Review of

Systematic underestimation of type-specific ecosystem process variability in the Community Land Model v5 over Europe

 by Poppe-Teran et al.

General comments:

This is an interesting study that compares several gridded datasets of land energy, water and carbon fluxes with in situ observations over Europe. Simulations from the CLM5 land surface model are examined in more detail using two categories of simulations, one for each plant functional type and another aggregated to the grid cell level. It is found that CLM5 tends to underestimate the variability of water and carbon fluxes. A list of possible reasons is given in the Discussion section, but nothing is said about the representation of leaf area index (LAI) by CLM5 and how a misrepresentation of the LAI seasonal cycle and interannual variability could affect model performance. In the introduction, the authors give a very broad definition of phenology without ever mentioning LAI. In reality, LAI is certainly more directly related to phenology than any other variable considered in this work. Moreover, LAI strongly controls land surface fluxes and the evaporative fraction. How is LAI represented in CLM5? Because LAI responds to environmental conditions, it can exhibit large interannual variability. Failure to represent this variability would reduce the ability of the model to represent land surface fluxes.

Recommendation: major revisions.

Particular comments:

- L. 105-106 (warm winter 2020): Explain why it is called "warm winter".

- L. 143-144 (single soil column): This means that PFT-scale simulations are influenced by other PFTs. This weakens the rationale for PFT-scale simulations. It should be noticed than in other models, each PFT has its own soil column within a grid cell. This should be mentioned in the Discussion section.

- L. 151: How does soil moisture affect stomatal conductance? Given the scope of this work, this should be clearly and completely explained.

- L. 227 (warm winter 2020): For which time period are data available in the WARM-WINTER 2020 dataset? Only the 2019-2020 winter?

- L. 263 (1995-2018): Clarify the link to the WARM-WINTER 2020 data set.

- L. 300: Fig. 1c is not readable as many symbols overlap. This could be improved.

- L. 326 (Table 2): units are missing ; what is the meaning of the symbol of column 2, lines 6 and 11?

- L. 340 (Table 3): units are missing ; what is the meaning of the symbol of column 2, lines 6 and 11?

- L. 356 (Fig. 2): For a given PFT, is it a mean value across sites?

Synopsis and general comments: CLM5 ET and GPP are compared to ICOS sites in Europe, with RMSE and percent bias metrics. Model ET is often closer to the observations than remote sensing data, but model GPP is underestimated, particularly in deciduous forests.

Generally, the methods in this study seem robust, sources of uncertainty are carefully considered (Section 4), and the aims of the study are worthwhile. I certainly agree with the recommendations, especially with respect to optimizing PFT parameters and co-location of biodiversity and other data with the ICOS sites. The RMSE and bias metrics are well explained and appropriate. However, this study would be more accessible to a broader readership if metrics re the phenology and data distributions were better explained, with far less text given to describing the many details of the results and more to interpretation. Data-model comparisons (including for seasonal effects) should be quantified where possible, rather than just assessed by eye.

The authors present the RMSE and bias results in tables 2 and 3, and also in the text (with some mistakes; e.g. in Section 3.2). Please consider displaying these results in a single diagram, such as a modified Taylor diagram.

Seasonal effects are shown in Figures 2 and 3, but they are then only discussed qualitatively in the text. There is no attempt to quantify differences (or variability) between model and observed peak ET or GPP timings. For example, model vs observed phase lag or estimated day of max ET or GPP (calculations clearly explained, with appropriate error bars) could be plotted and assessed. In any case, it would be helpful if the second "hypothesis" at the end of the introduction states how goodness-of-fit for phenology is to be quantified. Likewise, for the third "hypothesis", briefly state how the variability will be quantified.

The introduction states that the statistical distributions can help "contextualize" model drought responses, but there is no analysis or discussion about interpretation of the higher moments, responses to drought, or the apparent bimodality seen in Figures 4 and 6 in this article. Do the ICOS data suggest drought conditions at any time at any station? If not, are there any other climate-related factors that could be discussed or quantified here? Please analyze/quantify/discuss drought, or another factor appearing in the ICOS data. In any case, it would be useful to know how drought (or other factor) affects skewness and kurtosis, and more broadly, what these moments will actually tell us or why we should care about them. For example, would we use the kurtosis to indicate changes in the frequency of extreme values, given kurtosis is a measure of the "heaviness" of the tails of a distribution?

Specific remarks

Abstract

The second sentence sounds odd. CLM5 quantifies fluxes and estimates the carbon and water budgets, potentially allowing for a better understanding of how climate change impacts ecosystems.

Line 30: reanalyses of what?

Figures and tables

Figure 1c (map of flux towers): Please show the extent of the CLM5 grid (1544x1592 gridcells), perhaps by using a different/lighter grey or white outside the grid area. Please state the number of stations shown in the caption.

Figures 2 and 3 (seasonality curves): Define the ICOS, GLASS and model acronyms, and clarify that these curves are means and standard deviations of data covering X-X years during the period 1995 through 2018. It is difficult to see the ICOS curves in some of the panels; please bring them to the foreground in these plots to make them more obvious.

Figures 4 and 6 (statistical distributions): It is rather difficult to see alignment of the main peaks in some of the panels, which is a point of discussion in the text.

Figures 5 and 7 (moments): Clarify that these are moments from the distributions shown in figures 4 and 6. The kurtosis appears an "excess" kurtosis, given the normal distribution has kurtosis=3. Please clarify.

Tables 2 and 3: Ideally, the model and remote sensing acronyms should be defined in the caption; this may be more important than those of the PFTs which are defined in the previous figure and table. Please also explain PFT ⌀ in the final rows for the RMSE and PBIAS sections.

2.1.2 Setup of the European CLM5

Line 178: did you mean sub models for ice rather than "stub" models for ice?

2.2.1 Station data

Line 229: Table S1 lists a single PFT for each ICOS station; please clarify here that this the dominant PFT as indicated in the last sentence of sec 2.3 ending line 267.

Line 234: Please state how many of the 73 stations were kept after wetlands, mixed forest, shrublands and indeterminate-land-cover stations were excluded; I assume 42 (the sum from Table 1).

3.2 General model performance

Line 333 bottom of page 16: The absolute value of PBIAS is smaller for CLM5_PFT than for CLM5_grid

but the actual PBIAS is lower, being more negative. Please clarify; at least replace the word "lower" with "smaller".

Line 336: In Table 2, ERA5L and GLASS RMSEs are largest for ENF and DBF as stated, but their RMSEs are lower than those of the CLM for GRA, rather than "similarly" as stated. Their PBIAS values are also much closer to zero than those of CLM5.

Section 3.3.1 ET

Line 371: Please refer back to Table 2 when discussing the PBIAS and RMSE for ET. "Conversely" is better than "Oppositely"; the latter sounds weird.

After this point, I stopped attempting to compare the text to the tables and figures. Rather than summarizing key points of a story, the text rambles on far too much about almost every detail of the results, and it is not always easy to see those details in the figures.