Comments to: gmd-2019-253, Iles et al. “The benefits of increasing resolution in global and regional climate simulations for European climate extremes”.
Overall recommendation: minor revision.
This is an interesting paper about the effect of increasing model resolution on extreme events, considering the added value of regional climate models with respect to the driving GCMs and different spatial resolutions for the same GCM. An analysis assigning the differences due to resolution to upscaling or downscaling effects is certainly interesting, although a bit too succinct. The first part of the paper makes use of an impressive set of simulations and the consideration of observational uncertainty in the evaluation of precipitation and wind speed is highly appreciated. The paper reads well and is well-structured. It adds valuable insights to the existing literature, which is nicely referenced in the discussion.
I would recommend the consideration for publication in Geoscientific Model Development after minor revision. Note that although labeled as minor, these issues are relevant and should be well addressed in a revised version of the manuscript.
My main concerns are:
1) Bias correction. Why was the bias correction applied directly to the indices instead to the daily input data before obtaining the indices (L278-279)? I think that a correction based on the indices themselves could be noisier since they are values in the upper tail of the distributions and the objective is to look at return values which might be even more sensitive to unstable corrections. Applying a simple correction, such as the mean of the daily distribution, prior to the indices calculation would be more robust and also better for consistency among the indices based on the same variable. As far as I understand, maps show biases of the raw data (that should be said explicitly) but bias-corrected values values were only used for the return values (that should be stressed), thus as expected, moving the models towards the observations. It should be clearly motivated why this is the case or done in a more appropriate way. I also do not understand why bias correction was not applied to the UPSCALE simulations because of being only one model (L455-457). I think these simulations should be corrected as well, due to the nature of the analyzed metrics and for the sake of comparability with the previous sections.
Also, was it also an additive correction for precipitation and wind indices? I would expect to have a multiplicative correction in such cases.
2) Inconsistencies in wind extremes (L374-376). I acknowledge the explanations and the sensitivity analysis carried out about the different temporal resolution of the wind speed variables in the models. However, I would recommend not to include the analysis with such caveats. A safer way to go would be to consider for wind extremes only the models which can provide the variable which is consistent with the observations (6-hourly). I think that this consistency is more important than keeping consistency with the temperature and precipitation extremes, or than having a larger ensemble. Also, I do not understand the reasoning that values depend on the timestep; wouldn’t the primary time step for a given model the same for the three variables? The differences for CMCC, CNRM and the IPSLs are massive in the sensitivity analysis.
3) Wind reanalyses. It is good to include three reanalyses as reference to sample observational uncertainty (L371). However the differences among the reanalyses for the considered wind extreme index are massive. Are there any studies comparing them, showing how similar are they to real measurements? Which one should we trust more? Their quality should be brought into question: CMIP5 (also true for the UPSCALE simulations) has small bias with respect to ERA5, but that seems to be very unrealistic.
Minor comments
L86-69 When referring to the added value of higher resolution RCMs with respect coarser counterparts, the authors could consider to mention that the added value of the high resolution is not so evident when evaluated on the coarse grid, in particular, the improvement in the spatial pattern of precipitation indices is not statistically significant after applying simple bias correction methods (Casanueva et al. 2016, https://link.springer.com/article/10.1007/s00382-015-2865-x).
L102-113 About wind extremes, the authors might consider to mention the added value of coupled regional climate simulations in terms of surface wind and coastal low-level jet (Soares et al. 2019, https://link.springer.com/article/10.1007/s00382-018-4565-9), although this work focuses on northern Africa.
L142 Which version of E-OBS was used?
L 148 A good illustration of the E-OBS limitations (including indices such as return values) can be found in Herrera et al. 2019 (https://essd.copernicus.org/articles/11/1947/2019/) for the Iberian Peninsula.
L169-170 I am not sure if “sub-sample” is correct here, since the process goes from 1 hour to 6 hours, wouldn’t it rather be “aggregated”? Was the 6-hourly mean or maximum value obtained? Otherwise, please give some further details of the subsampling. Also in L208.
L184 It could be worth to mention here that simulations at the two resolutions are carried out with the same model versions and parameterizations, except for REMO, where rain advection is used for 0.11 but not for 0.44 (Kotlarski et al. 2014, https://doi.org/10.5194/gmd-7-1297-2014).
L215 That is probably a too strong statement. Smoothing/upscaling the high resolution might lead to partial information loss, but if there is an added value that might be also present at a coarser resolution.
L270-282 This paragraph does not really fit here. L270-273 was explained already in Sect.2.1.1 (no need to repeat), where the details about EC-EARTHr3 and the combination of the GCMs of the common subset (L274-276) should be moved to. Bias correction (L277-282) does not fit here either, it could be included in a new little subsection after regridding.
L291-293 This paragraph does not fit in the section about return periods. Either this analogue approach is fully described in the Methods in an own section, or this is removed and entirely described in Sect. 4.3. I would go for the second option.
L313-314 The last sentence of the paragraph is probably the main conclusion of Fig.S3: the driving GCM seems to be the largest source of variability, which is in agreement with previous studies (e.g. Rummukainen, et al. 2001, https://doi.org/10.1007/s003820000109). But I do not understand what the authors mean in this sentence with consistent results for a GCM-RCM chain; consistent with what? The message is clear if the authors remove “GCM-RCM chain”. Also in Fig.3, it would be very helpful to draw a box or mark somehow the columns belonging to the common subset (also in Fig.S5).
Fig.2 is too complicated, I am no able to see the mentioned shadings in the caption, corresponding to the full set or to the subset. Such shadings (if present) could be omitted and I would recommend to show only the individual simulations with different colours and the multi-model median of each subset. Observations and their ranges are a bit difficult to distinguish, the authors could try using another colour.
L352 What does “models” refer to here? GCMs? RCMs? I think that there is a difference here to the extreme temperature index, since for RX1day results seem to be more consistent for a given RCM regardless of the GCM than for the RCMs with the same driving GCM (see RCA-011).
Fig. 5 Aren’t the colorbars in Fig.5 switched?
L464 If Figure 5 is based on bias-corrected data and Fig.6 is not, they cannot be compared.
Fig.6 and S9 RX1day: I noticed different values for MESAN, which seems to be closer to E-OBS in Fig.S9, was there a bias correction performed for the reanalysis products towards E-OBS? I think I miss that explanation, which should be included in the methods. This comment also applies to wind extremes.
Fig.6 and S9. Why are the Alps not shown for wind (also in Fig.9)?
L565-569 The analogues analysis is very interesting. One question about analogues recognition: how do you set that the analogue is a good one? I mean, by looking at the correlation of spatial patterns you can always find analogues which can be more or less similar to the target situation, but did you set a correlation limit below which there is not a good analogue for one day? Or did you always find high correlations?
L574-577 This approach of smoothing before calculating the analogues does not seem to be right. The analogue day should be obtained with the same criteria for all variables/indices, i.e. a given atmospheric circulation is related to a value of temperature, precipitation and wind speed. Calculating it differently brings inconsistent variables. Moreover, this approach seems to be responsible of the overestimation of the return periods of Tx5day, later in lines 601-603, where it is also said that doing it differently results shift downwards.
L603 “but otherwise gives the same results”, isn’t “otherwise” the way it is done in the paper (i.e. first averaging, second analogues)? Then of course it produces the same results as shown. This sentence needs some rephrasing.
L606 It would be nice to have a quantitative value of the downscaling and upscaling effect on the indices, such as the relative change with respect to the self-analogue for a given or several return periods.
L604 How are model biases treated in Sect. 4.3? In my view, following the above thoughts on bias correction, mean biases should be removed from the analogue series prior to the indices calculation.
L618-625 The obtained results should be discussed in the context of other studies which show that RCMs yield systematic reduction of temperature biases compared with the driving GCM (Soerland et al. 2018, https://iopscience.iop.org/article/10.1088/1748-9326/aacc77, where some reasons for this are also given).
L692 I would recommend to mention around here or somewhere in the the discussion the potential benefits of current projects such as the EURO-CORDEX flagship pilot studies, about land use change and convection permitting simulations (Jacob et al. 2020, https://doi.org/10.1007/s10113-020-01606-9, for an overview on EURO-CORDEX perspectives).
Spellings and typos
L69 EURO-CORDEX initiative instead of EUROCORDEX project. Use EURO-CORDEX throughout the manuscript (there are some inconsistencies).
L69 missing bracket after the reference.
L74 Maybe better “coarser” than “less”, since it refers to the resolution.
L164 Isn’t a word missing between “adaptation” and “a downscaled”?
Table S1, caption “their corresponding CORDEX simulations to the left”, shouldn’tit be to the right? When describing the crosses, are they really bold? I would say that those in the “common subset” are those with coloured (not bold) crosses.
L215-217 Is then the 0.5º common grid the E-OBS grid? Is so, say it explicitly.
L221 I would say “The sensitivity of the results to the regridding technique...”, also L223 “sensitive to the regridding technique”, L225 “the regridding technique did not make much difference to the results”; but check with a native English speaker.
L305 Wouldn’t “CORDEX subset” be “common subset”? Here the differences between the left and right panels are being compared.
L309 Capitalize Figure, also in other parts of the manuscript.
L318 What do you mean with “are representative of the subregions”?
Fig.2 Caption. British Isles are in the top right panel, not in the top left.
L350 “E-OBS”. Homogenize notation along the manuscript: it is sometimes Eobs or E-OBS or EOBS.
L475 heavier.
L511 dot missing at the end of the sentence.
L566 Should “lows” be “flows”?
In all figures of return period in which the region of Scandinavia is included, Scandinavia is badly spelled.
L602 “see Methods” should be “see above”, since the procedure is explained above in the same section.
L655 dot missing at the end of the sentence.
L660 “can overestimate” |
- Printer-friendly version
- Supplement
