Review of

Getting the leaves right matters for estimating temperature extremes

by Duveiller et al.

General comments:

This paper shows that the ERA5-Land product should be used with caution and that the ERA5-Land production chain should use a more modern approach for representing vegetation or include satellite-derived LAI into their LSM. I am not sure this paper has much practical value from a modelling point of view. "Getting the leaves right matters" is perfectly right but far from being a new requirement. The added-value of this work should be better explained. Interesting recommendations are given in the Discussion section. For example, the recommendation to use hourly LST data from geostationary satellites. It is well known that solutions to integrate LAI into LSMs do exist. They are not mentioned and not used in ERA5-land. Could ERA-Land incorporate interactive LAI at some stage? Assimilation of LAI observations? Why not using another more advanced LSM forced by ERA5 atmospheric variables? The joint use of LAI and LST is interesting and offers a good benchmarking framework for assessing model performance. However, this paper may give the wrong impression that LST biases are completely explained by LAI. Other factors include the absence of representation of irrigation, snow misrepresentation, altitude solar radiation bias in mountainous areas, and slope effects in complex terrain. Overall, the paper is well written and a few changes could be sufficient to address my remarks.

Recommendation: minor revisions.

Particular comments:

- L. 40: I don’t understand well the LAI definition used by the authors: “LAI is defined as half of the total green leaf area per unit horizontal ground surface area”. Why half? Because only one side of the leaf is counted? For the sake of clarity, I would recommend using this more precise definition: “LAI is the one-sided green leaf area per unit horizontal ground surface area”.

- L. 152 (GEOV2/AVHRR): The THEIA LAI data portal web page should be given. Not only the LTDR web page.

- L. 202 (GLEAM): Which satellite date are used in this version of GLEAM? Is LST used for example?

- L. 241 (“darker than”): I am not sure this can be considered as a general rule. At wintertime, wet soils might be darker than senescent vegetation. Is this represented in the model? Adding a reference showing what is occurring in the real world would be useful.

- L. 266 (Fig. 4): Figure 4 interpretation is not straightforward. This Figure does not show much more than Fig. 3 and has too many tiny sub-figures, difficult to read. The complete Figure could be moved to a Supplement and a selection of meaningful sub-figures could be left in the paper. Why not plotting the two HI and BD indices in a SM - Temperature space instead? I.e. replace Fig. 4 by the two left sub-figures of Fig. 5?

- L. 311 (Fig. 6): Connecting maps to the colour scale is difficult. I suggest plotting only 3 colour classes: one for significant positive correlation, one for significant negative correlation (significant meaning p-value < 0.01), and white for non-significant correlation.

- L. 324 (Fig. 7): The top Cfb subfigure is not readable (dark green is difficult to distinguish from the dark gray color used for ocean surfaces).

The manuscript aims at assessing the impact of LAI errors and missing inter-annual variability in the ERA5-Land reanalysis land surface temperature (LST). This is a very relevant study for users of the reanalysis, but more generally to model development as it clearly identifies several limitations of the reanalysis LST which can be associated with errors and missing variability in LAI. The methodology is very robust, in particular (i) the focus on the hysteresis behavior between LAI and LST biases and (ii) the focus on several recent European heatwaves. However, there are some points in the methodology and results discussion that could be clarified and are listed in the following points. 

Main comment: 

In this study LAI is presented as a key variable that could explain some of the errors and it is also suggested that the assimilation of LAI could reduce some of the biases. I think that this is a simplification of a very complex problem, and deserves a more detailed discussion. Land surface models are a set of parameterizations with different degrees of complexity and uncertainty. In the case of LST, it depends on the turbulent exchanges and ground heat flux. There we can identify 3 key components: (i) coupling with the underlying soil temperature; (ii) aerodynamic resistance and (iii) canopy resistance. It is expected that LAI modulates both the aerodynamic resistance (e.g. increasing drag/turbulence when LAI is higher) and canopy resistance (increasing this resistance when LAI is higher). In the case of ERA5-land only the canopy resistance is modulated by LAI linearly and normalized by the minimum canopy resistance. In other models this relation is different, and other models also account for the LAI in the aerodynamic resistance (e.g. roughness lengths in SURFEX). The message that I think it would be important to be clear is that LAI is not directly used by the model, but used as a predictor of some parameters in some parameterizations. This study makes a significant advance to our knowledge by clearly showing the relation between vegetation status (via the LAI) and the LST biases, and in my opinion, suggests the need to revise both (i) the actual LAI data used (including assimilation) and (ii) the way that the model uses this data in the different parameterizations. My suggestion to the authors, if they agree, is to have some of this discussion in the paper (discussion and conclusions).  

Results & discussion clarifications

Section 2.3: Selection of 5 warmest days: The selected days from MYD11A1 and ERA5-Land are the same, or where are the 5 warmest days selected independently from each dataset ? In the discussion “A possible improvement could be to work with daily values and select explicitly days in ERA5L that have clear-sky condition” suggests that the days are not the same. Please clarify. If the days are not the same, I see that this could be a limitation in the interpretation of some of the results, as they could be affected by the clear-sky bias. 

Lines near 240: “... LAI during winter, and this corresponds to an overestimation of LST by ERA5L in the northern latitudes. This relationship between the bias in LST and the bias in LAI is consistent for such energy limited situation where biophysical effects of vegetation on climate are dominated by radiative effects. In fact, the modelling framework assumes there is an excess of leaves covering the background, the former being generally darker than the latter (especially when the background is covered with snow) resulting in more heat accumulation than what would be observed in a situation with fewer leaves.” 

The authors suggest that the northern hemisphere winter warm bias could be related with excessive LAI. I have some concerns with this hypothesis : 

1. from a previous comment, are the 5 warmest days in each month the same in MYD11A1 and ERA5-Land ? In winter with snow conditions clear-sky conditions could be colder than cloudy-sky (stronger radiative cooling)
2. The bulk snow scheme in ERA5 and ERA5-land is known to have a warm bias due to a large thermal inertia which is partially addressed with a multi-layer snow scheme (see https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2019MS001725 )
3. The results in Figure 3b also show a rather constant LAI bias in winter with a wide range of LST biases.

Therefore I suggest the authors revise these arguments regarding the warm bias of LST in winter in the northern hemisphere, also in lines 260:265 (discussion of Figure 3b). 

Figure 6 : inter-annual correlations: From my understanding the correlations were computed for each point using 16 values (years). With this sampling a correlation of 0.5 will have a p-value near 0.05. Therefore only correlations above  0.5 (simple approximation here) should be considered to avoid miss-interpretation due to the very small samples. I recommend to show this in the figure, e.g. shading non-significant correlations, and treat it in the discussion with caution. 

Line 307: “In other words, the years when the seasonally prescribed LAI of ERA5L is further from the reality, e.g. in years where the LAI peak is lower or shifted due to particular growing conditions of that year, the underestimation of LST can be expected to be more severe.” I recommend caution on the interpretation of these results. The correlations show that in years with negative anomalies of LAI ERA5-Land has a positive LST bias. The LST warm bias can be due to the missing LAI anomaly, but such LAI anomalies are also likely reflected on other aspects of the land surface, e.g. drier soil moisture, changes in albedo, changes in surface roughness. But these results are very relevant, as they show that ERA5L LST has larger errors during extremes, which are key in our understanding of the present and future climate.  

Results and discussion in Figure 7: By mixing all seasons I find it hard to interpret the results. Considering the case studies focus in summer, would it be easier to interpret these results if only summer months were used ? I leave this to the author's consideration. 

375: “warning for users of both ERA5 and ERA5-Land that the magnitude of heatwaves in particular may be underestimated in these datasets” This is the case if a user uses LST to define/evaluate the heatwave event. However the most common approach in the literature is to evaluate 2-meters temperature. Please clarify here that these results only apply to Land surface temperature. 

376: “assess the sensitivity of LAI to temperature, not to…” To avoid misunderstanding, I suggest to change to “assess the sensitivity of LAI to land surface temperature, not to” 

Line:408 “ This was currently not done because the values provided in ERA5L only represent snow-free albedo, which do not reflect the same reality as the MODIS albedo covering all conditions.” A clean way to compute ERA5L albedo and compare with MODIS would be to compute it from the surface net solar radiation and Surface solar radiation downwards, which are available in the CDS for ERA5-Land . I don’t think it would make a large difference and be necessary to be done for this study, but could be mentioned in the discussion.