Modeling Carbon Turnover in Five Terrestrial Ecosystems in the Boreal Zone Using Multiple Criteria of Acceptance

Louise Karlberg, David Gustafsson and Per-Erik Jansson

Modeling Carbon Turnover in Five TerrestrialEcosystems in the Boreal Zone Using MultipleCriteria of Acceptance

Estimates of carbon fluxes and turnover in ecosystemsare key elements in the understanding of climate changeand in predicting the accumulation of trace elements inthe biosphere. In this paper we present estimates ofcarbon fluxes and turnover times for five terrestrialecosystems using a modeling approach. Multiple criteriaof acceptance were used to parameterize the model, thusincorporating large amounts of multi-faceted empiricaldata in the simulations in a standardized manner. Meanturnover times of carbon were found to be rather similarbetween systems with a few exceptions, even though thesize of both the pools and the fluxes varied substantially.Depending on the route of the carbon through theecosystem, turnover times varied from less than oneyear to more than one hundred, which may be ofimportance when considering trace element transportand retention. The parameterization method was usefulboth in the estimation of unknown parameters, and toidentify variability in carbon turnover in the selectedecosystems.

INTRODUCTION

In the light of recent concerns regarding future climate change,the contribution of ecosystem carbon exchange with theatmosphere has been the focus of many current research efforts(e.g., 1–4). Although both carbon budgets and carbon turnovertimes have been estimated for several ecosystems and at severallocations, it has been shown that the outcome is highlydependent on climate and the correct estimation of soilrespiration, both of which are highly site specific (5). However,the requirement of conducting site-specific field experiments canbe circumvented by the use of modeling tools. Fluxes of carbon,nutrients and water have been successfully estimated usingmany different integrated ecosystems models (e.g., 5–12). Oneof the major challenges will be to develop efficient parameter-ization strategies for these types of large ecosystems modelingpackages, integrating sub-models of many interacting processes,and to link them to large amounts of information on ecosystembehavior from field experiments (13).

Carbon turnover timescales can also contribute to estimatesof the accumulation of trace elements, such as radionuclidesoriginating from underground, nuclear waste repositories. TheSwedish Nuclear Fuel and Waste Management Co. (SKB) arecurrently investigating two sites (Forsmark and Oskarshamn) aspossible locations for a deep repository of radioactive waste(14). Estimates of carbon turnover in the ecosystems locatedwithin these areas facilitate an assessment of potential traceelement accumulation in biomass in the event of a leakage ofradionuclides into the groundwater from the repositories. Theaim of this study is to compare carbon budgets for fiverepresentative coastal terrestrial ecosystems in the Oskarshamnarea, southeast Sweden, using an ecosystem process model. Thismodel is calibrated with a Bayesian approach using multiplecriteria for model acceptance. Another aim is to identify

differences between ecosystems with respect to mean turnovertimes (MTT) of carbon, that may impact on the fate of traceelements entering into these systems.

METHODS

Ecosystem Description

Five terrestrial ecosystems were selected to represent thehypothetical systems included in the study (Table 1). Thesesystems were selected both because they are likely to differ interms of carbon turnover times, and also because together theyconstitute about 80% of the land cover at the study site atOskarshamn (578N 02605.0, 168E 038015.7). Average annualrainfall is 600–700 mm, and the yearly mean temperature is 6–7 8C (15). The first ecosystem, a semi-natural grassland, ischaracterized by the lack of a tree layer, and a field layerconsisting of a mixture of grasses and herbs growing on a claysoil. A forest dominated by alder (Alnus glutinosa) with a highgroundwater table was chosen to represent the secondecosystem. This deciduous tree has symbiotic nitrogen-fixingbacteria in its root nodules. Due to the ample supply ofnitrogen, alder retains only a small fraction of its nutrientsbefore shedding its leaves in the autumn. The field layer wascharacterized by nitrophilic grasses and herbs growing on a wetorganogenic soil type. A pine forest (Pinus sylvestris), growingon a thin layer of till with a field layer dominated by cowberry(Vaccineum vitis-idea), was chosen as the third ecosystem in thestudy. Another coniferous forest on till was selected to representthe fourth ecosystem; in this case Norway spruce (Picea abies).In this forest, the field layer consisted mainly of blueberry(Vaccineum myrtillis) and some broad-leaved grasses. Lastly, amanaged forest similar to the Norway spruce ecosystem incomposition was also included in the study as a comparison tothe natural ecosystem. Management was assumed to follow thegeneral practice for southern Sweden as recommended by theSwedish Forest Agency (16). Thus, the managed spruce forestwas cleaned after 15 years, resulting in a removal of 60% of thetree biomass, which was left in the forest as litter. In addition,the forest was thinned after 40 and after 80 years, affecting 25%of the tree biomass. While leaves, coarse roots and fine rootsremain at the site to form litter, 80% of the stem biomass (of thetrees affected by thinning) is removed. The field layer wasassumed to be unaffected by these operations.

Model Description

The CoupModel is a physically-based, ecosystems modelingpackage (17) that can be used to design ecosystem specificmodels (18). The model describes the interaction betweenbiogeochemical and hydrological processes in a one-dimension-al soil-plant-atmosphere system. Fluxes of water, heat, andmatter are calculated for a layered soil profile and one or severalvegetation layers above with time series of meteorological dataas the driving force.

The abiotic part of the model is based on two coupled partialdifferential equations for the water and heat flows in the soil:

448 Ambio Vol. 35, No. 8, December 2006� Royal Swedish Academy of Sciences 2006http://www.ambio.kva.se

the Richard’s equation (water) and the Fourier law of diffusion(heat), respectively (19). Surface boundary conditions, such asevapotranspiration, soil surface temperature, and snow melt arebased on energy balance calculations where net radiation isbalanced by turbulent fluxes of sensible and latent heat, andsurface heat flow (20–21). Water uptake from the soil is basedon a soil-plant-atmosphere-continuum approach, consideringthe flux of water from the soil through the plant as a response tothe demand of water from the atmosphere, i.e. the Penman-Monteith equation (22–24). Snow accumulation and melt aredescribed, as well as the partitioning between infiltration to thesoil or surface runoff at the uppermost soil boundary.

The biotic part of the model simulates plant growth, as well ascarbon and nitrogen turnover in the soil (25–26). Biomass ispartitioned into several above-ground and below-ground poolsof carbon and nitrogen (Fig. 1). Gross production of carbon(GPP), driven by solar radiation (27) and regulated by leafnitrogen content, water uptake, and air temperature, is allocatedto different compartments of the plant (leaves, stem, coarse rootsand fine roots) in different fractions (Table A1-A2). Theallocation to fine roots for spruce and pine is also a function ofroot mass (Table A1). At cleaning and thinning carbon isremoved from the plant storage compartments in the model.Each compartment is assumed to have a potential carbon tonitrogen ratio, which subsequently give rise to a nitrogendemand. Plant respiration is partitioned on growth andmaintenance respiration from all plant compartments, and is afunction of temperature (28). Daily litterfall is calculated as afraction of above-ground and below-ground parts of the plantentering the soil organic pools. Two pools with different turnoverrates were used to represent the soil organic material, called litterand humus. Decomposition of these pools are functions of soiltemperature and soil moisture content. The most importantinputs to the biotic part are thus characteristics governing the

plant life cycle such as allocation patterns, plant assimilation andrespiration, nutrient uptake by plants, external nitrogen inputs tothe soil, and finally decomposition and redistribution of differentdecomposition products in the soil profile.

Parameterization and Model Application

Parameter values characterizing the ecosystems were eithersynthesized from literature or determined in an automatedcalibration procedure based on Bayesian principles (29), usingprespecified criteria of acceptance including both site specific andgeneric data (Table A3–A5). This method quantifies parameteruncertainty and correlation rather than maximizing fit.

A number of parameter values characterizing the ecosystemswere derived from the literature and excluded from thecalibration; either values from field measurements or frommodeling studies (Table A1). These parameters described, forexample, carbon allocation in the plant, growth respiration,nitrogen demand and plant litterfall, and were called primaryparameters. Some of those were plant or soil specific, whereasthe rest were of a more general nature and were thereforeassumed to be the same for all systems. The parameterization ofthe field layer in the alder forest was used to represent thegrassland vegetation.

For each ecosystem there were also a number of parametersthat were not easily determined from the literature, so calledsecondary parameters (Table A2), which were used in thecalibration procedure. For each of these parameters, a range ofvalues was specified, assuming that the ‘‘true’’ value would liesomewhere within that range.

Essentially, the calibration method estimated probabilitydistributions for the calibrated parameters, characterized bymean values and variances. The basic principle is Bayes’theorem which, applied to the calibration problem, can beformulated as: the probability of a candidate parameter set tobe part of the posterior distribution, is equal to its priorprobability times the likelihood of the model output to be equalto the available data. It is further assumed that the likelihoodfunction can be chosen such that the difference between themodel output and the data can be attributed to additivemeasurement errors. This assumption is highly useful, since itallows the procedure to take into account observations ofdifferent output variables and error estimates. The followingfunction was used to calculate the logarithm of the likelihood L,based on the normal distribution:

logL ¼Xn

i¼1

�0:5ðOi � SiÞ2

M2i

� 0:5logð2pÞ � logMi

where Oi is the observed data, Si is the simulated data, Mi is themeasurement error, and n is the number of data points. It isassumed that the measurement errors are Gaussian and non-correlated. Since we have no other information on themeasurement error, it was further assumed that Mi was equalto 30% of the observed variable. For criteria that were assumedto be of particular importance, such as the carbon stock ofplants, the error was assumed to be smaller, which increasedtheir importance for the total likelihood.

Table 1. Description of the main characteristics of the ecosystems included in the study.

Name Tree layer Field layer Soil type Management

Grassland none Grass and herbs Clay NoneAlder Alder (Alnus glutinosa) Grass and herbs Organogenic NonePine Scots pine (Pinus sylvestris) Cowberry (Vaccineum vitis-idea) Till (thin) / bedrock NoneSpruce Norway spruce (Picea abies) Blueberry (Vaccineum myrtillis) Till NoneSpruce, managed Norway spruce (Picea abies) Blueberry (Vaccineum myrtillis) Till Cleaning, thinning and harvest.

Figure 1. Conceptual model of the main carbon pools and fluxes inthe CoupModel. If several vegetation layers are simulated, all plantstorage pools (the oval shaped symbol in the middle of the figure)are duplicated for each additional layer, as well as the flows to, fromand between these pools.

Ambio Vol. 35, No. 8, December 2006 449� Royal Swedish Academy of Sciences 2006http://www.ambio.kva.se

The parameter posterior distributions were identified byinvestigating the parameter space in an iterative manner, calledthe Metropolis-Hastings random walk. This is a version of theso-called Markov Chain Monte Carlo (MCMC) method. Thefirst simulation in the calibration procedure was started withparameter values in the center of the maximum and minimumvalues according to Table A2. A random change was thenadded to each parameter value in the subsequent simulations,assuming that the prior distributions of the calibrationparameters were uniform and non-correlated. The new candi-date values were accepted as part of the posterior distribution ifthe ratio of the corresponding likelihood value and the like-lihood of the previous point was larger than a random numberbetween 0 and 1. A total of 10000 iterations were included in thecalibration of each ecosystem and approximately 20–30 % of thecandidate values were accepted. Finally, for the calculation ofcarbon budgets and MTT, another 1000 simulations were runfor each ecosystem, using parameter values sampled from theposterior distributions.

Simulations were based on a one year dataset containinghourly climatic data from a meteorological station at thenorthern part of the Oland island (578N02201.3,178E05043.4)1981 (15), scaled to be representative for the SKB study area inAspo, Oskarhamn. This data was recycled and used as drivingdata for 10-year simulations for the natural ecosystems, and100-years for the managed spruce ecosystem. The reason forchoosing such long simulation periods was to insure that soiland plant carbon remained stable in the natural ecosystems.Another reason was that 100 years is the approximate rotationperiod for a spruce production forest in southern Sweden.Consequently, the initial carbon contents in the plant and soilwas parameterized according to average carbon plant and soillevels in natural ecosystems, except for in the managed forestecosystem, where the initial plant carbon content was chosen torepresent a newly planted tree.

RESULTS

The calibrated secondary parameters showed some similaritiesbetween ecosystems (Table A2). In the coniferous forests, thethreshold for reduction of photosynthesis due to leaf nitrogenstress was lower for the field layer compared to the trees;however, in the alder ecosystem, the trees had a comparativelylow threshold value. Humus decomposition rate tended to belower for the organogenic soils in the grassland and the aldersystems. The variability in the parameter estimations was large(CV.0.1, Table A2) for the respiratory flows, decompositionrate and uptake of organic nitrogen, in the pine, spruce andgrassland systems, while it was generally lower in the alder andproduction systems. A large variability in some of the allocationparameters and the threshold for reduction of photosynthesiswas common to all ecosystems.

There was a large degree of variation between ecosystemswith respect to how well the criteria of acceptance were met(Table A3 and A4). In all forest ecosystems but pine, the fieldlayer was about 10–20% too small, and in the alder ecosystemthe simulated tree carbon content was about 25% higher thanthe optimum value given in the corresponding criterion.Simulated carbon allocation to different parts of the plantgenerally showed good agreement with the relevant criteria,although the grass in the alder and the grassland ecosystems hadtoo much carbon in fine roots at the expense of the stem.Furthermore, plant respiration in relation to total respirationwas overestimated in all ecosystems except grassland, andconsequently the net primary production (NPP) was too smallcompared to GPP. Despite this, heterotrophic respiration wastoo large in relation to root respiration in the grassland, pine

and spruce ecosystems. The estimations of the mean values forthe criteria on the change in soil carbon pools, carbon contentin the field layer, plant carbon allocation, nitrogen balance,organic nitrogen uptake, net primary production, litterfall andleaf area index showed the largest degree of variability.

Carbon storage varied by a factor two between ecosystems(Fig. 2). The alder ecosystem had the highest total carboncontent (26 000 gC m�2) of all systems (Fig. 2b), followed by thenatural spruce forest (18 000 gC m�2) (Fig. 2d). Grassland, pineand managed spruce all had about the same total carbonstorage (around 13 000 - 15 000 gC m�2) (Fig. 2a, c and e).Looking instead at the distribution of carbon within thesystems, similarities exist between the grassland and the aldersystems, in which the majority of the carbon in the system waslocated in the soil biomass (97% and 70% respectively) (Fig. 2aand b). On the contrary, in the ecosystems dominated byconiferous trees, the soil biomass consisted of about half of thetotal carbon storage (Fig. 2c, d and e). A comparison betweenthe natural and the managed spruce ecosystems shows that thesoil carbon content was rather similar in the two systems,whereas the plant biomass was lower in the production systemaveraged over the 100 year rotation time (Fig. 2d and e). Totalecosystem plant biomass both above and below ground wassubstantially lower in the grassland ecosystem compared to therest, since this system lacks a tree layer (Fig. 3a and c). Thegrassland ecosystem was also the only system with the majorityof the plant carbon below ground. Litter content was rathersimilar in all systems, while the humus content was larger in thetwo organogenic soils (Fig. 3e).

Total fluxes of carbon in and out of the ecosystems alsovaried by a factor three between ecosystems (Fig. 2). Theproduction forest had the highest carbon flux (2600 gC m�2

yr�1), followed by the natural spruce forest (1800 gC m�2 yr�1)and the alder forest (1500 gC m�2 yr�1) (Fig. 2b, d and e). In thepine ecosystem, the carbon flux was rather low (1000 gC m�2

yr�1) and was even lower in the grassland system (690 gC m�2

yr�1) (Fig. 2a and c). Below-ground plant carbon flows werehigher than the corresponding flows above ground in theconiferous forests (Fig. 3b and d). However, in the grasslandand the alder ecosystems above ground, plant carbon flowswere instead higher than below-ground flows. Litterfall wastypically high below ground compared to the correspondingflow above ground in all ecosystems but alder (Fig. 3b and d).Finally, heterotrophic respiration from the soil litter fractionswas high compared to humus for all ecosystems (Fig. 3f).

MTT, that is the average time an assimilated carbonmolecule stays in a certain part of system before beingdischarged or respired, was calculated for the above- andbelow-ground parts of the tree and the field layers, as well as forthe litter and humus fractions of the soil (Fig. 4). Total fieldlayer biomass had the lowest MTT, less than one year, in allecosystems but pine (Fig. 4). In comparison, tree biomass hadan MTT of seven years, which thus was significantly higher.However, if the turnover time in trees is calculated for above-and below-ground parts separately, the picture is altered.Above-ground MTT for trees was 16 years on average, whilethe corresponding figure for the below-ground tree componentswas less than two years. There was also a large differencebetween the MTT for litter and humus (Fig. 4). While theformer had an MTT of less than two years, the MTT in humuswas found to be approximately 120 years. On average, totalMTT for the entire ecosystem was around 13 years.

Some differences in MTT between the ecosystems could beidentified (Fig. 4). For instance, the MTT of the below-groundtree biomass was more than three times higher in the alderforest (6.0 years) compared to the coniferous ecosystems (1.6years). Furthermore, MTT of the above-ground tree biomass in

450 Ambio Vol. 35, No. 8, December 2006� Royal Swedish Academy of Sciences 2006http://www.ambio.kva.se

the pine forest was about twice that of the other systems. Thepine forest also differs from the others in that the MTT of thefield layer vegetation was almost two years, compared to lessthan one year for all other ecosystems. Finally, MTT for thehumus pools varied between 60 years for the managed spruceforest to 230 years for the alder forest, whereas the MTT for thelitter pool varied very little between ecosystems (ranging from1.3 years to 1.8 years).

DISCUSSION

The small difference between mean simulated tree and soilcarbon storage pools, and the optimum values in thecorresponding criteria of acceptance, was expected. By choosingrelatively small error values, the model was forced to acceptonly those parameterizations that generated very accuratecarbon pools. This was done purposely since we believed thatthe correct estimation of these pools was fundamental to thecalculations of carbon flows within the systems. However, thehigh carbon content in the tree in the alder ecosystem wasproblematic. At present, few variables in the model limit thegrowth of the tree, since it is able to supply extra nitrogen fromthe nitrogen-fixing bacteria. The need for reducing tree growthin the alder ecosystem to meet the criteria of acceptance was

also reflected in the secondary parameter governing nitrogenstress, which in the case of alder was comparatively low. Inreality, growth in the alder is most likely limited by anotherfactor, such as potassium, but this was not accounted for in thepresent version of the model. Moreover, the model generallytended to underestimate the carbon content in the field layer,although the standard deviation of the simulated mean valuewas rather large. Lack of carbon in the stems of the grass in thealder and the grassland ecosystems was not realistic and couldhave been avoided by making the error values of thecorresponding criteria of acceptance smaller. However, thiserror in allocation patterns for the grass is not likely to have hada large impact on the total fluxes of carbon to and from the fieldlayer. In general, for the field layers, the relative amounts ofassimilates allocated to each plant carbon storage organ wassimilar to the relative actual amount of carbon in each storagepool. Trees, on the other hand, allocated a relatively largeproportion of new assimilates to fine roots and leaves, whereasmost of their standing stock biomass consists of stems andcoarse roots. The reason for this difference is that the turnoverrates of leaves and fine roots are substantially higher comparedto the turnover of stems and coarse roots.

All carbon storage pools are dependent on the fluxes to andfrom them. In the simulations, most secondary parametersgoverned fluxes of carbon, meaning that there was ample roomfor variation in the rates of all carbon flows. However, manycriteria of acceptance related both directly and indirectly to theflux, for example in the case of allowed changes in the carbonstorage pools over time, which thus contributed to limiting thisvariation. Some simulated flow variables were quite differentfrom the optimum values of their corresponding criteria,although the standard deviation for these variables wasrelatively low. This would indicate that they were in directconflict with one or more of the other criteria. Such was the casefor the criteria ‘‘plant respiration of total respiration’’ and‘‘heterotrophic respiration of total soil respiration’’. While theformer indicated that plant respiration generally was too highor that heterotrophic respiration was too low in the simulations,the latter instead indicated that simulated autotrophic respira-tion was too low or heterotrophic respiration too high. Theinability of the model to fulfill these criteria would thus implythat the model is either failing to describe some crucial process

Figure 2. Carbon budgets for different terrestrial ecosystems for thetree and the field layers separately: a) grassland b) alder forest c)pine forest d) natural spruce forest and e) managed spruce forest.Carbon storages (gC m�2) in bold and carbon fluxes (gC m�2 yr�1) initalics, including standard deviations. Plant carbon storage poolsinclude both above- and below-ground biomass for the tree and thefield layer respectively. The soil carbon storage pool representsboth the litter and the humus fraction. Litterfall and respiration fromplant storage pools take place both above and below ground, andrespiration from the soil carbon pool only includes heterotrophicrespiration.

Figure 3. Carbon budgets for different terrestrial ecosystems for theabove-ground and below-ground carbon pools and fluxes separate-ly. a) above-ground plant carbon pools b) above-ground plantcarbon flows c) below-ground plant carbon pools d) below-groundplant carbon flows e) soil organic pools, and f) soil organic carbonflows. Prod.¼managed spruce forest.

Ambio Vol. 35, No. 8, December 2006 451� Royal Swedish Academy of Sciences 2006http://www.ambio.kva.se

in the ecosystem, such as carbon allocation in the plant or thecompetition between the tree and the field layer, or that one ormore criteria are incorrect.

That carbon flows are higher below ground compared toabove ground in coniferous ecosystems is due to the highlitterfall rate of the fine roots. Similarly, the high litterfall rate inthe deciduous forest and the grassland above ground causes theabove-ground carbon flows to be higher than the correspondingflows below ground.

Because the size of the storage pools in relation to the fluxeswere of the same order of magnitude in all systems, the MTTsconsequently turned out to be rather similar. Although theMTT for the entire ecosystem was on average 13 years for allsystems, MTT varied drastically depending on the route ofcarbon through the ecosystem. Not surprisingly, the highestturnover rates were found in the field layer. The field layerturnover time was slightly higher in the pine forest, mainly as aresult of a relatively low field layer production. Generally,above-ground MTT was higher than below-ground MTT fortrees in all ecosystems, due to a slow turnover of carbon in thewoody material of the stems. Another interesting aspect onturnover in the different trees was the higher MTT of the below-ground biomass in alder. This might be a reflection of anunderestimation of below-ground litterfall in the model. Carbonthat does not leave the plants through respiration continuesthrough the system to the soil where it forms litter. Litter isquickly decomposed, resulting in another respiratory loss ofcarbon from the system. The remaining material forms humus,which has an estimated MTT of about 120 years. The variationof simulated MTT between ecosystems for the humus pools wasreflected in the difference in parameter values for humusdecomposition (Table A2). In conclusion, the turnover time foran atom of carbon entering the ecosystem can vary between lessthan one to more than one hundred years. If the fate ofradionuclides is of interest, it is thus of importance to know notonly how much that is assimilated into the biomass, but alsohow the trace elements are allocated to different parts of thesystem in order to predict the turnover time of the traceelements. To make it even more complex, the allocation anddecomposition patterns often differ between trace elements andcarbon. However, the present simulation of the carbon turnoverpatterns in the five selected ecosystems may be a starting pointfor a more elaborate analysis of trace element turnover.

The major uncertainties in the model parameterizations wererelated to carbon fluxes, e.g., photosynthesis, respiration,decomposition and organic uptake of carbon. This was evidentfrom the large standard deviation in the secondary parametersand criteria of acceptance governing these processes in the

model. Carbon budgets for boreal ecosystems are characterizedby a small net exchange as a result of large inflows andoutflows. For such ecosystems, it is of great importance to havewell-defined criteria, not only with regards to the relationshipbetween the inflows and outflows of carbon, but also on theabsolute level, to avoid over-parameterizations.

The estimated turnover times for the woody part of the trees(approximately 10–50 years), may be too low compared to otherstudies (13), which indicates that respiration as well asassimilation of carbon may have been overestimated. On theother hand, the turnover time of about 120 years in the slowlydecomposing soil organic matter (humus) is well within therange of reported values (13, 30–31). In the soil-surface layer,MTT was approximately five years in a spruce forest insouthern Sweden (Skogaby) (31), while in the terrestrialbiosphere in Australia it was less than two years (13). Assumingthat this layer consists mainly of what in the simulation wasdescribed as litter, this figure could be compared with the MTTof approximately two years for the simulated litter fraction.However, in the model the total litter pool consists of a mixtureof leaf, stem and root litter, which is not directly comparable tothese measurements, and could also explain why the simulatedvalue is slightly lower than the measurements in Sweden. Thus,it seems that the simulated values of soil MTT correspondedrather well with field measurements.

Net ecosystem carbon flux data were missing in this study,which could have improved the parameterizations of the plantand soil respiration. On the other hand, the resulting carbonbudgets for the spruce ecosystem were similar to those presentedby, for instance, Medlyn et al. (5), who reported a total systemrespiration of about 900-1600 gC m�2 yr�1 comparing spruceand pine forests in Sweden, UK, and France. Similarly, the totalsystem respiration from an alder forest in Germany wasestimated to be about 1800 gC m�2 yr�1, which is just slightlyhigher than those reported in this study (32).

This study demonstrated a method for how to identifycrucial ecosystem behavior using a detailed process-orientedmodel. The model was combined with climatic data from onesite and a number of different data sources to provide consistentdescriptions of carbon fluxes for different terrestrial ecosystems.Obviously, the criteria for model acceptance could be designedin many ways and the basic functional differences between theassumed parameter settings could be discussed. The criteriawere to some extent subjectively chosen, even though theBayesian calibration method was operational and transparentwhen synthesizing knowledge from many different sources andinvestigations. Obvious problems are that the simulation periodis very long and that the ecosystems are to a large extent onlysimplified representations of possible real ecosystems. Thismeans that a conventional validation and calibration of modelparameters is not a realistic alternative. The description ofcarbon turnover in the different ecosystems provided a framefor a dual understanding of the model design: first, of theconsequences of the interaction between processes described inthe model, and secondly, of the ability of the model to estimateimportant characteristics such as turnover time in differentcomponents of the ecosystems. Furthermore, the parameteriza-tion derived from this study could directly be used for differenttypes of simulation experiments, such as the impact of differentland use practices on the transport and retention of differentradionuclides. In order to understand trace element turnover interrestrial ecosystems it is important to recognize not only thedilution and allocation in the biomass following carbonassimilation by the plant and turnover in the soil, but alsoplant water uptake from the soil with its associated traceelement uptake. Investigations where the link between traceelement turnover and fluxes of both water and biomass in the

Figure 4. Mean turnover time of carbon in different parts of theecosystems; tree layer and field layer are separated into aboveground (AG) and below ground (BG) components, and the soilorganic carbon is separated into the fast ‘‘litter’’ pool and the slow‘‘humus’’ pool according to the CoupModel simulations. Verticalbars denote one standard deviation of the mean.

452 Ambio Vol. 35, No. 8, December 2006� Royal Swedish Academy of Sciences 2006http://www.ambio.kva.se

soil-plant system are crucial and could benefit from the type ofmodel descriptions presented in this study.

CONCLUSION

Carbon budgets and mean turnover times were estimated in allecosystems. The greatest uncertainties lie in the estimations offluxes such as photosynthesis, respiration and litterfall. Al-though both carbon storage and fluxes varied betweenecosystems, MTT was similar in all systems with minorexceptions such as longer turnover times of plant biomass inthe pine forest compared to the other systems. Depending onthe route of carbon through the system, turnover times variedfrom less than one year to over one hundred years, which maybe of importance when considering trace element transport andretention. The parameterization method based on multiplecriteria, synthesizing a wide range of empirical knowledge onecosystem behavior, was useful both in the estimation ofunknown parameters, and to identify possible variability incarbon turnover in the five selected systems.

References and Notes

1. Sellers, P.J., Hall, F.G., Kelly, R.D., Black, A., Baldocchi, D., Berry, J., Margolis, H.,Ryan, M., et al. 1997. BOREAS in 1997: Scientific results, experimental overview andfuture directions. J. Geophys. Res. 102, 28731–28770.

2. Aubinet, M., Grelle, A., Ibrom, A., Rannik, U., Moncrieff, J., Foken, T., Kowalski,A.S., Martin, P.H., et al. 2000. Estimates of the annual net carbon and water exchangeof forests: the EUROFLUX methodology. Adv. Ecol. Res. 30, 113–175.

3. Schulze, E.-D. 2000. Carbon and nitrogen cycling in European forest ecosystems,Ecological Studies 142. Springer: Berlin, Heidelberg, New York, 500 pp.

4. Baldocchi, D., Falge, E., Gu, L.H., Olson, R., Hollinger, D., Running, S., Anthoni, P.,Bernhofer, C., et al. 2001. FLUXNET: A new tool to study the temporal and spatialvariability of ecosystem-scale carbon dioxide, water vapor, and energy flux densities. B.Am. Meteorol. Soc. 82, 2415–2434.

5. Medlyn, B.E., Berbigier, P., Clement, R., Grelle, A., Loustau, D., Linder, S., Wingate,L., Jarvis, P.G., Sigurdsson, B.D. and McMurtrie, R.E. 2005. Carbon balance ofconiferous climates: Model-based analysis. Agr. Forest Meteorol. 131, 97–124.

6. Parton, W.J., Coughenour, M.B., Scurlock, J.M.O., Ojima, D.S., Gilmanov, T.G.,Scholes, R.J., Schimel, D.S., Kirchner, T.B., et al. 1996. Global grassland ecosystemmodelling: development and test of ecosystem models for grassland systems. In: GlobalChange Effects on Coniferous Forests and Grasslands. Breymeyer, A.I., Hall, D.O.,Melillo, J.M. and Agren, G.I. (eds.). Wiley and Sons, Ltd. Hoboken, New Jersey, USA,pp. 229–269.

7. Running, S.W. and Coughlan, J.C. 1988. A general model of forest ecosystem processesfor regional applications. I. Hydrologic balance, canopy gas exchange and primaryproduction processes. Ecol. Model. 42, 125–154.

8. Bossel, H. 1996. TREEDYN3 forest simulation model. Ecol. Model. 90, 187–227.9. Aber, J.D., Ollinger, S.V. and Driscoll, C.T. 1997. Modeling nitrogen saturation in

forest ecosystems in response to land use and atmospheric deposition. Ecol. Model. 101,61–78.

10. Bergh, J., McMurtrie, R.E. and Linder, S. 1998. Climatic factors controlling theproductivity of Norway spruce: A model-based analysis. Forest Ecol. Manag. 110, 127–139.

11. van Oene, H., Berendse, F., Persson, T., Harrison, A.F., Schulze, E.-D., Anderson, B.R.,Bauer, G.A., Dambrine, E., Hogberg, P., Matteucci, G. and Paces, T. 2000. Modelanalysis of carbon and nitrogen cycling in Picea and Fagus forests. Ecol. Studies. 142,419–467.

12. Ito, A. and Oikawa, T. 2002. A simulation model of the carbon cycle in land ecosystems(Sim-CYCLE): a description based on dry-matter production theory and plot-scalevalidation. Ecol. Model. 151, 143–176.

13. Barrett, D.J. 2002. Steady state turnover time of carbon in the Australian terrestrialbiosphere. Global Biogeochem. Cy. 16, 1108, 21 pp.

14. Lindborg, T., Lindborg, R., Lofgren, A., Soderback, B., Bradshaw, C. and Kautsky, U.2006. A strategy for describing the biosphere at candidate sites for repositories of nuclearwaste: Linking ecosystem and landscape modeling. Ambio 35, 418–424.

15. Larsson-McCann, S., Karlsson, A., Nord, M., Sjogren, J., Johansson, L., Ivarsson, M.and Kindell, S. 2002. Meteorological, Hydrological and Oceanographical Information andData for the Site Investigation Program in the Community of Oskarshamn. SKB ReportTR-02-03, Svensk Karnbranslehantering AB, Stockholm, Sweden.

16. Swedish Forest Agency, 2005. Grundbok for skogsbrukare. Skogsstyrelsens forlag,Jonkoping, Sweden 190 p. (In Swedish).

17. Jansson, P.-E. and Moon, D. 2001. A coupled model of water, heat and mass transferusing object orientation to improve flexibility and functionality. Environ. Modell. Softw.16, 37–46.

18. Jansson, P.-E. and Karlberg, L. 2004. Coupled Heat and Mass Transfer Model for Soil-Plant-Atmosphere Systems. Royal Institute of Technology, Dept of Civil andEnvironmental Engineering, Stockholm, Sweden, 435 pp.

A nutrient-rich semi-natural grassland, comprising a mixture ofgrasses and herbs. Photo: L. Modin.

Pine forest (Pinus sylvestris) is one of the five terrestrial ecosystemsmodeled. Photo: SKB photo archive.

Ambio Vol. 35, No. 8, December 2006 453� Royal Swedish Academy of Sciences 2006http://www.ambio.kva.se

19. Jansson, P.-E. and Halldin, S. 1979. Model for the annual water and energy flow in alayered soil. In: Comparison of Forest and Energy Exchange Models. Halldin, S. (ed).Society for Ecological Modelling, Copenhagen, Denmark, pp. 145–163.

20. Alvenas, G. and Jansson, P.-E. 1997. Model for evaporation, moisture and temperatureof bare soil: calibration and sensitivity analysis. Agr. Forest Meteorol. 88, 47–56.

21. Gustafsson, D., Lewan, L. and Jansson, P.-E. 2004. Modeling water and heat balance ofthe boreal landscape - comparison of forest and arable land in Scandinavia. J. Appl.Meteorol. 43, 1750–1767.

22. Penman, H.L. 1953. The physical basis of irrigation control. In: Report of the 13thInternational Horticultural Congress 1952. Vol:II., Synge, P.M. (ed). Roy. HorticulturalSoc., London, England, pp. 913–924.

23. Monteith, J.L. 1965. Evaporation and the atmosphere. In: The State and Movement ofWater in Living Organisms. 19th Symp. Fogg, G.E. (ed). Soc. Exp. Biol., The Companyof biologists, Cambridge, UK, pp. 205–234.

24. Lindroth, A. 1985. Canopy conductance of coniferous forests related to climate. WaterResour. Res. 21, 297–304.

25. Johnsson, H., Bergstrom, L., Jansson, P.-E. and Paustian, K. 1987. Simulated nitrogendynamics and losses in a layered agricultural soil. Agricultural Ecosystems andEnvironment 18, 333–356.

26. Eckersten, H., Jansson, P.-E. and Johnsson, H. 1998. SOILN Model, Ver. 9.2, User’sManual. Division of Hydrotechnics, Communications 98:6, Department of Soil Sciences,Swedish Agricultural University, Uppsala, Sweden, 113 pp.

27. Monteith, J.L. 1977. Climate and the efficiency of crop production in Britain. Philos. T.Roy. Soc. B. 281, 277–294.

28. Karlberg, L., Ben-Gal, A., Jansson, P.-E. and Shani, U. 2005. Modelling transpirationand growth in salinity-stressed tomato under different climatic conditions. Ecol. Model.190, 15–40.

29. van Oijen, M., Rougier, J. and Smith, R. 2005. Bayesian calibration of process-basedforest models: bridging the gap between models and data. Tree Physiol. 25, 915–927.

30. Schultze, E.-D., Beck, E. and Muller-Hohenstein, K. 2005. Plant Ecology. Springer-Verlag, Berlin, Germany, 702 pp.

31. Harrison, A.F., Harkness, D.D., Rowland, A.P., Garnett, J.S. and Bacon, P.J. 2000.Annual carbon and nitrogen fluxes in soils along the European forest transect,determined using 14C-bomb. In: Carbon and Nitrogen Cycling in European ForestEcosystems. Ecological Studies 142, Schulze, E.-D. (ed)., Springer, Berlin, Heidelberg,Germany and New York, pp. 237–256.

32. Kutch, W.L., Eschenbach, C., Dilly, O., Middelhoff, U., Steinborn, W., Vanselow, R.,Weisheit, K., Wotzel, J. and Kappen, L. 2001. The carbon cycle of contrasting landscapeelements of the Bornhoved Lake district. In: Ecosystem Approaches to LandscapeManagement in Central Europe. Tenhunen, J.D., et al. (eds.). Springer-Verlag, Berlin andHeidelberg, Germany, pp. 75–95.

33. Charles-Edwards, D.A., Doley, D. and Rimmington, G.M. 1986. Modelling PlantGrowth and Development. Academic Press, Australia, 235 pp.

34. Larcher, W. 2003. Physiological Plant Ecology: Ecophysiology and Stress Physiology ofFunctional Groups. 4th ed. Springer-Verlag, Berlin, German, 523 pp.

35. Strand, M., Lundmark, T., Soderbergh, I. and Mellander, P.-E. 2002. Impacts ofseasonal air and soil temperatures on photosynthesis in Scots pine trees. Tree Physiol 22,839–847.

36. Peng, Y.Y. and Dang, Q.-L. 2003. Effects of soil temperature on biomass productionand allocation in seedlings of four boreal tree species. Forest Ecol. Manag. 180, 1–9.

37. Hoffmann, F. 1995. FAGUS, a model for growth and development of beech. Ecol.Model 83, 327–348.

38. Wikstrom, F. and Ericsson, T. 1995. Allocation of mass in trees subject to nitrogen andmagnesium limitation. Tree Physiol. 15, 339–344.

39. Chung, H.H. and Barnes, R.L. 1977. Photosynthate allocation in Pinus taeda. I.Substrate requirements for synthesis of shoot biomass. Can. J. Forest Res. 7, 106–111.

40. Linder, S. and Troeng, E. 1981. The seasonal variation in stem and coarse rootrespiration in a 20-year old Scots pine (Pinus sylvestris L.). Mitteilungender ForstlichenBundes-Versuchsanstalt (Wein) 142, 125–139.

41. Penning de Vries, F.W.T. and van Laar, H.H. 1982. Simulation of plant growth andcrop production. Simulation monographs, Wageningen, the Netherlands. 307 pp.

42. Helmisaari, H.-S., Makkonen, K., Kellomaki, S., Valtonen, E. and Malkonen, E. 2002.Below- and above-ground biomass, production and nitrogen use in Scots pine stands ineastern Finland. Forest Ecol. Mana. 65, 317–326.

43. Johansson, T. 2000. Biomass equations for determining fractions of common and greyalders growing on abandoned farmland and some practical implications. BiomassBioenerg 18, 147–159.

44. Uri, V., Tullus, H. and L~ohmus, K. 2002. Biomass production and nutrientaccumulation in short-rotation grey alder (Alnus incana (L.) Moench) plantation onabandoned agricultural land. Forest Ecol. Manag. 161, 169–179.

45. Nordin, A., Nasholm, T. and Ericson, L. 1998. Effects of simulated N deposition onunderstory vegetation of a boreal coniferous forest. Funct. Ecol. 12, 691–699.

46. Nilsson, L.-O., Ostergren, M. and Wiklund, K. 2001. Hur paverkas traden ovan mark?In: Skogabyforsoket – effekter av langvarig kvave- och svaveltillforsel till ettskogsekosystem. Persson, T. and Nilsson, L.-O. (eds.) Naturvardsverket, Rapport5173, Stockholm, Sweden, 51–74. (In Swedish).

47. Ericsson, T. 1994. Nutrient dynamics and requirements of forest crops. New Zea. J. For.Sci. 24, 133–168.

48. Annika, Nordin Associate Professor, Unpublished data, Umea Plant Science Centre,Umea, Sweden.

49. Goverde, M., Arnone, J.A. and Erhardt, A. 2002. Species-specific reactions to elevatedCO2 and nutrient availability in four grass species. Basic and Applied Ecology 3, 221–227.

50. Olsson, Bengst Researcher, Department of Ecology and Environmental Research,Unpublished data. Swedish University of Agricultural Sciences (SLU). Uppsala,Sweden. 51. Persson, T. and Nilsson, L.-O. 2001. Skogabyforsoket – effekter avlangvarig kvave- och svaveltillforsel till ett skogsekosystem. Naturvardsverket, Rapport5173, Stockholm, Sweden. 220 pp. (In Swedish).

51. Nordin, A. and Nasholm, T. 1997. Nitrogen storage forms in nine boreal understoreyspecies. Oecologica 110, 487–492.

52. Middleton, E.M., Chan, S.S., Rusin, R.J. and Mitchell, S.K. 1997. Optical properties ofblack spruce and jack pine needles at BOREAS sites at Saskatchewan, Canada.Canadian Journal of Remote Sensing 23, 108–119.

53. Reich, P.B., Walters, M.B., Ellsworth, D.S., Vose, J.M., Volin, J.C., Gresham, C. andBowman, W.D. 1998. Relationships of leaf dark respiration to leaf nitrogen, specific leafarea and leaf life span: a test across biomes and functional groups. Oecologia 114, 471–482.

54. Waring, R.H. and Running, S.W. 1998. Forest Ecosystems. Analysis at Multiple Scales.Second Edition. Academic Press, California, USA. 370 pp.

55. Foster, T.E. and Brooks, J.R. 2005. Functional groups based on leaf physiology: arethey spatially and temporally robust? Oecologica 144, 337–352.

56. Tsialtas, J.T., Pritsa, T.S. and Veresoglou, D.S. 2004. Leaf physiological traits and theirimportance for species success in a Mediterranean grassland. Photosynthetica 42, 371–376.

57. Torbern, Tagesson, PhD student, Dept of Physical Geography and Ecosystems Analysis,Unpublished data, Lund University, Lund, Sweden.

58. Lindborg, T. 2005. Description of Surface Systems. Preliminary Site DescriptionSimpevarp Sub Area – Version 1.2. SKB Report R-05-01. Swedish Nuclear Fuel andWaste Management Co. Stockholm, Sweden, 273 pp.

59. Betts, A.K. and Ball, J.H. 1997. Albedo over the boreal forest, J Geophys Res 102,28901–28909.

60. Oke, T.R. 1987. Boundary layer climates. Methuen, London, pp. 12 and 138–139.61. Gustafsson, D., Stahli, M. and Jansson, P.-E. 2001. The surface energy balance of a

snow cover: comparing measurements to two different simulation models. Theor. Appl.Climatol. 70, 81–96.

62. Thornley, J.H.M. and Cannell, M.G.R. 1992. Nitrogen relations in a forest plantation –soil organic matter ecosystem model. Annals of Botany 70, 137–151.

63. Field, C. and Mooney, H.A. 1986. The photosynthesis-nitrogen relationship in wildplants. In: On the Economy of Plant Form and Function. Givnish, T.J. (ed) CambridgeUniversity Press, Cambridge, UK, pp. 25–55.

64. Kinerson, R., Ralstone, C.W. and Wells, C.G. 1977. Carbon cycling in a loblolly pineplantation. Oecologia 29, 1–10.

65. Penning de Vries, F.W.T. 1975. The cost of maintenance process in plant cells. Annals ofBotany 39, 77–92.

66. Olsson, M. 2000. Bidrar marken till vaxthuseffekten – eller kan den hjalpa oss att losaproblemen? Kungliga Skogs- och Lantbruksakademins tidsskrift 139, 41–50. (In Swedish).

67. Blomback, K. 1998. Carbon and Nitrogen in Catch Crop Systems – Modelling of Seasonaland Long-term Dynamics in Plant and Soil. Ph.D. Thesis. Acta Universitatis AgriculturaeSueciae, Agraria 134. Swedish University of Agricultural Sciences, Uppsala, Sweden, 31pp.

68. Skogsdata 2003. Aktuella uppgifter om de svenska skogarna fran Riksskogstaxeringen.SLU, Inst. for skoglig resurshushallning och geomatik. Uppsala, Sweden. Unpublisheddata, 114 pp. (In Swedish).

69. Lofgren, A. 2005. Oskarshamn/Forsmark Site Investigation. Estimation of Biomass andNet Primary Production in Field and Ground Layer, and Biomass in Litter Layer ofDifferent Vegetation Types in Forsmark and Oskarshamn. SKB Report P-05-80.Swedish Nuclear Fuel and Waste Management Co. Stockholm, Sweden, 37 pp.

70. Berggren, D., Bergkvist, B., Johansson, M.-B., Langvall, O., Majdi, H., Melkerud, P.-A.,Nilsson, A., Weslien, P. and Olsson, M. 2004. A Description of LUSTRA’s Common FieldSites.Report 87. Department of Forest Soils, Swedish University of Agriculture Sciences,Uppsala, Sweden, 41 pp.

71. Dilly, O., Bach, H.-J., Buscot, F., Eschenbach, C., Kutch, W.L., Middelhoff, U., Pritsch,K. and Munch, J.C. 2000. Characteristics and energetic strategies of the rhizosphere inecosystems of the Bornhoved Lake district. Appl. Soil Ecol. 15, 201–210.

72. Kutch, W.L., Staack, A., Wotzel, J., Middelhoff, U. and Kappen, L. 2001. Fieldmeasurements of root respiration and total soil respiration in an alder forest. NewPhytol. 150, 157–168.

73. Ericsson, T. 1995. Growth and shoot: root ratio of seedlings in relation to nutrientavailability. Plant Soil 168, 205–214.

74. Johnson, D.W. and Lindberg, S.E. 1992. Atmospheric deposition and forest nutrientcycling. Ecological Studies 91, Springer-Verlag, New York, 707 pp.

75. Nasholm, T., Ekblad, A., Nordin, A., Giesler, R., Hogberg, M. and Hogberg, P. 1998.Boreal forest plants take up organic nitrogen. Nature 392, 914–916.

76. Canham, C.D., Pace, M.L., Papaik, M.J., Primack, A.G.B., Roy, K.M., Maranger, R.J.,Curran, R.P. and Spada, D.M. 2004. A spatially explicit watershed-scale analysis ofdissolved organic carbon in Adirondack lakes. Ecol. Appl. 14, 839–854.

77. Raich, J.W. and Tufekcioglu, A. 2000. Vegetation and soil respiration: Correlations andcontrols. Biogeochemistry 48, 71–90.

78. Chapin, F.S., Matson, P. and Mooney, H.A. 2002. Principles of Terrestrial EcosystemEcology. Springer-Verlag, New York, 436 pp.

79. Flower-Ellis, J.G.K. 1985. Litterfall in an age series of Scots pine stands: Summary ofresults for the period 1973–1983. Rapport 19 Institutionen for ekologi och miljovard.Sveriges lantbruksuniversitet. Uppsala, Sweden, 75–94.

80. Marklund, L.-G. 1988. Biomassafunktioner for tall, gran och bjork i Sverige. Rapport 45.SLU, Inst. for skogstaxering, Umea, Sweden, 73 pp. (In Swedish).

81. Swedish Environmental Protection Agency, 2005. Sweden’s National Inventory Report2005 - Submitted under the United Nations Framework Convention on Climate Change.Report 5451, Swedish Environmental Protection Agency, Stockholm, Sweden, pp. 216.(http://www.naturvardsverket.se/bokhandeln)

82. We are grateful towards Annika Nordin, Tom Ericsson, Werner Kutch, MagnusSvensson and Anders Lofgren for providing data and useful references that facilitatedthe parameterization of the model. The funding of this work came from the SwedishNuclear Fuel and Waste Management Co (SKB).

83. All SKB reports mentioned in this article can be downloaded from http://www.skb.se

Louise Karlberg is a research fellow at the StockholmEnvironment Institute (SEI), in Sweden. Her main area ofinterest is water, nitrogen and carbon cycling in agriculturaland forest ecosystems in tropical and temperate regions. Heraddress: Stockholm Environment Institute (SEI), Box 2142,SE-103 14 Stockholm, Sweden.E-mail: [email protected]

David Gustafsson is an assistant professor at the Departmentof Land and Water Resources Engineering KTH, Sweden. Hismain research field is wintertime hydrometeorological pro-cesses in boreal and alpine landscapes.E-mail: [email protected]

Per-Erik Jansson is professor of Land and Water ResourcesSciences at the same department. He has long experience ofmodeling biogeophysical processes and has taken an activerole in the development of a number of widely used numericalmodels.E-mail: [email protected]

454 Ambio Vol. 35, No. 8, December 2006� Royal Swedish Academy of Sciences 2006http://www.ambio.kva.se

Appendix (Tables A1–A5)

Ambio Vol. 35, No. 8, December 2006 455� Royal Swedish Academy of Sciences 2006http://www.ambio.kva.se

456 Ambio Vol. 35, No. 8, December 2006� Royal Swedish Academy of Sciences 2006http://www.ambio.kva.se

Ambio Vol. 35, No. 8, December 2006 457� Royal Swedish Academy of Sciences 2006http://www.ambio.kva.se

458 Ambio Vol. 35, No. 8, December 2006� Royal Swedish Academy of Sciences 2006http://www.ambio.kva.se