Constraining a complex biogeochemical model for multi-site greenhouse gas emission simulations by model-data fusion

This study presents the results of a combined measurement and modelling strategy to analyse N2O and CO2 emissions from adjacent arable land, forest and grassland sites in Germany. The measured emissions reveal seasonal patterns and management effects, including fertilizer application, tillage, harvest and grazing. The measured annual N2O fluxes are 4.5, 15 0.4 and 0.1 kg N ha a, and the CO2 fluxes are 20.0, 12.2 and 3.0 t C ha a for the arable land, grassland and forest sites, respectively. An innovative model-data fusion concept based on a multi-criteria evaluation (soil moisture at different depths, yield, CO2 and N2O emissions) is used to rigorously test the LandscapeDNDC biogeochemical model. The model is run in a Latin Hypercube based uncertainty analysis framework to constrain model parameter uncertainty and derive behavioural model runs. The results indicate that the model is generally capable of predicting trace gas emissions, as evaluated with RMSE as the 20 objective function. The model shows a reasonable performance in simulating the ecosystem C and N balances. The modeldata fusion concept helps to detect remaining model errors, such as missing (e.g., freeze-thaw cycling) or incomplete model processes (e.g., respiration rates after harvest). This concept further elucidates the identification of missing model input sources (e.g., the uptake of N through shallow groundwater on grassland during the vegetation period) and uncertainty in the measured validation data (e.g., forest N2O emissions in winter months). Guidance is provided to improve the model structure and field 25 measurements to further advance landscape-scale model predictions.


Introduction
Carbon dioxide (CO2) and nitrous oxide (N2O) are two prominent greenhouse gases (GHG) contributing to global warming, the latter having a 300 times higher global warming potential (GWP) than CO2 considering a 100 year time horizon (Myhre et al., 2013).Terrestrial ecosystems play an important role in the global atmospheric budgets of both GHGs (Cole et al., 1997).
Global CO2 emissions from soils are five times higher than anthropogenic (mainly fossil fuel) CO2 emissions (Raich and Schlesinger, 1992; updated with recent fossil fuel data by Boden, et al., 2010), while agricultural land use release over 60% of the global anthropogenic N2O emissions in 2005 (IPCC, 2007).Besides the radiative forcing of both GHGs, N2O is currently the main driver of stratospheric ozone depletion (Ravishankara et al., 2009)  could result in skin cancer and other health problems (Graedel and Crutzen, 1989).While CO2 is exchanged with the soil (heterotrophic respiration) and the vegetation (photosynthesis and autotrophic respiration), N2O fluxes refer mainly to nitrification and denitrification processes occurring only in the soil (Butterbach-Bahl et al., 2013).
Emissions of both GHGs are highly variable in space, time and depend on a multitude of different interacting environmental factors, e.g., land use/management, nitrogen/carbon inputs, meteorological conditions and physical and chemical soil properties (Davidson, 1992;Smith et al., 2003).They are largely regulated by plant physiological (Rochette et al., 1999) and microbial processes (Burton et al., 2008).Field measurements of GHG emissions and environmental drivers have allowed a basic understanding of observed emissions patterns.Nevertheless, the large number and complexity of the processes involved in the production and consumption of CO2 and N2O still challenge the reliable quantification of GHG emission (Butterbach-Bahl et al., 2013).Therefore, various biogeochemical models have been developed in recent years for temporal as well as spatial up-scaling of GHG emissions, hypothesis testing of our process understanding, and, moreover, for scenario analyses and evaluation of efficient mitigation options (Kim et al., 2015;Molina-Herrera et al., 2016).These include, e.g.BASFOR (Van Oijen et al., 2005), CERES-EGC (Gabrielle et al., 2006), COUP (Jansson, 2012), DAYCENT (Parton et al., 1998) or DNDC and its descendant LandscapeDNDC (Haas et al., 2013).However, models remain as simplifications of the real world, driven by uncertainties due to intrinsic model structure, parameterization and the current model state (Vrugt, 2016).During model application, further uncertainties are added by applied forcing data (Kavetski et al., 2006).However, currently there is no method available dealing properly with all those sources of uncertainty at the same time (Vrugt, 2016).One promising way to reduce the magnitude of model output uncertainties is to use model-data fusion techniques, i.e. matching model prediction with multiple observations by varying model parameters or states using statistical uncertainty estimation (Keenan et al., 2011).
There are several statistical uncertainty estimation methods available, e.g., formal Bayesian approaches like DREAM (Vrugt, 2016) or informal Bayesian approaches like GLUE (Beven and Binley, 1992).However, they are mostly used to fit models to single types of observation (Giltrap et al., 2010).Innovative multiple observation data evaluation with model-data fusion is getting common in ecosystem carbon modelling (Wang et al., 2009) and is getting more and more important in the nitrogen modelling community, too (Wang and Chen, 2012).Gained knowledge can and should be used to guide further model improvements (Vrugt, 2016).
This work focuses on establishing model-data fusion in the biogeochemical community -i.e.showing their capability to gain knowledge with process-based model application.We present weekly measurements of CO2 and N2O emissions from a developed landscape including different land uses, i.e., arable, grassland and forest ecosystems, covering a two years period of observations.In addition to field measurements, we set up the biogeochemical LandscapeDNDC model for each of the three land uses.During model-data fusion with GLUE, we rigorously accept only model runs, which return concurrent, acceptable outputs for N2O, CO2, soil moisture in different depths and yields.Posterior model runs are not only evaluated fulfilling appropriate objective functions, but also regarding realistic simulations of GHG emissions for separate seasons, annual sums as well as before and after land management.

Study area
The study area is located in the catchment of a low-mountainous creek (Vollnkirchener Bach) in the municipality Hüttenberg, Hesse, Germany (50°29′56″ N, 8°33′2″ E).One kilometer north of the village Vollnkirchen, next to the creek, we established eight transects (oriented mostly vertical to slope) along a valley cross section covering different types of land uses (Fig. 1) for GHG emission measurements.See Table 1 for detailed information on soils characteristics.Three transects (A1-A3) are located on arable land westwards the creek and were cultivated with the same field management and crop rotations (Table 1).
Three transects are located in a light beech (Fagus sylvatica) forest (W1-W3) with young and old trees on a steep hillside (slope: 10%) eastwards the creek.A shallow 0.05 m litter layer characterizes the forest soils.Furthermore, there are two transects (G1, G2) located on grassland in 4 m distance on each side to the Vollnkirchener Bach.One of the two transects is managed and grazed (G1) and is mainly covered with brown knapweed (Centaurea jacea), meadow foxtail (Alopecurus pratensis), red clover (Trifolium pretense) and ribwort plantain (Plantago lanceolate).The second grassland transect (G2) represents a wetland and is mainly covered by meadowsweet (Filipendula ulmaria), common nettle (Urtica dioica), hoary ragwort (Senecio erucifolius) and field bindweed (Convolvulus arvensis).The groundwater table is close to the surface on both grassland sites.Mean annual wet deposition of nitrate and ammonium were measured from 2013-2015 with 1.66 kg N ha -1 and 3.45 kg N ha -1 , respectively.In the catchment mean annual precipitation is 588 mm and mean annual temperature is 10.5 °C for the hydrological year 01.11.201301.11. -31.10.201401.11. (Seifert et al., 2016))

Trace gas measurement
The weekly trace gas measurements started in November 2013.GHG exchange fluxes were manually measured with nonsteady state opaque chambers each covering a basal area of 0.12 m 2 .Chambers were placed on frames (both polypropylene) which were inserted approx.8 cm into the soil in order to facilitate gas tight sampling as well as to avoid soil structure damage and lateral trace gas leakage.Each chamber is equipped with an extraction septum, a counterbalance valve (in-box pressure balance) and a small fan/ventilator for homogenous mixing of the headspace air.During a 40 minutes closure period five air samples are taken from the chamber headspace at regular time intervals t0-t4 of ten minutes (0, 10, 20, 30 and 40 min.).
Samples are analyzed by gas chromatography (GC 8610C, SRI Instruments, Torrance, US) with an ECD for N2O and a methanizer and FID for CO2.Sampling was performed on a weekly basis with five replicated chambers per transect sampled by the gas sample pooling technique (Arias-Navarro et al., 2013).According to this approach, at any time interval (t0-t4) 10 ml headspace sample are collected subsequently from any of the five replicated chambers and are pooled into one gas tight glass vial (SRI Instruments).Trace gas fluxes are calculated from the rate of change in the headspace gas concentration over time by linear regression and were corrected by chamber temperature, atmospheric pressure and chamber volume according to Barton et al. (2008).All measurements with regression quality of r 2 < 0.7 for CO2 (using at least four individual samples) were rejected.
Soil emissions of CO2 and N2O can be subject to significant diurnal patterns with peak values observed in the early afternoon (Savage et al., 2014), making an up-scale process of hourly measured emissions (usually obtained at midday) to daily values difficult.We performed multiple linear regression (ordinary least square regression including air temperature, relative humidity and water filled pores space), to account for difference between, e.g., daytime (Wohlfahrt et al., 2005a) and nighttime respiration (Wohlfahrt et al., 2005b).In our dataset, only CO2 emissions showed significant correlations with mentioned environmental drivers on arable land (r 2 = 0.53), grassland (r 2 = 0.59) and forest (r 2 = 0.51).Following Subke et al. (2003), we derived an hourly integration formula in order to obtain daily representative mean values of CO2 emissions from our field measurements conducted mostly between 9am-5pm.N2O emissions are up-scaled to daily mean values with the common approach, i.e., by multiplying hourly emissions with 24.Annual CO2 and N2O emissions are calculated by linear interpolation between the measurements.All the underlying data of chapter 2.1 and 2.2 is available on request from a database (http://fb09pasig.umwelt.uni-giessen.de:8081/).

Model set up
We tested the biogeochemical model framework LandscapeDNDC (Haas et al., 2013) with the observed data of our study area.
Individual models were set up for arable, grassland and forest ecosystems.The modules describe different processes in ecosystem compartments, i.e. mathematical descriptions of microclimate, water cycle, plant physiology and soil biogeochemical processes.We applied the biogeochemical module MeTr x (Kraus et al., 2015) and the watercycle module watercycleDNDC (Kiese et al., 2011) for all land uses, but selected individual physiology modules, i.e. arableDNDC for arable, PSIM/TREEDYN for forest and grasslandDNDC for grassland simulations.As LandscapeDNDC is not yet capable for wetland simulation, we included only G1 within the modelling activities.The three different model setups were driven by the same meteorological data and initialized by land use specific soil and vegetation characteristics (Table 1).The thin and stony soil at the forest site is a Cambisol and modeled until bedrock (0.55 m) with a litter height of 0.05 m.Bulk density is increasing for every land use in depth, while soil organic carbon and nitrogen are decreasing with depth.We run simulations for any land use in daily time resolution for 6 years, starting on 1 st January 2010 with a model spin-up time of two years.

Model-data fusion
For the multi-objective Bayesian model calibration we used a two-tiered Generalized Likelihood Uncertainty Estimation (GLUE) approach (Beven and Binley, 1992).The model was started in both tier for 100,000 times by changing the parameter sets by Latin Hypercube sampling using the Python software SPOTPY (Houska et al., 2015).Parameters for the physiology and the water-cycle modules where treated land use-specifically, while the parameters of the biogeochemical module were calibrated using data of all land uses (Table A1).We assume no prior knowledge than the given parameter ranges, i.e. consider a uniform (non-informative) prior probability distribution for all parameters.We statistically judged the performance of every parameter set to reproduce measurements with a root mean squared error (RMSE).Similar to Bloom and Williams (2015) GLUE based multi-objective model calibration can result in very low acceptance rates down to 0.01%.We therefor considered a two-tiered GLUE approach, in order to increase identifiability and accuracy of accepted model runs: Tier I: In a first step we constrained the parameter space of the hydrology and plant physiology modules of LandscapeDNDC (Table A1).We accepted only model runs, which are within the best 5% of all simulated RMSEs Tier II: In order to achieve realistic GHG simulations from the biogeochemical module MeTr x of LandscapeDNDC, we took the posterior parameter boundaries of tier I and ran GLUE again.This time, we considered the best 5% of all RMSEs in terms of respective N2O and CO2 emissions for each land use (A1-3, G1 and W1-3).Again, only the 5% best parameter sets were accepted per land use.These results are shown in the following chapters.
(2) Management comparison of measured and modeled emissions, i.e. investigation of model performance within two weeks before and two weeks after management events to check model performance in generating hot moments, e.g. after fertilizer application.
(3) Model performance in simulating magnitude and uncertainty of C and N fluxes not measured in situ, like N2 or autotrophic and heterotrophic components of CO2 emissions.

Measured N2O fluxes
To determine the representativeness of each transect for a given land use, respective differences of measured N2O emissions were compared (Table 2).Temporal dynamics of N2O emissions are presented (Fig. 3), separated into different seasons (Fig. 4) and before/after management-events occur (Fig. 5).The mean annual N cycle is given in Table 4.
Arable N2O fluxes: Emissions on arable land vary between 0 and 0.3 kg N2O-N ha -1 day -1 .There were no statistical differences over time between the three arable transects.Highest emissions occur after management events.Especially mineral fertilizer application stimulates N2O emissions, causing hot moments from e.g., March to May 2014.Input of N through manure application has a minor influence on the magnitude of N2O emissions.Mean annual measured N2O emissions from arable land are comparably high with 4.5 kg N2O-N ha -1 a -1 (Jungkunst et al., 2006), equaling a GWP of 1,338 kg CO2-C equiv.ha -1 a -1 .
With yearly fertilizer application of 248.2 kg N a -1 a mean annual emission factor (EF) of 1.4% (varying between 1.2% for A2 and 1.8% for A3) can be calculated, where 1 kg N ha -1 a -1 is attributed to background emissions of unfertilized soil (IPCC,  1997).This EF is inside the IPCC assumed range 1.25 ±1% and close to the average EF (1.56%) of several (n=56) agricultural sites in Germany (Jungkunst et al., 2006).A robust finding throughout the literature is that reduced nitrogen input would lead to lower and therefore more climate-friendly agriculture (Bouwman et al., 2002).
Grassland N2O fluxes: N2O emissions vary significantly between the grazed site G1 and the wetland site G2, which can be attribute to differences in management, hydrological, soil and vegetation characteristics.Most likely, nitrate supply through groundwater and uptake by the rooting system of the plants is important (Liebermann et al., 2017).Even though the groundwater table (0.2 -0.4 m below ground) is rather shallow in winter/spring, uptake rates in summer/autumn (groundwater table 0.3 -1.0 m below ground) are supposedly larger due to the vegetation period.Here, capillary rise might play a relevant role (Orlowski et al., 2016).While G1, with a mix of Centaurea jacea, Alopercurus pratensis, Plantago lancelotata and Trifolium prantense, is grazed by sheep twice a year and cut once a year, the non-managed transect G2 is dominated by other species like Urtica dioca, Filipendula ulmaria and Senecio erucifolius.Typically, a deeper rooting system is found compared to G1 and accordingly, additional nitrate uptake from the groundwater is more prevalent.Mean measured emissions are higher on non-managed G2 than on grazed G1 throughout the year, especially during summer and autumn (Fig. 3).Emissions from grazed grassland vary between -0.0019 and 0.014 kg N ha -1 day -1 .High emissions were measured after grazing, e.g. in March 2014 when sheep dung was stimulating N2O emissions.Negative values depict N2O uptake, frequently found under prevailing wet conditions in spring, also reported by Glatzel and Stahr (2001).The grassland annual N2O emissions are much lower than observed for the arable system (A1-3).However, with 0.29 (G1) and 0.52 kg N2O-N ha -1 a -1 (G2) they are in line with a study site 12 km northeast of our site, where annual emissions amount to 0.18 to 0.79 kg N2O-N ha -1 a -1 on an unfertilized grassland with shallow groundwater table (Kammann et al., 1998).They also report a similar seasonal pattern we found, with emissions close to zero in the dry and colder autumn months.Measured annual emissions are below the assumed background level of N2O-N emissions of 1 kg N2O-N ha -1 a -1 from agricultural soils (IPCC, 1997).Our annual N2O emissions equal a GWP of 87 (G1) and 156 kg CO2-C equiv.ha -1 a -1 (G2).The EF through grazing is 5.4%, which is in line with usually found emissions factors from extensive grazed grasslands, ranging globally from 0.2 -9.9% (Oenema et al., 1997).
Forest N2O fluxes: Significant differences were found for the forest transects W2 and W3, which can be explained by natural variations along the steep hillslope: On the hillside (W2) the soil is potentially washed out through lateral transport leading to decreased nutrition availability, compared to the dryer top (W1, +300% N2O emissions) and the wetter hillfoot (W3, +430% N2O emissions).N2O emissions from the forest transects are most of the time low between -0.003 and 0.004 kg N ha -1 day -1 .
Higher emissions were measured only for several weeks in January 2014, with highest values observed at W1.We contribute this to freeze thaw effects, typically found when year-around measurements are considered (Papen and Butterbach-Bahl, 1999).
Negative fluxes were measured e.g. in March and May 2014.The underlying process of N2O uptake has been reported before (e.g.Flechard et al., 2005;Neftel et al., 2007) and is assumed to be a microbial process, in which denitrifier use N2O as an electron acceptor for respiration under wet/anaerobic conditions (Bremner, 1997).Negative emissions are positively correlated with WFPS (Fig. A3) being in line with Bremner (1997).Our annual measured emissions in forest are with 0.08 kg N2O-N ha -1 a -1 (GWP of 25 kg CO2-C equiv.ha -1 a -1 CO2 emissions) much lower than at adjacent grassland and arable sites.Even more they are a magnitude lower than the N2O emissions (5.1 kg N2O-N ha -1 a -1 ) measured from a beech forest in the Högelwald, Germany (Papen and Butterbach-Bahl, 1999).A likely reason is the substantially higher annual deposition rate of 25 kg N ha -1 a -1 , a five times higher N input as compared to our system.However, our measurements of N deposition only includes wet deposition.Additional dry depositions is often assumed to add another 30-60% to total atmospheric N deposition (Flechard et al., 2011).

Measured CO2 fluxes
Emissions measured using our closed chamber on arable and grassland include those from soil and vegetation as the entire plants are covered by the chamber.Therefore, we interpret these emissions as total ecosystem respiration (TER).In contrast, chambers in the forest were placed on the forest floor without any vegetation inside, thus, these measurements include soil (heterotroph) and root (autotroph) respiration, only.For the sake of a better read-flow, we decide to define emissions from the different land use as 'CO2 emissions', even though we considered the different flux components on different land use.To determine the representativeness of each transect for a given land use, respective differences of measured CO2 emissions were compared to each other (Table 3).Measured CO2 emissions are given in time (Fig. 6), separated into different seasons (Fig. 7) and before/after management-events occur (Fig. 8).The mean annual N cycle is given in Table 5.
Arable CO2 fluxes: Weekly measured values from our arable transects range between 0 and 199.6 kg C-CO2 ha -1 day -1 and are not significantly different.Emissions occur mainly during the growing season, starting in March and ending in November.For a comparable study site in southern Finland, reported daily TER values under barley during May and September were between 23.6 to 235.6 kg C-CO2 ha -1 day -1 (Lohila et al., 2003), which is in the same range as our observations.The annual sum of our TER emissions is 19.96 t C-CO2 ha -1 a -1 .This is slightly lower than yearly TER measured on a winter wheat study site in Belgium with 23.18 t C-CO2 ha -1 a -1 (Suleau et al., 2011) multiple linear regression, in our case based on our hourly measurements of air temperature and soil moisture.Such methods do not fully account for management effects like harvest (Subke et al., 2003).
Grassland CO2 fluxes: Emissions from grassland vary from 5.0 to 68.3 (G1) and from 0 to 92 kg C-CO2 ha -1 day -1 (G2), with no significant difference between the two transects.Emissions are close to zero in the winter months (December to February) and highest during the growing season.A distinct negative correlation of measured CO2 fluxes was found during wet conditions from end of June to July in 2014.In this time emissions drop to values of 41.0 kg C-CO2 ha -1 day -1 .The TER of a grassland CO2 is mainly driven by the growing season (Soussana et al., 2007).Emissions typically end with pasture growth during temperatures under 5°C (Parsons, 1988).Total yearly emissions are 11.79 t C-CO2 ha -1 a -1 , which agrees well with mean yearly emissions reported for 19 different grassland sites across Europe with mean annual emissions of 12.83 t C-CO2 ha -1 a -1 (Gilmanov et al., 2007).However, due to the many different grassland sites considered in their study, Gilmanov et al. report a much wider range of observed annual TER values from 4.9 to 16.4 t C-CO2 ha -1 a -1 .They also found that management as a main influence on TER, where intensively managed grassland produce higher emissions than extensively managed grasslands.
With regard to grazing, we found only a minor direct impact on measured flux rates (Fig. 7).
Forest CO2 fluxes: Mean measured soil respiration span from 2.1 to 19.9 kg C-CO2 ha -1 day -1 .While we found higher emissions in the summer months, seasonal differences have a lower magnitude as TER on arable and grassland.This was expected, as we do not measure above ground biomass respiration on our forest study site.Overall, rewetting has the strongest influence on changes of soil respiration in our forest study sites.Highest emissions occur in July 2014 after several rewetting events of the uppermost soil layer (Fig. A1).Xiang et al. (2008) reported that multiple rewetting led up to eight-times higher respiration rates.Total yearly soil emissions are with 2.98 t C-CO2 ha -1 a -1 at the lower end of other European forest ecosystems, e.g.6.6 ± 2.9 t C-CO2 ha -1 a -1 as reported by Janssens et al., (2001).The uphill transect W1 has the highest emission rates throughout the year and shows significant differences to W2 and W3.This transect is less shaded through trees, resulting in a 1.3°C higher annual mean soil temperature compared to W2 and W3, likely causing higher CO2-emissions (Table 3).

Modeled N fluxes
After selecting the posterior model runs according to chapter 2.3.2,we found the model generally capable in reproducing the measured data and consequently investigate the modeled C and N cycle in more detail.Modeled N2O emissions are shown for the different land use over time (Fig. 2), separated into different seasons (Fig. 3) and before/after management-events occur (Fig. 4).The complete modeled N cycle is given in Table 4.
Arable N cycle: Arable simulations consider an annual N input of 198 kg N ha -1 a -1 .This input is balanced by 109 kg N ha -1 a -1 gaseous (primarily N2), 30 kg N ha -1 a -1 nitrate leaching and 99 kg N ha -1 a -1 harvest losses (Table 4), meaning that modeled outputs are higher as the given inputs.This gap in the annual N cycle is fed by storages of the soil in the model, indicating N depletion over time.Even though N losses of NO3 -and particularly N2O emission (7 kg N ha -1 a -1 ) have only a minor percentage in the total N balance, both rates are high regarding their environmental impact as a GHG contributing to global warming and as a water pollutant regarding eutrophication and drinking water supply, respectively.However, the uncertainty related to our estimated NO3 -leaching rate is overall the second largest uncertainty source in our N balance.These estimates cannot be sufficiently constraint with the given observation data, but they are in line with other reported N leaching rates on arable land in Germany (Siemens and Kaupenjohann, 2002).
The simulated N2O emission contribute 3.1% to total simulated N losses.The underlying model runs follow the trend of the observation data.Hot moments can be observed after fertilizer applications, which are predicted by the model in time, but sometimes not in magnitude (e.g.March to May 2014).During these events, soil moisture is often not modeled accurately: The model predicts rewetting processes that have not been measured in the same magnitude (Fig. A1), which might explain the overestimated fluxes.A possible reason can also be uncertain rainfall model input data.Kavetski et al. (2006) found the measurements of precipitation within a catchment to be uncertain, as the trajectory of storm cells through a catchment may be different for each storm and may not have its center at the rain gauge where traditionally rainfall inputs are measured.Our rainfall data is measured 4 km northeast of the trace gas study area and is likely effected by such uncertainties.For both, simulated and measured emissions on the arable site are highest in spring (Fig. 3).While the transects A1 and A2 vary with 95% of the values between 0 and 0.05 kg N2O-N ha -1 day -1 , A3 shows a higher variation up to 0.15 kg N2O-N ha -1 day -1 .As A3 is located at the hill toe, we attribute this effect to lateral transport of nitrate from uphill.
However, our one-dimensional model set up does not cover lateral water and nutrition transport, accordingly the model is not able to predict the higher emissions at A3 in spring.While such a process is part of complex integrated hydro-biogeochemical catchment models (Haas et al., 2013;Klatt et al., 2017;Wlotzka et al., 2013), it has not yet been confirmed experimentally.
The distributions of the summer, autumn and winter seasons measured emissions are well in line with the modeled emissions.
Furthermore, the modeled emissions are also in agreement with measured emissions before and after manure applications (Fig. 4).This result corresponds with a study of Molina-Herrera et al., (2016) who found LandscapeDNDC to be capable of simulating agricultural N2O emissions.However, in our case the model overestimates peak emissions before fertilizer applications, which leads to higher mean annual modeled emissions (7.33 kg N2O-N ha -1 a -1 ).This is 2.8 kg N2O-N ha -1 a -1 higher than our observed emissions and even outside the large model uncertainty of 2.3 kg N2O-N ha -1 a -1 .Hence, future research should particularly investigate the reason for this overestimation of peaks, either by revising the model structure or identification of other sources of model uncertainty.
Grassland N cycle: Grassland simulations consider an annual N input of 12.7 kg, with 7.6 kg coming from modeled biomass that is transferred into applied dung and urine through grazing sheep.Simulated N loss is substantially larger with 22.3 kg N ha - 1 a -1 gaseous losses (primarily N2), 1.5 kg N ha -1 a -1 occurring as nitrate leaching and 29.8 kg N ha -1 a -1 by biomass removal through grazing sheep and harvest (hay making).Comparing inputs and outputs, we simulated a mean nitrogen gap of 41 kg N ha -1 a -1 .Decreasing soil organic N stocks in model simulations indicate that the model is currently mimicking an additional N source, which is not included in the current modelling approach.We assume uptake of additional N in form of nitrate by the shallow groundwater as a potential dominating process that is not included in the current LandscapeDNDC version we used.
Liebermann et al. ( 2017) used a revised LandscapeDNDC set up for hypothesis testing to identify potential additional N sources in groundwater-dominated grasslands and showed, that groundwater N uptake is a likely contributor.
Taking a closer look at the modeled N2O emissions, one can see that the model did not reproduce high as well as negative (N2O uptake) emissions.Currently, LandscapeDNDC does not consider any N2O uptake, accordingly negative fluxes cannot be simulated by the model.The peaky dynamics of simulated N2O emissions, especially from August 2014 to January 2015 are not confirmed by the measurements, indicating possible measurement errors in this time.One has also to consider the temporal mismatch of our weekly N2O measurements and the hourly simulations, making a full match of observations versus simulations difficult.
There is no clear effect of grazing on the N2O emissions on the grassland site in both measurements and modeled results (Fig. 4).Mean modeled annual emissions overestimate the observations by 0.4 kg N2O-N ha -1 a -1 , and even the simulated uncertainty bounds of 0.27 kg N2O-N ha -1 a -1 do not capture the measured dynamics.
Forest N cylce: N input is given for the forest model only by atmospheric deposition with an annual amount of 5.1 kg N ha -1 a -1 .
Gaseous losses amount to 1.8 kg N ha -1 a -1 .Leaching contributes 0.6% of the N output.The rest (3.3 kg N ha -1 a -1 ) is allocated  2).Parameterizing and simulating the forest transects independent from each other, would improve the simulations.One limiting factor is that both N2O uptake and frost-thaw cycles are not included in the current LandscapeDNDC version.We therefore recommend to particularly include frost-thaw cycles into the model (De Bruijn et al., 5 2009) as this process can have a major influence on N2O inventories, e.g. up to 73% of the annual N2O loss of a forest site in Högelwald, Germany (Papen and Butterbach-Bahl, 1999) was occurring during such cycles.The mean modeled annual emissions (0.33 kg N ha -1 a -1 ) overestimate the observed emissions, but capture the mean observed annual emissions with their uncertainty bands of 0.15 kg N2O-N ha -1 a -1 .10

Modeled C fluxes
Modeled CO2 emissions are shown for the different land uses over time (Fig. 5), separated into different seasons (Fig. 6) and before/after management-events occur (Fig. 7).The complete modeled C cycle is given in Table 5.
Arable C cycle: LandscapeDNDC simulations for the arable system predict a mean annual gross carbon uptake of 25.7 ± 1.3 t C-CO2 ha -1 a -1 .20.5 t C-CO2 ha -1 a -1 leave the system through respiration, from which maintenance respiration contributes the largest proportion (65%).This is perfect in line with annual measured losses (Table 3).Harvest output is with 4.7 t C ha -1 a -1 in good agreement with the observed yields (Fig. A4).However, the temporal dynamic of the modeled TER on the arable study site underestimates the emissions in the summer season (Fig. 6) and mean modeled fluxes are substantially lower than measured ones (Fig. 7).
Tillage and harvest events fall into the summer season.While the observed emissions drop after harvest by 25%, the modeled emissions drop by even 50%.The reason for this is either an underestimation of the emissions through LandscapeDNDC (after harvest events until tillage occurs), or the uncertainty in the upscaling method of the measured CO2 emissions (discussed in chapter 3.1.2).As microbial processes can oxidize more soil carbon after harvest (resulting in higher heterotrophic respiration), we assume that the discrepancy is rather stemming from the model simulations.There are also studies, e.g.Buyanovsky et al. (1986), who report highest soil respiration rates after harvest.Modelled and measured soil CO2 emissions agree well after tillage.
Grassland C cycle: LandscapeDNDC simulations for the grassland system predict a mean annual gross carbon uptake of 16.9 ± 1.7 t C-CO2 ha -1 a -1 and an annual loss of 13.2 t C-CO2 ha -1 a -1 through respiration.Further minor outputs are related to grazing (0.2 t C-CO2 ha -1 a -1 ), harvest (2.0 t C-CO2 ha -1 a -1 ) and allocation in the soil (1.4 t C-CO2 ha -1 a -1 ).Mean annual as well as temporal dynamics of modeled emissions are well in line with measured emissions.Effect of grazing has a minor influence on the total ecosystem respiration (Fig. 7), resulting in a wider range of both measured and modeled emissions.
Grazing, i.e. reduction of root biomass, results in two contrary processes: reduction of maintenance respiration and increasing autotrophic respiration (Raich and Tufekciogul, 2000).
Forest C cycle: The forest model predicts an annual C input of 8.9 ± 0.6 t C-CO2 ha -1 a -1 which is quite low compared to estimations for old-growth beech forests in Europe with reported rates from 14.4 to 18.3 t C-CO2 ha -1 a -1 (Molina-Herrera et al., 2015).However, C uptake rates vary in magnitude with values presented from 3 to 34 t C-CO2 ha -1 a -1 for different forests in different growing stages (Waring et al., 1998).As our study site is a mixture of young and old beech trees, we assume 40-50% less biomass compared to an old beech forest.Of the modeled C input, 6.6 t C-CO2 ha -1 a -1 leave the system as gaseous CO2.The rest is accumulated in the biomass and soil.Mean annual sum and dynamic of modeled emissions are in line with measured emissions.We expected to see rising emissions with litter fall in autumn (Raich and Tufekciogul, 2000), but cannot report this effect, neither with measurements, nor with model results (Fig. 6).

Biogeosciences
Furthermore, posterior model runs allowed quantifying magnitude and uncertainty of not measured fluxes of the C and N cycle.Investigated forest site is in general acting as the largest sink for C and N, with annual sequestration rates of 2.4 t C ha -1 and 3.3 kg N ha -1 .The extensive grazed grassland is also acting as a sink for C with 1.4 t C ha -1 per year, while the N cycle of the grassland model cannot be closed with the given settings.Shrinking N soil pools indicate a missing input, which we assume from shallow groundwater with additional N supply of around 40 kg N ha -1 a -1 .
Current land use in this catchment is dominated by forests (37%) and arable land (35%), whereas grassland sites (11%) are mainly distributed along the stream.Under the viewpoint of climate smart landscapes, measured data suggests the benefit of forests in a landscape, having the least GHG emissions.Riparian zones can act as sinks of N, but only during the vegetation period and times when roots have access to groundwater.Arable land use produces high amounts of N2O, but not throughout the year, rather in spring after fertilizer application.
Potential interactions of land use pattern cannot be quantified with the current one dimensional model approach.However, the dataset could be used in future studies to quantify nitrate uptake of riparian zones in more detail, e.g. by coupling LandscapeDNDC to a hydrological model as done by Klatt et al. (2017).Such a model setup would also allow an upscaling in space, e.g. for generation of GHG inventories or an analysis of more detailed management scenarios in time.
The model is finally used to estimate the magnitude and uncertainty of C and N fluxes, like N2 emissions or autotrophic and heterotrophic CO2 emissions, which are experimentally not yet quantifiable in Biogeosciences Discuss., doi:10.5194/bg-2017-96,2017 Manuscript under review for journal Biogeosciences Discussion started: 17 March 2017 c Author(s) 2017.CC-BY 3.0 License.situ.Remaining model and data errors are traced back to their potential sources to improve ongoing measurements and future model applications.

Figure 1 :
Figure 1: Map of the study area.Red squares represent GHG chamber positions at the different transects.Dark grey contour lines represent 5 m differences in elevation, light grey areas are outside of the catchment area.
in terms of the respective variable (WFPS in different depths [arable in 0.2, 0.4 and 0.6 m, grassland in 0.1 and 0.25 m and forest in 0.15 and 0.25 m], as well as yield on arable land).Parameter sets were accepted, if they belong to the 5% best model runs for each land use.Results of tier I are summarized in supplement Fig. A1-A4 and are not further discussed in this paper, as they belong to the initialization of the model.
Biogeosciences Discuss., doi:10.5194/bg-2017-96,2017 Manuscript under review for journal Biogeosciences Discussion started: 17 March 2017 c Author(s) 2017.CC-BY 3.0 License.into biomass and soil.By taking a closer look on the N2O emissions (Fig. 2), we see the model failing to reproduce the observed emission dynamics.Observed N2O emissions have high error bars and not all transects are driven by frost-thaw cycles or N2O uptake at the same time (Table

Figure 2 :
Figure 2: Measured and modeled N2O emissions from different land use.Measurements are given as grey error bars showing the variance between the replicated transects and the mean value as a black dot.Posterior model uncertainty is given in light color for the 5 and 95 percentile and dark color for the 25 and 75 percentile.Vertical lines indicate management events.In the uppermost panel, blue colored vertical bars indicate fertilizer application, while brown color indicate manure application.

Figure 5 :
Figure 5: Modeled CO2 emissions and management.Measurements are given as grey error bars showing the variance between the replicated transects and the mean value as a black dot.Posterior model uncertainty is given in light color for the 5 and 95 percentile and dark color for the 25 and 75 percentile.Vertical lines indicate management events.Brown colored bars in the uppermost panel indicate manure application.

Figure 7 :
Figure 7: Management effects on CO2 emissions.Measured and modeled emissions where selected in a time window of 2 weeks 5 Biogeosciences Discuss., doi:10.5194/bg-2017-96,2017 Manuscript under review for journal Biogeosciences Discussion started: 17 March 2017 c Author(s) 2017.CC-BY 3.0 License.The model-data fusion approach allowed us to derive model structural deficiencies that would likely increase model performances if implemented in Landscape DNDC: (1) missing N2O uptake processes; (2) missing NO3 -(and potentially dissolved organic nitrogen) uptake through shallow groundwater; (3) missing lateral interaction at hillslopes due to 1D model set up.

Figure A6 :
Figure A6: Posterior parameter distribution of the LandscapeDNDC modules wcDNDC and physiology.Orange line = arable, light green line = grassland, dark green line = forest model set up.

Table 5 : Simulated carbon fluxes given by posterior model runs and their uncertainty on different land use in [t C ha -1 a -1 ]. C manure on grassland includes input by sheep's dung. Arable model assumes 20% return of stubble to field.
We found high emissions of N2O and CO2 on our arable sites, low emissions on grassland sites and lowest emissions on the forest sites.These observations enable us to investigate underlying effects of plant growth, temperature and WFPS, land use effects, seasonal patterns and management effects.Respiration amounts rise in less shaded (warmer) areas of the forest, while N2O emissions increase toward the foot of the hills of forest and arable sites due to nitrogen accumulation.Highly variable N2O emissions in forest resulted in large uncertainty of model verification data and was translated in large uncertainty of model results for forest.
regard to land use, simulations for grassland sites work best, followed by those for arable land.Simulations for N2O emissions on arable land outperform those for CO2, and vice versa for grassland.Low emissions on forest sites were generally difficult to depict by our modeling approach.