ORCHILEAK ( revision 3875 ) : a new model branch to simulate carbon transfers along the terrestrial – aquatic continuum of the Amazon basin

Lateral transfer of carbon (C) from terrestrial ecosystems into the inland water network is an important component of the global C cycle, which sustains a large aquatic CO2 evasion flux fuelled by the decomposition of allochthonous C inputs. Globally, estimates of the total C exports through the terrestrial–aquatic interface range from 1.5 to 2.7 Pg C yr−1 (Cole et al., 2007; Battin et al., 2009; Tranvik et al., 2009), i.e. of the order of 2–5 % of the terrestrial NPP. Earth system models (ESMs) of the climate system ignore these lateral transfers of C, and thus likely overestimate the terrestrial C sink. In this study, we present the implementation of fluvial transport of dissolved organic carbon (DOC) and CO2 into ORCHIDEE (Organising Carbon and Hydrology in Dynamic Ecosystems), the land surface scheme of the Institut PierreSimon Laplace ESM. This new model branch, called ORCHILEAK, represents DOC production from canopy and soils, DOC and CO2 leaching from soils to streams, DOC decomposition, and CO2 evasion to the atmosphere during its lateral transport in rivers, as well as exchange with the soil carbon and litter stocks on floodplains and in swamps. We parameterized and validated ORCHILEAK for the Amazon basin, the world’s largest river system with regard to discharge and one of the most productive ecosystems in the world. With ORCHILEAK, we are able to reproduce observed terrestrial and aquatic fluxes of DOC and CO2 in the Amazon basin, both in terms of mean values and seasonality. In addition, we are able to resolve the spatio-temporal variability in C fluxes along the canopy–soil–water continuum at high resolution (1, daily) and to quantify the different terrestrial contributions to the aquatic C fluxes. We simulate that more than two-thirds of the Amazon’s fluvial DOC export are contributed by the decomposition of submerged litter. Throughfall DOC fluxes from canopy to ground are about as high as the total DOC inputs to inland waters. The latter, however, are mainly sustained by litter decomposition. Decomposition of DOC and submerged plant litter contributes slightly more than half of the CO2 evasion from the water surface, while the remainder is contributed by soil respiration. Total CO2 evasion from the water surface equals about 5 % of the terrestrial NPP. Our results highlight that ORCHILEAK is well suited to simulate carbon transfers along the terrestrial– aquatic continuum of tropical forests. It also opens the perspective that provided parameterization, calibration and validation is performed for other biomes, the new model branch could improve the quantification of the global terrestrial C sink and help better constrain carbon cycle–climate feedbacks in future projections. Published by Copernicus Publications on behalf of the European Geosciences Union. 3822 R. Lauerwald et al.: ORCHILEAK (revision 3875)

Abstract.Lateral transfer of carbon (C) from terrestrial ecosystems into the inland water network is an important component of the global C cycle, which sustains a large aquatic CO 2 evasion flux fuelled by the decomposition of allochthonous C inputs.Globally, estimates of the total C exports through the terrestrial-aquatic interface range from 1.5 to 2.7 Pg C yr −1 (Cole et al., 2007;Battin et al., 2009;Tranvik et al., 2009), i.e. of the order of 2-5 % of the terrestrial NPP.Earth system models (ESMs) of the climate system ignore these lateral transfers of C, and thus likely overestimate the terrestrial C sink.
In this study, we present the implementation of fluvial transport of dissolved organic carbon (DOC) and CO 2 into ORCHIDEE (Organising Carbon and Hydrology in Dynamic Ecosystems), the land surface scheme of the Institut Pierre-Simon Laplace ESM.This new model branch, called OR-CHILEAK, represents DOC production from canopy and soils, DOC and CO 2 leaching from soils to streams, DOC decomposition, and CO 2 evasion to the atmosphere during its lateral transport in rivers, as well as exchange with the soil carbon and litter stocks on floodplains and in swamps.We parameterized and validated ORCHILEAK for the Amazon basin, the world's largest river system with regard to discharge and one of the most productive ecosystems in the world.
With ORCHILEAK, we are able to reproduce observed terrestrial and aquatic fluxes of DOC and CO 2 in the Amazon basin, both in terms of mean values and seasonality.In addition, we are able to resolve the spatio-temporal variability in C fluxes along the canopy-soil-water continuum at high resolution (1 • , daily) and to quantify the different terrestrial contributions to the aquatic C fluxes.We simulate that more than two-thirds of the Amazon's fluvial DOC export are contributed by the decomposition of submerged litter.Throughfall DOC fluxes from canopy to ground are about as high as the total DOC inputs to inland waters.The latter, however, are mainly sustained by litter decomposition.Decomposition of DOC and submerged plant litter contributes slightly more than half of the CO 2 evasion from the water surface, while the remainder is contributed by soil respiration.Total CO 2 evasion from the water surface equals about 5 % of the terrestrial NPP.Our results highlight that ORCHILEAK is well suited to simulate carbon transfers along the terrestrialaquatic continuum of tropical forests.It also opens the perspective that provided parameterization, calibration and validation is performed for other biomes, the new model branch could improve the quantification of the global terrestrial C sink and help better constrain carbon cycle-climate feedbacks in future projections.
Published by Copernicus Publications on behalf of the European Geosciences Union.

Introduction
The Fifth Assessment Report (AR5) of the Intergovernmental Panel on Climate Change (IPCC) acknowledges the transport of carbon (C) across the inland water network as a key component of the global C cycle (Ciais et al., 2013), involving a significant lateral C transfer along the flow path and stimulating vertical C fluxes in the form of greenhouse gases.However, Earth system models (ESMs) of the climate system and biogeochemical cycles used for the IPCC 5th Assessment currently omit lateral C transfers and simulate only local vertical exchange of C between atmosphere, vegetation and soils from photosynthesis, respiration and fires (Regnier et al., 2013).This is a major knowledge gap because recent evidence, from multiple disciplines, has highlighted that anthropogenic disturbances likely increase the lateral C transfers along hillslopes of upland catchments and through streams and rivers (Battin et al., 2009;Cole et al., 2007;Regnier et al., 2013).This perturbation may significantly reduce the estimated carbon stored in terrestrial vegetation and soils (Regnier et al., 2013) and increase the C evasion from inland waters to the atmosphere.Thus, it is suggested that lateral carbon transfers induce a positive feedback on the coupled carbon cycle-climate system, enhancing atmospheric CO 2 levels and global temperature.
Despite this important paradigm shift in carbon cycle science, it must be recognized that the quantitative significance of inland waters for the global C budget entails large uncertainties.In particular, the horizontal flux of organic C through the terrestrial-aquatic interface is poorly constrained (Regnier et al., 2013).Global first-order estimates of this flux, calculated as the sum of estimates of fluvial total organic C (TOC) exports to the coastal ocean, particulate organic C (POC) burial in aquatic sediments and net-CO 2 evasion through the air-inland water interface of the landocean aquatic continuum (LOAC, Fig. 1), range from 1.5 to 2.7 Pg C yr −1 (Battin et al., 2009;Cole et al., 2007;Tranvik et al., 2009), i.e. of the order of 2-5 % of the terrestrial NPP.It is now broadly accepted that the CO 2 outgassing from inland waters is the major export path in the LOAC C budget (Battin et al., 2009;Ciais et al., 2013;Le Quéré et al., 2014;Regnier et al., 2013;Tranvik et al., 2009), highlighting the highly reactive character of continental aquatic systems.However, it remains challenging to attribute and quantify the sources of the CO 2 evasion, as it is generally not known how much of the evading CO 2 originates from terrestrial soil respiration, from in-stream respiration of terrestrially derived organic C or from other sources such as root respiration of wetland plants (Abril et al., 2014).This is not only true on the global scale, but also on the regional scale of large river catchments like the Amazon basin.Budget calculations from observations alone have limited capabilities to constrain such C exports from terrestrial ecosystems, in particular with regard to temporal and spatial variability.In this study, we present an integrated, physical-based modelling approach, which incorporates the various allochthonous sources of DOC and CO 2 to the inland water network, the lateral transfers of C along the inland water network, and transformation of C in transit and CO 2 exchange with the atmosphere in a temporally resolved and spatially explicit manner.We parameterize and develop the model for the Amazon basin, although it is intended to be generalized in future works to be applied on a global scale.We consider the Amazon basin as an appropriate but challenging benchmark test, as it is the world's largest river system with regard to discharge (206 000 m 3 s −1 , Callede et al., 2010) and one of the most productive ecosystems in the world (Grace, 2004).Richey et al. (2002) estimated the CO 2 evasion from the Amazon River system and its connected floodplains at 0.47 Pg C yr −1 , about 13 times the fluvial TOC exports to the Atlantic Ocean from this catchment.Such evasion flux corresponds to about 6 % of the average terrestrial NPP within the Amazon basin.In the Amazon River and its major tributaries, in-stream respiration of allochthonous OC is likely the dominant source of CO 2 .The study by Mayorga et al. (2005) further revealed that a small pool of labile organic carbon maintains high CO 2 levels in the water column, likely linked to inputs from the riparian zone, while the bulk of TOC transported in the river channel is older and more refractory.Richey et al. (2002) also showed that the intense seasonal flooding in the central Amazon basin is a major control of river CO 2 dynamics, suggesting that submerged leaf litter in flooded forests and root respiration of floating and emergent plants are important sources of CO 2 .In a more recent study, Abril et al. (2014) estimated that riparian wetlands in the Amazon River system export about half of their gross primary production (GPP) to rivers as TOC and dissolved CO 2 produced by autotrophic root respiration in wetland plants, while terrestrial ecosystems export only a few percent of their GPP.Vascular wetland plants, including flooded forests and floating grasses, clearly dominate primary production in the flooded areas, the autochthonous contribution from phy-toplankton and periphyton being negligible (Melack et al., 2009).Another specific challenge is the reproduction of the different DOC loadings from the different sub-basins of the Amazon.While most of the major tributaries are white or clear-water rivers with low to moderate average DOC concentrations of up to 6 mg C L −1 , the Rio Negro, which after the Rio Madeira is the second largest tributary of the Amazon, is a black-water river with twice the concentrations of DOC (Moreira-Turcq et al., 2003).
Recently, one of the first steps in modelling the Amazon River C dynamics was performed using a river carbon model (RivCM) coupled to the land surface scheme LPJmL (Lund-Potsdam-Jena managed land, Bondeau et al., 2007) to simulate fluvial C transfers in the Amazon basin (Langerwisch et al., 2016).While the model was able to roughly reproduce the annual DOC export to the coast, it still largely underestimated the CO 2 evasion from the inland water network to the atmosphere, indicating that C inputs into the river network and their subsequent transformation would need to be reassessed.In our study, we go a step further with the direct implementation of the non-conservative transport of C through the inland water network into the ORCHIDEE land surface model (Krinner et al., 2005).This approach has the advantage of accounting for the effects of the lateral exports on the carbon budgets of terrestrial ecosystems and could thus help refine the assessment of the terrestrial C sink and its feedback on the climate system.The newly developed model branch, called ORCHILEAK, represents DOC production from soils and canopy, DOC and CO 2 leaching from soils to river headstreams, DOC decomposition and CO 2 evasion to the atmosphere during its lateral transport in rivers, as well as exchange with the soil carbon and litter stocks in riparian wetlands.The production and leaching of DOC relies on a new soil carbon module ORCHIDEE-SOM (Camino Serrano, 2015) with a vertically resolved soil column.We simulate all C fluxes and stocks at half-hourly to daily time steps, which allows the representation of seasonal and inter-annual variations.We focus on the lateral transfer of dissolved CO 2 and dissolved organic C (DOC), which represents the major and more reactive proportion of TOC exports to the coasts in the Amazon basin (Moreira-Turcq et al., 2003).Although we neglect the lateral transport of POC, we simulate decomposition of submerged litter in floodplains and rivers as an important source of DOC and CO 2 to the water column.While it is of importance for the greenhouse-gas exchange, CH 4 evasion is assumed to be negligible with regard to C exports (Wilson et al., 2016).Further, we ignore the fluxes of carbonate alkalinity as, at average pH values of 6.5 to 7.2 typical of the Amazon basin (Richey et al., 1990), the concentrations of CO 2− 3 are negligible and, thus, the carbonate-buffering of CO 2 is limited.

Model developments
ORCHILEAK is based on the recent model branch ORCHIDEE-SOM (Camino Serrano, 2015) which relies on a novel module representing the vertical distribution of soil organic carbon (SOC) and associated transport and reaction processes.These processes include the production, consumption, adsorption, desorption and transport of DOC within the soil column as well as DOC exports from the soil column by drainage and surface runoff.In this study, the module is upgraded to represent DOC cycling in tropical rainforests, in particular by adding fluxes of DOC from the atmosphere and canopy with throughfall and by distinguishing soil carbon processes on non-flooded and flooded soils, including the direct input of DOC and CO 2 from the decomposition of submerged litter and soil carbon to the water column.The trunk version of ORCHIDEE, as well as the branch ORCHIDEE-SOM, includes a river routing module (Guimberteau et al., 2012;Polcher, 2003) that simulates the lateral transfer of water from one grid to another, representing the river channel as well as connected wetlands.Here, this routing module has been upgraded with a tracer transport equation to simulate the fluxes of DOC and CO 2 along the fluvial network, distinguishing two pools of DOC: labile and refractory DOC.In addition, the representation of the floodplain dynamics is improved in this study to better reproduce the seasonal flooding in the Amazon basin, which is a major controlling factor of the water (Guimberteau et al., 2012) and carbon flow dynamics along the river network (Richey et al., 1990).ORCHIDEE can be run at different spatial and temporal resolutions.Here, in line with Guimberteau et al. (2012), the model runs for calibration and model testing were performed at 1 • spatial resolution over the period 1980-2000, using the regional climate and wetland forcing for the Amazon from Guimberteau et al. (2012); forcing of land cover and land use change after Belward et al. (1999), Olson et al. (1983) and Hurtt et al. (2006); river flow directions from Vörösmarty et al. (2000); and soil parameters after Reynolds et al. (1999) and the Harmonized World Soil Database (FAO/IIASA/ISRIC/ISS-CAS/JRC, 2009).The necessary forcing data are listed in Table 1.As temporal resolution, we use the default 30 min time step for all vertical exchanges of water, carbon and energy between atmosphere, vegetation and soils, as well as the default 1-day time step for the lateral routing of water.In the following, the model description will be based on these spatial and temporal resolutions.To obtain initial soil carbon pools which are in steady state with the model set-up for the 1980-2000 period, the model was first run for 5000 years, looping over the full set of climate forcings and using the land use and an atmospheric pCO 2 as representative for the year 1980.The terrestrial C pools simulated during this initialization phase were subsequently used for the simulation over the period 1980-2000 with changing land cover and increasing atmospheric pCO 2 .This section starts with the representation of the soil hydrology and  the river routing scheme in ORCHIDEE and ORCHILEAK (Sect.2.1).Here, we give an overview of the features that are shared between the original version of ORCHIDEE (the configuration used by Guimberteau et al., 2012) and OR-CHILEAK and we then highlight the improvements that have been implemented in ORCHILEAK.In the second part, the mathematical formulation of DOC production and leaching from the soil as well as transport and transformation of DOC and CO 2 along the fluvial network is described (Sect.2.2).

Hydrology
Like most land surface schemes of ESMs, ORCHIDEE distinguishes two kinds of surface hydrology processes: (i) the water budget processes, which are mostly vertical and control the partitioning of precipitation into evapotranspiration, infiltration, production of surface runoff and drainage (Sect.2.1.1);(ii) the horizontal transfer, or routing, of grid-based simulated surface runoff and drainage along the river network (Sect.2.1.2,with improvements described in Sect.2.1.3).

Water budget and soil hydrology
In the vegetation canopy, rainfall is partitioned between interception loss and throughfall according to the leaf area index (LAI).The throughfall (possibly increased by snowmelt in cold climates and by return flow from the floodplains; see Sect.2.1.2) is then further subdivided into infiltration into the soil and surface runoff produced by infiltration excess.
In ORCHIDEE, the infiltration rate depends on precipitation rates, local slope and vegetation and is limited by the hydraulic conductivity of the soil, which defines a Hortonian surface runoff (d'Orgeval et al., 2008).The corresponding

Outline of basins Flow direction
Figure 2. Schematic representation of 4 ORCHIDEE grids x at 1 • spatial resolution for a simulation using a river routing scheme running at 0.5 • resolution.
parameterization is tightly linked to the soil moisture redistribution scheme, which is ruled by the Richards equation, solved here over a 2 m soil profile, using an 11-layer discretization, with layers of geometrically increasing depth (de Rosnay et al., 2002;Campoy et al., 2013).The redistribution of soil moisture is controlled by the soil hydraulic properties, transpiration and evaporation within the soil column, as well as a gravitational drainage at the soil bottom.All these processes are simulated at a 30 min time step and a 1 • resolution.
In addition, a bottom return flow feeding the soil is also accounted for in the presence of swamps, simulated at the daily time step of the routing scheme (Sect.2.1.2).

Routing of water along the river network, floodplains and swamps
The river routing module simulates the water exports from the soil column as river discharge along a distributed routing scheme, and it is possible to simulate lateral flows at a higher spatial resolution than the rest of the model to better describe the borders of watersheds within each grid box and   the directions of incoming or outgoing water from distinct basins (Fig. 2).For that, each ORCHIDEE grid cell x is divided into multiple subunits named "basins".As, in our case, we run simulations at 1 • resolution and use a routing scheme at 0.5 • resolution (Vörösmarty et al., 2000), each grid cell is simply sub-divided into four basins (Fig. 2).Note that all information derived from the forcing files or computed in the other modules has the resolution of the grid cell and is then downscaled to the basins within the routing module.In the following, variables on the grid scale are denoted by the index grid,x, while information on a basin scale are denoted by the index i.For a full overview of the variables and the system of indices used here, consult Table A1 in Appendix A.
The river routing aggregates the 30 min surface runoff and drainage computed by the soil hydrology module to the daily time step t of this module.As shown in Fig. 3, surface runoff and drainage initially feed a "fast" (S fast,H 2 O ) and a "slow" (S slow,H 2 O ) water reservoir, respectively (Eqs. 1, 2).The proportions of runoff (F RO,H 2 O,grid x,t ) and drainage (F DR,H 2 O,grid x,t ) assigned to each basin i within the grid x are scaled to the area of the basin (A total,i ) relative to that of the grid cell (A total,grid x ).S fast and S slow have distinct linear response timescales in each basin of the simulation domain, which are defined by a topographic index Topo grid x extracted from a forcing file (values range between 1 and 4 in our study area) and a factor τ which translates Topo grid x into a water residence time of each reservoir (Eqs. 3,4).Following the calibration of Guimberteau et al. (2012), both τ fast and τ slow are set to a value of 3.0 days.The river reservoir (S river ) in each basin i is mainly fed by the outflows of S fast , S slow and S river of the basins i − 1 lying immediately upstream (Eqs.5, 6, 7), but can, in addition, interact with two kinds of hydraulic sub-systems (the floodplains and the swamps), the maximum extents of which are defined by forcing files.Swamps are intended to mimic groundwater-fed wetlands.Where swamps are present, a constant fraction of the upstream inflow F up (Eq.7), which is scaled to the areal proportion of swamps (%swamp) in a given basin i, is diverted from the S river and added to the bottom of the soil column of the grid x containing the basin i (F up2swamp , Eq. 8).

Improved floodplain dynamics
Seasonal flooding in the Amazon is a major control of the hydraulic and C dynamics of the river system (Abril et al., 2014;Melack et al., 2009;Rasera et al., 2013;Richey et al., 1990Richey et al., , 2002)).This is particularly true in the central basin where the extent of flooded areas can increase from 4 to 16 % of the total area (Hamilton et al., 2002;Hess et al., 2003;Richey et al., 2002)

Original trunk version
When floodplains are present in a given basin, all water inputs from upstream basins (F up ) which are not infiltrating in swamps (F up2swamp ) are routed to S flood instead of S river (Eq.9).After floodplain and river reservoirs have been updated with in-and outflows for each basin (Eqs.5,11), the inundated fraction %flood is calculated first for each grid cell, and second for each basin within the grid cell.This sequential procedure is necessary, because the maximum floodable proportion (%flood max ), which is prescribed by the forcing file, is given at the resolution of the grid cells.The %flood per grid x is calculated from the total water storage in the floodplain reservoirs (S flood,H 2 O,grid x,t , Eq. 13) of all basins i contained in that grid cell, assuming a slightly convex slope of the floodable area (Eqs.14, 15), as this shape is typical of large lowland rivers like the Amazon (Hamilton et al., 2002;Huggett, 2016).In the original version of OR-CHIDEE (Fig. 5), the computation is performed as follows: first, a potential fraction of flooded area (%flood pot ) is calculated based on the total area of the grid cell (A total,grid x ) and a potential water level height on the floodplain (floodcri, set to 2 m by default) for which it is assumed that the whole grid cell is inundated (Eq.14, Fig. 5).The maximum flooded proportion (%flood max ) of the grid cell is defined by values reported in the PRIMA forcing file (see below), that is, %flood cannot exceed %flood max (Eq. 15).Second, the actual water level over the floodplain area (floodh) is calculated from %flood and the water storage in the floodplain reservoir S flood,H 2 O (Eq. 16).Finally, the %flood of each basin i within the grid x is calculated based on the S flood,H 2 O of the basin compared to that of the grid box and A total of the basin i compared to A total of grid x (Eq.17  % flood i,t = % flood grid x,t The PRIMA forcing file was introduced by Guimberteau et al. (2012) to represent the maximum spatial extent of swamps and floodplains on the scale of the entire Amazon basin.The available global wetland (swamps and floodplains) forcings (Lehner and Döll, 2004) are underrepresenting swamp and floodplain areas in this region, and were thus not sufficient to simulate water retention needed to reproduce the hydrograph of the Amazon River.The PRIMA data set was obtained using the maximum floodable areas derived from satellite imagery (Prigent et al., 2007), after subtraction of the vegetated proportion reported by Martinez and Le Toan (2007).The vegetated part of the maximum floodable area was assigned to "swamp" areas, which, as stated above, do not include a specific water body in ORCHIDEE.

Changes in ORCHILEAK
Although water retention in floodplains was validated by reproducing the water height over the floodplains (Guimberteau et al., 2012), the seasonality in the flooded area is still not well captured in the trunk version.Furthermore, according to the PRIMA forcing, the maximum floodable area in the central Amazon basin is < 5 %, while according to Richey et al. (2002) the areal proportion of inundated area is comprised of between 4 and 16 %, leaving a temporarily flooded proportion of 12 %.For the simulations with ORCHILEAK, we merged back the swamp and floodplain areas, thus relying directly on the maximum inundated area of Prigent et al. (2007), while, at the same time, keeping swamp areas as www.geosci-model-dev.net/10/3821/2017/Geosci.Model Dev., 10, 3821-3859, 2017 zone of return flow from the river to the bottom layer of the soil column (Fig. 6).With this modified forcing, %flood max increases to 10 % within the central Amazon basin, in better agreement with observations.To improve the representation of seasonal flooding using updated values of %flood max , the original equations to calculate the inflow of water to the floodplains and the extent of flooded area in each grid cell were altered as follows.First, floodplains are now only inundated when a threshold in river discharge is exceeded (F up lim , Eq. 18), and it is only the excess part of the river discharge that contributes to the flooding while the remainder is directly entering the next river reservoir (Eq.19).The threshold is defined for each grid by the median river reservoir water storage of each grid cell over the simulation period , which is derived in a first simulation with flooding deactivated, and then used as a forcing file for the model (Fig. 6).The choice of the median as threshold provides the advantage of a robust statistical measure and is similar to threshold of 90 % of long-term mean discharge used by Vörösmarty et al. (1989) for the Amazon.This modification assumes that a fraction of river water continues to be transported by the river instead of being entirely diverted to the floodplains.
While the default value for floodcri, like that used in global modelling, was set to 2 m in the trunk version, this value is not applicable to the Amazon, where water levels of up to 12 m have been reported in the central Amazon floodplain (Trigg et al., 2009).Thus, instead of using a single value for floodcri as was previously done, we now first compute for each grid cell the 95th percentile of all simulated water level heights over the floodplain area for the simulation period 1980-2000 (floodh 95th , Eq.21, see Fig. 5).We used the regional data set of monthly inundated areas from Hamilton et al. (2011) for validation in the Roraima and Llanos de Moxos wetland areas, which cover part of our simulation period.For inundation in the central Amazon basin, we used the data from Hess et al. (2003) as summarized in Richey et al. (2002) for validation.
Following the changes in the flooding scheme, we recalibrated two parameters in order to reproduce the monthly discharges from the Amazon and its major tributaries: (1) we decrease the water residence time on the floodplains by changing τ flood from 2.5 days as used by Guimberteau et al. (2012) to 1.4 days (Eq.12), and (2) we halved the proportion of water diverted to swamps by setting f swamp from 0.2 to 0.1 (Eq.8), while using the same forcing for %swamp as Guimberteau et al. (2012).In addition, because %flood can now take values close to 100 % in some areas, we modified the equation to calculate the outflow from the river reservoir, which is not decreased anymore depending on %flood (Eq.22).The simulated river discharges were validated against gauging data from ORE HYBAM (Cochonneau et al., 2006) and mean monthly discharges provided by the Global Runoff Data Centre (GRDC, 2016).
In ORCHILEAK, for the purpose of calculating CO 2 evasion from the river network, the river reservoir is now assigned a surface area as well (A river ).The base surface area A river (A river basic ) per grid cell is extracted from a forcing file derived from the global river surface maps of Lauerwald et al. (2015).Following the findings by Rasera et al. (2013), we assume that the surface area of small rivers (A river small , width < 100 m) can increase by about 20 % from low to high water stages, whereas the area of larger rivers (A river large , width ≥ 100 m) increases by about 10 %.Assuming the 10th and 90th percentile of S river,H 2 O over the simulation period 1980-2000 (S river,H 2 O,grid x,10th , S river,H 2 O,grid x,90th , Fig. 6) as representative for the low and high water stages, an actual A river (A river act ) is calculated at each time step depending on S river,H 2 O .As the A river forcings likely underestimate the total A river (Lauerwald et al., 2015), it is assumed that A river basic represent A river at low water stage.A river act per basin i is calculated from A river per grid x containing that basin, scaling to the square root of S river,H 2 O , because S river,H 2 O is linearly related to discharge (Eq.27) and it was empirically shown that stream width scales roughly with the square root of discharge (Raymond et al., 2012(Raymond et al., , 2013)).Assuming that stream length does not change significantly, the relative change in stream width equals the relative change in A river act .
A river basic,grid x = A river small,grid x + A river large,grid x The difference between A river act and A river basic gives a seasonally flooded area directly adjacent to the river (%flood river , Eqs. 28, 29).This flooded area induced by changes in water levels in the river was then added to the total flooded proportion of soils (%flood total , Eqs. 30, 31).Note, however, that for the calculation of C inputs from flooded soils to the water column (Sect.2.3), S flood and S river need again to be distinguished.

% flood river,grid x,t =
A river act,grid x,t − A river basic,grid x A total,grid x (28) % flood river,i,t = A river act,grid x,t − A river basic,grid x • S river,H 2 O,i,t % flood total,grid x,t = % flood grid x,t + % flood river,grid x,t (30) % flood total,i,t = % flood i,t + % flood river,i,t (31) 2.2 Carbon dynamics along the vegetation-soil-water continuum 2.2.1 Overview of the DOC transport scheme DOC and CO 2 are exported through the terrestrial-aquatic interface by runoff (F RO ) and drainage (F DR ), respectively (Fig. 3).Part of the terrestrial DOC stems from throughfall (F TF = F WD2ground +F can2ground , see below), whereas the other part stems from the decomposition of litter and soil organic carbon (F dec terr ).DOC exports from flooded areas to the river network are another important source, because F TF and the decomposition of submerged litter and soil carbon in the floodplains (F soil2flood ) add directly to the DOC storage in the overlying water column and, from there, a delayed flux (F flood out ) feeds S river .In addition, streams and rivers extend laterally during high flow periods (see Sect. 2.1.3)and there is thus a direct input of DOC from litter and SOC decomposition on or in seasonally inundated soils immediately adjacent to the stream bed into S river (F soil2river ).DOC and CO 2 are transported as passive tracers with the fluxes of water through the different reservoirs of the routing scheme (see Sect. 2.1) and can feed back into the soil system via two mechanisms: (1) re-infiltration from the floodplain reservoir into the first layer of the soil column (F flood2soil ); (2) infiltration of DOC into the bottom layer of the soil column entrained with water entering swamps (F up2swamp ) (Fig. 3).In addition, DOC is mineralized to CO 2 in transit and CO 2 is evading to the atmosphere from the water surface.Depending on the relative magnitude between inputs, outputs and in situ transformations, the storage of DOC in the canopy, soil, fast, slow, river and floodplain reservoirs (S can , S soil , S fast , S slow , S river and S flood ) can thus increase or decrease over different time periods.For the routing of DOC, we distinguish two pools: a labile and a refractory pool.Like the cycling of water and C in vegetation and soils, the allochthonous inputs of DOC from S can and S soil into the inland water network (F RO , F DR , F soil2flood , F soil2river , see Fig. 3) are computed at a temporal resolution of 30 min and at the spatial resolution of the grid cell.The lateral transfer between the S fast , S slow , S river and S flood and the transformation of C within those storage reservoirs are only simulated at a daily time step and at the spatial resolution of the basin.Therefore, to simulate the lateral transfers, the allochthonous DOC and CO 2 inputs are first aggregated over 48 30 min time steps until 1 full day is over.The fluxes from the water column back into the soil column (F flood2soil , F up2swamp in Fig. 3) are simulated at the daily time step of the routing module, but are used as inputs in the soil carbon module, which runs at a 30 min temporal resolution.This is achieved by downscaling the daily fluxes uniformly over the 48 30 min time steps of the following day of simulation.The evasion of CO 2 from river and floodplain water surfaces (F river2atm , F flood2atm ) is also simulated at the daily time step of the routing module, but to approximate the continuous interplay of CO 2 inputs and CO 2 evasion controlling the water-air gradient in CO 2 partial pressures (pCO 2 ) a much shorter time step of 6 min is used, and the CO 2 inputs to the water column are thus uniformly distributed over the 240 6 min time step contained in each day.The following subsections describe in more detail the simulation of DOC in precipitation and throughfall (Sect.2.2.2), production of DOC, and its export through the terrestrial-aquatic interface (Sect.2.2.3),CO 2 inputs through the terrestrial-aquatic interface (Sect.2.2.4), and in-transit DOC mineralization and CO 2 evasion along the inland water network (Sect.2.2.5).

DOC in precipitation and throughfall
Reported average rain DOC concentrations in the Amazon basin are significant with 1.3 to 3.9 mg C L −1 (  rivers of the region (Moreira-Turcq et al., 2003).The spatial variation in rain DOC concentration is unknown and we thus assumed a constant value of 2.4 mg C L −1 throughout the Amazon basin, from the average of reported literature values (Table 5).Observed average DOC concentrations in throughfall are higher than in precipitation because of the DOC enrichment of leaf-intercepted water due to evaporation losses and dissolution of organic carbon from leaf leachates and dry deposition.Reported annual throughfall DOC flux (F TF ) in the Amazonian rainforest varies little, from 14.8 to 19.0 g C m −2 yr −1 (see  (Johnson et al., 2006).Here, we used the time-series data on throughfall DOC fluxes in southern Amazonia from Johnson et al. (2006) to set up and calibrate a simple model of throughfall DOC fluxes.
In ORCHILEAK, the wet deposition of DOC, F WD , is calculated from precipitation and the prescribed constant concentration of 2.4 mg C L −1 , which also equals the minimum throughfall concentration in the time series by Johnson et al. (2006).For each of the 13 ORCHIDEE plant functional types (PFTs) which are potentially present in a grid cell, the wet deposition of DOC onto the canopy (F WD2can ) and the direct precipitation of DOC onto the ground (F WD2ground ) directly scales to the corresponding water fluxes simulated in the hydrology module.According to our simulation, F WD contributes to only about one-third of the F TF at our calibration site (14.9 g C m −2 yr −1 ; Johnson et al., 2006).Thus we assumed that the unaccounted flux of 10 g C m −2 yr −1 must originate from dry deposition onto the canopy or leaf leachates.We further assumed that this dry addition of soluble organic carbon (F add2can ) does not vary over time and scales to the leaf biomass (which, in the model, is directly related to leaf area).Based on the simulated leaf biomass of 457 ± 1 g C m −2 for tropical rainforests at the field-site location, we calibrated F add2can at 6 × 10 −5 g C per day and per gram of carbon in the leaf biomass (Eq.32).For agricultural and grasslands, we set F add2can to zero.
Whenever intercepted water from the canopy falls to the ground (F can2ground ), the related flux of DOC (F can2ground ) will empty the storage of DOC in the canopy (S can ) at once unless a maximum concentration DOC max of 100 mg DOC kg H 2 O −1 (Eq.33) in F can2ground is exceeded.This value corresponds to the maximum concentration observed by Johnson et al. (2006).Beyond this threshold, F can2ground is set as the product of the water flux and the maximum concentration, and the DOC in excess is assumed to remain in the canopy reservoir S can .This threshold prevents unreasonably high DOC concentrations in the first throughfall events after dry periods and allows simulation of progressive depletion of the S can reservoir after a time of significant DOC accumulation.At each 30 min time step, F WD2can , F add2can and F can2ground are calculated and subsequently used to update the DOC storage in the canopy at each grid x and PFT v (Eq.34).F add2can,DOC,grid x,v,t = leaf biomass grid x,v,t • 10 −5 dt day (32) S can,DOC,grid x,v,t+1 = S can,DOC,grid x,v,t F TF is calculated as the sum of the non-intercepted wet deposition F WD2ground and F can2ground (Eq.35).Based on the range of values reported in the literature (Aitkenhead-Peterson et al., 2003), we assume that half of the DOC reaching the ground is labile (DOC lab ) while the other half is refractory (DOC ref ) (Eq. 36).F TF then infiltrates into the topsoil or adds to S flood in areas where it falls on inundated land (see Sect. 2.2.4).

Production and export of soil DOC through the terrestrial-aquatic interface
ORCHILEAK is largely based on ORCHIDEE-SOM, the new soil carbon module simulating microbial production and consumption of DOC, its adsorption and desorption onto or from mineral surfaces, the vertical advective and diffusive fluxes of DOC within the soil profile, and the exports of DOC from the soil via surface runoff and drainage (Camino Serrano, 2015).Consistent with the soil hydrology module (Campoy et al., 2013;de Rosnay et al., 2002), the carbon dynamics are resolved using a discretization of a 2 m soil profile into 11 layers geometrically increasing in depth and running at a 30 min time step (Camino Serrano, 2015).DOC is produced from the decomposition of litter and SOC (Eqs.37-40), and consumed by further decomposition (Eqs. 41,42).Here, the soil carbon module has been modified to better represent the soil DOC dynamics in the Amazon.First, decomposition on non-flooded (F dec terr ) and flooded (F dec flood ) soils is distinguished, with decomposition rates of the litter, SOC and DOC pools 3 times slower when soils are flooded (Rueda-Delgado et al., 2006).Second, in "poor soils" characterized by low pH and low nutrient levels such as Podzols, Arenosols or soils located in black-water swamps (referred to as igapo in the Amazon basin), decomposition rates are significantly reduced.Here, we assume a reduction by a factor of 2, following findings from the literature (Bardy et al., 2011;Vitousek and Hobbie, 2000;Vitousek and Sanford, 1986).This feature was implemented in the model by adding a layer defining the areal proportion of "poor soils" in the soil-forcing file.The spatial distribution of Podzols and Arenosols was derived from the Harmonized World Soil Database (FAO/IIASA/ISRIC/ISS-CAS/JRC, 2009).To determine the spatial distribution of igapo forest soils, we used www.geosci-model-dev.net/10/3821/2017/Geosci.Model Dev., 10, 3821-3859, 2017 the PRIMA forcing for swamps in combination with the boundaries of the Rio Negro catchment as derived from the 0.5 • river network (Fig. 6).
F dec terr,SOC pool,grid x,v,l,t = S soil,SOC pool,grid x,v,l,t F dec flood,SOC pool,grid x,v,l,t = S soil,SOC pool,grid x,v,l,t F dec terr,litter pool,grid x,v,l,t = S soil,litter pool,grid x,v,l,t • F dec flood,litter pool,grid x,v,l,t = S soil,litter pool,grid x,v,l,t F dec terr,DOC pool,grid x,v,l,t = S soil,DOC pool,grid x,v,l,t F dec flood,DOC pool,grid x,v,l,t = S soil,DOC pool,grid x,v,l,t The soil carbon module distinguishes 3 different pools of DOC depending on the source material: active, slow and passive (Camino Serrano, 2015).The DOC derived from the active SOC pool and metabolic litter is assigned to the active DOC pool, while the DOC derived from the slow and passive SOC pools are assigned to the slow and passive DOC pools, respectively .A part of DOC derived from structural plant litter, which is related to the lignin structure of the litter pool (Krinner et al., 2005), is allocated to the slow DOC pool, while the remainder feeds the active DOC pool.The proportion of the decomposed litter and SOC that is transformed into DOC instead of CO 2 depends on the carbon use efficiency (CUE), set here to a value of 0.5 (Manzoni et al., 2012).Taken that the same residence time for the slow and passive DOC pools is used in ORCHIDEE-SOM (Camino Serrano, 2015), we merge these two pools when computing throughfall and lateral transport of DOC.Thus, the labile pool is identical to the active pool of the soil carbon module, while the refractory pool combines the slow and passive pools.The labile (F TF,DOC lab ) and refractory (F TF,DOC ref ) proportions of throughfall DOC are added to the active and slow DOC pools of the first soil layer, respectively.S soil,DOC active,grid x,v,t = 11 l=1 F dec terr,litter str,grid x,v,l,t + F dec flood,litter str,grid x,v,l,t Along with decomposition, DOC is lost from the soil column through lateral exports with surface runoff and/or drainage, which occur at the top and bottom of the soil column, respectively.The DOC export by drainage at the bottom of the soil is proportional to the DOC concentration in the deepest (11th) soil layer (Eq.46).Surface runoff occurs when the maximum infiltration rate is exceeded, beyond which the excess water does not enter the soil column anymore.Because the first soil layers are extremely thin, it is assumed here that surface runoff can entrain DOC from the first five layers of the soil column, which together have a thickness of 4.5 cm (Eq.47).In each basin, the DOC release is proportional to the mean DOC concentration in this zone of the soil column as well as to the areal extent of the saturated zone around headwaters, as detailed below.The usually higher DOC concentration in the topsoil compared to the subsoils is mainly due to the higher inputs of plant litter into and onto the topsoil.However, DOC is efficiently transported between the soil layers along with the vertical flow of water through the soil matrix (F soil adv , Eqs. 51-52).Therefore, a part of the DOC exported with the drainage is not produced in situ but rather originates from percolation across the entire soil column.The vertical DOC transport within the soils, as well as for the export of DOC with surface runoff, is not directly computed as the product of water flux and DOC concentration.Instead, a reduction factor (red DOC ) is applied to account for the effect of preferential vertical flow paths, e.g.along macrospores produced by the root system (Karup et al., 2016) and in zones of reduced flow rates which increase the DOC residence time in the remaining parts of the soil.Only in "poor soils" is the flow of DOC not reduced relative to the flow of water (no reduction, Eq. 54).This allows for their poor filtering capacity, which is the cause of the very high DOC concentrations in groundwater below Podzols and black-water swamps, to be accounted for (Brinkmann, 1984;McClain et al., 1997).While the effect of preferential flow paths should be envisioned as a general concept in ORCHILEAK, the introduction of "poor soils" is specific to tropical black-water systems.It remains to be shown in future work how their effects will have to be parametrized in other climate zones, for instance in the Boreal zone, where Podzols are abundant.
DOC exports with surface runoff is even further reduced, because the riverine DOC mostly derives from saturated soils in the direct vicinity of surface waters (Idir et al., 1999).As we do not have direct information on the density of headwater streams on a small scale and the extent of the saturated, riparian zone, the reduction in DOC exports with surface runoff (red connect ) was scaled to the storage of water in S fast and S slow (Eq.55).We assumed these reservoirs to represent the water stored in groundwater and headwater streams (S river being attributed to wider water bodies due to the coarse resolution (0.5 Although mineralization of litter, SOC and DOC in the soil is simulated in ORCHIDEE, the CO 2 partial pressure in the soil air and soil solution of the different layers is not represented.Thus, we implemented simple estimates of these soil-derived CO 2 inputs in order to reproduce the observed CO 2 evasion fluxes from the water surface of the fluvial network.For simulating the export of CO 2 with surface runoff and drainage, we use fixed concentrations of 20 mg C L −1 (pCO 2 of 50 000 µatm at 25 • C) and 2 mg C L −1 (pCO 2 of 5000 µatm at 25 • C), respectively, derived from reported literature values (Davidson et al., 2010;Johnson et al., 2008;Saunders et al., 2006).The lateral exports of CO 2 dissolved in soil water are then calculated by multiplying these CO 2 concentrations with the water fluxes from surface runoff and drainage simulated at half-hourly time step in the soil hydrology module (Eqs.56, 57).Next, the computed lateral fluxes of CO 2 exported out of soils are subtracted from the total soil respiration and the remainder, by far the dominant fraction (Davidson et al., 2010), is assumed to evade directly to the atmosphere through the topsoil (Eq.58).Carbonate chemistry and export of alkalinity are neglected.
In floodplains, mineralization of submerged litter and soil carbon are considered to be sources of CO 2 to S flood (Eq.59).In addition, we allocated the root respiration in inundated areas to the "CO 2 inputs to S flood " term.The lateral transfer of CO 2 by advection and the re-infiltration of dissolved CO 2 into swamps and on floodplains are simulated following the approach implemented for DOC (Fig. 3, and preceding subsections).

Carbon transport and transformation along the inland water network
Transport and transformation of terrestrially derived C in the river system are implemented into the river routing module.
The lateral transport of DOC and CO 2 between reservoirs are assumed to be proportional to the water fluxes, that is, the exports from each reservoir to the next have the same concentration of DOC and CO 2 as in the reservoir from which they originate (Eq.60).The same holds true for infiltration on the floodplains (F flood2soil , Eq. 61).The inputs from upstream F up are the sum of F fast out , F slow out and F river out of all basins i − 1 lying directly upstream (Eq.62), and inputs into swamps (F up2swamp , Eq. 63), S flood (F up2flood , Eq. 64) and S river (F up2river , Eq. 65) have all the same concentrations as F up .
* : "fast", "slow", "stream", or "flood" reservoir; C spec: As discussed above, in the routing scheme, we distinguish two pools of DOC: the labile (DOC lab ), which corresponds to the active DOC pool of the soil carbon module, and the refractory pool (DOC ref ), which combines the slow and passive pool of the soil carbon module.For each pool, the DOC stocks in S fast and S low are then updated from the balance between the C inputs simulated in the soil carbon module at a 30 min time step and aggregated to the 1-day time step of the routing module, and the outflows of C which are proportional to the water fluxes (Eqs.66, 67).S river in basin i is augmented by the sum of outflows from the fast, slow and river reservoirs of the basins located directly upstream (i−1), minus the flows diverted to the subsoil of swamps and into floodplains (Eq.68).The floodplains (S flood ) receive inputs from upstream (F up2flood ) and transfer C to the river reservoir (F flood out ) and via infiltration into the soil (F flood2soil ) (Eq. 69).The inputs of DOC from the decomposition of inundated SOC and litter are added to S river and S flood accord- For Eqs. ( 68), ( 69): F soil2flood only for DOC; for CO 2 , see Eqs. ( 83), (84).
At each daily time step, after the lateral transfers along the flow path have been calculated, DOC decomposition and CO 2 evasion within the river and floodplain reservoirs are simulated.The continuous CO 2 production and CO 2 evasion from the aquatic network are computed using a much finer integration time step of 1/240 day (6 min) than the one of the river routing scheme to ensure precision of our numerical scheme.In addition, CO 2 inputs from the decomposition from flooded SOC and litter are also added at the same time step to represent the continuous additions of CO 2 during the water-atmosphere gas exchange.
For each 6 min time step, the pCO 2 in the water column is calculated from the concentration of dissolved CO 2 and the temperature-dependent solubility of CO 2 (K CO 2 ) (Eq. 70).The water temperature (T water ) needed to calculate K CO 2 (Telmer and Veizer, 1999) (Eq.71) is derived from the average air temperature close at the ground (T ground ) over the whole 1-day time step of the routing scheme (Eq.72, R 2 = 0.56, σ = 0.91 • C).This equation was empirically derived using values from the ORE HYBAM data set (Cochonneau et al., 2006) observed at a 10-day interval over the years 1999 and 2000 at three sampling locations (Fig. 7, see Fig. 4 for location).As the linear fits for each sampling location are quite similar (Fig. 6a), we consider the prediction equation derived for the total of observed data to be representative.Note that the slope is quite similar to that (0.82) found by Lauerwald et al. (2015) for average monthly T water using a global data  set.Furthermore, we investigated whether the correlations could be improved by introducing a time lag between T water and T ground , as suggested in the literature (Ducharne, 2008;Van Vliet et al., 2011).However, no significant improvement could be achieved (Fig. 7b), and we thus maintained Eq. ( 72 The same water temperature is used for the calculation of the Schmidt number (Sc) (Wanninkhof, 1992) (Eq.73), which is needed to calculate the actual gas exchange velocity from the standard conditions k 600 (Eqs.74, 75).We used distinct values of k 600 for rivers (k river,600 ) and for swamps (k swamp,600 ) to account for the reduced effect of the wind in flooded forests.The value k swamp,600 = 0.65 m d −1 is taken from Richey et al. (2002) while the value for k river,600 = 3.5 m d −1 corresponds to the average of the values reported in Alin et al. (2011).For the calculation of k flood,600 on the floodplains, we assumed that open floodplains have the same gas exchange velocity as the rivers, while within flooded forests (represented by %swamp), the gas exchange velocity is set to k swamp,600 .As the gas exchange is calculated for the • pCO 2flood,i,t − pCO 2atm,t • 12.011 The instream decomposition of terrestrial DOC is calculated using base rate constants for labile and refractory DOC, k DOC lab = 0.3 day −1 and k DOC ref = 0.01 day −1 , respectively (Eqs.79, 80).These values correspond to half-life times of 2 days and 80 days respectively.The value for k DOC lab is thus in agreement with Devol and Hedges (2001), who conclude that DOC lab in the Amazon River must have a very short half-life of hours to a few days.The value of k DOC ref also corresponds to the lower range of respiration rates found for Rio Solimoes of 0.2 µM h −1 (Amon and Benner, 1996) if an average concentration of about 5 mg C L −1 is assumed (see Moreira-Turcq et al., 2003).We assumed that the values for the rate constants are valid for an average T water of 28 • C (consistent with experiments of Amon and Benner, 1996, and the average temperature simulated here) and apply a temperature sensitivity factor on decomposition rates after Hanson et al. (2011) At each 6 min time step, the CO 2 produced from the decomposition of DOC is added to the relevant reservoirs (Eqs.81-84).For S fast , S river and S flood , the amount of evading CO 2 is subtracted from the CO 2 stocks (Eqs.82-84).For Geosci.Model Dev., 10,2017 www.geosci-model-dev.net/10/3821/2017/S river and S flood , the inputs of CO 2 from the decomposition of inundated SOC and litter are added to these reservoirs, based on the relative contribution of swollen rivers (%flood river ) and floodplains (%flood) on the total fraction of inundated soils (%flood total ) (Eqs.83-84).

Model calibration and evaluation
The main strategy was to start with the calibration of the hydrology, before calibrating the fluxes of carbon.We started from the forcing data and parametrization used by Guimberteau et al. (2012), and thus already had an initial calibration for that model.As we changed the flooding scheme and increased the maximum floodable area, we had to recalibrate discharge, in particular the residence time of water in the floodplains τ flood .Due to the increased floodable area, more water is infiltrating into the topsoil on the floodplain and, thus, we had to reduce the water infiltrating into the subsoil (f swamp ) in order to reproduce the total amount of discharge.The recalibration of discharge focused mainly on reproducing the river discharge at Obidos, the most downstream discharge gauge.The idea is that the discharge dynamics at the basin outlet integrates all hydrological processes in the basin and determines the exports of water and matter to the coast.Nevertheless, the discharges from major sub-basins are evaluated as well.
For the fluxes of C along the terrestrial-aquatic continuum, we build on the default calibration of vegetation processes in ORCHIDEE and on the calibration of soil C processes in ORCHIDEE-SOM (Camino Serrano, 2015), and based on that we tried to reproduce observed DOC exports from the soil to the river network by F RO and F DR , before evaluating the model performance with regard to reproducing observed DOC concentrations in the river (Table 2).The main parameters controlling the DOC concentration in F RO and F DR relative to DOC concentrations in the soil solu-tion are S fast+slow,H 2 O,ref and red DOC,base .As empirical data for calibration and validation are limited, we started with parameter values taken from the literature or based on assumptions.The parameter red DOC,base was set to a value of 0.2 following Braun (2002).The S fast+slow,H 2 O,ref was set to 160 mm, which is about the 90th percentile of S fast,H 2 O + S slow,H 2 O within the Amazon basin.The decomposition rates for labile and refractory DOC within the inland water network, k DOC lab = 0.3 day −1 and k DOC ref = 0.01 day −1 , were also taken from the literature (see Sect. 2.2.5).Nevertheless, we made sure that the simulated DOC concentrations in F RO and F DR are comparable to values reported in the literature, and that deviations between simulated and observed DOC concentrations in the rivers are minimal.In that context, we performed a sensitivity analysis with regard to model performance for changes in S fast+slow,H 2 O,ref , red DOC,base , k DOC lab and k DOC ref .
3 Model results and discussion

Evaluation of simulated seasonal flooding and river discharge
The upgraded river routing scheme allows us to reproduce seasonal inundation in the Amazon basin (Fig. 8).The improvement using ORCHILEAK instead of the trunk version of ORCHIDEE is in particular visible for the central Amazon basin (Fig. 8a, see Fig. 4 for location).However, compared to the observed inundation reported by Richey et al. (2002), our simulation underestimates the total areal extent of inundation.This is not surprising as our forcing data derived from space-borne microwave remote sensing (Prigent et al., 2007) excludes flooded forests with dense canopies covering free water surfaces and does not capture small water bodies.In contrast, the observed inundation from Richey et al. (2002) www.geosci-model-dev.net/10/3821/2017/Geosci.Model Dev., 10, 3821-3859, 2017  Guimberteau et al., 2012) versus ORCHILEAK.In addition, it is shown how ORCHILEAK would perform with the τ flood = 2.5 and f swamp = 0.2 used in Guimberteau et al. (2012).As performance measures, the RMSE and Nash-Sutcliffe efficiency (NSE) are reported * .In most cases, only the performance in reproducing the seasonality is reported, as it is presented in Fig. 10.For 4 stream gauges with time-series data, the performance in reproducing these time series is additionally reported.For locations and full names of gauging stations, see Fig. 4 and its caption.An NSE = 1 would mean a perfect fit between observed and simulated values.A NSE = 0 means that using the mean observed value as constant simulated value would lead to as much deviation between observed and predicted values as using the actual simulated values.If NSE is negative, there is more deviation between simulated and observed values than between the observed values and their mean.
was derived from airborne radar imagery, which is able to detect flooded areas in more detail and at higher resolution (Hess et al., 2003).Nevertheless, the simulated spatial pattern inundation throughout the Amazon basin correlates well with the high-resolution airborne observations (Hess et al., 2015) (Fig. 9) The observed inundation data for the Roraima and Llanos de Moxos wetlands (Hamilton et al., 2002;Hamilton et al., 2011) were derived from space-borne microwave imagery, and are thus, in terms of spatial resolution and detail, more directly comparable to our forcing data.Therefore, the good match between observed and simulated inundation in these regions highlights the good performance of our new flooding module in ORCHILEAK (Fig. 8b).Nevertheless, it is important to keep in mind that while the overall seasonality of inundation is well reproduced in all regions, the total inundated area across the Amazon basin is likely underestimated because of our choice of forcing data.
After recalibrating the outflow velocity from the floodplains and reducing the amount of water redirected to swamps (τ flood = 1.4,f swamp = 0.1), the simulated discharges are in general quite close to those simulated by Guimberteau et al. (2012) (Fig. 10, Table 3).In the southern tributaries of the Amazon basin (Rio Jurua, Purus, Madeira, Tapajos, Xingu), we overestimate the discharge during highflow periods (February to April) while for the rest of the year our simulation is well in line with observations.This might be due to a bias in the meteorological forcing data, which could give too much weight to very rainy spots during the interpolation process, or to an underestimation of simulated evapotranspiration compared to flux tower measurements (Guimberteau et al., 2012).For the northern tributaries (Rio Japura and Rio Negro), such an overestimation during high flows is not visible.Along the main stem (Amazon at Obidos, Rio Solimoes at Sau Paulo de Olivenca), the seasonality is reproduced very well except for Rio Solimoes at Manacapuru where the simulated discharge peak occurs 1 month too early, due to backwater effects by Rio Negro and Rio Madeira (Meade et al., 1991), processes which are not accounted for in ORCHILEAK nor in the trunk version of ORCHIDEE.

Vegetation, litter and soil carbon
The Amazon basin is largely dominated by tropical rainforest.Other notable plant functional types (PFTs) in the study area are tropical dry forest, i.e. deciduous tropical forest with litter fall during the dry period, and tropical C 3 and C 4 grasslands (Table 4).C 3 cropland contributes with an areal proportion of 1 %, mainly in the form of sugar cane plantations.All other PFTs have an areal proportion smaller than 1 %.Over the simulation period , the land use forcings give a slight increase in C 4 grasslands and croplands at the expense of tropical rainforest (Table 4).Table 4a summarizes yearly-mean NPP per PFT reported in the literature and simulated with ORCHILEAK, using the default settings for vegetation simulation.Overall, simulated values are in good agreement with those reported in the literature, especially for the dominant PFTs (rainforests).Values for C 3 grassland are compared to a study in the Andes, as most C 3 grassland in the Amazon basin is found in high altitudes at the western rim of the study area.For C 4 grassland, a rather wide range of NPP has been reported, with the highest values for grass-dominated wetland systems which are important for the C biogeochemistry in the Amazon floodplains (Melack et al., 2009).In that specific area, the average annual NPP for this PFT is simulated at around 2900 g C m 2 yr −1 , i.e. still at the lower end of the reported value range for C 4 wetland grasses.In the southernmost part of our study areas, the average simulated NPP for simulated C 4 grassland goes below 1500 g C m 2 yr −1 .Figure 11 shows the spatial heterogeneity in simulated average NPP 1980 to 2000.The spatial pattern reflects the relatively low NPP of rainforest compared to tropical grasses.Within the Amazon basin, the tropical grasses in the lower Amazon floodplains and in the Llanos de Mojos show the highest average NPP.The simulated soil carbon stocks in the Amazon basin are in good agreement with the Harmonized Worlds Soil Database (Table 4b).

DOC in precipitation and throughfall
Figure 11 shows the spatial patterns in simulated averages of DOC production in the canopy (sum of dry deposition of soluble organic C and leaching of DOC from leaves, F add2can,DOC ), wet deposition of DOC (DOC in rain, F WD,DOC ) and throughfall DOC flux (F TF,DOC ).In most parts, F add2can,DOC contributes more to F TF,DOC than F WD,DOC .The patterns in F add2can,DOC are mainly controlled by the distribution of tropical rainforest and tropical dry tropical forests, because, due to limitations in calibration data, we do not simulate this flux for grasslands or croplands.F WD,DOC follows the patterns of precipitation, as we use fixed DOC concentrations for this flux.Simulated average values for F TF,DOC range from 0 to about 20 g C m 2 yr −1 , with the highest fluxes to be found where dense rainforests coincide with highest average precipitation, like in the northwest Amazon basin.
Our simple representation of throughfall DOC fluxes is able to reproduce the yearly-mean and seasonal variations www.geosci-model-dev.net/10/3821/2017/Geosci.Model Dev., 10, 3821-3859, 2017   4) Williams et al. (1997): July 1989 to June 1990; here throughfall is assumed to equal rain, as there is no vegetation on the lake. in throughfall DOC concentrations observed by Johnson et al. (2006) in southern Amazonia (Fig. 12).Although the throughfall DOC was calibrated only for this study area, it reproduces the observed yearly-mean fluxes in north-west and central Amazonia (Filoso et al., 1999;Tobón et al., 2004) in a satisfying way as well (Fig. 12a, Table 5).Interestingly, the annual throughfall DOC fluxes do not differ much among these very different regions of the Amazon.In particular, the average annual precipitation differs substantially from 3400 mm yr −1 in the north-west part of the basin (locations 1, 2 and 3 correspond to points TF1, TF2 and TF3 in Fig. 4) to only about 2000-2200 mm yr −1 at the other two locations in the central and southern part of Amazonia (see Table 5).Similar throughfall flux has also been reported for tropical rainforest in Indonesia (12.6 to 16.4 g C m −2 yr −1 , Fujii et al., 2011) as well as for primary, sub-tropical rainforests in Puerto Rico (13.2 g C m −2 yr −1 , Heartsill-Scalley et al., 2007) and Taiwan (18.9 g C m −2 yr −1 Liu and Sheu, 2003).

Exports of DOC from soils to headwaters and floodplains
Comparing our simulation results to observed export fluxes through the terrestrial-aquatic interface is rather difficult, because studies with robust data are rare and the coarse resolution of our simulation in combination with the global soil forcing data may not reproduce the soil hydrology on the plot scale.Nevertheless, we attempted such comparison for three headwater catchments located far apart in the Amazon basin (Table 6, RO1-3, Fig. 4).All three case studies have more recent sampling times than our simulation period, and we thus compared observations with simulation results averaged over the 1980-2000 period.The first basin used for comparison is a small black-water and headwater catchment in the lower Rio Negro basin (RO1 in Table 6) (Waterloo et al., 2006).While our forcing data agree with the reported annual precipitation in the region, ORCHILEAK underesti-  For the stations at Rio Negro, Rio Purus and Rio Tabajos as well as for the Amazon at Obidos, observed discharges are derived from ORE HYBAM gauging data for the same period.For the other stations, long-term average monthly discharges from GRDC data set have been used, which cover a longer period: Amazon at Sao Paolo de Olivenca , Rio Madeira at Porto Velho , Rio Japura at Acanaui , Rio Jurua at Gaviao (1972Gaviao ( -2010)), Rio Xingu at Atamira .See Table 3.
mates the contribution of surface runoff to total runoff by a factor of 2. Nevertheless, the simulated DOC concentrations in F DR and F RO agree well with the observed values (Table 6).We can also compare to reported concentrations in headwater catchments, which are not represented by S river due to the coarse resolution of the routing scheme, but which can roughly be estimated from the concentration associated with the summed flux of F fast out and F slow out .Here, we underestimate the DOC concentration in the headwaters by a factor of about 2, which is consistent with the underestimation of F RO contributions with high DOC concentrations.In the second case for comparison, a small headwater in the Peruvian mountains (RO2, Table 6, Fig. 4), our simulated headwater DOC concentrations are close to observed values.The third case study, RO3 (Johnson et al., 2006(Johnson et al., , 2008)), is for two neighbouring headwater catchments in southern Amazonia, and was also used for calibrating the throughfall DOC component.At this location, we have again good agreement for  the annual precipitation, but, for the grid cell corresponding to the sampling location, we overestimate the contribution of F RO to total runoff, due to the contribution of swamps.Thus, we also compare observations to the simulation results for a neighbouring grid cell without swamps (Simulation b, Table 6).Here, the simulated contribution of F RO is closer to the observations.The simulated DOC concentration in F RO is about the same for both cells and lies between the values observed for the two headwater catchments.The simu-lated headwater DOC concentration agrees well with the observed values for the second cell, for which the simulated F RO contribution is more in agreement with the observation.For the first grid cell, for which the contribution of F RO is overestimated, the headwater DOC concentration is overestimated accordingly.From the sensitivity analysis in Ta-   (3) southern Amazonia (TF3 in Fig. 4) - Johnson et al. (2006).(b) Seasonality in throughfall DOC concentrations for the site in southern Amazonia (TF3, Johnson et al., 2006).Note the sharp concentration increase during dry season from May to September.As the sampling period is outside of our simulation period, we compare the observed concentration with simulated average DOC concentrations over the entire run .concentration in F RO relative to the DOC concentration in the top 4.5 cm of the soil column (Eqs. 47,(53)(54)(55).The parameter red DOC,base has in addition an influence on DOC concentrations in drainage, as it controls the advection of DOC relative to water fluxes within the soil column .The decomposition rate of labile DOC (k DOC,lab ) exerts a moderate control on the simulated DOC concentrations in the headwaters (here the combined outflows from S fast and S slow ), while the decomposition rate of refractory DOC (k DOC,ref ) does not have a significant effect due to the rather short residence time in S fast and S slow .
With an arithmetic mean of about 21 g C m −2 yr −1 , the simulated total DOC inputs to the inland water network of the Amazon are significant (Table 8, Fig. 11), and about 5 times larger than the lateral DOC export from the Amazon basin at Obidos (4.6 g C m −2 yr −1 , Moreira-Turcq et al. 2003).More than half of the inputs are delivered by surface runoff (F RO ) (Table 8).More specifically, the total DOC input associated with F RO is more than 3 times higher than that originating from drainage (F DR ) although the simulated F RO contributes only to 44 % of the total runoff.This result can be explained by the much higher basin-scale average Geosci.Model Dev., 10,2017 www.geosci-model-dev.net/10/3821/2017/DOC concentration in F RO than in F DR (see Table 8).The simulated DOC inputs from F RO can reach very high values (Table 8) in the presence of swamps, where a constant fraction of river water is redirected to the soil column, leading to a very high runoff from the topsoil that can be several times higher than the precipitation flux.Note that a substantial part of this DOC export from swamps is fed by the DOC from the infiltrating river water.Thus the very high basinscale DOC input associated with F RO of 362 g C m −2 yr −1 (Table 8) is reduced to 71 g C m −2 yr −1 when swamp areas are excluded from the analysis.The simulated return flow of river water into the soil column in swamps (F up2swamp ) averages 2.1 g C m −2 yr −1 throughout the Amazon basin.The simulated infiltration of DOC on floodplains reaches a similar value of 2.4 g C m −2 yr −1 (F flood2soil ).Subtracting these fluxes from the inputs, we obtain an average net input from the soil-vegetation system into the inland water network of about 16.5 g C m −2 yr −1 .Although the maximum floodable area in the Amazon basin does not exceed 6.4 % according to our forcing files (Fig. 6), the simulated DOC input from submerged litter amounts to one-third of total DOC inputs to the inland waters.
As explained in the method section, a "poor soils" forcing was implemented to represent coarsely textured, acidic and nutrient-depleted soils in which DOC decomposition is reduced and vertical advection is more effective.For nine grid cells in the Amazon basin where the areal proportion of "poor soils" is higher than 75 % (Fig. 6), the simulated DOC export is dominated by such soils.Here, the DOC export flux is associated with F DR averages at 22.7 g C m −2 yr −1 , i.e. nearly 9 times the basin average value.The average DOC concentration in drainage (21.6 mg C L −1 ) is more than 6 times the basin average.For the two grid cells having 100 % "poor soils", the average DOC concentration reaches 24.7 mg C L −1 , which is, however, still substantially lower than the value of 36 mg C L −1 reported for groundwater seeping through the Podzols of the Rio Negro basin (McClain et al., 1997).

Transport and decomposition of DOC in the river network
To evaluate the simulated DOC concentrations and fluxes, we used data from the CAMREX (Carbon in the Amazon River Experiment) program (Richey et al., 2008)  al ., 2006), which was designed to capture the land-ocean matter transfer through the Amazon River network from the Andes down to Obidos with regular sampling campaigns; and the data from the study of Moreira-Turcq et al. (2003).
Comparing observed versus simulated DOC concentrations, we were able to reproduce the average concentrations at least in the main stem of Rio Solimoes-Amazon River and in the Rio Negro (Fig. 13).However, apart for the Rio Negro, we generally underestimate the seasonal variability of DOC concentrations.For Obidos, the most downstream sampling location for which we have data, the mean simulated DOC concentration deviates by only −2 % from the observed ones (Table 9).For the whole set of observed data, the deviation of simulated from observed average concentrations is −1 % (Table 9, "Final set-up").For Rio Jurua, concentrations are generally underestimated, while they are overestimated in Rio Japuru.These discrepancies could likely result from the coarseness of the river routing scheme, soil and wetland forcing files, thereby limiting our ability to reproduce the contributions of a specific flow path (F RO high in DOC versus F DR low in DOC) to stream flow and additional inputs from riparian wetlands.The simulated DOC concentrations are sensitive to the parameters controlling DOC exports with sur-face runoff from the topsoil, F fast+slow,H 2 O and red DOC,base , as well as the decomposition rate of labile DOC, k DOC,lab , but not to the decomposition rate of refractory DOC, k DOC,ref , which is very low and does not contribute much to in-stream respiration (Table 9).The simulated DOC fluxes (Figs. 14,15) mainly follow the dynamics in simulated discharge (Fig. 10), while the simulated DOC concentrations are less variable.In Fig. 14, we compare our simulations to data from the CAMREX project.We restrict our validation to the period 1982 to 1986, during which sampling frequency was highest (9 of the 13 cruises in that first half of the total period).In Fig. 15, we collate various data sources (CAMREX, the ORE HYBAM sampling network, Cochonneau et al., 2006, andthe data from Moreira-Turcq et al., 2003) to validate the simulated seasonality in DOC fluxes at the sampling location Manacapuru (Rio Solimoes) and Porto Velho (Rio Madeira).Overall, just as for discharge, the simulation reproduces the observed mean and seasonal variability in DOC fluxes quite well (Figs. 14,15).We find very good agreement for the Rio Solimoes at Sao Paulo de Olivenca, which drains the Andes in the western part of the Amazon basin, the Rio Negro as the major black-water tributary, and Rio Jurua Geosci.Model Dev., 10,2017 www.geosci-model-dev.net/10/3821/2017/ (Fig. 14, see Fig. 4 for locations).For Rio Solimoes at Manacapuru, the simulated peak in DOC fluxes occurs 1 month too early (Fig. 15), consistent with the simulation of discharge (Fig. 10).This slight time lag can be attributed to backwater effects from the two main tributaries, Rio Negro and Rio Madeira, which are not accounted for in our simulation (see Sect. 3.1).For Rio Japura, we overestimate the DOC fluxes although the simulated discharge agrees quite well with observations (Fig. 10), because we generally overestimate the DOC concentrations (Fig. 13, Table 9).For the Rio Madeira (Fig. 15), we have only observed DOC fluxes for years (2003)(2004)(2005)(2006) beyond our simulation period .Comparing the mean monthly fluxes for the respective periods, we observe that simulated fluxes are generally overestimated, particularly during high flow periods, a result which is consistent with the overestimation of river flow (Sect.3.1).
Combining the fluxes at Obidos with that of Rio Tapajos, which is entering the Amazon just below Obidos, the integrated yearly DOC export fluxes during our simulation period are in the range 19-27 Tg DOC yr −1 , with a mean value of 23.4 Tg C yr −1 .Our estimate is very close to that of 22.4 Tg DOC yr −1 (710 kg C s −1 ) calculated by Richey et al. (1990) and slightly lower than the 27 Tg C yr −1 (856 kg C s −1 ) estimated by Moreira-Turcq et al. (2003).This mean simulated annual DOC export flux corresponds to a flux of about 4 g C m −2 yr −1 if normalized to the whole catchment area, a value which is 80 % lower than the simulated net input flux of DOC from precipitation, vegetation and the soil system (see Sect. 3.3.1).The Amazon

Transport and evasion of CO 2
The simulated total inputs of CO 2 to the inland waters is significantly higher than that of DOC (Table 8).However, for inputs via F RO only, the CO 2 load is 1 order of magnitude lower than that of DOC.This is compensated by the inputs via F DR , where the simulated CO 2 exports are more than 5 times higher than that of DOC.Overall, F DR is responsible for about 90 % of the CO 2 exports from nonflooded soils to inland waters, in agreement with the relative CO 2 concentrations set for the two export pathways (see Sect. 2).Similarly, the CO 2 inputs from root respiration and heterotrophic respiration in the flooded soils gives an average flux of 39.5 g C m −2 yr −1 , nearly twice as large as the input from non-flooded soils.Abril et al. ( 2014) estimate the C inputs (CO 2 + DOC) to the water column per floodable area to be 1100 ± 455 g C m −2 yr −1 for the central Amazon basin.
Relating our simulated F soil2flood to %flood max , we obtain a similar average flux rate 1036 g C m −2 yr −1 within the central Amazon basin.The spatial pattern in our simulated CO 2 evasion (Fig. 16) naturally correlates strongly with %flood max (Fig. 6), because floodplains represent the largest contribution to the total inland water surface area.Thus, the highest average fluxes are found in the central Amazon floodplain and the Llanos de Moxos.As we use constant gas exchange velocities and do not account for in-river autotrophic production by algae, our simulated CO 2 evasion cannot reproduce short-term variation in evasion fluxes.However, our average CO 2 evasion rate per water surface area are in good agree- Note that due to the coarse resolution of our model, only data from the largest rivers (catchment area > 100 000 km 2 ) are taken into account.The simulated values refer to the average evasion rate during low (monthly avg.discharge < yearly avg.discharge) and high flow periods (monthly avg.discharge > yearly avg.discharge) across the whole simulation period .The whiskers represent the standard deviations from the inter-annual variations.ment with average observed evasion rates from several large rivers of the Amazon basin (Fig. 17).In addition, the simulated CO 2 evasion can be compared to the values reported by Richey et al. (2002).For the central Amazon basin (see Fig. 4), our simulation results give an average CO 2 evasion of 229 Tg C yr −1 , which is close to Richey et al.'s (2002) estimate of 210 ± 60 Tg C yr −1 .In addition, the simulation reproduces well the observed seasonal variations in CO 2 fluxes (Fig. 18).According to our results, floodplains contribute half (51 %) of the yearly-mean CO 2 evasion and rivers contribute another 39 %, while the remainder (10 %) evades from the fast reservoir.The latter can be regarded as small headwaters without inputs of CO 2 rich groundwater, which, in our model, do not exchange CO 2 with the atmosphere until they enter the river reservoir.
The fact that we simulate a total CO 2 evasion similar to the one reported by Richey et al. (2002) is somewhat surprising given that our mean water surface area is substantially lower (see Sect. 3.1).In other words, we simulate a higher CO 2 evasion rate per water surface area than estimated by Richey et al. (2002).These authors used relatively low gas exchange velocities k 600 of 1.2 to 2.3 m day −1 to calculate CO 2 evasion from rivers, while we applied a significantly higher value of 3.5 m day −1 , following more recent observations (Alin et al., 2011;Rasera et al., 2013).Note that in our physically based model approach, the total CO 2 evasion is not very sensitive to the gas exchange velocity, but rather to the simulated CO 2 sources.Reducing or increasing the gas exchange velocities k river,600 and k swamp,600 by 50 % leads to a change in simu-  4).The simulation result reports the mean monthly CO 2 evasion during the simulation period 1980-2000 as well as the standard deviation of monthly mean simulated values during the same period.The CO 2 evasion from headwaters is here represented by the CO 2 evasion from S fast .Simulation results are compared with the observation-based estimate by Richey et al. (2002), given here as the sum of the evasion from the Amazon main channel, the tributaries and the floodplains.R 2 = 0.85, RMSE = 23 %).lated total CO 2 evasion of only −4 and 1 %, respectively.On the contrary, in a data-driven approach to calculating CO 2 from observed river pCO 2 values, the calculated CO 2 evasion will change linearly with changes in the gas exchange velocity.Rasera et al. (2013) finds higher gas exchange rates than Richey et al. (2002) and thus suggests that the total CO 2 evasion must be considerably higher.As the results summarized in Fig. 16 suggest, our CO 2 evasion rates per water surface area are comparable to those of Rasera et al. (2013).Assuming that we underestimate the average flooded area, we conclude that we likely underestimated the CO 2 inputs from flooded soils and vegetation and the CO 2 evasion from the water surface to the atmosphere.In the future, improved floodplain forcings and simulations at higher spatial resolution might help to overcome these underestimations.
Although our estimates of CO 2 evasion from inland waters of the central Amazon basin are slightly higher than those of Richey et al. (2002), the same conclusion does not hold when assessing the CO 2 budget for the whole Amazon basin.The upscaling of Richey et al. (2002) led to a total CO 2 evasion estimate of 470 Tg C yr −1 while our simulation, which explicitly accounts for spatial heterogeneities across the basin, leads to a total CO 2 evasion of only about 379 ± 46 Tg C yr −1 .

Synthesis of simulation results
Figure 19 summarizes the simulated fluxes of dissolved C, i.e. the sum of DOC and CO 2 , through the river network of the Amazon basin.The total simulated export of carbon from the basin amounts to 413.9 ± 50.0 Tg C yr −1 , to which lateral exports to the coast contribute only 8.3 %, while the remainder is contributed by CO 2 evasion from the inland water surface.A total of 57 % of the total dissolved carbon inputs is contributed by flooded soils and litter.Surface runoff and drainage contribute 14 and 28 %, respectively.It is interesting that the flux carbon via throughfall onto the topsoil is as high as the lateral exports of dissolved C from the topsoil, although it is not necessarily its source.According to our simulations, about 8 % of the dissolved C mobilized into the water column are re-infiltrating into the soil column in swamps (F up2swamp ) or on floodplains (F flood2soil ).

Simplification of biogeochemical processes in the river network
The representation of biogeochemical transformation processes between different C species in the water column of the inland water network is rather simplistic.In the light of the limited empirical basis for calibration and validation on the one hand, as well as the rather uncertain boundary conditions provided by the forcing data and structural model uncertainties to represent terrestrial biogeochemical processes for tropical forests on the other hand, a more detailed representation of in-river processes is, for the time being, hardly achievable.Moreover, the validation supports the idea that ORCHILEAK represents the dominant aquatic C cycle processes on the scale of the major sub-basins in a rather satisfactory way.In the following, we briefly discuss the main limitations and future perspective towards improving the simulation of in-stream biogeochemical processes.One of the major future steps would be the implementation of particulate organic C fluxes in ORCHILEAK.Of the TOC fluxes at Obidos, the most downstream sampling location on the Amazon main stem, POC contributes less than one-quarter of the total flux (Moreira-Turcq et al., 2003;Ward et al., 2015) and was reported to further decrease to only about 10 % downstream to the river mouth (Ward et al., 2015).The decomposition of this POC, which is mainly derived from floodplain litter, has been reported to contribute substantially to the in-river CO 2 production in the lower part of the Amazon (Ward et al., 2013).Our simulation results also highlighted the substantial contribution of submerged leaf litter to the CO 2 evasion.However, in our simulation, POC is not transported downstream with the water flow, i. sition in the model, it is impossible to conclude whether the lack of representation of POC transport explains part of the discrepancy between observed and simulated DOC concentrations (Fig. 13), or whether a too-low DOC decomposition rate compensates for the bias.Mayorga et al. (2005) found that there must be a small, rapidly cycling pool of young organic matter from terrestrial vegetation close to the river that sustains high CO 2 concentrations of a young 14 C age, while the majority of the transported POC is substantially older.The actual effect of POC transport shifting CO 2 evasion downstream is thus likely rather limited.Nevertheless, a more complete representation of fluvial POC and DOC exports would be highly beneficial to constrain dynamic boundary conditions for an ocean biogeochemical model of the Amazon plume.The application of ORCHILEAK to rivers with substantial soil-erosion-driven POC exports will require the implementation of soil erosion and sediment transport modules (Naipal et al., 2015(Naipal et al., , 2016)).
The next major simplification in ORCHILEAK is the exclusion of autochthonous sources of TOC.In most parts of the Amazon River system, in-river autotrophic production is inhibited by the high water turbidity due to sediment fluxes from the Andes and, thus, most of the exported TOC is from allochthonous sources (Moreira-Turcq et al., 2003).For the application to more eutrophic, heavily dammed rivers, autotrophic production plays a non-negligible role in the aquatic organic C cycle (Maavara et al., 2017).However, the simulation of in-river autotrophic production requires the synchronous simulation of potentially limiting nutrients, nitrogen (N) and phosphorous (P), as well as of the light conditions as another limiting factor of algae growth (Billen et al., 1994).Taken the recent efforts in coupling the terrestrial C-N-P cycles in ESMs (e.g.Goll et al., 2012), the simulation of nutrient lateral transfers along the land-water continuum seems a realistic target in the coming years.The implementation of dams into a river routing scheme (Lehner et al., 2011;Maavara et al., 2017;Zarfl et al., 2014) could also support this development.
For the decomposition of DOC in transit, we considered here two pools of DOC with different, water-temperaturedependent decomposition rates.So far, our approach does not distinguish between heterotrophic respiration of DOC and photo-oxidation, which would make the simulation of the DOC fate more complex.For heterotrophic respiration, inclusion of the priming effects of more labile organic carbon on the decomposition of more refractory fractions could also be implemented (Guenet et al., 2014;Ward et al., 2016).Here, in particular, the labile pools produced by autotrophic processes could be of importance.Moreover, the production and decomposition of organic C, N and P would need to be coupled if the effect of the C : N : P ratios of organic matter on its degradability is to be accounted for.In addition, particularly where POC is concerned, a representation the heterotrophs in the ecosystem could be useful, including the "shredders" responsible for the physical breakdown of POC (Yoshimura et al., 2010) and "grazers" that feed on algae (Billen et al., 1994).Finally, photo-oxidation plays an important role in the breakdown of chromatic dissolved organic matter (CDOM), which is usually highly resistant to heterotrophic degradation.This process is likely important in black-water systems such as the Rio Negro (Amon and Benner, 1996).If this process was to be simulated, one would need to distinguish CDOM as a distinct species, and precise information on light-penetration depth and river-channel geometry would be required.For further developments in the modelling of DOC and POC decomposition in transit, a stronger empirical basis is needed, in particular for tropical river systems like that of the Amazon.

Conclusion and outlook
ORCHILEAK reproduces observed DOC and CO 2 fluxes in the Amazon basin, and their seasonal to inter-annual variability, at least on the scale of the major sub-basins.As highlighted in the introduction, we consider that the explicit simulation of the lateral export of soil and litter material to river headstreams and further down to the tropical ocean using an approach consistent with existing representations of terrestrial ecosystem carbon and water budgets, is a major step forward in physically based, integrated modelling approaches of the global C cycle.Currently, the empirical basis for calibration and validation of these lateral fluxes and their fate within the aquatic system is still limited for the Amazon basin.Nevertheless, the simulated terrestrial inputs are within the ranges reported in the literature, and the basinscale export fluxes agree well with observed fluxes.An improved representation of spatial heterogeneities and peculiar environments such as black-water systems will require even higher spatial resolution (0.25 • or less), improved regional soil, wetland and climate forcings as well as observations with higher spatial and temporal coverage for calibration and validation.
In this study, ORCHILEAK was applied to the Amazon using upgraded regional wetland and climate forcing files.In order to apply ORCHILEAK to other river systems, similar forcings will have to be constructed using the methodology described in Guimberteau et al. (2012) and in this study.In the future, ORCHILEAK is intended for global-scale applications.Before this objective can be reached, however, the new model branch will have to be tested on a regional scale in other river basins pertaining to different climate zones and ecosystem types.Adaption of the parameterization and, if required, implementation of additional key processes will need to be considered.The latter will, for instance, be important in high-latitude rivers under the influence of permafrost, an ecosystem subject to distinct physical and biogeochemical processes currently not included in ORCHILEAK.
Geosci.Model Dev., 10,2017 www.geosci-model-dev.net/10/3821/2017/ORCHILEAK will in future be augmented with additional transported species, in particular POC and nutrients.The simulated export fluxes to the coast will also provide useful time-dependent boundary conditions for ocean biogeochemistry models.Finally, ORCHILEAK will be useful to better assess the terrestrial C sink in ESM simulations by taking into account the permanent leakage of C from the plant-soil system.In the long run, our new model could also help better constrain terrestrial C cycle-climate feedbacks, future atmospheric CO 2 levels and temperature for different scenarios of anthropogenic CO 2 emissions.Gas exchange velocity (m day −1 ) for CO 2 in * = river,swamp or flood at 20

Figure 1 .
Figure 1.Representation of C exports from terrestrial ecosystems through the land-ocean aquatic continuum (LOAC).

Figure 3 .
Figure 3. Simulated flows of water and C along the vegetation-soil-water continuum.For reasons of simplicity, the fluxes (F ) and storages (S) are characterized by subscripts indicating path or environmental compartment only (see Table A1).Basin i − 1 is the basin upstream of basin i, basin i + 1 is the basin downstream of basin i.In this hypothetical example, swamps and floodplains are only present in basin i + 1.The depiction of water and soil-river C fluxes in basins i +1 and i −1 were omitted for reasons of readability.Straight arrows represent water and C fluxes between the canopy (S can ), soil (S soil ), fast (S fast ), slow (S slow ), river (S river ) and flood (S flood ) reservoirs.Circular arrows represent carbon transformations within the reservoirs.See text for further details.
Figure 3. Simulated flows of water and C along the vegetation-soil-water continuum.For reasons of simplicity, the fluxes (F ) and storages (S) are characterized by subscripts indicating path or environmental compartment only (see Table A1).Basin i − 1 is the basin upstream of basin i, basin i + 1 is the basin downstream of basin i.In this hypothetical example, swamps and floodplains are only present in basin i + 1.The depiction of water and soil-river C fluxes in basins i +1 and i −1 were omitted for reasons of readability.Straight arrows represent water and C fluxes between the canopy (S can ), soil (S soil ), fast (S fast ), slow (S slow ), river (S river ) and flood (S flood ) reservoirs.Circular arrows represent carbon transformations within the reservoirs.See text for further details.

Figure 4 .FloodcriFigure 5 .
Figure 4. Overview of the Amazon basin, with highlighted boundaries (thick grey) between the three major sub-basins (R. Solimoes, Madeira and Negro).The central Amazon basin (green box) and the sampling locations discussed in this study are also shown.River sampling locations and discharge gauges include Rio Japura at Acanaui (AC), Rio Xingu at Altamira (AL), Rio Araguaia (AR), Rio Jurua at Gaviao (G), Rio Tapajos at Itaituba (I), Rio Purus at Labrea (L), Rio Solimoes at Manacapuru (M), Amazon River at Obidos (O), Rio Madeira at Porto Velho (PV), Rio Negro at Serrinha (SE), Rio Solimoes at Sao Paulo de Olivenca (SP) and Tabatinga (T).The contributing areas are shown by the different colour codes on the map, except for location T as it is very similar to location SP.The remaining ungauged terrestrial area is represented in yellow.Sampling locations for throughfall DOC are indicated by "TF" and report data from Tobon et al. (2004) (TF1), Filoso et al. (1999) (TF2), Johnson et al. (2006) (TF3) and Williams et al. (1997) (TF4).Sampling location for DOC concentration in surface runoff and/or head waters are indicated by "RO" and report data from Waterloo et al. (2006) (RO1), Saunders et al. (2006) (RO2) and Johnson et al. (2006) (RO3).The red box and red line represent large floodplain areas outside the central Amazon basin for which observations are available.

Figure 6 .
Figure 6.Overview of forcing files (see Table2).Climatic forcings are comprised of, among others, variables like precipitation (F WD,H 2 O ) and air temperature (T air ).The climatic forcings used here are based on the NCC(Ngo-Duc et al., 2005) data set, except that F WD,H 2 O was replaced by a regional data set created byGuimberteau et al. (2012).The forcing of maximum floodable areas %flood max was adopted fromGuimberteau et al. (2012) after merging swamp areas (%swamp) into %flood max .Simulations of inundation in ORCHILEAK are based on 10th, 50th and 90th percentiles of water storage in the river reservoir S river (S river,H 2 O,10th , S river,H 2 O,50th ,S river,H 2 O,90th ), here given in millimetres which equals kilograms of H 2 O per squared metre assuming a density of water of 10 −3 kg m −3 , and the 95th percentile of water table level over the floodplains floodh (floodh 95th ), all derived from simulation results over the period 1980 to 2000.Surface areas of small (width < 100 m) and large (width ≥ 100 m) rivers (A river small , A river large ) are taken fromLauerwald et al. (2015).Of importance for representation of DOC cycling in watersheds of black-water rivers is the identification of "poor soils" (Podzols, Arenosols and soils in black-water swamps), which we derived from the Harmonized World Soil Database (HWSD, FAO/IIASA/ISRIC/ISS-CAS/JRC, 2009) and %swamp.

Figure 7 .
Figure 7. Predictability of water temperature (T water ) from simulated ground temperature (T ground ).(a) Linear regressions between T water and T ground recorded on the same day.The black line represents the linear fit through all data combined, while the coloured dashed lines represent the linear fits per sampling location.(b) Changes in RMSE (σ ) of the prediction equation per sampling location after applying different time lags to the predictor, T ground .

*
Calculated from the (flux-weighted) average concentration and throughfall DOC flux.(1) Tobón et al. (2004): based on samples taken from January 1995 and August 1997; simulation results for 1995-1997.(2) Filosos et al. (1999): based on samples March to December 1991, rainy season is from December to May; here, rainy season is from March to May + December 1991, dry season is from June to November 1991.(3) Johnson et al, 2006: based on samples taken during 2003-2004; here simulation results for 1980-2000 (no newer forcing file available).(

Figure 8 .
Figure 8. Simulated versus observed flooded area in the Amazon basin.(a) Central Amazon basin.Observed data from Richeyet al. (2002)  afterHess et al. (2003).Inundation corresponds to the sum of water surfaces of main channel, tributaries and floodplains recorded during the period October 1995 to September 1996.(b) Llanos de Moxos and Roraima floodplains over the period January 1980 to September 1987.Observed data fromHamilton et  al. (2011).RMSE is expressed as relative to the mean observed value per area.

Figure 10 .
Figure10.Simulated versus observed monthly discharge in the Amazon River and its major tributaries.The simulated discharge represents the average over the simulation period 1980-2000.For the stations at Rio Negro, Rio Purus and Rio Tabajos as well as for the Amazon at Obidos, observed discharges are derived from ORE HYBAM gauging data for the same period.For the other stations, long-term average monthly discharges from GRDC data set have been used, which cover a longer period: Amazon at Sao Paolo de Olivenca, Rio Madeira at Porto Velho, Rio Japura at Acanaui, Rio Jurua atGaviao (1972Gaviao ( -2010)), Rio Xingu at Atamira.See Table3.

Figure 11 .
Figure 11.Averages of simulated net primary production (NPP), dry deposition of soluble organic C onto the canopy and leaching of DOC from leaves (F add2can,DOC ), wet deposition of DOC (DOC in rain, F WD,DOC ) and throughfall DOC flux (F TF,DOC ), as well as total DOC and CO 2 exports into the inland water network (F RO + F DR + F soil2flood + F soil2river ) over the simulation period 1980-2000.

Figure 12 .
Figure 12.Simulated versus observed DOC in throughfall (F TF ).(a) Yearly-mean throughfall DOC flux versus literature values for the following three locations: (1) north-west Amazonia (TF1 in Fig. 4) -Tobon et al. (2004); (2) lower Rio Negro (TF2 in Fig. 4) -Filoso et al. (1999);(3) southern Amazonia (TF3 in Fig.4) -Johnson et al. (2006).(b) Seasonality in throughfall DOC concentrations for the site in southern Amazonia (TF3,Johnson et al., 2006).Note the sharp concentration increase during dry season from May to September.As the sampling period is outside of our simulation period, we compare the observed concentration with simulated average DOC concentrations over the entire run.

Figure 13 .
Figure 13.Observed versus simulated DOC concentrations (R 2 = 0.45, RMSE = 1.45 mg C L −1 ).For simulated values, each point represents the average during the year and month for which field data are available.The dashed line represents the 1 : 1 line.

Figure 14 .
Figure14.Simulated versus observed DOC fluxes in the Amazon main stem and its major tributaries.Observed data are taken from the CAMREX data set(Richey et al., 2008).

Figure 15 .
Figure 15.Seasonality in DOC fluxes in rivers at two sampling locations with more than 10 samples: Rio Solimoes at Manacapuru (RMSE = 29.4%, NSE = 0.17) and Rio Madeira at Porto Velho (RMSE = 89 NSE = −0.06).Simulated data report the mean of simulated values per month over the simulation period 1980-2000, including standard deviations of monthly means over the same period.Observed data are from Moreira-Turcq et al. (2003), Cochonneau et al. (2006) andRichey et al. (2008).For the observed data, we report median values (instead of the mean, which is more sensitive to single outliers).

Figure 16 .
Figure 16.Simulated average CO 2 evasion from rivers, floodplains and headwaters (summed up as F water2atm ) for the period 1980-2000.The evasion flux is reported relative to the total area of each grid cell.

Figure 18 .
Figure18.Seasonality in CO 2 evasion from inland waters (rivers plus floodplains, including swamps) within the central Amazon basin (see map in Fig.4).The simulation result reports the mean monthly CO 2 evasion during the simulation period 1980-2000 as well as the standard deviation of monthly mean simulated values during the same period.The CO 2 evasion from headwaters is here represented by the CO 2 evasion from S fast .Simulation results are compared with the observation-based estimate byRichey et al. (2002), given here as the sum of the evasion from the Amazon main channel, the tributaries and the floodplains.R 2 = 0.85, RMSE = 23 %).

Figure 19 .
Figure 19.Simulated fluxes of dissolved carbon (DOC + CO 2 ) through the inland water network of the Amazon basin.Numbers are average annual fluxes ± standard deviations over the simulation period 1980-2000.
e. it is assumed to decompose locally, and only the DOC and dissolved CO 2 produced from this decomposition are transferred laterally.The representation of POC transport would induce a downstream shift in the simulated DOC and CO 2 production from POC.The lack of this representation might have induced a bias in the simulated longitudinal pattern of DOC concentrations, pCO 2 and CO 2 evasion with an overestimation of upstream values compared to downstream values.With the limited availability of evaluation data and the rather simplified representation of POC and DOC decompowww.geosci-model-dev.net/10/3821/2017/Geosci.Model Dev., 10, 3821-3859, 2017

Figure A1 .
Figure A1.Land cover composition of the study area as representative for the years 2005-2006 derived from GLOBCOVER data(Arino et al., 2008).The black outline represents the Amazon watershed.

Table 1 .
List of forcing data needed to run ORCHILEAK.See text for explanations and Fig.6for an overview.

WD F RO F DR S soil S can S can F add2can F add2can F TF F flood2soil F soil2flood F dec terr S slow F fast out F slow out F WD2can F WD2ground F WD F WD2ground F can2ground F up2flood F flood out S flood S river S river S fast F river out F up F up2river F up2swamp BASIN i BASIN i+1 F up2river F up BASIN i-1 F river2atm F river2atm F fast2atm F flood2atm F soil2river F can2ground F TF
Geosci.Model Dev., 10, 3821-3859, 2017www.geosci-model-dev.net/10/3821/2017/F 30 min time step Fluxes computed at … D ily time step a 6 min time step ). • S river,H 2 O,grid x,t − S river,H 2 O,grid x,10th S river,H 2 O,grid x,90th • 0.2 If S river,H 2 O,grid x,t ≥ S river,H 2 O,grid x,90th : A river act,grid x,t = 1.2 • A river small,grid x ) If S river,H 2 O,grid x,t ≤ S river,H 2 O,grid x,10th : A river act,grid x,t = A river basic,grid x (24) If S river,H 2 O,grid x,10th <S river,H 2 O,grid x,t <S river,H 2 O,grid x,90th : Geosci.Model Dev., 10, 3821-3859, 2017 www.geosci-model-dev.net/10/3821/2017/A river act,grid x,t = 1 + + 1.1 • A river large,grid x (26) A river act,i,t = A river act,grid x,t • S river,H 2 O,i,t S river,H 2 O,grid x,t

Table 5
amount of throughfall, which acts as a dilution factor, and by the duration of preceding dry periods, which favours the accumulation of soluble organic C on the canopy ).The temporal variability in throughfall DOC concentrations is mainly controlled by the Geosci.Model Dev., 10, 3821-3859, 2017 www.geosci-model-dev.net/10/3821/2017/ • 1 − % lignin grid x,v,l,t + F dec terr,litter met,grid x,v,l,t + F dec flood,litter met,grid x,v,l,t + F dec terr,SOC active,grid x,v,l,t + F dec flood,SOC active,grid x,v,l,t • CUE +F dec flood,DOC active,grid x,v,l,t + F TF,DOC lab ,grid x,v,t • 1 − % flood grid x,t − F RO,DOC active,grid x,v,t − F DR,DOC active,grid x,v,t • % lignin grid x,v,l,t + F dec terr,SOC slow,grid x,v,l,t + F dec flood,SOC slow,grid x,v,l,t + F dec terr,SOC passive,grid x,v,l,t+F dec flood,SOC passive,grid x,v,l,t • CUE − 11 l=1 F dec terr,DOC slow,grid x,v,l,t +F dec flood,DOC slow,grid x,v,l,t + F TF,DOC ref ,grid x,v,t • 1 − % flood grid x,t − F RO,DOC slow,grid x,v,t − F DR,DOC slow,grid x,v,t − F soil2flood,DOC slow,grid x,v,t(44)S soil,DOC passive,grid x,v,t = 11 l=1 F dec terr,SOC passive,grid x,v,l,t +F dec flood,SOC passive,grid x,v,l,t • CUE − 11 l=1 F dec terr,DOC passive,grid x,v,l,t +F dec flood,DOC passive,grid x,v,l,t − F RO,DOC passive,grid x,v,t − F DR,DOC passive,grid x,v,t − F soil2flood,DOC passive,grid x,v,t production in flooded areas, we assume that the DOC produced from the decomposition of litter and SOC within these same 5 topsoil layers adds directly to the DOC storage in the overlying surface water body S flood (seeFig.3,.Accordingly, the inputs of DOC to the non-flooded soils via F dec terr are estimated using the non-flooded proportion of the grid cell (1 − % flood i,t ) (Eqs.37, 39, 41).
To simulate the DOC Geosci.Model Dev., 10, 3821-3859, 2017 www.geosci-model-dev.net/10/3821/2017/F soil2flood,DOC active,grid x,v,t = 5 l=1 F dec flood,litter str,grid x,v,l,t • 1 − % lignin grid x,v,l,t + F dec flood,litter met,grid x,v,l,t + F dec flood,SOC active,grid x,v,l,t • CUE (48) F soil2flood,DOC slow,grid x,v,t = 5 l=1 F dec flood,litter str,grid x,v,l,t • % lignin grid x,v,l,t + F dec flood,SOC slow,grid x,v,l,t • CUE (49) F soil2flood,DOC passive,grid x,v,t = 5 l=1 F dec flood,SOC passive,grid x,v,l,t • CUE (50) • ) of the river network).Next, based on model calibration, we set a threshold value for the sum of S fast,H 2 O and S slow,H 2 O (S fast+slow,H 2 O,ref ) at which a 100 % connection between topsoils and headwaters is achieved.When S fast+slow,H 2 O,ref does not reach the threshold, a lower proportion of topsoil is in connection with the headwaters.Consistent with our approach in Sect.2.1.3,weassumedhere that the extent of saturated soils around headwaters (i.e. the connected topsoils) increases linearly with the square root of the sum of S fast,H 2 O and S slow,H 2 O .Finally, the maximum amount of DOC that can be exported through surface runoff and drainage is limited by the storage of DOC in the top and bottom soil layers(Eqs.46,47).Export of dissolved CO 2 through the soil-water network interface www.geosci-model-dev.net/10/3821/2017/Geosci.Model Dev., 10, 3821-3859, 2017 2.2.4 F soil2atm,CO 2 ,grid x,t = F soil resp,CO 2 ,grid x,t − F RO,CO 2 ,grid x,t ing to their contribution to the total fraction of inundated soil (% flood total ).S fast,C spec,i,t+1 = S fast,C spec,i,t + F RO,C spec,i,t − F fast out,C spec,i,t(66) S slow,C spec,i,t+1 = S slow,C spec,i,t + F DR,C spec,i,t − F slow out,C spec,i,t (67) S river,C spec,i,t+1 = S river,C spec,i,t + F up2river,C spec,i,t + F flood out,C spec,i,t − F river out,C spec,i,t Geosci.Model Dev.,10, 2017www.geosci-model-dev.net/10/3821/2017/S flood,C spec,i,t + F up2flood,C spec,i,t − F flood2soil,C spec,i,t + F TF,C spec,i,t • % flood i,t − F flood out,C spec,i,t + 13 v=1 F soil2flood,C spec,grid x,v,t • dt day • % flood i,t • A total,i % flood total,grid x,t • A total,grid x ) as predictor of water temperature.T water,grid x,t = 6.13 • C + 0.80 • T ground,grid x,t and is thus a combination of open-water floodplain and swamps, the average k flood is calculated according to the vegetated and open proportions (Eq.75).In rivers and floodplains, the CO 2 evasion is calculated based on the pCO 2 , the gas exchange velocity and the surface water area available for gas exchange, which changes at the daily time step(Eqs.76,77).The maximum possible CO 2 evasion per time step is constrained by the amount of dissolved CO 2 in excess of the hypothetical equilibrium with the atmospheric pCO 2 .For S fast , for which a surface area is not known, full equilibration with the atmosphere is assumed (Eq.78).For S slow , which we consider as groundwater storage even though a groundwater table itself is not simulated, no gas exchange is assumed.

Table 2 .
Data sets used for model evaluation.

Table 3 .
Performance of discharge simulations in the trunk version of ORCHIDEE (parametrization by

Table 4 .
(a)Yearly-mean simulated NPP in the Amazon basin(period 1980-2000)reported for the five dominant plant functional types (PFTs).(b) Simulated and observed mean soil organic carbon (SOC) stocks in the Amazon basin.Values are reported for the top 30 cm, the top 100 cm and the whole 200 cm profile used in the simulation.

Table 5 .
Simulated versus observed DOC concentrations (conc.),water and DOC fluxes in precipitation (rain) and throughfall (TF).Ditto marks ( ) denote that the value is the same as the one above.

Table 6 .
Observed and simulated DOC concentrations in overland flow (= F RO ) and headwater streams (= F fast out + F slow out ).The surface runoff is reported as percentage of total runoff, in the literature, and, for comparison, also for simulated values.

Table 7 .
Sensitivity of simulated average DOC concentrations (mg C L −1 ) in surface runoff, drainage and headwater streams to changes in key parameters in calibration.Final set-up S fast+slow,H 2 O,ref red DOC,base k doc,lab k doc,ref

Table 8 .
Statistical distributions of simulated export fluxes and concentrations within the Amazon basin.

Table 9 .
Sensitivity analysis on the performance of simulating DOC concentrations.Performance measures, root mean squared errors (RM-SEs) and mean signed deviation (MSD), both relative to the mean observed concentration, are reported per sampling location, and for the whole set of observed DOC concentrations.

Table A1 .
Continued.Other acronyms (subscripts i/x, v, l, and t correspond to 3rd to 6th subscript described above)Aflood Water surface area of S flood (m 2 ) A river River surface area (m 2 ) A river small Area of rivers with a width ≤ 100 m (m 2 ) A river large Area of rivers with a width >,100 m (m 2 ) A river basic River surface area (m 2 ) at low water stage (m 2 ) A river act Actual A river (m 2 ) that can be larger than A river basic A total Area of the grid cell or basin (dependent on subscript) (m 2 ) b Parameter describing shape of floodplain (see text) CUE Carbon use efficiency (fraction of organic C that is transformed to another form of organic C) dt Time step used for soil C and vertical fluxes (= 30 min) floodcri Constant (m) (default 2 m) used in TRUNK version in simulation of actual flood extent, in ORCHILEAK replace by floodh 95th floodh Water level over floodplain (m) floodh 95th 95th percentile of floodh i,t over simulation period (m) f swamp Fraction of F up that is diverted to the bottom soil layer K CO 2 Solubility constant of CO 2 (mol L −1 atm −1 ) k DOC lab Decomposition rate of labile DOC at T water = 28 • C (day −1 ) k DOC ref Decomposition rate of refractory DOC at T water = 28 • C (day −1 ) k flood Gas exchange velocity for CO 2 (m day −1 ) from floodplains, mix of the k river and k swamp k river Gas exchange velocity for CO 2 (m day −1 ) from open water k swamp Gas exchange velocity for CO 2 (m day −1 ) from flooded forests k * ,600 • C k SOC pool Decomposition rate of the active, slow or passive SOC pool k litter pool Decomposition rate of the metabolic or structural litter pool k DOC pool Decomposition rate of the active, slow or passive DOC pool leaf biomass i,v,t Biomass allocated to leaves (g C m −2 ) pCO 2 atm Atmospheric partial pressure of CO 2 (atm) pCO 2 fast Aquatic partial pressure of CO 2 in S fast (atm) pCO 2 river Aquatic partial pressure of CO 2 in river (atm) pCO 2 flood Aquatic partial pressure of CO 2 in floodplain (atm) red RO Combined reduction factor for exports with runoff red DOC Reduction factor for vertical, advective DOC fluxes and lateral DOC export from soil column (set to 0.2) red connect Reduction factor for exports with runoff depending on extends of saturated soils around headwaters Sc Schmidt number S fast+slow,H 2 O,ref Reference storage of water (mm) in S fast and S slow , at which red connect = 1.0 (set to 160 mm) T ground Mean daily air temperature near the surface ( • C) river + %flood %lignin Lignin content (mass fraction) in the structural litter %poorsoils Areal proportion of Podzols + Arenosols + black-water swamps %swamp Area proportion of swamps in grid box www.geosci-model-dev.net/10/3821/2017/Geosci.Model Dev., 10, 3821-3859, 2017