The trajectories of soil carbon in our changing climate are of the utmost importance as soil is a substantial carbon reservoir with a large potential to impact the atmospheric carbon dioxide (
The terrestrial carbon cycle consists of the uptake of
One way to evaluate soil carbon models has been to use observations of soil carbon stocks
An alternative, regionally integrated approach is using observations of atmospheric
In this study, we present a framework of how to use atmospheric
To obtain the atmospheric
We use in this work JSBACH, the land surface model of the Max Planck Institute's Earth system model, one of the models participating in CMIP6. The JSBACH model has two distinct soil models implemented in it (CBALANCE, CBA, and Yasso, YAS). We are interested in seeing if the two soil carbon models lead to markedly different How can we use a land surface model together with a transport model to evaluate soil carbon models and what problems do we face when doing that? What is the role of soil carbon stocks, the variables driving their decomposition and the functional dependencies of those variables on modelled heterotrophic respiration at global scale and how does this lead to differences in the atmospheric
We used the land surface model JSBACH
JSBACH is the global land surface model of the Max Planck Institute's Earth system model
Independent of the sub-model used for soil carbon, JSBACH uses three carbon pools for living vegetation: a wood pool containing woody parts of plants and green and reserve pools that contain the non-woody parts. JSBACH simulates different processes that lead to losses from the vegetation pools, such as grazing, shedding of leaves, and natural or anthropogenic disturbances. Depending on the process, some of the vegetation carbon is lost as
CBALANCE (CBA) is the original soil carbon sub-model of JSBACH
The function for soil temperature dependence,
Together these functions are modulating the rate of decomposition so that the heterotrophic respiration ( two green litter pools, one above ground and one below ground, in which the non-woody plant parts decompose with turnover times of between 1.8 and 2.5 years two woody litter pools, one above ground and one below ground, in which the woody plant parts decompose with turnover times of several decades; one slow pool receiving its input from the four litter pools and having a turnover time in the order of a century.
The original soil carbon model of JSBACH was replaced by Yasso (YAS)
The decomposition dependency on air temperature is as follows:
The decomposition depends on precipitation
Similar to CBA, YAS has slowly and rapidly decomposing pools, but its pool dynamics are more structured. First, all the pools are divided into woody and non-woody materials. The difference in the calculation of the decomposition rates between non-woody and woody pools is an additional parameter that decreases the turnover rates of the woody litter, which is dependent on its plant-functional-type-specific (PFT) size parameter
YAS takes the chemical composition of the incoming litter into account. The incoming litter is divided into different chemical pools according to the PFT-dependent factors. Information on the PFT-dependent factors for the litter decomposition has been derived from observations
JSBACH model simulations followed the TRENDY v4 protocol in terms of the JSBACH version, simulation protocol and forcing data
To estimate atmospheric
We fed TM5 daily biospheric, weekly ocean and annual fossil fuel fluxes to obtain realistic atmospheric
Fossil fuel emissions are from the EDGAR4.2 database
Surface observations of atmospheric
Locations of Global Atmosphere Watch stations, denoted by black dots, and different TransCom regions (different numbers denote the different TransCom regions in this study), denoted by different colours.
GOSAT (Greenhouse Gases Observing Satellite) from Japan Aerospace Exploration Agency (JAXA) was launched in 2009 and observes column
For the evaluation of the JSBACH model results, we additionally used data from two soil carbon databases and the FLUXCOM project
Since the two different model formulations differ only in their soil carbon module formulation, the incoming flux to the ecosystem from photosynthesis is the same in both cases. We analysed results for 2000–2014, and we show here averaged values for that period. The main target variable of our analysis is the heterotrophic respiration, but, to better elucidate how it influences the atmospheric
Even though annual total global values of heterotrophic respiration are close between the different model formulations (Table
The annual cycles of net primary production (NPP)
In addition to the comparison of the global results, we investigated how the two soil modules differed for broad latitudinally separated regions. The NPP is the same in the different latitudinal regions (Fig.
The annual cycle of net primary production (NPP)
The variation in
For the YAS model, on the other hand,
We also investigated whether the seasonal cycle of the heterotrophic respiration is correlated with litter fall. The only significant correlation occurred at 30–60
The global simulated GPP of 167
The annual net
The soil carbon stocks simulated by the two models differed in magnitude and also in their latitudinal distributions. The global estimate for total soil carbon by CBA was 4.5-fold larger than that by YAS (Table
Global C storage in the two different model formulations averaged over 2000–2014. For the YAS model, the eight above-ground pools are summed to obtain the litter pool, while the remaining 10 pools (below ground and humus) represent the soil pool.
The global distribution of soil carbon is very different between the model formulations (Figs. S11c, d, S12). The CBA model has large values of soil carbon at the mid-latitudes of the Northern Hemisphere. YAS predicts larger values in the temperate region of the Northern Hemisphere, but the highest values of soil carbon are located in arctic regions. The data-based estimates from SoilGrids and HWSD also predict the highest values at high northern latitudes (Figs. S11a, b and S12). The CBA model predicts higher values and differing latitudinal patterns south of 60
The turnover times of the two formulations must differ since the soil carbon pools are of very different magnitudes, but the annual
The turnover times for different temperature and precipitation regimes for the CBALANCE
Turnover time,
To assess whether the larger seasonal cycle amplitude in
The amplitudes of the seasonal cycle of
Seasonal cycle amplitudes of atmospheric
The seasonal cycle amplitudes of atmospheric
The capability of the model formulations to simulate the amplitude of the seasonal cycle differs within latitudinal regions (Fig.
Four surface observation sites in the Northern Hemisphere illustrate a similar behaviour in the seasonal cycle and its amplitudes as described above (Fig.
The detrended seasonal cycles of atmospheric
When comparing the overall bias in atmospheric
This evaluation of the two soil modules against satellite column
To further illustrate the results from this comparison, we show data for two regions that have a clear seasonal cycle. In TC region 2, the southern part of North America, CBA is more successful in capturing the observed seasonal cycle amplitude than YAS (Fig.
The seasonal cycles of detrended atmospheric
Overall, observed
There is bias in absolute
In this work, our aim was to use atmospheric observations to assess whether soil carbon models of a land surface model can be evaluated with this kind of framework. Our main finding was that the two models predicted different annual cycles of global
Annual heterotrophic respiration was 66.1
Global terrestrial C fluxes from the two different model formulations averaged over 2001–2014.
Moving to monthly timescales, we can see that the global seasonal
When heterotrophic respiration is compared by latitudinal zones, differences between the model formulations are visible in the variability and timing of the seasonal cycles in many regions (Fig.
The Pearson correlation
The observations show that litterfall has strong influences on heterotrophic respiration
Different moisture dependencies of
Simulated global GPP (165
The two soil models predicted different global soil carbon stocks (Table
The distribution of soil carbon stocks was also more realistic in YAS than in CBA (Fig. S12, Table S2). The large soil carbon stocks at mid-latitudes predicted by CBA (Figs. S11c, S12) are unrealistic compared to current data-based estimates of the global soil carbon distribution (Fig. S12). The large carbon stocks at high latitudes predicted by the YAS model (Figs. S11d, S12) are more in line with the observations but miss the high values observed from peatlands and permafrost in high-latitude regions. The version of JSBACH used does not include peatlands and is modelling only mineral soils. Therefore, the large carbon reservoirs of peatlands are not captured by the model. This JSBACH version also did not have permafrost described. If permafrost would be modelled, the seasonal cycle of heterotrophic respiration at high latitudes would likely be dampened as the depth of the active layer determines the amount of soil capable of respiring. The YAS model has been used in a JSBACH version containing permafrost in a study concentrating on the Russian far east
The environmental responses of the turnover times have quite different forms for the two soil carbon models (Fig.
The study by
The global turnover time of soil carbon by CBA was somewhat larger than in an earlier study, where it was estimated to be 40.8 years
The differences between the two models in the seasonal cycle of atmospheric
The differences in absolute
The space-borne observations give a similar message as the surface observations in TransCom regions, which showed a clear seasonal cycle. Niwot Ridge is located in TransCom region 2 (southern part of North America); there YAS also showed an amplitude that was too low, and CBA performed better, in a similar way as seen in the Fig.
The transport model itself also brings uncertainty to the result. The modelling of atmospheric transport is a challenging task as open scientific questions in the field remain
We demonstrated how atmospheric
The comparison of the two soil carbon models revealed large differences in their estimates. The YAS model better captured the magnitude and spatial distribution of soil carbon stocks globally. However, it was biased in its atmospheric
The evaluation was done within a land surface model that overestimates GPP in comparison to an upscaled GPP product, and this hampers doing benchmarking using this modelling system. Since the model is run to a steady state during the spin-up procedure, it also leads to other biases in the modelling system (influencing, for example, autotrophic respiration). Overestimated GPP leads to an enhanced litter input to the soil. This causes the comparison of the magnitudes of the soil carbon pools to the actual observations to be cumbersome as the overestimated litter fall causes biases in the model estimates. In this study, the magnitudes of simulated soil carbon are therefore not as good as the spatial patterns as an indicator of the model performance (such as latitudinal gradient). The other downside of the GPP biases is their influence on the estimated NEE. Due to the biases in the timing and magnitude of the other carbon fluxes, it is challenging to use
Soil carbon models have several development needs
In this study, we used space-born
A simple box model calculation was performed to evaluate the importance of the dependencies of environmental drivers and the soil carbon pool sizes on the larger global seasonal cycle amplitude in
Different annual cycles of the heterotrophic respiration (
The equations used monthly heterotrophic respiration, environmental drivers and soil carbon stocks averaged over 2000–2014 to estimate the turnover times for each grid point for YAS using Eq. (
Since the driving variables of soil moisture and annual precipitation differed in magnitude by approximately 4-fold, soil moisture was multiplied by 4 when using the function for annual precipitation (
The amplitude of global heterotrophic respiration within a year in different box model formulations. The input variables or functions that differ from the original model formulation are in bold letters.
The site level data from Global Atmospheric Watch – network is available via Obspack (2016) (
The supplement related to this article is available online at:
TT designed the experiment with the help of SZ. JEMSN performed the JSBACH model simulations. AT did the CarbonTracker Europe (CTE2016) runs with the JSBACH biospheric fluxes with the
Sönke Zaehle is an associate editor for
Tea Thum was funded by the Academy of Finland (grant no. 266803). Tea Thum and Sönke Zaehle were funded by the European Research Council (ERC) under the European Union's Horizon 2020 research and innovation programme (QUINCY; grant no. 647204). Sönke Zaehle was furthermore supported by the European Union's Horizon 2020 project funded under the programme SC5-01-2014 (CRESCENDO; grant no. 641816). Ingrid T. Luijkx received funding from the Netherlands Organisation for Scientific Research (NWO) under contract no. 016.Veni.171.095. Julia E. M. S. Nabel and Julia Pongratz were supported by the German Research Foundation's Emmy Noether Programme (PO1751/1-1). JSBACH simulations were conducted at the German Climate Computing Centre (DKRZ; allocation bm0891). We acknowledge JAXA/NIES/MOE for the GOSAT data. We thank Janne Hakkarainen for helping in analysing the GOSAT data and averaging kernel calculations. We thank Martin Jung for access to the FLUXCOM results and the FLUXCOM initiative. We are grateful for Naixin Fan for sharing the preprocessed SoilGrids and HWSD data with us. We thank Wouter Peters for constructive comments on an earlier version of this paper. We thank Willy R. Wieder and one anonymous reviewer whose constructive comments improved this paper.
This research has been supported by the Academy of Finland, Biotieteiden ja Ympäristön Tutkimuksen Toimikunta (grant no. 266803), the European Research Council's H2020 Research Infrastructures (QUINCY; grant no. 647204) and Horizon 2020 (CRESCENDO; grant no. 641816), the Netherlands Organisation for Scientific Research (NWO) (grant no. 016.Veni.171.095), and the Deutsche Forschungsgemeinschaft (grant no. PO1751/1-1)The article processing charges for this open-access publication were covered by the Max Planck Society.
This paper was edited by Kirsten Thonicke and reviewed by William Wieder and one anonymous referee.