This study evaluates the performance of the ORCHIDEE dynamic global vegetation model (DGVM) in accurately depicting the effects of land-use change (LUC) on biospheric carbon stocks across Europe. Through a systematic evaluation, the authors compare ORCHIDEE's predictions with observation-based estimates of key variables such as net and gross primary productivity (NPP and GPP), biomass growth patterns, and soil organic carbon (SOC) stocks. Additionally, they employ interpretable machine learning techniques to uncover factors contributing to discrepancies between model simulations and observational data.

A notable aspect of this study is its comprehensive analysis of ORCHIDEE's ability to reproduce spatial patterns and temporal trends in biospheric carbon stocks. The results demonstrate encouraging agreement between model predictions and observed data, particularly in capturing biomass accumulation with age and the general trends in SOC responses to LUC. Moreover, the authors adeptly discuss both the strengths and weaknesses of ORCHIDEE, providing a balanced assessment of its performance.

Overall, this paper makes a significant contribution to the fields of biogeochemistry and dynamic vegetation model development. Its meticulous methodology, insightful findings, and implications for model refinement enhance our understanding of the complex interactions between land-use change and biospheric carbon dynamics. I commend the authors for their rigorous approach and eagerly anticipate further advancements in this important area of research.

review of gmd-2024-42:

Impacts of land-use change on biospheric carbon: an oriented benchmark using orchidee land surface model

Summary

The authors perform a series of land model simulations for Europe and evaluate model performance against observations for key carbon variables, including soil carbon change due to land use change. They conclude that the model’s performance is reasonable, given its limitations.

Overall comments

I appreciate this effort to validate key carbon outputs of ORCHIDEE. In particular, the approach for assessing soil carbon change due to land use change is clever. The paper is also well organized and relatively clear. I do have the following main concerns, however:

1) Some of the methods are unclear; see below for details. in particular, it isn’t clear how you obtained your forest age outputs from the listed simulations.

2) I think that the experimental design for land use change is not adequate. Steady-state assumptions and inappropriate transition years introduce error into the desired relationships. You have information on the age of transitioned sites or the year of transition. And you have historical driving data for the model. You may even have site-specific meteorological driving data. While it may be difficult to simulate hundreds of individual sites, you can at least set up historical simulations that represent site-specific transition years, and don’t assume steady-state values. You can use the transient land cover up to the specified year in a cell, then make the desired PFT transition and then keep it constant for the rest of the simulation. Since SOC is PFT specific, just set the other PFTs to something else and then hold them constant also. This way the climate data are also more aligned with your observations. I think this could improve your results.

3) The conclusions regarding model performance are overstated. If these results are comparable to other models, this needs to be referenced.

Specific comments and suggestions

Abstract

Clarify that carbon stock change analysis is just for SOC.

Introduction

Materials and Methods

line 142:

but you include data with the forest floor. so this statement about excluding this is not true at all sites.

lines 219-227:

are the subsequent sims (eg, FG2F) done without wood harvest?

line 220:

this does not seem to be in steady state to me. climate is changing and did wood harvest stop only 50 year prior?

line 255:

what is the soil depth in the model? I don’t see it in the methods.

line 256:

should there be different beta^30 and beta^d0?

lines 265-282:

not sure how comparable the simulated and observed response functions are. the simulated one depends on particular climate forcing during the 20th century before and after transition. the observed one is based on contemporary measurements using space for time and transition years that do not correspond with your fixed 1950 model transition year, and there is a general and possibly invalid assumption that everything is in steady-state prior to transition.

lines 284-290:

unclear. how do you do model biases per site when you have an aggregate observed SOC change function? is the subset taken from model cells corresponding to the sites?

orchidee performance

lines 299-301:

the mode alignment to observations is overstated. 40% is not a relativey small difference. figure 1 clearly shows that some model medians outside of the 25-25th percentile boxes.

lines 310-328:

It isn’t clear how you obtained the model data for the different age classes. your BM simulations appear to be spun up with the PFTS, and then continue with these PFTs. Where is the starting point for a PFT to determine its age? see lines 197-200.

lines 330-337, figure 4:

it is difficult to see the differences in figure 4. i suggest replacing 4b with the difference plot: model - obs

SOC change

lines 373-381:

I suspect these results are affected by a site-specific training to simulations that are not site specific.

Discussion

line 394:

this does not seem like a positive ‘noteworthy’ here. you focused on sites that matched the model PFTs, and while several processes may not be represented, the correlations are not very good.

If this is comparable to other models, you need to reference their evaluations.

lines 399-400:

this is actually a feature of averaging, not climate.

Conclusions

lines 447-449:

this agreement is overstated.