Review Reports

Round 1

Reviewer 1 Report

I recommend the publication of this manuscript in the present form.

Author Response

Thank You for accepting the manuscript in the original form.

Please note that we revised the original version of the manuscript according to suggestions and comments of other two Reviewers.

Reviewer 2 Report

The study by Ostrogović Sever et al. presents an evaluation of the soil carbon (C) stocks simulated by the widely applied Biome-BGCMuSo ecosystem model by comparison to an extensive soil survey in Croatia, Southern Europe. Such a model testing is urgently needed for the soil modeling community and also for policy makers who want to use such models for national greenhouse gas (GHG) accounting. The paper shows how the apparent suitability of the model at the highly aggregated land-use-level degrades when dis-aggregating the results.

 

General comments

 

The manuscript is well written and introduces the motivation, data, and methodology in a concise and reproducable manner. The results are clearly presented and I enjoyed reading the paper.

 

However, from the presented results I draw a different conclusion: they do -not- indicate the models usefulness for modeling the average mineral SOC content.

To my reading, the apparent agreement on national land-use levels is a lucky compensation of two opposing biases. A large part of the modeled variation is not found in observations (Fig. 3, Fig A2) and the observed variation is not predicted by the model (Fig. 4). Hence, just taking the average observed C-stocks is better than the model prediction at biome-level (negative model efficiency in Table 4).

 

This different conclusion is already supported by the provided results and the authors also discuss ways forward (biome-specific parameterization). Hence, the paper with this negative result could be published.

 

On the other hand, to show the usefulness of the model, an uncertainty analysis for single plot-predictions is lacking. How much uncertainty in the prediction is due to uncertain model inputs and what due to uncertain model parameterization?

Moreover, for the GHG accounting, the changes in stocks are more relevant than the current stocks. For this changed objective the issue of model initialization and unknown management and disturbance history becomes even more severe. This needs to be discussed.

 

Specific comments

 

Please, use a consistent naming/abbreviations of land use categories and indicate the abbreviations, currently introduced in 2.4) already in Table 4 and earlier in the introduction (Line 143ff).

 

Please, give some indication of variability inside LU classes or LU-x-Bio classes in addition to the standard error. Are those classes indeed the main source of variability?

 

Some recent literature on changes, uncertainty, and the problem of model initialization:

 

Mao, Z., Derrien, D., Didion, M., Liski, J., Eglin, T., Nicolas, M., Jonard, M., and Saint-André, L.: Modeling soil organic carbon dynamics in temperate forests with Yasso07, Biogeosciences, 16, 1955–1973, https://doi.org/10.5194/bg-16-1955-2019, 2019.

 

Smallman, T. L., Milodowski, D. T., Neto, E. S., Koren, G., Ometto, J., and Williams, M.: Parameter uncertainty dominates C cycle forecast errors over most of Brazil for the 21st Century, Earth Syst. Dynam. Discuss. [preprint], https://doi.org/10.5194/esd-2021-17, in review, 2021.

 

 

 

 

 

Author Response

Please see attached document.

Author Response File: Author Response.pdf

Reviewer 3 Report

In this paper, the authors seek to model SOC stocks under different land uses and climates to support IPCC Tier 3 methodology in national C inventories. However this methodology requires not SOC stocks, but rather changes in these stocks caused by changes in land use over specified time periods, typically in units of g C m-2 y-1. Such changes are not addressed in this paper, nor how these changes are caused by land use practices. Rather the authors present different SOC associated with different ecosystems (cropland, grassland, deciduous and coniferous forests) without changes in land use.

The method by which these different SOC were modelled is not clear. How was SOC initialized for the 6000 year spinup runs? Were the SOC values modelled for the different ecosystems in different biogeographic regions independent of the those with which they were initialized? If not, the results in Figs 2 and 3 are not meaningful. No explanation of how these different SOC were modelled was offered. Was NPP modelled for the different ecosystems consistent with those measured at the various sites in the study? What about crop and grassland harvest removals?

The failure of some of the model runs due to N deficiency suggests a problem with the N cycle in the model, as N cycles continuously in these ecosystems as does C. This problem is important because N determines primary productivity and hence litterfall and thereby SOC. Does BIOME-BGC have fully coupled ecosystem C and N cycles? If not, then publication cannot be considered.

To contribute to IPCC Tier 3 methodology, the authors should examine the processes by which land use determines changes in SOC, how these processes should be modelled, and whether the results of such modelling are consistent with changes in SOC measured over specified periods of land use change.

Author Response

Please see the attached document.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

The updated version of the study does not well address my main concern. Although the difference in conclusions between the authors and me is shortly discussed, still the evidence presented does not support the conclusions drawn.

 

The conclusions drawn by the authors: “model is suitable” (or the similarly strong “useful”) needs more work. The results presented are necessary but not sufficient. For this conclusion, the laborious work of analyzing different (sub)sets of data and performing a parameter uncertainty analysis has to be endeavored.

 

First suggestions: 3 sites from each of the 3 biomes could be randomly selected to analyze the effects of parameter uncertainty with constrained labor. Computations could be done in parallel to constrain necessary computing time.

 

Second suggestion: A more cautious conclusion: The results do not refute the suitability of the model for …. While these results are a first step towards corroborating the suitability, we plan to tackle …