However, including chalk hydrology in large-scale land surface modelling using the contemporary dual-porosity concept can be complicated due to the large number of additional parameters. In this context, we propose a new parameterization, namely the Bulk Conductivity (BC) model, as a first step towards a simple chalk representation suitable for land surface modelling. In order to test the proposed parameterization, the BC model is included in JULES (version 4.2), which, by default (i.e. uniform soil column representation using a general soil database as typically applied in land surface models), does not represent any chalk feature. In this study, the BC model (included in JULES) is applied at two distinct spatial scales (i.e. point and catchment). At the point scale, the proposed parameterization is calibrated using observed soil moisture profile data. This is achieved by randomly sampling the parameter space and extensively running the model in order to minimize the differences between observed and simulated soil moisture variability at different depths. Finally, the proposed model is applied to the Kennet catchment in southern England and the fluxes and states of the hydrological cycle are simulated for multiple years. The simulation results are evaluated using observed latent heat flux (LE) and runoff data to assess the performance of the BC model in simulating land surface processes at the catchment scale.
In this study, the BC model based on the work by Zehe et al. (2001) is
incorporated into JULES to represent the flow of water through the fractured
chalk unsaturated zone. According to this approach, if the relative
saturation (
At the first step of evaluation, the
Finally, in Zehe et al. (2001),
Field measurements and remote sensing data.
The study area encompasses the Kennet catchment located in southern England,
with an area of about 1033 km
The solid geology of the Kennet catchment is dominated by chalk, which is overlain by a thin soil layer. While lower chalk outcrops along the northern catchment boundary, progressively younger rocks are found in the southern part. In general, surface runoff production is very limited over the regions of the catchment where chalk outcrops. The flow regime shows a distinct characteristic of slow response to groundwater held within the chalk aquifer (Le Vine et al., 2016). According to Ireson and Butler (2013), the unsaturated zone of chalk shows slow drainage over summer and bypass flow during wet periods in this catchment.
Table 1 summarizes the field measurements and remote sensing data used in
this study. We use in situ soil moisture and runoff measurements along with
remotely sensed LE data to assess model performance in simulating the mass
and energy balance components of the hydrological cycle. Point-scale soil
moisture measurements at two adjacent sites (
The National River Flow Archive (NRFA) coordinates discharge measurements from the gauging station networks across the UK. These networks are operated by the Environmental Agency (England), Natural Resources Wales, the Scottish Environment Protection Agency, and the Rivers Agency (Northern Ireland). We use discharge measurements provided by the NRFA to assess model performance in simulating runoff over the Kennet catchment in this study.
The MOD16 product of the Moderate Resolution Imaging Spectroradiometer (MODIS) is a part of the NASA/EOS project that provides estimation of global terrestrial LE. The LE estimation from MOD16 is based on remotely sensed land surface data (e.g. Mu et al., 2007). In this study, the 8-day and monthly LE data products from MODIS are used to evaluate the model performance in simulating land surface energy fluxes.
In this study, we use the Joint UK Land Environment Simulator (JULES; e.g. Best et al., 2011; Clark et al., 2011) version 4.2. JULES is a flexible modelling platform with a modular structure aligned to various physical processes developed based on the Met Office Surface Exchange Scheme (MOSES; e.g. Cox et al., 1999; Essery et al., 2001). Meteorological data including precipitation, incoming short- and long-wave radiation, temperature, specific humidity, surface pressure, and wind speed are required to drive JULES. Each grid box in JULES can comprise nine surface types (broadleaf trees, needleleaf trees, C3 grass, C4 grass, shrubs, inland water, bare soil, and ice) represented by respective fractional coverage. Each surface type is represented by a tile and a separate energy balance is calculated for each tile.
Subsurface heat and water transport equations are solved based on finite difference approximation in JULES as described in Cox et al. (1999). Moisture transport in the subsurface is described by the finite difference form of Richards' equation. The vertical soil moisture flux is calculated using Darcy's law. While the top boundary condition to solve the Richards' equation is infiltration at the soil surface, the bottom boundary condition in JULES is free drainage that contributes to subsurface runoff.
Surface runoff is calculated by combining the equations of throughfall and grid box average infiltration in JULES. In order to direct the generated runoff to a channel network, river routing is implemented based on the discrete approximation of a one-dimensional kinematic wave equation (e.g. Bell et al., 2007). In this approach, a river network is derived from the digital elevation model (DEM) of the study area and different wave speeds are applied to surface and subsurface runoff components and channel flows (e.g. Bell and Moore, 1998). A return flow term accounts for the transfer of water between subsurface and land surface (e.g. Dadson et al., 2010, 2011).
In this study, simulations are performed at two distinct spatial scales, namely point and catchment. At the point scale, JULES is configured to simulate the mass and energy fluxes at the Warren Farm site (Fig. 1a). A total subsurface depth of 5 m is considered in the model with a vertical discretization ranging from 10 cm at the land surface to 50 cm at the bottom of the model domain. Note that this discretization is consistent with the soil moisture measurement depths mentioned in Sect. 3.2. The vegetation type is implemented as C3 grass using the default parameters in JULES. Point-scale simulations were performed over 2 consecutive years from 2003 to 2005 at an hourly time step. Except for precipitation, hourly atmospheric forcing data to drive JULES were obtained from an automatic weather station operated by the CEH at Warren Farm. In order to estimate hourly precipitation data to run JULES, rain gauge measurements from the Met Office (Met Office, 2006) were used. The inverse distance interpolation technique (e.g. Garcia et al., 2008; Ly et al., 2013) was applied to rainfall measurements from the 13 gauges closest to Warren Farm (the distance varies from 25 to 60 km) to obtain hourly precipitation for the point-scale simulations.
At the catchment scale, JULES is configured over a study area encompassing
the Kennet catchment (Fig. 1a) considering a uniform lateral grid resolution
of 1 km with 70
We estimate the soil hydraulic properties based on texture (Table 2). At the point scale, loam soil is dominant at the Warren Farm site. At the catchment scale, the Harmonized World Soil Database (HWSD) from the Food and Agriculture Organization of the UNO (FAO) is used to obtain the texture of different soil types over Kennet (Fig. 1c). The saturation–pressure head relationship for different soil types is described using the Van Genuchten (1980) model with parameter values (Table 2) obtained from Schaap and Leij (1998).
The hydraulic properties for chalk used in this study are summarized in
Table 3. These properties are obtained based on the existing literature as a
first step when evaluating the uncalibrated BC model. The BC model parameters
are subsequently calibrated to minimize the differences between observed and
simulated
In this study, we consider two different model configurations, namely default
and macro (Fig. 2). The default configuration corresponds to the standard
parameterizations of JULES that does not represent chalk hydrology in the
model. In this configuration, each soil column in JULES is considered to be
vertically homogeneous with the soil properties defined in Table 2, which is
motivated by the Met Office JULES Global Land 4.0 configuration described in
Walters et al. (2014). The macro configuration, in contrast, explicitly
represents chalk by applying the BC model starting at 30 cm below the land
surface to the bottom of the model domain (i.e. 500 cm). Therefore, the soil
column in the macro configuration can be divided into topsoil (0–30 cm) and
chalk (30–500 cm). The topsoil depth of 30 cm is defined based on several
augured soil samples collected during a field campaign at Warren Farm in 2015
(Fig. 2). This depth is corroborated by additional information from the
British Geological Survey (BGS) operated borehole records
(
Hydraulic properties for different soil types (refer to Fig. 1c).
Saturated hydraulic conductivity (
Hydraulic properties of chalk.
We calibrate the BC model at the point scale to minimize the differences
between observed and simulated soil moisture variability (
Equation (1) reveals that the calibration of the BC model involves optimizing
three parameters, namely the saturated hydraulic conductivity of the chalk
matrix (
Comparison between observed and simulated
At the point scale, the simulation results are evaluated using soil moisture
observations at the Warren Farm site. Figure 3a compares observed and
simulated soil moisture (
The results show that despite the macro configuration improves
simulated
The RMSE between observed and simulated
Optimizing both
The lower three panels in Fig. 4 presents the BC model parameter values for
the default and uncalibrated macro cases as well as for different
combinations of parameters calibrated. The red bars in Fig. 4b–d highlight
the cases in which a given parameter is constrained by optimization. In those
cases, the calibrated parameter values are obtained from model runs producing
the lowest RMSE. An interesting feature in Fig. 4b (calibrating
Figure 5 compares
As mentioned earlier, efficiently reproducing soil moisture variability over
the profile is important due to the fact that
In this section, the BC model was evaluated at the point scale. The results
showed that, in general, the macro configuration outperforms the default case
in simulating
Comparison between observed and simulated
Catchment average 8-day composites of MODIS estimated
LE (LE
At the catchment scale, simulation results from the default and
macro
Figure 7 shows considerable differences between LE
(a) Spatially averaged monthly latent heat flux (LE) from MODIS,
default and macro
In addition, Fig. 8b compares the observed and simulated monthly average
discharge from the two model configurations at the Kennet at Theale
gauging station (Fig. 1a). This figure shows that the default configuration
generally overestimates discharge at this gauging station, which is improved
considerably in the case of macro
In order to summarize the results at the catchment scale, Table 4 compares
observed and simulated runoff from the two model configurations over the
Kennet catchment from 2006 to 2011. The runoff ratio (RR; see the Appendix),
which is equal to the mean volume of flow divided by the volume of
precipitation (e.g. Kelleher et al., 2015), assesses the partitioning of
precipitation into runoff over the catchment. The default configuration
(RR
In Table 4, the relative bias (
Comparison between observed and simulated daily average runoff from
the two configurations over the Kennet catchment. Metrics include the runoff
ratio (RR), relative bias (
In this section, the BC model is evaluated using observed mass and energy
fluxes over the Kennet catchment. The default configuration suggested
relatively low summertime LE over the catchment. The agreement between
observed and simulated LE was improved in the case of the macro
In this study, we proposed a simple parameterization, namely the Bulk Conductivity (BC) model, to simulate water flow through the matrix–fracture system of chalk in large-scale land surface modelling applications. This parameterization was implemented in the Joint UK Land Environment Simulator (JULES) and applied to the Kennet catchment located in the southern UK to simulate the mass and energy fluxes of the hydrological cycle for multiple years. Two model configurations, namely default and macro, were considered, with the latter using the BC model to simulate chalk hydrology.
The proposed BC model is a single-continuum approach to modelling
preferential flow (e.g. Beven and Germann, 2013) that involves only
three parameters, namely the saturated hydraulic conductivity of the chalk
matrix (
At the catchment scale, the proposed BC parameterization improved simulated latent heat flux (especially in summer) and the overall runoff compared to the default. Note that the complexity (i.e. the number of parameters) of the BC model for simulating water flow through a chalk unsaturated zone is substantially lower compared to more commonly used models for this purpose (e.g. dual-porosity models). Despite its simplicity, the proposed parameterization considerably improves the key hydrological fluxes simulated by JULES at the catchment scale. Therefore, the BC model can potentially be useful for land surface modelling applications over large-scale chalk-dominated areas.
The soil moisture profiles and atmospheric information used for the point-scale simulations (Warren Farm site) are available from the Centre for
Ecology and Hydrology (CEH) upon request. Rain gauge measurements to obtain
the precipitation estimates at the Warren Farm site are available upon
request from the NCAS British Atmospheric Data Centre
(
The coefficient of determination (
Runoff ratio (RR) assesses the portion of precipitation that generates runoff
over the catchment. RR is defined as
The relative bias (
The relative difference in standard deviation (
The authors declare that they have no conflict of interest.
We gratefully acknowledge the support by the A MUlti-scale Soil moisture Evapotranspiration Dynamics study – AMUSED – project funded by Natural Environment Research Council (NERC) grant number NE/M003086/1. The authors would also like to thank Ned Hewitt and Jonathan Evans from the Centre for Ecology and Hydrology (CEH) for providing the data for the point-scale analyses at Warren Farm. We would also like to thank Miguel Rico-Ramirez (University of Bristol) for helping prepare the precipitation data from the rain gauge network used for the point-scale simulations, Thorsten Wagener (University of Bristol) for his valuable suggestions on model diagnostics, and Joost Iwema (University of Bristol) for helping with the soil samples collected during the 2015 fieldwork campaign. Finally, we would like to thank the reviewers for their comments and suggestions that added to the quality of this paper. Edited by: N. Romano Reviewed by: N. Le Vine and three anonymous referees