You may read the attached "response to reviewer 1".  It is more clear.

Review of gmd-2020-329 by Xue et al.

"Impact of Initialized Land Surface Temperature and Snowpack on Subseasonal to Seasonal Prediction Project, Phase I (LS4P-I): Organization and Experimental design"

This paper introduces the LS4P project including its motivation, goals, and instructions on the modelling experiments. The key idea is to study the effect of soil temperatures in mountainous areas on subsequent precipitation in downstream areas through remote effects of land-atmosphere interactions. Models and observations hint that such effects could be existing, and could consequently be exploited for weather/climate forecasting. An accurate representation of these effects across models is challenging, and the project is aiming to improve this.

Recommendation:

I think the paper requires major revisions.

This paper comprehensively presents the LS4P project which brings together the land- climate modelling community. The key idea about exploiting remote effects of land- atmosphere interactions for weather predictability is promising, and fits well with other recent studies illustrating so-far largely overlooked effects of land surface status and fluxes in downstream areas. Also considering soil temperature in this context is innovative.     as it also reflects to some extent the moisture/energy state of the land surface, and quite some satellite and ground-based data are available which are partly insufficiently exploited, particularly in comparison with the more prominent soil moisture. Nevertheless, I also see some shortcomings in this paper which should to be addressed to make the paper suitable for publication in Geoscientific Model Development:

Response:  The reviewer has very carefully reviewed the manuscript, and provided very insightful, constructive, critical, and encouraging comments and detailed suggestions.  The manuscript has been revised based on the comments/suggestions.  We sincerely appreciate the reviewer’s efforts and have acknowledged the reviewer and editor in the revised manuscripts.

(1)

I have some doubts about the application of the mask as described in section 3.2.   While equation 1a is clear, I do not understand why equation 1b is needed;

in this context also lines 403-408 and Figure 2 are unclear to me.

More generally, I think that the models' memory is a dynamic feature which should

be addressed through adapting the modelled (soil temperature and moisture) *dynamics* rather than their *states*.

This way, I feel if the tuning parameter "n" could exaggerate initial corrections which could degrade the simulations in the early forecasting period

Why not simply correcting the model/forecast biases through post-processing and without interferring with the actual model simulations?

Finally, as the mask correction requires observed soil temperature information, maybe I missed that but I was wondering how this is done in places where this is not available?

Response: The reviewer raises several issues for clarifications of our soil temperature initialization methodology.  The following is our responses. 

1).  The LS4P project pursues a new approach, i.e., using the LST/SUBT anomaly in high mountain areas, to improve the S2S prediction.  The current start-of-the-art models, however, are unable to properly produce the observed anomalies, and then by extension, this anomaly-induced dynamic and the associated physical processes, in their simulations.   As such processes are not existed in the model simulation; the bias correction in post-processing is unable to generate these processes.  In fact, the LS4P deals with the S2S prediction, which is essentially the same as weather forecasting.   A bias correction in the post-process is normally not employed.

2).  It is a good idea to improve the model dynamic and physical processes to overcome the modeling problem.  However, improving Earth system model/land model and reanalyses data (which problem is shown in Figure 4 of this paper) in order to overcome the deficiency in modeling high mountain land surface temperature anomaly is not a simple task and may take decades of effort (today’s land temperature model development has more than 70-year history), but proper S2S prediction, including drought/flood prediction, is an urgent World Meteorological Organization (WMO) task with societal implications.  On line 377 of the revised manuscript, we pointed out that preliminary research suggests that “prescribing both LST and SUBT initial anomalies based on the observed T-2m anomaly and model bias is the only way for the current ESMs to reproduce the observed May T-2m anomalies”.  This is current status in the LS4P modeling groups.  Of course, we welcome any research group to find a new way to produce observed anomaly, but it is not a simple task.  On other hand, using initialization to improve meteorological prediction is nothing new but a traditional meteorological approach.  We believe when this issue gets more attention and more data from measurement are available, more methodologies may be developed.  But the scientific development always takes time.  We have to undertake any development by making step-by-step improvements. 

3).  The reviewer has raised question about the schematic diagram (Figure2), which is designed to help other modeling groups to reproduce the mask that we use to test our approach.  We acknowledge that the presentations of the Figure 2, its caption, and relevant text were not comprehensive and had shortcomings in helping readers to understand our approach.  We have had several iterations among our co-authors and revised the figure, the caption, and the related presentation in the text to help readers better understand our approach as well as the procedure to produce the mask for initialization.  In particular, we have clarified in the revised Figure 2 which initial temperature is for Task 1 and which one is for Task 3 with more details and explanations in figure note.  We also reorganize Section 3.2(3).  We believe these should help readers to understand the overall ideas in this figure and in the LS4P-I experimental design.

Meanwhile,  the original Figure 2 includes both warm and cold years.  In the revised version, to make the thing simple and less confusion for readers who are unfamiliar with LS4P, we only include cold year (same as the case in the LS4P Phase I) in the text and move the schematic diagram for the warm case to appendix for readers who want to pursue this issue further for their own research. 

4).  Regarding the observational data, under the current big data system, it should be available very soon.   The general public can obtain last month’s observational data.  In a major climate center, the real time analysis data are generated in a very timely manner.  In terms of the model bias information, if it is unavailable in some cases, then the model climatology bias can be applied.  Our paper (Section 5 and Figure 7) indicates that the bias is very consistent if comparing climatology and year 2003.  The methodology does not require a precise bias value.   But the general signs (positive or negative) are important.

5).  The reviewer thought we may impose artificial large forcing because of a tuning parameter “n”.   It is not true.  In the following paper in a Climate Dynamics special issue, we will show every model’s-imposed forcing.   They are not that extreme.  The LS4P-I includes most of major climate centers in the world.  If our approach is totally different from their normal practices, they would not endorse the LS4P-I and participate in this project. By the way, many parameterizations, such as convective schemes, aerosol parameterizations, etc. also have the tuning parameters.         

(2)

The description and motivation for selection of the ground truth data is insufficient in my opinion. The text in lines 249-273 lists many datasets but does not indicate

how they are applied. A summary table of all employed datasets would be nice, including their spatial and temporal extent, advantages, and variables provided.

I understand that you are using the CMA dataset as this is based on a relatively large number of ground measurements. This makes sense, but it would be interesting how (well)

this dataset extrapolates between these measurements and how many (fewer?) measurements

are employed by other datasets such as the state-of-the-art ERA5 reanalysis. Next to this, I was wondering why you are not employing available satellite-based land surface temperature products http://data.globtemperature.info/ ?

Response: The data sets described in this section are produced by the LS4P-I data group, and we intend to provide this information for scientists who are interesting in conducting further LS4P-I researches.  These measurements are the foundation for the LS4P-I research.  For instance, in the reviews of earlier papers/proposals of the LST/SUBT effect, some reviewers just use “there are no large scale SUBT measurements in high mountain areas to confirm the exist/presence of such anomaly” to suggest for rejection.  Now the LS4P-I data groups have provided comprehensive data sets to support the community for the research.  That’s why we choose the Tibetan Plateau as the focus area for the LS4P Phase I and would like to provide relevant information for the community to introduce these data sets.  In the revised manuscript, per the reviewer’s comments, the role of the data group is clearly presented (lines 186-187; 251-254). We have also added a table in the appendix (Table B1) listing all the data that we used.

As to the station number in China, for other data sets, such as ECMWF data set, they normally obtain the data from GTS system and through collaborative agreements.  Normally, they have less than 400 stations, much less than what we have listed in the text.

Remote sensing over the Tibetan Plateau is challenging because of a lack of validation data (for testing the algorithms).  Since the Tibetan Plateau is one of the focus areas of GLASS satellite products, the GLASS group has more experience on this area’s remote sensing with better quality there. In addition, the GLASS group is participating in the LS4P-I project.  That’s why we mainly use the GLASS products in the project.   But we certainly do not exclude other satellite products and will use them if they can provide useful information for the project.   

Thank you for sharing the link (http://data.globtemperature.info/) for surface temperature data. Unfortunately, all these datasets have a short coverage period and do not cover the climatology period (1981-2010 and 1980-2013), which we have considered in our study.

(3)

The manuscript is comprehensive but also quite long. To make it more concise it would be helpful to shorten where possible I think. Below, I have indicated some examples where content

is repeated and where I would see potential to shorten the text.

Moreover, a summary table listing the main information regarding tasks 1-5 in section 3.2 would improve the readability.

Response: Thanks for the reviewer’s suggestion.  We will make the revision according to every suggestion that you list in specific comments (see our response in Specific Comments below for detail).  In the revised manuscript we have added a table (Table 1) to clearly provide information’s regarding Tasks 1-4 as suggested.

(4)

The authors refer a lot to snow effects in sections 1 and 2, and I like these ideas. However, snow is not mentioned at all in sections 3, 4 and 5, and apparently only implicitly (through LST) part of the analyses.

This should be clarified, and the role of snow in the project as described in sections 1 and 2 should be toned down.

Response:  The snow part of the work is an important component in LS4P. Snow is one of the major driver to produce LST/SUBT anomalies.  That’s why it is included in the project’s name.  In Phase 1, however, we are mainly looking for first order effects most related to the soil surface and deeper layers; but indeed, we mentioned snow effects as we (agree with this reviewer) think it is important and related to the LST/SUBT anomaly, and plan to examine it in the Phase II.  Since this is the main paper introducing the LS4P, we have to present the importance of snow in this project in section 1.  Otherwise, the readers may immediately raise issue about where the LST/SUBT anomalies come from. 

The shortcoming in our previous presentation was not to provide a clear expectation how much snow related activity will be discussed in this paper, and make some readers keep waiting till the end of the paper.  We now make a clarification at the end of Section II that this part of research will be considered in Phase II papers and will not be presented further.  So readers (who are interest in this) will have a proper expectation. 

(5)

As with the snow, also predictability is prominently mentioned in sections 1 and 2 and even in the abstract and title, but the detailed description of the project and the simulations

does not refer at all to this. So also here I would suggest to either include details

on how the predictability could/will be assessed, or tone it down in the beginning of the manuscript.

Response:  Please see our response above.

I do not wish to remain anonymous - René Orth.

Response:  Thanks.  We have acknowledged the reviewer in the revised manuscript.

Specific comments:

line 29 & 179: "Data groups" is not clear.

Response: Sorry, we failed to clearly provide the relevant information. In the response to your main comment 2, we had a discussion on this issue.  In the revised paper, we have clarified their contribution to the LS4P on lines 186-187; 251-254. 

line 35: Summer precipitation in which area?

Response: We have added “beyond East Asia” on line 35

line 49: "stubbornly low", not everywhere, there are quite some regional variations of  precipitation forecast skill

Response: The S2S prediction is associated with drought/flood/heatwave prediction. In N. America, Europe, West Africa, and East and South Asia, the Earth System Models have difficult to reproduce these events.  That’s why the WCRP and the WRRP/WMO list S2S as the current high priority.   In some areas, such as California, the weather and climate have low variations in some seasons.  But this is not the focus of our S2S prediction research.      

lines 60-80: in this context you could cite Orth and Seneviratne (2017) where we compare the impacts of SST vs soil moisture

for land climate globally

Response:  Added and thanks for the suggestion

lines 83-85, and elsewhere: I am missing some justification why you chose to focus on (usually data sparse) mountain areas, and why their downstream impacts are expected to be higher than that of flat regions with possibly larger surface heat fluxes.

Response: Lines 85-111 (the revised version) in the introduction and the 1^st paragraph in Section 2 discuss the justification to use high mountain areas as focus for S2S prediction.  The LST’s effect in high mountains have been long overlooked until scientists in the LS4P groups conducted preliminary researched discovering its important role in S2S predictability.  The preliminary sensitivity experiments and relevant papers presented in this part show the effect of spring LST/SUBT in high mountain areas on downstream summer precipitation.  The papers cited here also discussed the mechanisms why the impacts are in downstream.  Because of the high elevation, the perturbations induced from the land surface process could propagate for long distances through Rossby wave interactions affecting the downstream areas.  We have modified the paper in the introduction to help the reader to better relate the discussions there to the justification as to why we choose a high mountain area as the focus.

Yes, the mountain area measurements are overlooked because of the difficulty making measurements in the environmental condition there and because of failing to understand the importance of the measurement in these areas. We hope our research will stimulate more measurements in these areas.

line 86: "very close", you mean strongly related here?

Response: On line 89 of the revised version, the text has been modified to clarify the meaning here.

line 89 & Figure 1 caption: I do not fully understand how this was computed. Do you select years with warm/cold Mays, respectively, and then you track the temperature difference between these years through all months?

Response:  Yes, we use the approach as you mentioned.  This approach has been pointed out in the revised note after the caption of figure 1.  We further emphasize it on line 95 of the revision

line 100: SSiB, abbreviation not explained

Response: Added

line 122: insert "the" before "snow darkening effect"

Response: done

lines 127/128: could you please specify/explain "large diversity" a bit more? 

Response: we have modified the sentence and in the revised version of the manuscript it reads as “this could be one of the major reasons for the large discrepancies in simulated T-2m and its anomaly in current Earth System Models (ESMs)”.  

Line 146: insert "the initiative" after "historical development of"

added

lines 178/179: there is no need to put both written-out and numeric numbers there

Response: agree. We have eliminated the written-out numbers.

lines 195/196: I think this is an important part of the project, can you give some more details on this data base?

Response. We have added relevant information for the data base on line 203 of the revised version.

lines 210/211: Sorry for mentioning an own study again, but Orth and Seneviratne (2017) can be instructive here I think

 Here just summarize questions, no any citation listed here.  The relevant citations are in Introduction and the reviewer’s paper has been added in citation there.

lines 221-225: repetition, could be removed line 246: insert "the" before "Tibetan Plateau"

done

line 248: please be more specific which of these datasets are useful for the project, and why

The data used for this project is clarified on lines 249-251 with a table in Appendix B in the revised version.  This section mainly presents data sets produced by the LS4P data groups.  Only some of these data are used by the LS4P phase I experiment.  Most data introduced here can be used for the LS4P related research, such as the causes of LST/SUBT anomalies, Tibetan Plateau land surface characteristics (for instance, snow, frozen soil) associated with land memory and land surface energy and water balances, and land/atmosphere interaction there.  The multi-model testing in the LS4P project can only investigate some key issues and intends to stimulate more research on this aspect from the community, which is needed to ultimately understand this issue.  This section provides the relevant data information for the community, which is very useful for them to conduct high mountain-related researches. We have clarified this in the revised section 3.1 of the manuscript.    

lines 267-269: could be removed

 As indicated earlier, during the early LST/SUBT research, this was the precise reason some reviewers reject our approach.  We make the statement here is to show although some of the data do not directly use in LS4P phase I experiment, but they do provide basic information/evidence to support the crucial justification for the LS4P activity.

lines 282-283: you mean anomaly precipitation rates of +1.32 mm/day?

Response: Yes. Anomaly is added.

line 294: "around late April", why is this not more specific?

Response: The LS4P requests for at least 6 members for each Task, which normally means 6 different starting dates (for initial conditions), such as April 25, April 26, April 27, April 28, April 29, and April 30.  Every modeling group may select different days based on their normal practice for numerical prediction (for instance, some days from April 26-May 1, and some from April 28-May 3).  That’s why we use “around late April” here.

line 316: "the models' performances are then checked" can be removed

done 

line 321: what is a "proper" lapse rate?

Response:  This issue is discussed in Xue et al. (1996) and Gao et al. (2017).  We modify the sentence to make it clear.

lines 323-327: repetition, can be removed

Okay.

lines 334-335: Not sure if the project is obsolete if models would do a good job, as you could still study LST/SUBT downstream effects, as stated in the project goals in e.g. the  abstract

Response.  The downstream effect is a new scientific discovery for the S2S predictability regardless whether the models are able to produce the LST anomalies on the mountain areas.   However, if the models were able to produce proper temperature anomaly over the Tibetan Plateau, the focus would not be how to generate the observed LST anomaly.  We have modified the statement on line 344 and add lines 624-625 to make it more adequate and precise.

line 349: "around 2010", why is this not more specific?

Response: Every climate modeling center has made long term climatological runs but with slightly different starting and ending years.  We are unable to request these big centers to redo their climatological simulation because of the huge amount of work load and computer time, so as a best approximation we only ask them to send their climatology which is an average around certain time period in order to be compatible with our experiment. We believe a climate based on the average 1980-2010 and another from 1979 to 2009 should have no fundamental difference.   

line 426: what are the "sensitivity" and "control" runs?

Response:  The sensitivity is Task 3 run and control is Task 1 run.  We have clarified this.

line 450: here you could point to Table S1

Done.  Thanks.

lines 478-479: ERA-Intermin should be updated to ERA5 (https://climate.copernicus.eu/climate-reanalysis)

 We did not update due to the time period that ERA5 covers.  The GMD also does not allow to put website address in the paper, so we still use the previous citation.

lines 473-475: I get the point that you would use the dataset that uses most ground station measurements as reference, but I am wondering how many stations ERA5 is using over this area?

Response:  I do not know exactly how many stations ERA5 has.  But based on my information (pers. Comm.) it should not be more than 400 stations in China, which is much less than what we use.

lines 590-591: While it becomes more clear later, it would be good to already motivate here why you are comparing biases with anomalies.

Response.  This is a good suggestion.  We have added lines 624-625 to indicate why we want to compare bias and anomalies.

line 594 & Figure 6: I do not understand why you do this multiplication with -1.

Response:  Results in Figure 6 provide the underpinnings for the LS4P conjecture: if the May land temperature anomaly on the Tibetan Plateau does contribute to the June East Asian precipitation anomaly, then improving the May land surface temperature simulation over the Tibetan Plateau through an improved initialization should make Earth System models to produce better June East Asian precipitation.

For instance, May 2003 was a cold month for the Tibetan Plateau and June 2003 was dry to the south of the Yangtze River in the observations.  If we postulate the May cold T2m in the Tibetan Plateau caused the drought, then a model with cold (warm) bias in May in the Tibetan Plateau should produce a dry (wet) bias to the south of the Yangtze river if these Earth System models’ dynamics and physics reflect such linkage that in the real world. 

Because some models have a cold and dry bias and some have a warm and wet bias, when we make the composite in Figure 6, we have to multiply “-1” for the models with warm/wet bias to integrate them with the models with cold and dry bias (to avoid their biases cancel each other), and to compare them with observed anomalies.   

line 605: How is this correlation computed? Is it a spatial correlation? Over which domain?

Response: it is a spatial correlation over the figure domain, which now is specified in the revised version (lines 638-639).

line 615: Please specify the area over which the subsequent drought occurred.

It is clarified on lines 647-649

line 637: Why these methods, and not a similar approach as for example in Koster et al. (2016)

Response:  Here we deal with the prediction.  Smith et al.’s work on this line of the original manuscript is a statistical prediction.  Koster et al.’s work (2016) conducted series of stationary wave model (SWM) experiments in which the boreal summer atmosphere is forced, over a number of locations in the continental United States, with an idealized diabatic heating anomaly.   As such, Koster’s work is an ideal sensitivity study and is different from what we pursue (prediction study). 

lines 643-647: Why and how is the fore-restore method causing inaccurate soil memory?

Response:  We have two published studies (Liu et al., 2020; Li et al., 2021) addressing how the force restore method causes the problem and how to improve the memory through the proper parameterizations.  Since a comprehensive discussion on these issues are out of the scope for this paper, in the revised text we added these two citations on line 680. 

Liu Y., Y. Xue, Q. Li, D. Lettenmaier, and P. Zhao, 2020: Investigation of the variability of near-surface temperature anomaly and its causes over the Tibetan Plateau.  J. Geophy. Res. 125, e2020JD032800. https://doi.org/10.1029/2020JD032800.

Li, Q., Xue, Y., and Liu, Y.: Impact of frozen soil processes on soil thermal characteristics at seasonal to decadal scales over the Tibetan Plateau and North China, Hydrol. Earth Syst. Sci., 25, 2089–2107, https://doi.org/10.5194/hess-25-2089-2021, 2021.

line 666: Correct tense, 2020 is in the past now :)

Corrected

line 669-670: "A possible ... will also be prepared" can be remove

Done

 Figure 5:

- the quality/resolution is very low, please improve

Done

References:

Orth, R. and S.I. Seneviratne

Variability of Soil Moisture and Sea Surface Temperatures Similarly Important for Warm- Season Land Climate in the Community Earth System Model

Clim. Dyn. 30(6), 2141–2162 (2017).

Koster, R., et al.

Impacts of Local Soil Moisture Anomalies on the Atmospheric Circulation and on Remote Surface Meteorological Fields during Boreal Summer: A Comprehensive Analysis over North America

1. Climate 29(20), 7345-7364 (2016).

You may use the attached response to reviewer 2.  It is more clear.

Comment on gmd-2020-329

Anonymous Referee #2

Referee comment on "Impact of Initialized Land Surface Temperature and Snowpack on Subseasonal to Seasonal Prediction Project, Phase I (LS4P-I): Organization and Experimental design" by Yongkang Xue et al., Geosci. Model Dev. Discuss., https://doi.org/10.5194/gmd-2020-329-RC2, 2021

This paper describes the international / multi-organization LS4P project and provides some initial analyses of data submissions. It is essentially an “introduce the community to the project” paper, with most of the scientific findings to be documented later, as the project progresses.

While the paper is well written, in my opinion it has enough issues to rate a review of

“major revision”. On the one hand, I do applaud the experiment organizers for bringing together such a wide-ranging group of participants to address such an important problem. With this diverse a group and a corresponding collection of model outputs from such a diverse set of models, I don’t doubt that the project will bear useful scientific fruit, and a paper like this that introduces the project to the broader community is certainly of value. On the other hand, the paper’s write-up glosses over several critical aspects of seasonal and subseasonal prediction that call some of the project’s long-term strategies into question, at least in terms of how they’re currently described. A revised version should address these shortcomings through substantial qualification (not just a sentence or two here and there) or, better yet, through a substantial rethinking of the approaches to be applied.

Response:  We thank for the reviewer’s acknowledgment that we have a wide-ranging group of participants to address an important problem and that the project will bear scientific fruit. We have taken into careful considerations of the reviewer’s comments/concerns on our writing and questioning of our strategies.  The following is our point-by-point responses to address the issues that the reviewer raises.  We believe some issues are rooted in the reviewers understanding in our experimental design, thus in our revisions, we have made a great deal of effort to improve the presentation of our experimental design section.  The following responses explain and address the reviewer’s questions.

1. The underlying assumption of the project appears to be that if a model does not produce an accurate temperature over the Tibetan Plateau, the fault lies with the land model (e.g., in how long the land model maintains an initial condition). The paper states this explicitly on line 465. The truth is, all models have biases in both air temperature and precipitation across the globe, and these biases could have any number of sources. A temperature bias over the Tibetan Plateau might have nothing to do with land model It might instead result from deficiencies in the reproduction of the general circulation, for example, or from some problems with the radiation balance. Forcing the model to have a low temperature bias by imposing a stronger initial temperature anomaly (perhaps even an unrealistic anomaly, through eq. 1) may amount to “getting the right answer for the wrong reason”, which is not a good basis for a forecast experiment. It’s quite possible that forcing a correct temperature through such an initialization when the model wants to do something else for reasons unrelated to land processes might have unexpected negative consequences – especially if the model is artificially modified in one region and not in surrounding regions. Substantial discussion regarding this is needed.

Response: we fully agree with the reviewer’s comments that “all models have biases in both air temperature and precipitation across the globe, and these biases could have any number of sources,” and when you correct the bias you may be “getting the right answer for the wrong reason”.

But we have to clarify a few issues:

(1). Numerous modeling studies, since the very beginning of meteorological model development, have worked on correcting model-produced precipitation and temperature errors by improving some model parameterizations and through improving initial and boundary conditions.  Our approaches and statements are based on a number of published papers in our field’s major journals.  We cannot simply speculate these peer-reviewed research’s results are produced due to the wrong reason unless there is an evidence to support such statement.  Otherwise, we may eliminate any modeling improvement studies because, at least at its early stage, they cannot prove that improvement is not due to a wrong reason.  Only the community collective efforts with long term exercise can prove it. 

(2).  The reviewer speculates the problem in simulating the LST anomaly may not be due to the surface models.  In general, when we try to correct a model deficiency in one variable, it is normal to check the dynamic and physical processes relevant to this variable first.  For instance, for the precipitation simulation errors, we naturally check the convective and cloud process modeling and surface evaporation parametrization first.  Unless there is an evidence to show these are other reasons, we cannot claim it is a wrong approach.  Our statement on land model deficiencies are based on published papers and analyses from the LS4P research.  On lines 126-129, we present a publication (Liu et al., 2020), which focuses on exploring the causes of model deficiency in properly producing the observed surface temperature anomaly in high mountains.  That study demonstrated this deficiency IS associated with the land surface process model, including snow/albedo and soil subsurface memory effects.  Recently, we had another published paper (Li et al., 2021) address this issue and further confirms Liu et al.’s conclusion.  In addition, we have also analyzed the reanalysis data, which are used for model initialization, and indicated that the deficiencies in reanalysis data in high mountain areas also contributes to the simulated surface temperature bias.  We believe we have caught the main (if not all of the) reasons for the model deficiency for this aspect.  We have more clearly indicated our statements are based on the published papers in abstract and conclusion, and welcome more research to explore this issue in the revised version (Lines 493-496 )

Liu Y., Y. Xue, Q. Li, D. Lettenmaier, and P. Zhao, 2020: Investigation of the variability of near-surface temperature anomaly and its causes over the Tibetan Plateau.  J. Geophy. Res. 125, e2020JD032800. https://doi.org/10.1029/2020JD032800.

Li, Q., Xue, Y., and Liu, Y.: Impact of frozen soil processes on soil thermal characteristics at seasonal to decadal scales over the Tibetan Plateau and North China, Hydrol. Earth Syst. Sci., 25, 2089–2107, https://doi.org/10.5194/hess-25-2089-2021, 2021.

(3).  However, to improve land model and reanalyses data for this aspect are not simple tasks and may take decadal effort (today’s land temperature model development has a more than 70-year history), but proper S2S prediction, including drought/flood/heat wave prediction, is a WMO task requiring urgent attention owing to a significant societal demand.  On line 377-379, based on the LS4P modeling group’s practice, we have pointed out that preliminary research suggests “prescribing both LST and SUBT initial anomalies based on the observed T-2m anomaly and model bias is the only way for the current ESMs to accurately produce the observed May T-2m anomalies”.  In fact, using initialization to improve meteorological prediction is nothing new but a traditional meteorological approach.  We believe when this issue gets more attention and more data are/become available, more methodologies may be developed to address this issue.  But the scientific development always takes time.  We have to allow any development to take step-by-step improvement. 

(4).  We agree with the reviewer that the statement on line 465(previous version) may be misleading and therefore we have modified the text to more properly reflect the ideas that we present in this paper (Lines 493-501).

1. The overall strategy seems to ignore the fact that forecast models generate their forecasts relative to their own A model that is known to be biased warm

may produce an anomalously cold 2003 over the Tibetan Plateau (compared to what it usually produces there), and that would be useful information even if the forecasted temperatures are still warmer than the average observed TP temperature. The point is that people know how to account for long-term model biases. They would properly consider a forecast model’s result to be “2003 will be colder than usual by 5 degrees” rather than “the temperature will be 20C”. The emphasis here on matching the observed temperature in absolute magnitude seems inappropriate to a discussion of forecast systems. See, e.g., the NMME forecast anomaly pages [https://www.cpc.ncep.noaa.gov/products/NMME/seasanom.shtml], which show forecasted anomalies relative to each model’s climatology.

Response: The reviewer raises several issues here.

(1).  Because of the model systematic bias, some groups indeed applied the anomaly prediction in their normal practice.  We understand the justification for these groups, including some LS4P groups, used the anomaly predictions, but predicting the temperature in the real world is always our ultimate goal because the public needs the forecast for the real world, not the forecast relative to one group’s model climatology. 

(2).  However, in the LS4P experimental design, as the first step, we only intend to see if there is any relationship between the observed Tibetan Plateau spring LST/SUBT anomaly and downstream summer precipitation anomalies.  We mainly look at the difference between Task 3 and Task1.  In this way, the model systematic bias has been eliminated so what we really look is indeed the anomaly, which is just what the reviewer tries to emphasize here.  Because of this approach, our goal here does not emphasize producing the best initialization for May 2003 per say, although the methodology present here should serve this purpose. We have made major revision for Section 3.2 (3).  In the modified manuscript, we have clarified our idea/approach.

We acknowledge that our schematic diagram and relevant text do not emphasize this idea clearly as pointed out in the Reviewer’s third minor comment.  We have modified the schematic diagram (Figure 2), figure captions, and reorganize the “Section 3.2(3) Task 3” in the revised manuscript to make the idea clearer.  Original Figure 2 includes both warm and cold years.  In the revised version, to make the thing simple and less confusion for readers who are unfamiliar with LS4P, we only include cold year (same as the case in the LS4P Phase I) in the text and move the schematic diagram for the warm case to appendix for readers who want to pursue this issue further for their own research.  In our response to the reviewer’ 3^rd minor comment below, we will have some more detailed explanation. 

1. Forecast systems also produce a range of forecasted values through the running of ensembles, and any one ensemble member could represent what happens in nature. The experimental analysis protocol, however, emphasizes the importance of having the

*ensemble mean* match the observed anomaly. This is inappropriate. The key question is, do any of the ensemble members look like the observations? (And, in conjunction with point #2 above, the truly key question is, do any of the ensemble member anomalies

*relative to the forecast model’s climatology* look like the observed anomaly?) A model cannot be considered wrong if one of its ensemble members looks like the observations. Insisting that an ensemble mean match a specific year’s temperature seems wrong.

Response: In this paper, we do not emphasize the model intercomparison as well as each model’s evaluation because the major focus for the LS4P is whether the LST /SUBT can provide S2S predictability through the multi-model efforts, which idea has never been tested before.  Since many multi-model projects, such as CMIP, WAMME, and many others, find the ensemble means normally produce better results than any individual model’s performance, we use the ensemble mean and the range of individual model results to assess whether there is S2S predictability and its uncertainty.  The reviewer may have different opinion on this, but this is a common approach currently used in multi-model studies in the community, such as CMIP, AMIP, WAMME and endorsed by the LS4P modeling groups. In addition, the LS4P has more than 20 ESMs and many RCMs.  A comprehensive analysis of each model performance is not that closely related to our main focus at this stage.

1. The model results concerning May Tibetan Plateau temperature anomalies and June precipitation anomalies in east Asia is perhaps suggestive but far from indicative of a causal relationship. Even if the agreements in 6a/6c and 6b/6d do suggest that one pattern led to the other (it could very well be coincidental), I don’t see how the Tibetan Plateau in 6a/6c can be isolated as the source of the agreement in 6b/6d. Significant qualification of this figure’s implications is

Response.  We fully agree with the reviewer’s comments and apologize not to have presented our ideas more clearly.  Figures 6a/6c and 6b/6d are NOT intended to present the causal relationship.  These figures only intend to explain how we develop our hypothesis. 

Observational results in Figure 6A, B (from various observational data sets) along with the remarkable consistency of modeling results in Figure 6C, D (compared to Fig. 6A, B) together provided the underpinnings for the LS4P conjecture that if the May land surface temperature anomaly on the Tibetan Plateau does contribute to the June East Asian precipitation anomaly, then improving the May land surface temperature simulation over the Tibetan Plateau through an improved initialization should allow Earth System models to better predict June East Asian precipitation.

The scientific development normally starts from scientists’ curiosity based on some preliminary discoveries.  The reviewer apparently is an expert in the Earth system modeling and should understand such consistency in Figure 6 from various observational data sets and various Earth system models (with very different dynamic processes and physical parameterizations) is not by chance, and is worth to propose a hypothesis then explore this issue further. 

By and large, Figure 6 is an important step to motivate the hypothesis.  Only Task 3 (with Task 1) is designed to prove the casual relationship.  We have more explicitly emphasize this point in the revised paper to avoid confusion (Lines 349-353, 624-625).

Minor comments:

-- Just to clarify: Are the warm and cold years the same for each month shown? If not, it’s not clear what this figure says about the persistence of warm and cold anomalies (line 92).

Response:  Yes.  The years are the same for each month.  We have clarified this in the Figure 1 caption of the revised paper.

-- lines 106-108. This sounds strange. Can the authors clarify the link between temperature and water amount? While I see that more water leads to a slower seasonal transition, it’s not obvious that at a single point in time, temperature tells you something about water amount.

Response: This part (lines 112-115 in the revised version) has been revised in the revision. The word linking temperature and water amount is indeed potentially confusing and has been eliminated.

-- I studied Figure 2 for a long time and still can’t make sense of it. Why, for a cold year initialization, does the initial condition for a model with a cold bias get set to climatology whereas that with a warm bias does not? Also, please clarify in the caption: are the biases discussed here errors for the particular year of simulation, or are they long term climatological biases? I’m guessing the former, since Task 2 would need to be done for the latter; in that case, though, the use of the term “bias” is confusing here. Bias should refer to a long-term climatological error (reflecting a model deficiency), not to the error at a specific time (which should reflect both bias and random error). Overall, Figure 2 is not helpful for explaining the approach. And again, based on my earlier comments, I’m not convinced the Tmask strategy is appropriate anyway.

Response:  This paper is for the LS4P experimental design and Figure 2 shows how we tried to generate the observed T-2m anomalies using the imposed mask in initialization then checked whether this anomaly improves the S2S predictability and leads to better prediction of the observed drought and flood events.  Based on the reviewer’s comments, we recognize that we did not explain the idea clearly in the previous figure, figure caption, and text.  Although the researchers (co-authors of this paper) working in the LS4P know the idea from Figure, but probably not the readers who are unfamiliar with the project.  We have performed several iterations within co-authors to improve the figure, figure caption, and presentation in the text.  The section 3.2(3) is reorganized.  We hope that the revised figure and text clear-up the confusion.

In the revised figure, we have clarified which initial temperature is for Task 1 and which is for Task 3 with more explanations and indicate that the LS4P phase I’s goal is to prove the causal relationship between the Tibetan Plateau spring LST/SUBT anomaly and large-scale summer precipitation anomaly. Figure 2 and Equation 1 show how to produce observed surface temperature anomaly through Task 3 initialization (relative to Task 1’s initial condition).  We believe these should help readers to understand the ideas better in this figure.  With the revised figure, the readers can see that when we use the difference between Task 3 and Task 1, we actually try to avoid the model systematic bias which was precisely suggested by the Reviewer in the main comments.

Regarding the bias, here we did not clearly indicate whether this should be a specific year or a climatology because it depends on case and data availability.  “Bias” is not always associated with climatology. As Pan et al (2001, JGR) state in their analysis that “Both GCM and RCM fields can exhibit substantial systematic differences from gridded observational data. Such discrepancies between simulated and observed fields are commonly referred to as biases”, although “some differences clearly are not biases in the strict sense, but for simplicity we use the term "bias" to refer to the entire set of comparisons”.  Such interchanges are also used in other studies/fields.  A recent paper “Precipitation Biases in the ECMWF Integrated Forecasting System” by Lavers et al (2021) discusses using the IFS control forecast from 12 June 2019 to 11 June 2020 to show that in each of the boreal winter and summer half years, the IFS has an average global wet bias”.  The way they use bias is similar to what we use.  In remote sensing, “bias correction” terminology is also commonly used.  Moreover, as discussed in the paper, from Task 2, we know the climatological T-2m bias and year 2003 T-2m bias are very consistent.    Therefore, we point out this terminology issue but still keep “bias’ in our paper for simplicity as did in other current practices on lines 362-365.

-- Equation 1 appears to be a means to impose an artificially large temperature anomaly at the start of a simulation so that the anomaly is maintained realistically during the forecast. As far as I can see, there’s no physical basis for the equation; it’s fully empirical and could lead to initial temperatures that make little physical sense (e.g., colder than the model ever gets). More qualification is needed regarding how artificial this construct is. (And again, based on my earlier comments, it may not be appropriate to fix the temperature error in this way, since it may have a source other than the land model.)

Response:  The LST/SUBT approach is a new development and is at a very early stage. The importance of memory of surface temperature has still not been fully recognized by the community.  Currently, no ESMs, including reanalyses, are capable to reproduce observed high mountain T-2m anomaly.  Developing adequate numerical methods and physical parameterizations to permanently solve the issue may take decades or longer.  In meteorology, using the initialization scheme before we develop better models and better data sets to improve the prediction is a very common approach, as done by Yeh et al. (1984, MWR) and Yang et al (1994 MWR).  Those initialization strategies were always based on some empirical relations and not a strict physically-based approach.  Especially in the early stages, some approaches are highly idealized.  For instance, in Koster at al. (2004), which is a rather famous study, were used an approach for which a soil moisture value is artificially imposed for every time step.  But that did not prevent that paper from receiving more than 2250 citations and from becoming a classical paper in meteorology.

On other hand, our approach is not that extreme.  The reviewer thought we may impose an artificially large forcing because of a tuning parameter.  In fact, this is not the case. In the follow-up paper in a Climate Dynamics special issue, we will show every model’s-imposed forcing.   in fact, they are not extreme.  The LS4P includes most of major climate centers in the world.  If our approach is totally different from their normal practices, they would not endorse the LS4P and participate in this project.        

If we wait until the best dynamic and physical method are developed, nothing will happen.

Reference

Koster, R. D., P. A. Dirmeyer, Z. Guo, G. Bonan, E. Chan, P. Cox, C. T., Gordon, S. Kanae, E. Kowalczyk, D. Lawrence, P. Liu, C.-H. Lu, S. Malyshev, B. McAvaney,K. Mitchell, D. Mocko, T. Oki, K.. Oleson,  A. Pitman, Y. C. Sud, C. M. Taylor, D. Verseghy, R.Vasic, Y. Xue, T. Yamada, 2004: Regions of strong coupling between soil moisture and precipitation. Science, 305, 1138-1140.

Lavers, D. A., S. Harrigan, and C. Prudhomme, 2021: Precipitation Biases in the ECMWF Integrated Forecasting System, JHM, 22, 1187-1198.  DOI: https://doi.org/10.1175/JHM-D-20-0308.1

Pan, Z., J. H. Christensen, R. W. Arritt, W. J. Gutowski Jr., E. S. Takle, and F. Otieno, 2001: Evaluation of uncertainties in regional climate change simulations. J. Geophys. Res., 106, 17 735–17 751, https://doi.org/10.1029/2001JD900193.

Yang, R., M.J. Fennessy and J. Shukla, 1994: The influence of initial soil wetness on medium range surface weather forecasts, Mon. Wea. Rev., 122, 471-485.

YEH, T.-C., WETHERALD, R.T. and MANABE, S. (1984). The effect of soil moisture on the short-term climate and hydrology change—A numerical experiment. Mon.Wea.Rev. 112; 474-490

-- line 211 (and elsewhere): Replace “SST” with “ocean state”, since SST is only a small part of what seasonal forecast systems rely on from the ocean. Arguably, subsurface ocean temperature distributions are more relevant.

Response:  Per reviewer’s suggestion, we have replaced “SST” with “ocean state” in most places.  However, in the historical review part, since those studies really used SST for analyses, we still keep SST there.   In addition, for some discussions on the AMIP type runs, we also keep SST. 

-- Clarification regarding figure 7: is this the average of the 2003 anomalies relative to the different models’ climatologies, or is it the average (over all years) model T-2m and

precipitation minus the average (over all years) observations? I’m guessing the latter, given the remarkable agreement with figure 6c,d. The latter can truly be considered a bias, but the term bias was used differently elsewhere in the paper.

Response: Figure 7 shows the average (over all years) model T-2m and precipitation minus the average (over all years) observations.  In the “Section 3.2 (2) Task 2” of the revised paper (lines 362-365), we have clearly indicated that we use the “bias” for both climatology and 2003 for simplicity as was done in Pan et al (2001) and Lavers et al. (2021).