the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 08 Apr 2026
The first offline land carbon simulation over Europe driven by the atmospheric forcing of a global storm-resolving climate model
Abstract. This study is motivated by the hypothesis that the drizzle problem of coarse-resolution climate models, whereby convective precipitation preferentially falls as light precipitation rather than short-lived and intense storms, leads to low gross primary productivity (GPP). To test this hypothesis, we perform an offline land carbon simulation over Europe using a terrestrial biosphere model driven by atmospheric forcing from a global km-scale climate simulation with explicitly resolved convection. This simulation is compared with a coarse-resolution simulation derived by atmospheric forcing from a coarse-resolution climate model with parameterized convection. The km-scale forcing leads, on average, to higher GPP. We find that shorter, more intense daily precipitation events when convection is explicitly resolved allow for stronger downward shortwave radiation on rainy days, thereby enhancing photosynthesis. At the same time, differences in the precipitation climatology between the two atmospheric forcing datasets, with a deficit of precipitation over eastern Europe in the km-scale forcing, result in soil moisture falling below the wilting point and reduced GPP in that region. Consistent with these GPP changes, autotrophic respiration is larger in the km-scale simulation, whereas heterotrophic respiration is smaller, the latter due to drier conditions.
- Preprint
(1928 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2026-1493', Anonymous Referee #1, 24 Jun 2026
-
RC2: 'Comment on egusphere-2026-1493', Anonymous Referee #2, 10 Jul 2026
Lee and Hohenegger investigate the hypothesis that differences in precipiation patterns and frequency are the dominant cause for carbon cycle differences between coarse and high resolution climate models. They analyse outputs of one land surface model driven by output from two different climate models, which also operate of different spatial resolutions. While the hypothesis is interesting, the manuscript suffers from a number of major and minor flaws. I am uncertain whether these can be addressed in a satisfactory manner by revisions.
Major comments:
The authors acknowledge that the climate field used as input differ in nearly all aspects of climate (temperature, radiation, humidity, rain), however, their method to disentangle these co-variate effects) is at best cursory. Figure 2, for instance, analyses the dependency of simulated gross primary productivity to precipitation intensity, without controlling for covarying factors such as average and seasonality of temperature, air humidity etc. As the authors do not provide any analysis of the seasonality of precipitation intensity, it is impossible to evaluate to what extend seasonal variations in temperature (and possibly phenology) confound this finding.
The authors interpret small (a few percent) regional differences in average carbon fluxes in terms of the hypothesised effect of rainfall intensity, but fail to ascertain that the difference do not simply stem from difference in the spatial extend of land mass resulting from the obviously radically different land-sea mask. No evaluation is presented that rules out that difference presented are simply an expression of slight shifts in the contribution of different plant functional types resulting from the different resolutions. No assessment is presented in terms of the different average climatologies of the simulations. It is for instance very conceivable that the strong different in East-European gross primary production (GPP) between the two simulation is an expression of different seasonal or annual mean precipitation and thus moisture stress, rather than of the frequency of rainfall per-se. However, an evaluation of these climate characteristics and spatial patterns is not given.
The authors imply that the decline in heterotrophic respiration (Rh) between coarse and high resolution is a function of rain fall intensity. However, they fail to evaluate the more likely reason that the stronger increase in autotrophic respiration (Ra) compared to GPP results from a shift in carbon allocation (the split between respiring and non-respring biomass), which potentially results from small shifts in the relative contribution of herbaceous and wood PFTs to continental productivity. Given that the authors spun the model into equilibrium for both configuration, the stronger increase in Ra over GPP requires a decline in Rh for the carbon budget to be balanced.
It is questionable that the results obtained by the authors are applicable to any other Earth system model. Although the JSBACH model accounts for the diffuse radiation in the canopy radiation scheme, this information is not used to account for the effect of diffuse radiation on gross primary production. Thus, unlike other land surface models, JSBACH does not account for the change in light use efficiency with cloud cover and hence likely overestimates the GPP sensitivity to shortwave downward radiation.
JSBACH employs a soil representation unlike most other land-surface models, with the notable difference that it assumes a direct dependence of soil respiration on precipitation, while most other land surface models tend to assume that the moisture control on decomposition acts as a function of soil moisture (and therefore a quantity that integrates rainfall, evapotranspiration, surface runoff and drainage). YASSO’s precipitation response has originally been devised to account for long-term spatial differences in moisture and not to represent the impact on rainfall variability on Rh. It is unlikely that many land surface model would exhibit a similar rainfall sensitivity
The method description contains a number of inaccuracies that need to be resolved. The authors claim they use MPI-ESM-LR for their coarse resolution simulation. However, the output is presented not on the spherical grid of MPI-ESM-LR, but on a triangular grid of ICON. Either, the input forcing was regridded to the triangular grid (which is missing from the methods description, but is relevant because this will tend to spatially smear out the native climate model output), or a different climate model has been used to generate the forcing.
Minor Comments
L19, but note Georgievski & Hagemann (2019), who show stronger effects of land-use uncertainty.
L29: This is not fully correct, as many of the studies quoted are either driven by observed climate or re-analysis, but not Earth system model simulations.
L68ff: Please stick to commonly accepted terms. The carbon cycle is a biogeochemical cycle and not a biogeophysical process .
L98: This is not correct, amonst others, one model is discretised using a spherical grid whereas the other employs a icosahedral-triangular grid.
Section 2.1 describes a version of JSBACH that differs from the version referred to in Section 2.2. This is confusing, because the version referred to in Section 2.2. does not consider PFT or carbon processes
L104-105: (a) it is inconsistent to use 1850s climate state together with 1990s CO2 concentration and a justification for this set-up is not given. (b) The statement that the carbon balance between land and atmosphere is closed is misleading because in the simulations presented, there is no interactive carbon cycle between land and atmosphere and therefore the carbon balance is not closed at all. The atmospheric concentration is prescribed and does not react to changes in the land carbon balance. This set-up is the opposite of presenting a closed carbon cycle.
L171: correct, but why do you then not investigate the alternative differences stemming from the different radiation schemes?
References
Georgievski, G., Hagemann, S. Characterizing uncertainties in the ESA-CCI land cover map of the epoch 2010 and their impacts on MPI-ESM climate simulations. Theor Appl Climatol137, 1587–1603 (2019). https://doi.org/10.1007/s00704-018-2675-2
Citation: https://doi.org/10.5194/egusphere-2026-1493-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 262 | 77 | 25 | 364 | 17 | 21 |
- HTML: 262
- PDF: 77
- XML: 25
- Total: 364
- BibTeX: 17
- EndNote: 21
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
Overall, this is an interesting study. The idea that precipitation duration in km-scale climate forcing may affect shortwave radiation and land carbon fluxes is worth exploring. However, the manuscript still needs substantial additional support. At this stage, many results are based on comparisons between two model setups, but it is still not clear whether the SRM forcing is more realistic, or how much of the GPP difference can really be attributed to the treatment of convection.
1. The paragraph around L50, where the authors explain that the two forcing datasets differ in many aspects beyond convection, is important. However, it may fit better in the Methods or Discussion rather than in the Introduction. The two forcing datasets differ in more aspects than just the treatment of convection. However, acknowledging this limitation does not resolve the uncertainty in the interpretation. Since SRM and LR differ in model resolution, climate model, forcing frequency, land-surface representation, and background climate, it remains difficult to determine how much of the GPP difference is actually caused by explicitly resolved convection or shorter rainfall duration. The additional sensitivity tests should be provided to better support this attribution.
2.The reason for fixing atmospheric CO₂ at 367 ppm in both simulations is not sufficiently clear. This choice may be reasonable, but the motivation should be explained more directly.
3.The Results section mainly compares LR and SRM, and most of the results show differences between LR and SRM. This is useful, but it does not really tell us whether SRM is better or more realistic. The paper needs more quantitative evaluation, not only maps and mean differences between the two simulations. Otherwise, the results show that the two simulations are different, but not necessarily that the SRM forcing improves the land carbon simulation. The paper mentions FLUXNET, MODIS, and MsTMIP, but this is mostly a broad comparison of carbon flux magnitudes. Why not directly compare the key variables with observations or high-resolution products?
4.The increase in domain-mean GPP in SRM is quite small compared with the mean GPP value. Given that SRM and LR differ in many aspects, it is difficult to know whether this small difference really comes from the mechanism highlighted in the paper or from other differences between the two setups. A more direct test would be to focus on the specific days or events that are most relevant to the hypothesis and compare precipitation duration, shortwave radiation, and GPP against observations or high-resolution products. This would help show whether SRM actually captures the observed rainy-day relationship between rainfall, radiation, and ecosystem productivity better than LR. Without this kind of event-based evaluation, the small mean GPP difference is hard to interpret.
5.There seems to be a numerical inconsistency in Section 3.2.2. The text says that the mean Ra value is 479.4 g C m⁻² yr⁻¹ in SRM, but Table 1 shows that 479.4 is the LR value, while the SRM value is 519.2. This should be corrected.
6.In Summary section, the statement that the land surface in SRM “takes up more carbon” should be clarified.