the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Simulations of the Holocene Climate in Europe Using Dynamical Downscaling within the iLOVECLIM model (version 1.1)
- 1Department of Natural Sciences and Environmental Health, University of South-Eastern Norway, Bø, Norway
- 2Faculty of Science, Cluster Earth and Climate, Vrije Universiteit Amsterdam, Amsterdam, the Netherlands
- 3Laboratoire des Sciences du Climat et de l'Environnement, LSCE/IPSL, CEA-CNRS-UVSQ, Université Paris-Saclay Gif-sur-Yvette, France
- 4School of Geography, Earth and Environmental Sciences, University of Plymouth, Plymouth, UK
- 1Department of Natural Sciences and Environmental Health, University of South-Eastern Norway, Bø, Norway
- 2Faculty of Science, Cluster Earth and Climate, Vrije Universiteit Amsterdam, Amsterdam, the Netherlands
- 3Laboratoire des Sciences du Climat et de l'Environnement, LSCE/IPSL, CEA-CNRS-UVSQ, Université Paris-Saclay Gif-sur-Yvette, France
- 4School of Geography, Earth and Environmental Sciences, University of Plymouth, Plymouth, UK
Abstract. This study presents the application of dynamical downscaling in Europe using iLOVECLIM (a model of intermediate complexity), increasing its resolution from 5.56° to 0.25° latitude-longitude. A transient simulation using the appropriate climate forcings for the entire Holocene (11.5–0 kyr BP) was done for both the standard version of the model and with dynamical downscaling applied. Our results show that, simulations from dynamical downscaling present spatial variability which agrees better with proxy-based reconstruction and other climate models as compared to the standard model. The downscaling scheme simulates much higher (by at least a factor of two) precipitation maxima and provides detailed information in mountainous regions. We focus on examples from the Scandes Mountains, the Alps, the Scottish Highlands and the Mediterranean. The higher spatial resolution of the dynamical downscaling provides a more realistic overview of the topography, gives local climate information such as precipitation and temperature gradient that is important for paleoclimate studies. The results from the downscaling show in some cases similar magnitude of the precipitation changes reconstructed by other proxy studies (for example in the Alps). There is also a good agreement for the overall trend and spatial pattern than the standard version. Our downscaling tool is numerically cheap which can perform kilometric-multi-millennial simulations and suitable for future studies.
Frank Arthur et al.
Status: final response (author comments only)
-
RC1: 'Comment on cp-2022-21', Anonymous Referee #1, 20 Apr 2022
Summary
The authors apply a “dynamical” downscaling technique to an orbital-only Holocene-long climate simulation and obtain a resolution of 0.25 degree over Europe. The downscaling is performed for temperature and precipitation. The authors were able to show that the downscaled precipitation matches precipitation and temperatures in mountain regions more realistically. In particular, the simulated trends of the downscaled data resemble reconstructions from different proxy archives.
General
The study touches a highly important topic in paleo climate, namely the mismatch between local climate proxy information and the rather coarsely resolved paleo climate modelling. The authors present an approach to bridge the gap of scales by a downscaling approach. Clearly the topic desires publication in the Climate of the past, but the current study shows a number of short comings. Besides some structural problems (see below) there is a lack in presenting the state of knowledge in the introduction as a large number of recent studies on dynamical downscaling for paleo climatic studies is missing. Then the results lack a clear discussion of seasonality differences in the proxy data. Also the method itself is to my opinion named wrongly: the authors called it dynamical downscaling and explain that the basic idea is to reproduce the model physics and NOT the dynamics. Due to my rather long list of comments (not sorted into major or minor) I recommend at least major revision of the manuscript.
Comments
Title: The authors use only orbital and GHG forcing so I would not call it a Holocene Climate simulation, rather an orbital-GHG-only simulation for the Holocene period.
L 60 and the following paragraphs: There is growing literature on real dynamical downscaling using RCMs on the Paleo perspective so please make a reasonable review on the existing knowledge. Here is a collection of possible publications:
Bromwich, et al. (2004). Polar MM5 simulations of the winter climate of the Laurentide ice sheet at the LGM. Journal of Climate, 17(17), 3415–3433.
Gómez-Navarro, J. J. et al.: Internal and external variability in regional simulations of the Iberian Peninsula climate over the last millennium, Clim. Past, 8, 25–36, doi:10.5194/cp-8-25-2012, 2012.
Gómez-Navarro, et al.: A regional climate palaeo simulation for Europe in the period1500–1990 – Part 1: Model validation, Clim. Past, 9, 1667–1682, doi:10.5194/cp-9-1667-2013, 2013
Gomez-Navarro, et al. 2015: A regional climate palaeosimulation for Europe in the period 1501-1990. Part II: comparison with gridded reconstructions. Climate of the Past, 11, 1077-1095
Ludwig, P et al. (2016), Regional atmospheric circulation over Europe during the Last Glacial Maximum and its links to precipitation, J. Geophys. Res. Atmos., 121, 2130–2145
Ludwig, P., et al. 2017: Impacts of North Atlantic Surface Temperatures on European Climate during the Last Glacial Maximum in a regional climate model simulation. Geophys. Res. Lett., 44, 5086-5095.
Velasquez P., et al. 2021: The role of land cover on the climate of glacial Europe. Climate of the Past, 17, 1161-1180.
Velasquez P., et al. 2020: A new bias-correction method for precipitation over complex terrain suitable for different climate states. Geoscientific Model Development, 13, 5007-5027.
Russo, E., and U. Cubasch, 2016: Mid-to-late Holocene temperature evolution and atmospheric dynamics over Europe in regional model simulations. Clim. Past, 12, 1645–1662
Russo, M., et al. 2022: The long-standing dilemma of European summer temperatures at the Mid-Holocene and other considerations on learning from the past for the future using a regional climate model, Climate of the Past, https://doi.org/10.5194/cp-2021-101
L61: There a several approaches to statistical downscaling simulation. An example is Latombe et al. 2018, but there are several more publications so I encourage the authors to make a lit. search to add it to the introduction.
Latombe, G. et al., 2018: Comparison of spatial downscaling methods of general circulation model results to study climate variability during the Last Glacial Maximum. Geosci. Model Dev., 11, 2563–2579
L65: The publication Feser et al. is not dealing with paleo research questions so why is it cited?
L80-85: Here is another place to add model publication of real dynamical downscaling.
L95: Units are not in italic.
L100 and following: There are several more studies doing Holocene long simulation. A recent one is Bader et al. but please check again the literature. There are also approaches which first use a coarse resolved AO GCM and then a high resolved A GCM forced by the SST and sea ice distributions, please check Merz et al. and Hofer et al. publications.
Bader, J., et al. 2020: Global temperature modes shed light on the Holocene temperature conundrum. Nat. Commun., 11, https://doi.org/10.1038/s41467-020-18478-6.
Hofer, D., et al. 2012: The impact of different glacial boundary conditions on atmospheric dynamics and precipitation in the North Atlantic region, Climate of the Past, 8, 935-949
Hofer, D., et al. , 2012: Simulated winter circulation types in the North Atlantic and European region for preindustrial and glacial conditions, Geophys. Res. Lett., 39, L15805
Merz, et al. 2013: Greenland accumulation and its connection to the large-scale atmospheric circulation in ERA-Interim and paleo-climate simulations, Climate of the Past, 9, 2433-2450
Introduction in general: A discussion on the so-called ‘Holocene temperature conundrum’ is missing. Please check Bader et al 2020 and Liu et al. 2014.
Liu, Z., et al. 2014: The Holocene temperature conundrum. Proc. Natl. Acad. Sci., 111, 3501–3505, https://doi.org/10.1073/pnas.1407229111.
L125: Please change in the caption to “extent”.
L 129: Why do you go from section 2 to subsubsection 2.1.1? This makes no sense. This is happening several times. There are also sections where there is only one subsection which is again awkward. Please correct the structure according to the rules of the journal.
L140: The ECBilt model is a quasi-geostrophic model so the most important mode of variability in the climate system ENSO is not included in the model by definition. So how does this shortcoming impact your results knowing that ENSO has an influence on the Europe?
Section 2.1.2: The authors do NOT apply a dynamical downscaling as they correctly say that the only try to reproduce the model physics and not the dynamics so it is awkward to call the method “dynamical downscaling. This is an important point as real dynamical downscaling implies the application of a regional climate model with incudes dynamics. So I recommend that the authors change the wording in the entire manuscript.
L158-59: If I understand this correctly the method conserves the precipitation amount so that I would average over the same area as the coarse grid I would obtain the same precipitation also in the fine grid. If this is correct I do not understand why precipitation is different in Figure 5and e.g. in Fig 4 if we look at the grid point over Scotland. So either the figure is wrong or the description of the methods is incorrect.
L159-164: Well isn’t this logical as the method does not include a dynamical part (only physics is changed) one would expect that it is not able to change the biased large scale atm. circulation.
Section 2.1.3: To my understanding I would not call this a Holocene simulation as important external forcing agents are missing, i.e., solar forcing and volcanic eruptions. It is clear that the authors cannot rerun the simulations using all forcings so I suggest to make it clear that the authors performed a r an orbital-GHG-only simulation for the Holocene period. So name the simulation always “an orbital-GHG-only simulation for the Holocene period.” Just for curiosity why do you only use orbital and GHG forcing and not include the other two?
L245: Why do the authors compare their results to PMIP2 and not to PMIP3 or 4? There are newer studies e.g. Liu et al. 2014, Russo et al. 2022 and PMIP4 studies.
L253-54: please change to “Overall the native grid (T21/11.5_Standard) is still seen on the 11.5K_Down model results in many regions for all times slices. ”
Fig.4: The downscaled data looks weird, e.g. in panel d we see at 50N a clear boundary with a change from +100 mm/yr to -100 mm/yr with no gradient in between. This makes no sense.
3.2.1 is the only subsubsection which makes no sense.
4.1.1 The authors compare their results to proxy data which is good. Still I miss a clear discussion on the seasonality of the proxy data which might play an important role in interpreting the proxy data especially the trends. E.g. tree rings and pollen data are biased to the growing season but these data are compared to yearly means of the simulation. Check out the Bader et al 2020 publication on this.
L442: Brayshaw et al. does not simulate the entire Holocene. He rather simulated time slices distributed during the Holocene. So it is not a transient simulation he performed. Please be more specific about this.
L475 and paragraph: Again only PMIP2 is used, why not using the updates of PMIP3 and 4 ?
4.1.2 This subsection is rather short compared to the 4.1.1 so just merge it to one section 4 Discussion.
L485-87: I think there is a caveat which makes the data not so useful as the authors think as the coarse grid sometimes remains preserved in the downscaled data leading to boundaries (see Fig. 5). I think the authors need to be more cautious about this and not overrate their results.
Reference list contains a lot of errors please correct them.
The quality of the figures is bad please use at least 300 dpi.
-
AC1: 'Reply on RC1', Frank Arthur, 05 Jul 2022
Reply to Reviewer #1
We appreciate the reviewer's thorough and constructive feedback on our article. We have responded to all his/her comments below (in italic font).
Summary
The authors apply a “dynamical” downscaling technique to an orbital-only Holocene-long climate simulation and obtain a resolution of 0.25 degree over Europe. The downscaling is performed for temperature and precipitation. The authors were able to show that the downscaled precipitation matches precipitation and temperatures in mountain regions more realistically. In particular, the simulated trends of the downscaled data resemble reconstructions from different proxy archives.
Reply: We thank the reviewer for a very good summary of what our aim and objective of this paper is about.
General
The study touches a highly important topic in paleo climate, namely the mismatch between local climate proxy information and the rather coarsely resolved paleo climate modelling. The authors present an approach to bridge the gap of scales by a downscaling approach. Clearly the topic desires publication in the Climate of the past, but the current study shows a number of short comings. Besides some structural problems (see below) there is a lack in presenting the state of knowledge in the introduction as a large number of recent studies on dynamical downscaling for paleo climatic studies is missing. Then the results lack a clear discussion of seasonality differences in the proxy data. Also the method itself is to my opinion named wrongly: the authors called it dynamical downscaling and explain that the basic idea is to reproduce the model physics and NOT the dynamics. Due to my rather long list of comments (not sorted into major or minor) I recommend at least major revision of the manuscript.
Reply: We thank the reviewer for pointing to us the gaps that are missing in our study. The reviewer gave some suggestions on recent studies related to dynamical downscaling. In our revised version, we provide a more thorough literature review on climate downscaling for paleoclimate studies. This literature review gives us the opportunity to explain better the originality of our approach and why we initially refer to our downscaling as «dynamical». We used "dynamical downscaling" in our case mostly because the downscaling is performed at each model time step, and it is consistent between the two grids (the precipitation at the coarse resolution interacts with the sub-grid). However, we understand that “dynamical downscaling” can be confusing for our approach since the regional climate model (RCM) community generally uses this terminology to refer to their approach. For this reason, we will change the terminology in the revised version. Following your suggestion, we have also added a discussion on the seasonality of the data.
.
Comments
Title: The authors use only orbital and GHG forcing so I would not call it a Holocene Climate simulation, rather an orbital-GHG-only simulation for the Holocene period.
Reply. We agree that short-term forcings (volcanic and solar) were not taken into consideration. We focus on the long-term multicentennial to millennial trends, so we consider it appropriate to have used forcings that are important at this time scale. Solar and volcanic forcings are known to be of particular importance for annual to decadal timescales. Simulations in the iLOVECLIM model show little impact of volcanic forcing in the Holocene transient simulations, but the impact is shown in the late Holocene (1.5kyr BP) (Bügelmayer et al. 2016). We prefer to keep “Holocene Climate simulation” to inform the reader we cover the entire Holocene.
L 60 and the following paragraphs: There is growing literature on real dynamical downscaling using RCMs on the Paleo perspective so please make a reasonable review on the existing knowledge. Here is a collection of possible publications:
Reply: Reply: We thank the reviewer for providing us with more suggestions about previous studies related to dynamical downscaling using RCMs and we also agree with him/her that our studies need more examples to motivate our work. In particular to highlight the pros and cons of the different approaches. In our revised version, we will include them in the introduction.
L61: There are several approaches to statistical downscaling simulation. An example is Latombe et al. 2018, but there are several more publications so I encourage the authors to make a lit. search to add it to the introduction
Reply: We agree that, similarly to RCM downscaling approaches, statistical downscaling can also provide useful information for paleoclimate studies. We now include such studies in the introduction.
L65: The publication Feser et al. is not dealing with paleo research questions so why is it cited?
Reply: The publication of Feser et al. was about how Regional climate models add value to Global model data. We cited the paper because we were trying to make a point on how RCMs are useful for dynamical downscaling because their output data may be processed to produce higher resolution atmospheric fields, allowing for the modelling of small-scale processes and a more complete description of features (such as mountain ranges, coastal zones, etc).
L80-85: Here is another place to add model publication of real dynamical downscaling.
Reply: Although we do not necessarily agree with the term “real dynamical downscaling” we do agree that indeed RCM studies should be mentioned here.
L95: Units are not in italic
Reply: This will be corrected.
L100 and following: There are several more studies doing Holocene long simulation. A recent one is Bader et al. but please check again the literature. There are also approaches which first use a coarse resolved AO GCM and then a high resolved A GCM forced by the SST and sea ice distributions, please check Merz et al. and Hofer et al. publications.
Bader, J., et al. 2020: Global temperature modes shed light on the Holocene temperature conundrum. Nat. Commun., 11, https://doi.org/10.1038/s41467-020-18478-6.
Hofer, D., et al. 2012: The impact of different glacial boundary conditions on atmospheric dynamics and precipitation in the North Atlantic region, Climate of the Past, 8, 935-949
Hofer, D., et al. , 2012: Simulated winter circulation types in the North Atlantic and European region for preindustrial and glacial conditions, Geophys. Res. Lett., 39, L15805
Merz, et al. 2013: Greenland accumulation and its connection to the large-scale atmospheric circulation in ERA-Interim and paleo-climate simulations, Climate of the Past, 9, 2433-2450
Reply: Thanks for these important publications. They will be part of the revised introduction.
Introduction in general: A discussion on the so-called ‘Holocene temperature conundrum’ is missing. Please check Bader et al 2020 and Liu et al. 2014.
Liu, Z., et al. 2014: The Holocene temperature conundrum. Proc. Natl. Acad. Sci., 111, 3501–3505, https://doi.org/10.1073/pnas.1407229111.
Reply: This is a very good point; We thank the reviewer for providing us very good literature suggestions. We will include the literatures proposed by the reviewer in the revised manuscript.
L125: Please change in the caption to “extent”.
Reply: This will be corrected.
L 129: Why do you go from section 2 to subsubsection 2.1.1? This makes no sense. This is happening several times. There are also sections where there is only one subsection which is again awkward. Please correct the structure according to the rules of the journal
Reply: The structuring of the paper will be done by following the Journal’s rule.
L140: The ECBilt model is a quasi-geostrophic model so the most important mode of variability in the climate system ENSO is not included in the model by definition. So how does this shortcoming impact your results knowing that ENSO has an influence on the Europe?
Reply: The impact of ENSO is expected to be minor in our simulations because we are investigating long (multi-millennial) trends while ENSO’s impact is normally on short timescale.
Section 2.1.2: The authors do NOT apply a dynamical downscaling as they correctly say that the only try to reproduce the model physics and not the dynamics so it is awkward to call the method “dynamical downscaling. This is an important point as real dynamical downscaling implies the application of a regional climate model with incudes dynamics. So I recommend that the authors change the wording in the entire manuscript.
Reply: We used "dynamical downscaling" in our case mostly because of two reasons. First, the downscaling is performed at each model time step during run time. Second, there is a two-way coupling between the coarse grid and the sub-grid which ensures consistency (the precipitation at the coarse resolution is the sum of the sub-grid precipitation). As such, there is a strong difference with standard offline downscaling techniques. We thought that the term "dynamical" is good to convey these two ideas. However, we will change the terminology in the revised version and explain better the originality of our downscaling procedure.
L158-59: If I understand this correctly the method conserves the precipitation amount so that I would average over the same area as the coarse grid I would obtain the same precipitation also in the fine grid. If this is correct I do not understand why precipitation is different in Figure 5and e.g. in Fig 4 if we look at the grid point over Scotland. So either the figure is wrong or the description of the methods is incorrect.
Reply: We do not fully understand this comment because Figure 5 shows the precipitation trend for the whole of Europe whiles figure 4 shows simulated precipitation anomaly in spatial distribution in Europe. The figures 4a and 4d are different because we have two different experiments: (an experiment with downscaling and one without the downscaling). The experiment with downscaling modifies the amount of precipitation with respect to the standard model given that it accounts for the sub grid orography and the non-linear relationship between temperature and humidity.
L159-164: Well isn’t this logical as the method does not include a dynamical part (only physics is changed) one would expect that it is not able to change the biased large scale atm. circulation.
Reply: Yes, we agree with the reviewer here on the fact that we cannot correct the biases related to atmospheric circulation. However, the downscaling changes the whole pattern of precipitation, generally producing higher precipitation over mountain ranges and as a result some drying over the continental plains. In this way it could have corrected some of the biases (or equally worsen them, depending on the bias considered).
Section 2.1.3: To my understanding I would not call this a Holocene simulation as important external forcing agents are missing, i.e., solar forcing and volcanic eruptions. It is clear that the authors cannot rerun the simulations using all forcings so I suggest to make it clear that the authors performed an orbital-GHG-only simulation for the Holocene period. So name the simulation always “an orbital-GHG-only simulation for the Holocene period.” Just for curiosity why do you only use orbital and GHG forcing and not include the other two?
Reply: As replied earlier, we focus on the long-term multicentennial to millennial trends, so we consider it appropriate to have used forcings that are important at this time scale. Solar and volcanic forcings are known to be of particular importance for annual to decadal timescales. Simulations in the iLOVECLIM model show little impact of volcanic forcing in the Holocene transient simulations, but the model responds to volcanic forcings in the late Holocene (1.5 kyr BP) (Bügelmayer et al. 2016).
L245: Why do the authors compare their results to PMIP2 and not to PMIP3 or 4? There are newer studies e.g. Liu et al. 2014, Russo et al. 2022 and PMIP4 studies.
Reply: We will compare our results with new figures of precipitation and temperature to the latest PMIP papers as well in the revised version.
L253-54: please change to “Overall the native grid (T21/11.5_Standard) is still seen on the 11.5K_Down model results in many regions for all times slices. ”
Reply: This will be corrected, thank you.
Fig.4: The downscaled data looks weird, e.g. in panel d we see at 50N a clear boundary with a change from +100 mm/yr to -100 mm/yr with no gradient in between. This makes no sense
Reply: The downscaling still uses the T21 grid, creating some unrealistic gradients. This is a limitation to our methodology. We will discuss this in the revised version.
3.2.1 is the only subsubsection which makes no sense.
Reply: We will correct this.
4.1.1 The authors compare their results to proxy data which is good. Still I miss a clear discussion on the seasonality of the proxy data which might play an important role in interpreting the proxy data especially the trends. E.g. tree rings and pollen data are biased to the growing season but these data are compared to yearly means of the simulation. Check out the Bader et al 2020 publication on this.
Reply: Seasonality is important, we intend to add a figure on seasonality and we will discuss this in our revised manuscript.
L442: Brayshaw et al. does not simulate the entire Holocene. He rather simulated time slices distributed during the Holocene. So it is not a transient simulation he performed. Please be more specific about this.
Reply: We thank the reviewer for this comment. yes, I think we will need to do re-wording here to explain to the reader that the work of Brayshaw et al was not transient simulations.
L475 and paragraph: Again only PMIP2 is used, why not using the updates of PMIP3 and 4 ?
Reply: We will add the updates of PMIP 3 and 4 to our revised manuscript.
4.1.2 This subsection is rather short compared to the 4.1.1 so just merge it to one section 4 Discussion
Reply: We will merge them together in the revised version.
L485-87: I think there is a caveat which makes the data not so useful as the authors think as the coarse grid sometimes remains preserved in the downscaled data leading to boundaries (see Fig. 5). I think the authors need to be more cautious about this and not overrate their results.
Reply: We thank the reviewer for this comment, there is a limitation to the downscaling methodology. Regardless of this limitation, the downscaling data is at least able to show more spatial details which is lacking in the course resolution. Yes, we agree with the reviewer to be cautions in exaggerating our results. We will revise in the next manuscript with the choice of words.
Reference list contains a lot of errors please correct them.
Reply: We will correct this.
The quality of the figures is bad please use at least 300 dpi.
Reply: We will improve the quality of the figures.
-
AC1: 'Reply on RC1', Frank Arthur, 05 Jul 2022
-
RC2: 'Comment on cp-2022-21', Anonymous Referee #2, 18 May 2022
Summary
The authors present a Holocene climate model simulation for Europe at a high-spatial resolution using dynamical (as opposed to statistical) downscaling. This regional model simulation uses inputs from the global iLOVECLIM EMIC. The authors compare this simulation with climate reconstructions and conclude that their higher-resolution model better matches the data than the lower-resolution EMIC. The main innovation is applying this method to a transient simulation that encompasses the whole Holocene and not just time-slices. I am with the authors here, in that I think that scaling down model results to the spatial scale of the proxy data is a potentially good idea, and one well worth investigating.
General comments
The premise of the paper is interesting and is clearly appropriate for publication in Climate of the Past. The main problem is the nature of the evaluation which is qualitative and anecdotal rather than being rigorously quantitative. The authors really need to include a better designed evaluation process where proxy reconstructions are compared with the model results on a site/record basis, or for a region, in the case of gridded reconstructions. Improvements in the evaluation should also be extend to comparisons between the EMIC and downscaled modeling, for instance plotting both in the time-series plots. It would also be useful to compare the iLOVECLIM EMIC that is used with other models (eg PMIP3 GCM’s) to see what particular biases this particular model has over the study region.
The authors also appear to be a bit loose with their commentary. When referring to temperatures and temperature changes it is essential to state what temperature variable it is that they are referring to, for instance whether it is mean annual, winter, summer or some other aspect of temperature. This appears to be a source of confusion throughout, and particularly when citing other studies as supporting evidence. Similarly, in the discussion of proxy data, the authors need to distinguish between studies that provide evidence from single sites, and those that provide evidence from many hundreds of sites, since they are not equivalent. Also, it is important to understand the different studies cited; In Southern Europe, Bartlein et al is essentially a synthesis of the data of Wu et al 2007 and Davis et al 2003, Brewer et al 2007 takes data from Davis et al 2003, and Mauri et al 2014/15 is an improved version of Davis et al 2003 and Brewer et al 2007, both of which it supersedes. All of these latter studies use gridded data, not site data, although the gridding itself is based on site data.
I will also just add here the importance of isostatic uplift over Scandinavia during the Holocene, which in some areas has been substantial (100m+). Data-model comparisons over Scandinavia should be treated with caution where the model uses modern topography but is compared with proxy reconstructions that were much lower in the early Holocene.
Detailed comments
67-69 (and 160-164) This is a critical point.. the regional downscaling cannot correct major errors and biases in the global model simulation, including atmospheric dynamics which have been suggested as the source of much of the data-model discrepancy over Europe during the Holocene (Mauri et al 2014). The authors should note how the iLOVECLIM model generally compares with other global model simulations (eg PMIP3), for instance if it is generally cooler/warmer or wetter/drier than average, or comparable.
77-78 The main ‘sensor’ area of a proxy-based climate reconstruction is rarely greater than ~20 km radius for pollen and can be as small as a couple of hectares for lake-based proxies such as chironomids. It therefore makes sense to undertake data-model comparisons at comparable spatial scales (see my opening comments about improving the data-model comparison).
101-106 This is misleading. While Brewer et al does suggest that climate models can simulate cooler temperatures over Southern Europe and the Mediterranean during the mid-Holocene, this is only in WINTER and the signal is very weak. In contrast, reconstructions of SUMMER temperatures are much cooler than the models, which all show warmer summer temperatures (ie not even the same sign). This is not discussed by Brewer et al, but is clearly shown in the more recent reconstruction by Mauri et al 2014. Both Brewer et al (who uses the data from Davis et al 2003) and Mauri et al use pollen data, but the problem with cooler summer temperatures is also shown in SST reconstructions for the Mediterranean, as shown in Hessler et al 2014 (doi:10.5194/cp-10-2237-2014) figure 4. In fact, the data-model discrepancy shown in Mauri et al 2014 is a very good justification for the authors to have undertaken their study.
119-121 I don’t really understand why the authors have chosen specific areas (and variables, eg precipitation) where they then say they don’t actually have proxy records, if their stated aim is to make comparisons with proxy records. It seems that they did the model analysis first, and then looked for proxy records afterwards.
194-195 The authors use a pre-industrial climate baseline to calculate anomalies to compare with climate reconstructions. I hope that the authors are aware that anomalies shown in almost all proxy-based reconstructions are based on a modern baseline (apart from for instance Davis et al 2003 used in Brewer et al 2007, and Mauri et al 2014/5 that use a pre-industrial baseline of ~1850)
234+ Please be very careful, do not use the unspecified term ‘temperature/s’. Please always state if this is annual, seasonal (JJA, DJF) etc. The authors appear to be conflating winter (Brewer et al 2007) and annual temperatures (Wu et al 2007) in the data, while the temperatures you are referring to in the model results are unspecified.
246 Again this is misleading. Brewer et al only considered winter temperatures. Better to refer to Mauri et al 2014/15 which is a more recent and more comprehensive study that includes summer winter and annual temperatures (and precipitation).
249 Wu et al uses an inverse modelling method, so represents a very different pollen-climate reconstruction to the MAT method used by Davis et al (in Brewer et al) and Mauri et al, although both show the similar results (see Davis 2017 https://doi.org/10.22498/pages.25.3.16). Note also that Wu et al is for individual sites, while Brewer et al use a gridded reconstruction where the site data has been interpolated onto a 1 degree spatial grid. There are also considerably more sites in Brewer et al than in the Wu et al reconstruction, while the sites used in Wu et al are poorly dated, use truncated taxa assemblages (a lot of data is from Huntley and Birks 1983), and have large uncertainties.
258 Fig.3 What aspect of temperature is the figure showing? Mean annual, summer, winter etc ? please specify
264+ Again, as with temperature, please specify what aspect of precipitation you are talking about.. I presume its mean annual (units are in mm/yr), but please state this clearly at the start
280-288 This is interesting, if not surprising. A question arising from this would be if the downscaling simply spatially redistributes the average precipitation of the EMIC grid box, or does it potentially increase/decrease the average precipitation that would occur in the EMIC grid box?
323 Fig. 5. & 371 Fig. 6, Section 4.1.1 Why have the authors chosen not to do a data-model comparison for the precipitation time-series for the Alps, Scandes etc? This would be very straight forward using the data from Mauri et al 2015 which is freely available. Also, why not show the EMIC result for the same spatial areas? This would illustrate the difference between the EMIC and the downscaling (and data)?
399-413 The Furlanetto et al paper consists of only one site from the Alps, the Mauri et al analysis consists of hundreds of sites from the Alps (this has been gridded, but there is also the underlying site data that could be used). Why did you pick only this particular study? In any case, it would be useful to plot the Furlanetto et al precipitation reconstruction against the model result (both high and low resolution) so that the reader can see for themselves. It is also notable that the authors identify the strong spatial variance of the precipitation signal, but compare this single proxy site with the average precipitation of the entire Alps. Surely it makes more sense to compare the proxy record with the nearest point in the model grid?
415-429 Again, as with the Alps, it would be better to directly show the proxy reconstructions plotted against the model result, and even better to show this at the model grid point closest to the site (or interpolated to the site location). If the authors really want to compare using the entire Scandes region, then at least compare against the same area using the Mauri et al 2015 gridded data, since this is designed to avoid spatial sampling bias associated with simply averaging site records together. It is also important to note that (presumably) the model uses modern topography and does not take into account the substantial changes in elevation that has occurred during the Holocene due to isostatic uplift. This is important when comparing with proxy records that actually include this isostatic change (See Mauri et al 2015).
431-435 Refer to Mauri et al 2015, there are quite a few quantitative precipitation records from this region. I am not sure why the authors say otherwise unless they want to exclude the Mauri et al data for some reason. There are also qualitative bog surface wetness records that may show trends eg Anderson 2008 (doi 10.1111/j.1502-3885.1998.tb00880.x)
448-465 The authors are conflating proxy reconstructions here across all kinds of spatial scales, some from individual sites, some based on the synthesis of large numbers of sites to represent individual regions, and some where the site records are projected onto a spatial grid. Again, it would be better if at least some of these records were compared explicitly with the model (ie one plotted over the other) rather than resorting to a rather vague ‘one thing looks like another thing’ statement, which is open to interpretation. This is particularly important because the ability to compare model and proxy record at the scale of the proxy site is supposed to be one of the main advantages of the model downscaling that the authors are proposing. Including the results of the EMIC in the same way would also help demonstrate this.
475 Southern Europe in the mid-Holocene in the PMIP2 simulations is warmer not cooler (only winter is a little cooler in the far east) and with little change in precipitation. This is shown in detail in Mauri et al 2014.
475-478 Please do not write about climate model results as if they are some kind of reality. For instance “we can infer from their work that southern Europe was wetter and cooler.” Should read something like “eg we can infer from their work that southern Europe was wetter and cooler in PMIP2 model simulations.”
475-482 The PMIP2 results encompass a large number of different models, each sometimes showing quite different results. Are you talking about individual PMIP2 models, the ensemble mean or something else? Please be more specific. Also, Braconnot et al 2007 does not show any detail for Southern Europe or the Mediterranean (but is shown in detail in Mauri et al 2014) so I am not sure how the authors are making their comparison unless they have been plotting the data separately (it would be great to actually show this). In any case PMIP2 has been superseded by PMIP3.
485 Grammar needs correcting: “in the change pattern for”
485-495 What about temperature lapse rates? Changes in temperature lapse rates as a result of the downscaling will also lead to change in temperatures at different altitudes, perhaps better reflecting the proxy data.
496+ Conclusions- see my opening comments. The study needs a more rigorous approach to the data-model and model-model comparison.
-
AC2: 'Reply on RC2', Frank Arthur, 05 Jul 2022
Reply to reviewer #2
We appreciate the reviewer's thorough and constructive feedback on our article. We have responded to all his/her comments below (in italic font).
Summary
The authors present a Holocene climate model simulation for Europe at a high-spatial resolution using dynamical (as opposed to statistical) downscaling. This regional model simulation uses inputs from the global iLOVECLIM EMIC. The authors compare this simulation with climate reconstructions and conclude that their higher-resolution model better matches the data than the lower-resolution EMIC. The main innovation is applying this method to a transient simulation that encompasses the whole Holocene and not just time-slices. I am with the authors here, in that I think that scaling down model results to the spatial scale of the proxy data is a potentially good idea, and one well worth investigating.
Reply: We thank the reviewer for summarising the motive behind our studies in a concise and clear understanding to readers.
General comments
The premise of the paper is interesting and is clearly appropriate for publication in Climate of the Past. The main problem is the nature of the evaluation which is qualitative and anecdotal rather than being rigorously quantitative. The authors really need to include a better designed evaluation process where proxy reconstructions are compared with the model results on a site/record basis, or for a region, in the case of gridded reconstructions. Improvements in the evaluation should also be extend to comparisons between the EMIC and downscaled modeling, for instance plotting both in the time-series plots. It would also be useful to compare the iLOVECLIM EMIC that is used with other models (eg PMIP3 GCM’s) to see what particular biases this particular model has over the study region.
The authors also appear to be a bit loose with their commentary. When referring to temperatures and temperature changes it is essential to state what temperature variable it is that they are referring to, for instance whether it is mean annual, winter, summer or some other aspect of temperature. This appears to be a source of confusion throughout, and particularly when citing other studies as supporting evidence. Similarly, in the discussion of proxy data, the authors need to distinguish between studies that provide evidence from single sites, and those that provide evidence from many hundreds of sites, since they are not equivalent. Also, it is important to understand the different studies cited; In Southern Europe, Bartlein et al is essentially a synthesis of the data of Wu et al 2007 and Davis et al 2003, Brewer et al 2007 takes data from Davis et al 2003, and Mauri et al 2014/15 is an improved version of Davis et al 2003 and Brewer et al 2007, both of which it supersedes. All of these latter studies use gridded data, not site data, although the gridding itself is based on site data.
I will also just add here the importance of isostatic uplift over Scandinavia during the Holocene, which in some areas has been substantial (100m+). Data-model comparisons over Scandinavia should be treated with caution where the model uses modern topography but is compared with proxy reconstructions that were much lower in the early Holocene
Reply: We thank the reviewer for this comment. We would like to note that the main intention of our paper is to evaluate our dynamical downscaling result against the standard version of the model which has a much lower spatial resolution. It was not our intention to make a full model-data comparison here, but we agree that it is very valuable to compare our high-resolution results with proxy-based reconstructions for specific mountainous regions where the potential of our downscaling method is especially clear. Still, in the revised version, we will try and do a quantitative evaluation as suggested by the reviewer and look for site-based proxy data in some specific regions and compare to our model results. In the discussion part we compared some GCMs model with our model results, However, we agree with the reviewer here that more GCMs model are needed to make a rigorous evaluation with our work. In the revised version, we will give a more detailed explanation regarding how we use the terms (such as annual and seasonal temperature results to avoid confusing the readers. We think the reviewer’s comment regarding isostatic uplift is very important and we will put that into consideration when preparing our revised version of the manuscript.
Detailed comments
67-69 (and 160-164) This is a critical point.. the regional downscaling cannot correct major errors and biases in the global model simulation, including atmospheric dynamics which have been suggested as the source of much of the data-model discrepancy over Europe during the Holocene (Mauri et al 2014). The authors should note how the iLOVECLIM model generally compares with other global model simulations (eg PMIP3), for instance if it is generally cooler/warmer or wetter/drier than average, or comparable.
Reply: This is well noted, and we will discuss this in the revised version.
77-78 The main ‘sensor’ area of a proxy-based climate reconstruction is rarely greater than ~20 km radius for pollen and can be as small as a couple of hectares for lake-based proxies such as chironomids. It therefore makes sense to undertake data-model comparisons at comparable spatial scales (see my opening comments about improving the data-model comparison.
Reply: Yes, a very important point here, data-model comparison is hindered by the different characteristics of each dataset at small spatial scales, model output is less reliable. To avoid unbiased comparison between our model results and the proxy data, it is important we consider the spatial scales. We will incorporate this in our revised version of the manuscript.
101-106 This is misleading. While Brewer et al does suggest that climate models can simulate cooler temperatures over Southern Europe and the Mediterranean during the mid-Holocene, this is only in WINTER and the signal is very weak. In contrast, reconstructions of SUMMER temperatures are much cooler than the models, which all show warmer summer temperatures (ie not even the same sign). This is not discussed by Brewer et al, but is clearly shown in the more recent reconstruction by Mauri et al 2014. Both Brewer et al (who uses the data from Davis et al 2003) and Mauri et al use pollen data, but the problem with cooler summer temperatures is also shown in SST reconstructions for the Mediterranean, as shown in Hessler et al 2014 (doi:10.5194/cp-10-2237-2014) figure 4. In fact, the data-model discrepancy shown in Mauri et al 2014 is a very good justification for the authors to have undertaken their study
Reply: Thanks for the comment, we will rephrase this paragraph and be more specific in the revised version of the manuscript.
119-121 I don’t really understand why the authors have chosen specific areas (and variables, eg precipitation) where they then say they don’t actually have proxy records, if their stated aim is to make comparisons with proxy records. It seems that they did the model analysis first, and then looked for proxy records afterwards.
Reply: With this study we provide the first transient simulations of the last 12k with interactive downscaling at 0.25° resolution over Europe. For this reason, we first wanted to assess the major changes induced by the downscaling with respect to the standard non-downscaled version of the model. The downscaling has a major impact over mountainous Region, so we focus our analysis there.
194-195 The authors use a pre-industrial climate baseline to calculate anomalies to compare with climate reconstructions. I hope that the authors are aware that anomalies shown in almost all proxy-based reconstructions are based on a modern baseline (apart from for instance Davis et al 2003 used in Brewer et al 2007, and Mauri et al 2014/5 that use a pre-industrial baseline of ~1850)
Reply: Yes, we are aware of this, but we thank the reviewer for bringing our attention to this. We will discuss this in our revised manuscript.
234+ Please be very careful, do not use the unspecified term ‘temperature/s’. Please always state if this is annual, seasonal (JJA, DJF) etc. The authors appear to be conflating winter (Brewer et al 2007) and annual temperatures (Wu et al 2007) in the data, while the temperatures you are referring to in the model results are unspecified.
Reply: This will be resolved in the revised version.
246 Again this is misleading. Brewer et al only considered winter temperatures. Better to refer to Mauri et al 2014/15 which is a more recent and more comprehensive study that includes summer winter and annual temperatures (and precipitation).
Reply: We thank the reviewer for this comment. This will be done in the revised version,
249 Wu et al uses an inverse modelling method, so represents a very different pollen-climate reconstruction to the MAT method used by Davis et al (in Brewer et al) and Mauri et al, although both show the similar results (see Davis 2017 https://doi.org/10.22498/pages.25.3.16). Note also that Wu et al is for individual sites, while Brewer et al use a gridded reconstruction where the site data has been interpolated onto a 1 degree spatial grid. There are also considerably more sites in Brewer et al than in the Wu et al reconstruction, while the sites used in Wu et al are poorly dated, use truncated taxa assemblages (a lot of data is from Huntley and Birks 1983), and have large uncertainties
Reply: Thank you for this informative comment. As we have replied earlier on, we will look for more proxy data which uses a lot of sites and good spatial scale for better representation.
258 Fig.3 What aspect of temperature is the figure showing? Mean annual, summer, winter etc ? please specify
Reply: This represents annual mean temperature anomaly. It will be corrected in the revised version.
264+ Again, as with temperature, please specify what aspect of precipitation you are talking about.. I presume its mean annual (units are in mm/yr), but please state this clearly at the start
Reply: Thanks for the comment, this is annual mean precipitation, so this will also be clarified in the revised version.
280-288 This is interesting, if not surprising. A question arising from this would be if the downscaling simply spatially redistributes the average precipitation of the EMIC grid box, or does it potentially increase/decrease the average precipitation that would occur in the EMIC grid box?
Reply: The downscaling does not simply redistribute the precipitation, as can be seen by a comparison for three mountainous regions in Figure 6. This clearly shows that with downscaling the precipitation is higher.
323 Fig. 5. & 371 Fig. 6, Section 4.1.1 Why have the authors chosen not to do a data-model comparison for the precipitation time-series for the Alps, Scandes etc? This would be very straight forward using the data from Mauri et al 2015 which is freely available. Also, why not show the EMIC result for the same spatial areas? This would illustrate the difference between the EMIC and the downscaling (and data)?
Reply: This will be done in the revised version.
399-413 The Furlanetto et al paper consists of only one site from the Alps, the Mauri et al analysis consists of hundreds of sites from the Alps (this has been gridded, but there is also the underlying site data that could be used). Why did you pick only this particular study? In any case, it would be useful to plot the Furlanetto et al precipitation reconstruction against the model result (both high and low resolution) so that the reader can see for themselves. It is also notable that the authors identify the strong spatial variance of the precipitation signal, but compare this single proxy site with the average precipitation of the entire Alps. Surely it makes more sense to compare the proxy record with the nearest point in the model grid?
Reply. Our downscaled results provide precipitation results at a relatively high spatial resolution. To evaluate this kind of model result, especially when evaluating against a low resolution model version, it is of crucial importance to use proxy-based records that capture this same high spatial resolution. This is the reason why we use the Furlanetto et al. paper. It is not easy to find other proxy data with such high resolution. The Mauri et al. dataset is very valuable for model evaluation, but it presents gridded precipitation at 1° resolution, which in our view does not represent a high enough resolution to represent specific conditions in the Alps or the other mountainous regions we focus on.
415-429 Again, as with the Alps, it would be better to directly show the proxy reconstructions plotted against the model result, and even better to show this at the model grid point closest to the site (or interpolated to the site location). If the authors really want to compare using the entire Scandes region, then at least compare against the same area using the Mauri et al 2015 gridded data, since this is designed to avoid spatial sampling bias associated with simply averaging site records together. It is also important to note that (presumably) the model uses modern topography and does not take into account the substantial changes in elevation that has occurred during the Holocene due to isostatic uplift. This is important when comparing with proxy records that actually include this isostatic change (See Mauri et al 2015)
Reply: Yes, a very important comment and we will work on all the constructive suggestions to improve our paper.
448-465 The authors are conflating proxy reconstructions here across all kinds of spatial scales, some from individual sites, some based on the synthesis of large numbers of sites to represent individual regions, and some where the site records are projected onto a spatial grid. Again, it would be better if at least some of these records were compared explicitly with the model (ie one plotted over the other) rather than resorting to a rather vague ‘one thing looks like another thing’ statement, which is open to interpretation. This is particularly important because the ability to compare model and proxy record at the scale of the proxy site is supposed to be one of the main advantages of the model downscaling that the authors are proposing. Including the results of the EMIC in the same way would also help demonstrate this
Reply: We agree with the reviewer that the ability to compare our model results and data at the scale of the proxy site should be treated as a high importance. We will do so in the revised version.
475 Southern Europe in the mid-Holocene in the PMIP2 simulations is warmer not cooler (only winter is a little cooler in the far east) and with little change in precipitation. This is shown in detail in Mauri et al 2014.
Reply: We will clarify this paragraph in the revised manuscript.
475-478 Please do not write about climate model results as if they are some kind of reality. For instance “we can infer from their work that southern Europe was wetter and cooler.” Should read something like “eg we can infer from their work that southern Europe was wetter and cooler in PMIP2 model simulations.”
Reply: Thanks for the suggestion, this will be corrected in the next version.
475-482 The PMIP2 results encompass a large number of different models, each sometimes showing quite different results. Are you talking about individual PMIP2 models, the ensemble mean or something else? Please be more specific. Also, Braconnot et al 2007 does not show any detail for Southern Europe or the Mediterranean (but is shown in detail in Mauri et al 2014) so I am not sure how the authors are making their comparison unless they have been plotting the data separately (it would be great to actually show this). In any case PMIP2 has been superseded by PMIP3.
Reply: We agree with the reviewer here and this will be taken into consideration.
485 Grammar needs correcting: “in the change pattern for”
Reply: We will revise as suggested.
485-495 What about temperature lapse rates? Changes in temperature lapse rates as a result of the downscaling will also lead to change in temperatures at different altitudes, perhaps better reflecting the proxy data.
Reply: The vertical lapse rate in temperature is computed in our model and shows a representative of the free-atmosphere temperature variations. So, we agree with the reviewer here that the lapse rate had an effect with temperature changes with elevation and could better represent the poxy data.
496+ Conclusions- see my opening comments. The study needs a more rigorous approach to the data-model and model-model comparison.
Reply: We thank the reviewer for taking time to make suggestions on ways to get an improved version of our manuscript. We will work on his/her major suggestions which is more based on the evaluation of our results with proxy data and other model data.
-
AC2: 'Reply on RC2', Frank Arthur, 05 Jul 2022
Frank Arthur et al.
Data sets
Simulations of the Holocene Climate in Europe Using Dynamical Downscaling within the iLOVECLIM model (version 1.1) Frank Arthur, Didier M. Roche, Ralph Fyfe , Aurélien Quiquet , Hans Renssen https://doi.org/10.23642/usn.19354082
Frank Arthur et al.
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 421 | 179 | 15 | 615 | 9 | 8 |
- HTML: 421
- PDF: 179
- XML: 15
- Total: 615
- BibTeX: 9
- EndNote: 8
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1