Current global models of the carbon (C) cycle consider only vertical gas exchanges between terrestrial or oceanic reservoirs and the atmosphere, thus not considering the lateral transport of carbon from the continents to the oceans. Therefore, those models implicitly consider all of the C which is not respired to the atmosphere to be stored on land and hence overestimate the land C sink capability. A model that represents the whole continuum from atmosphere to land and into the ocean would provide a better understanding of the Earth's C cycle and hence more reliable historical or future projections. A first and critical step in that direction is to include processes representing the production and export of dissolved organic carbon in soils. Here we present an original representation of dissolved organic C (DOC) processes in the Joint UK Land Environment Simulator (JULES-DOCM) that integrates a representation of DOC production in terrestrial ecosystems based on the incomplete decomposition of organic matter, DOC decomposition within the soil column, and DOC export to the river network via leaching. The model performance is evaluated in five specific sites for which observations of soil DOC concentration are available. Results show that the model is able to reproduce the DOC concentration and controlling processes, including leaching to the riverine system, which is fundamental for integrating terrestrial and aquatic ecosystems. Future work should include the fate of exported DOC in the river system as well as DIC and POC export from soil.
An estimated 1.9 Pg C yr
The main sources of DOC in terrestrial ecosystems are plant residues (Khomutova et al., 2000) and humus and root exudates (Kalbitz et al., 2000; Van den berg et al., 2012; Marschner, 1995). DOC within the soil can be the product of in situ production or can be brought in by advective fluxes with soil water transport. It has been hypothesized that the loss of the carbon from the soil by leaching has to be taken into account to reasonably reassess the terrestrial C budget of Europe (Siemens, 2003). The fate of this DOC within inland water networks, i.e. the proportion transported to the coast or respired and emitted to the atmosphere, is the key to understanding the link to the other compartments of the Earth system (Cole et al., 2007; Battin et al., 2009). Nevertheless, it is a difficult task to link riverine and terrestrial fluxes by empirical methods because (1) riverine fluxes are integrating fluxes from different land use systems (Kindler et al., 2011; Boyer and Groffman, 1996) with different leaching rates and DOC quality, (2) in-stream transformation makes it difficult to trace back terrestrial DOC sources, and (3) it is difficult to separate natural and anthropogenic perturbation fluxes (Schelker et al., 2013; Regnier et al., 2013).
A physical-based modelling approach explicitly representing different terrestrial sources and processes involved in DOC cycling within the soil column and DOC leaching from the soil can help overcome these difficulties. Representation of DOC cycling within the soil column is also a major step toward simulating deep soil SOC formation (Rumpel and Kögel-Knabner, 2011). Physical-based models help us to understand the processes involved in soil DOC cycling and leaching as well as the biogeochemistry of SOC in general. So far several models have been developed that simulate DOC with different temporal and spatial resolution, from 15 min as in SOLVEG-II (Ota et al., 2013) to monthly as in ECOSSE (Smith et al., 2010) or RivCM (Langerwisch et al., 2016), and from the site scale as in DyDOC (Michalzik et al., 2003) to the global scale as in TEM (Kicklighter et al., 2013). Some of these models represent DOC leaching, whereas others do not. Each model has its own particular definition for carbon pools (including DOC) and DOC production processes which can be based on turnover time as in TERRAFLUX (Neff and Asner, 2001) or based on chemical composition as in the DyDOC model (Michalzik et al., 2003). Although all these models have been evaluated, with the exception of the TEM model which was tested for arctic rivers, none of them has demonstrated an ability to represent DOC production, processing, and transport on the global scale.
In general, most of the models containing decomposition are based on first-order kinetics (Olson, 1963). Frequently, models tend to represent the topsoil layer as the major source for DOC production and export (Koven et al., 2013); other studies (Rumpel and Kögel-Knabner, 2011; Braakhekke et al., 2013) highlight the importance of DOC for SOC production in deeper soil layers.
Here we present an original representation of DOC processes in the Joint UK
Land Environment Simulator (JULES-DOCM) that integrates a representation of
DOC production in terrestrial ecosystems based on the incomplete decomposition of
organic matter, DOC decomposition within the soil column, and DOC export to
the river network via leaching. JULES has been used to evaluate the global C
cycle (e.g. Le Quéré, et al., 2015; Sitch et al., 2015) and its role
in the Earth system, but to date lacks the critical processes of DOC
production and export. The aim of this study is to include a representation
of DOC produced in terrestrial soils down to 3 m (as soil hydrology and
carbon are simulated over a 3 m soil profile in JULES). We assume an incomplete
decomposition of organic matter and its subsequent fate as DOC, including
(i) DOC decomposition and release as CO
JULES is a process-based model which represents energy, water, and C cycling
among vegetation, soil, and atmosphere as described in Best et al. (2011)
and Clark et al. (2011). Vegetation processes in JULES are represented in a
dynamic vegetation model (TRIFFID) distinguishing nine plant function types
(PFTs) on the global scale: tropical and temperate broadleaf evergreen trees,
broadleaf deciduous trees, needle-leaf evergreen trees and deciduous trees,
C
The representation of SOC in JULES follows the formulation of the RothC soil carbon scheme (Jenkinson et al., 1990; Jenkinson and Coleman, 2008) in distinguishing four carbon pools: decomposable plant material (DPM), resistant plant material (RPM), heterotrophic microbial biomass (BIO), and long-lived humified material (HUM). DPM and RPM pools receive litter inputs directly from the vegetation due to defoliation, mortality and disturbance, the allocation to DPM or RPM depending on the PFT characteristics with a higher fraction of decomposable litter provided from grasses, and a higher fraction of resistant litter provided from trees (Clark et al., 2011). HUM and BIO each receive inputs from the other two soil carbon pools as a fraction of the decomposition that is not respired to the atmosphere.
JULES-DOCM is an extension of JULES based on version 4.4 (vn4.4 documentation
in
SOC is specified as the main source of DOC in JULES-DOCM. In JULES v4.4, each
of the four SOC pools is treated as a single box down to 3 m, without any
representation of its vertical distribution. This absence of vertical
distribution has consequences in terms of simulating DOC fluxes, but also
potential impacts on soil CO
In order to calculate the fraction of SOC that is used as input for DOC
production in each layer of the DOC model (see Eq. 4 below), the weighting
factors are normalized (
In JULES-DOCM, four new DOC carbon pools have been added. First the model accounts for a labile and a recalcitrant DOC pool based on their decomposition rate (Aguilar and Thibodeaux, 2005; Thibodeaux and Aguilar, 2005). The labile pool is readily available for decomposition in soil solution at all times and the recalcitrant pool is subject to a slower decomposition rate (Smith et al., 2010). DOC produced from plant material pools (DPM and RPM) and microbial biomass (BIO) is directed to the labile pool, while DOC from humus (HUM) is directed to the recalcitrant pool. Second, both the labile and the recalcitrant DOC pools have a dissolved and an adsorbed form, with only the dissolved pool being subjected to decomposition and leaching.
DOC production (
JULES-DOCM model structure.
The DOC production rate is further modified by
After decomposition, carbon pools (
We assume that the decomposition of DOC pools (
Part of decomposed DOC is respired (
Hence the distribution of decomposed DOC to the BIO pool and respiration will be
These terms for DOC labile and recalcitrant pools in JULES-DOCM are as
follows
(arrows i and j, Fig. 1):
DOC-relevant parameters in the JULES-DOCM model.
Symbol definitions and units.
DOC diffusion (
Hence the dissolved and adsorbed DOC pools are updated as follows:
Two data levels were provided in order to test the model performance. Level 1, for Hainich, Carlow, and Brasschaat, included the carbon fluxes and continuous DOC measurements from soil water from a 3- to 10-year period, and Level 2, for Turkey Point 89 (TP89) and Guandaushi, showed fewer C flux measurements and discontinuous DOC measurements (Table 3). The locations of the sites are given in Fig. 2.
Data availability for model evaluation at different sites.
The site Hainich, located in Germany within Hainich National Park
(51
The site Carlow is located in Ireland in County Carlow
(52
Evaluation Level 1 site characteristics.
Study sites.
The site Brasschaat is located in Belgium and covered by mixed
coniferous and deciduous (De Inslag) forest, (51
DOC samples were collected at three horizons of Al–Ap, A–E, and Cg (Soil Classification Working Group, 1998) referring to 10, 35, and 75 cm of depth by means of ceramic suction cups on a biweekly interval. Two days prior to sample collection a tension of 600 hPa was applied to each suction cup. Samples were collected at three locations and pooled into one composite sample per layer for analysis (Gielen et al., 2011).
The site Turkey Point 89 (TP89), located in southern Ontario, Canada
(42
The site Guandaushi is located in central Taiwan (23
Model performance was tested against observed data from Guandaushi and four
FLUXNET sites (Hainich, Carlow, Brasschaat, and Turkey Point 89). The FLUXNET
database provides on-site meteorological data for each site that could be
used as forcing for simulations in JULES. However, we had to use the global
WATCH dataset (Weedon et al., 2010) as forcing for the Guandaushi site where no
on-site data were available. Forcing data were checked for any
missing information and they were gap filled by linear interpolation. The
meteorological forcing is provided at the measurement site level (no explicit
spatial resolution) and includes the downward shortwave and longwave
radiation at the surface (W m
For Brasschaat, additional model parameters such as BK and clay were taken from Janssens et al. (1999). The model was first spun up looping over the period 1996–2014 until all the soil variables reached a steady state. For Hainich, site parameters were taken from Kutsch et al. (2010). The spin-up was run looping 300 times over the years 2004–2014. For Carlow, site parameters were taken from Walmsley (2009) and Kindler et al. (2010). The spin-up was run looping 300 times over the years 2004–2009. For Turkey Point 89, site parameters were taken from Peichl and Arain (2006) and spin-up was run looping 300 times over the years 2002–2007. For Guandaushi, site vegetation parameters were taken from Liu and Sheu (2003) and soil parameters from HWSD global data (Nachtergaele et al., 2010); spin-up was run looping 300 times over years 1990–2000. The evaluation of the model was performed on the plot scale using climate forcing data, soil, and land cover consistent with the site, and no horizontal spatial dimension was involved.
In order to test the sensitivity of DOC-related model parameters on the DOC
concentration in different depths of the soil profile, simulations were
performed with varying values for
In total, 16 runs were performed by modifying each parameter once by
increasing it 50 % and once by decreasing it by 50 %, except for the
slope parameter controlling
In order to test the model performance with regard to simulated C stock and fluxes, we used an ANOVA (analysis of variance) test to compare the model results from the default set of parameters against measurements. In order to test the parameter impact on the simulated DOC concentrations, we computed the RMSE values from each set of model parameter configurations.
The measured (Obs.) versus the modelled (Mod.) carbon fluxes, SOC concentration, and soil DOC concentration at different soil depths in five study sites.
To examine the performance of soil DOC simulations, it is first necessary to
explore other carbon fluxes which link to soil DOC pools. The first flux to
be validated is the gross primary production (GPP), for which we have
observed values (Table 3). The modelled mean GPP for Brasschaat and Carlow
was significantly lower than measurements with
Total soil respiration measurements were available for Brasschaat, Hainich,
and Turkey Point 89 (Table 3) and were compared with the modelled outputs.
The simulated values were close to observed values at Hainich, while the
modelled values for Brasschaat were significantly higher
(
DOC concentration (mg C L
Finally, we compared the SOC in measurements and model outputs; measurements from Brasschaat for 100 cm, Hainich for 60 cm, Carlow for 50 cm, and Turkey Point 89 for 15 cm (A horizon) of soil were available. The modelled SOC stock for Brasschaat in the first 100 cm and for Hainich down to 60 cm were slightly lower than the observations, while for Carlow the simulated stocks down to 50 cm and for Turkey Point 89 the simulated stocks down to 15 cm were higher than the observed stocks (Table 5).
In general, JULES-DOCM was capable of reproducing the DOC concentrations at
all the tested sites using the default set of parameters (Table 1) chosen as
representative of the topsoil (Fig. 3 Level 1 sites, Fig. 4 Level 2 sites).
For Hainich, the simulated average values and value range were close to
observed values at 10 and 20 cm (Table 5; RMSE values for 10 and 20 cm are
3.0 and 2.5 mg L
DOC concentration (mg C L
Overall, the model was capable of reproducing the seasonality of DOC concentrations for the European sites where long-term observation data are available (Fig. 5). However, at Brasschaat the simulated DOC peaked from April–July, while observed DOC peaked from July–September.
We also examined the hydrology of the model and its interaction with DOC concentration and leaching (e.g. Hainich Fig. 6; other sites are plotted in Fig. S3 in the Supplement). It can be seen for the period 2005–2014 that during heavy precipitation, high run-off was produced, which caused the higher leaching, and the consequence was a drop in the DOC concentration in 3 m of soil.
Sensitivity to model parameters was tested on the three European sites where
a representative time series of observed DOC concentrations was available
(e.g. Hainich; 10 cm, Fig. 7). The results indicate that among all the
parameters in all three sites, the model shows the highest sensitivity to SOC
vertical profile controlled by parameter
The sensitivity of the model to each of these parameters was different at
each site. For Hainich, the highest sensitivity was assigned to
In Brasschaat, the highest sensitivity was to
For Carlow, the most sensitive parameters were
Observed precipitation, simulated run-off, DOC leaching, and DOC concentration in Hainich from 2006 to 2013 indicating the relation between the averaged DOC concentrations in 3 m of soil with leaching as a result of run-off that follows large precipitation events.
DOC concentration (mg C L
Relative change in simulated DOC (%) for a
DOC concentration (mg C L
Overall, JULES-DOCM reproduced the range of GPP for most of our sites to an acceptable degree. At some sites, due to overestimated or underestimated autotrophic respiration, the NPP and total respiration values were slightly different than measurements. Consequently, the modelled carbon stocks were different from the measurements in most of the sites, yet were capable of representing the general patterns that were observed in the measurements.
In Brasschaat, the modelled SOC was lower than the measurements, which could be due to the underestimated NPP (Table 5) and, as a consequence, the underestimated litter input, but also due to the overestimated soil respiration and SOC decomposition rates. The underestimation of SOC as a source of DOC led to a general underestimation of DOC. Nevertheless, the decrease in relative DOC concentration through the soil is consistent with the observations.
In Hainich, a slightly overestimated NPP partly counterbalanced the
overestimated soil respiration. Nevertheless, the SOC concentration simulated
down to 60 cm was lower than the measurement at this depth. As we did not
have observations of SOC down to 3 m, we cannot definitively say if the
simulated total SOC stock (13.7 kg C m
In Carlow, the slight overestimation of GPP led to the overestimated SOC concentrations down to 50 cm, whilst again we cannot say with certainty that the whole SOC stock is overestimated, as the SOC stock has not been measured down to 3 m. Some sources suggest that the SOC in Carlow grassland could be higher than the reported value in our reference if we calculate the C in soil based on the fraction of loss of ignition (LOI; Walmsley, 2009; Hoogsteen et al., 2015). As Carlow is our only grassland biome site, additional data from different study sites would be valuable to achieve a more representative parameterization of soil carbon processes under grassland. One of the parameters to be optimized for such sites could be CUE, which has a strong impact on the stocks and fluxes. Also, since the measured values for NPP or soil respiration for this site were not available to us, we were unable to assess whether we overestimated or underestimated these fluxes and if this could have potentially biased our SOC stock simulations. DOC measurements were provided from two plots which were placed on different terrain positions. The measurements from plot 2 (150 m in the south-westerly direction from plot 1) at 10 to 28 cm of depth had a higher DOC concentration than plot 1 at 10 cm (Walmsley, 2009). This could be the result of small-scale variations related to terrain position, which can be related to different soil moisture regimes and the lateral import of DOC. It is not possible to represent such small-scale variation in global models like JULES-DOCM.
At Turkey Point 89, the simulated GPP is close to the observations, while NPP is slightly overestimated. The simulated soil respiration and decomposition rates are higher than observed values. The overestimated SOC concentration in the topsoil could be the result of an overestimated depth gradient in SOC concentration, which in our simulations is derived from global data (Jobbágy and Jackson, 2000). Also, we simulated the steady-state SOC profile for forest vegetation, whereas the forest stand at the site is relatively young and succeeded agricultural land use in 1989; thus, the SOC profile is likely not representative of a forest site. The overestimated DOC concentration for 100 cm of depth may be due to this change in land use, which was not taken into account during simulations as providing more C input for DOC production. At this site, the observed higher soil moisture in the deeper profile could indicate a potentially high advection of DOC to the lower layers (Peichl et al., 2010a). This could be another reason for the lower DOC in 100 cm from the measured compared to the modelled results.
In Guandaushi, due to the lack of SOC or vegetation carbon flux measurements from the site, we have no information on SOC concentrations and stocks. The lower values of DOC from our model compared to the measurements could be due to the high temporal variability of observed concentrations (large standard deviation for all the depths from the three stands). It could also be due to the high value of DOC input from rainfall, which is not represented in JULES-DOCM (Liu and Sheu, 2003). Recent studies have indicated that including this flux in models can have a significant impact on the DOC in soil (Lauerwald et al., 2017).
As there are no measurements of lateral leaching of DOC from soil to the river, our evaluation of this flux is based on the simulated DOC concentration and run-off. Hence as the simulated hydrology of the JULES model has been evaluated previously (Gedney and Cox, 2003; Clark and Gedney, 2008), in this study we assume that we will get robust estimates of DOC leaching by multiplying the simulated concentration by run-off as long as simulated DOC concentrations can be validated.
Overall, aside from overestimation or underestimation of DOC at some sites, the model was capable of representing the trend of DOC concentration at different depths compared to the measurements at all the sites.
The sensitivity tests indicate that the parameters controlling SOC
concentrations in the soil profile (
One limitation in our simulation is that we use a single, calibrated value for recalcitrant DOC residence time, which is the most sensitive DOC controlling parameter. It has been shown that this parameter can vary with the biodegradability of SOC and litter under different PFTs and at different sites (Kalbitz et al., 2003; Turgeon, 2008). However, more detailed data for different biomes are needed for calibrating different residence times for different PFTs. We note that our sensitivity analysis, by changing one parameter at a time, does not investigate the potential interactions among different parameters.
Applying a carbon cycle model that integrates the
whole continuum from land to ocean to atmosphere provides a better
understanding of the Earth's carbon cycle and makes more reliable future
projections. In this study, we presented DOC-related processes in JULES,
JULES-DOCM, which includes the DOC produced in the soil down to 3 m
and its subsequent fate including its decomposition and release as CO
The code written for this version of JULES can be found at
The authors declare that they have no conflict of interest.
The research leading to these results received funding from the European Union's Horizon 2020 research and innovation programme under Marie Sklodowska-Curie grant agreement no. 643052 (C-CASCADES project). We want to thank Altaf Arain, Tim Moore, and Gerd Glexiner for providing the DOC measurements. Ronny Lauerwald received funding from the European Union's Horizon 2020 research and innovation programme under grant agreement no. 703813 for the Marie Sklodowska-Curie European Individual Fellowship “C-Leak”. Jing Tang is financed by a Marie Sklodowska-Curie Action Individual Fellowship (MABVOC: 707187) and supported by the Danish National Research Foundation (CENPERM DNRF100). Marta Camino-Serrano acknowledges funding from the European Research Council Synergy grant ERC-2013-SyG-610028 IMBALANCE-P. Edited by: Christoph Müller Reviewed by: two anonymous referees