Emergent constraints on future projections of the western North Pacific Subtropical High

The western North Pacific Subtropical High (WNPSH) is a key circulation system controlling the summer monsoon and typhoon activities over the western Pacific, but future projections of its changes remain hugely uncertain. Here we find two leading modes that account for nearly 80% intermodel spread in its future projection under a high emission scenario. They are linked to a cold-tongue-like bias in the central-eastern tropical Pacific and a warm bias beneath the marine stratocumulus, respectively. Observational constraints using sea surface temperature patterns reduce the uncertainties by 45% and indicate a robust intensification of the WNPSH due to suppressed warming in the western Pacific and enhanced land-sea thermal contrast, leading to 28% more rainfall projected in East China and 36% less rainfall in Southeast Asia than suggested by the multi-model mean. The intensification of the WNPSH implies more future monsoon rainfall and heatwaves but less typhoon landfalls over East Asia.

the bias and thus, improve future projections. While a full investigation of these factors is outside the scope of this study, a preliminary analysis would be highly beneficial. Given that space is limited, some simple scatterplots related to this (between simulated historical SST patterns and relevant factors/processes) could be shown as Supplemental Figure(s).
3) Role of internal variability: The authors do not state whether they are using the first ensemble member for each CMIP5 model or not? Assuming that this is the case, a note should be added to the Methods. Furthermore, it brings up the question of how much sensitive the results are to internal variability. I am interested in seeing how sensitive a model's placement on Figure 2 is to the choice of ensemble member. This simple analysis could be performed for a single model with several CMIP5 realizations and added as a supplemental figure if the results warrant inclusion.
Specific comments: L40: Change "despite of intensive" to "despite a number of intensive" L47: Remove "state-of-the-art" here, this is already stated in the paragraph above. L50: Add "the" before "surface warming pattern" L56 and L64: Change "projection" to "projections" L57: Change "75% uncertainty" to "75% of the uncertainty" L66: Change "consistence" to "consistency". I am also confused what is meant by model consistency in this context, does it have to do with the sign of future change? Please clarify and rephrase in the text. L70: Remove "totally". FigS1: Add a panel showing the historical SLP over the same region to give the reader more context. L83: change "present-day simulation" to "historical simulation" throughout. State what the interval between contours represents, is it 1 mm/day and 1% change in cloud fraction? Also, why is the mean SST being taken over different areas (30-30 and 60-60) before being regressed onto the principal components? L88: Use "It has been shown" or "It is known". L89: Change "too few low-cloud covers" to "a lack of low clouds" L92: Add some information about the pre-industrial simulation(s) to the Methods. Is this the PIcontrol scenario from each CMIP5 model? L99-100: I thought the stratocumulus cloud regions had a cold bias (as stated on L85-86)? Please clarify. Calculation of T1 and T2 indices: I would like to see more detail regarding how these indices are derived. In the Fig 2 caption the authors note that "Negative T1 denotes a reduction of cold tongue bias…", but the language used is confusing. Is the index centered on the model ensemble mean value? If so, "anomaly" is more appropriate than "bias", which is often reserved when comparing models to observations. Fig 1ab,3,4: It is difficult to see the land boundary on many of these maps. Perhaps switch the SLP contour color to grey and the land boundary to black for figs 1a and b? L125: I appreciate this discussion of the physical mechanisms behind the constraint, but would change the subtitle to "Physical mechanisms backing the constraint".
L129: "The positive SST warming" should say "The positive SST anomalies" or "The SST warming" L148: Change "can result larger decrease" to "can result in larger decreases" L149: change "respond" to "response" L175: "Reduce the CMIP5 intermodel spread by 75%" 75% of what? Please clarify and rephrase. L179: Fix "father" L209-211: Expand on why the global-scale bias in SSTs must be subtracted.
L212-213: There are more robust approaches to estimating the optimal PC1 and PC2 than just using the mean of observed T1 and T2 (see main comment #2).
Reviewer #3 (Remarks to the Author): COMMENT SHEET for PAPER NUMBER: NCOMMS-19-3009875 JOURNAL: Nature Communications TITLE: Emergent constraints on future projections of the western North Pacific Subtropical High The manuscript by Chen et al. present an interesting analysis and interpretation on future projection of western North Pacific Subtropical High (WNPSH). The study describes two inter-model modes that can explain uncertainties of future changes in WNPSH simulated by CMIP5 climate models. They found that the first leading mode is linked to the models' cold tongue biases in the tropical Pacific and the second mode is linked to the warm biases associated with marine stratocumulus. They corrected future projection of WNPSH using observational constraints estimated through the principal component and the related sea surface temperature pattern. The study further suggested the physical mechanisms behind the two leading inter-model modes. The paper is well focused, and the study objectives are clear. The finding is new and could be potentially important to understand future changes in WNPSH and east Asian monsoon. However, although the authors try to present a compelling case, the amount of information provided and the failure to support their interpretation weakens their case. The present study is mostly based on descriptive analysis, lacks important details, and must be strengthened before it can be accepted for publication. The major and minor issues are detailed below.
1.The critical point is the appropriateness of the analysis used in this study. The key analysis in this paper is to find optimal principal components (PCs) from the inter-model relationship between leading PCs and sea surface temperature (SST) pattern indexes as shown in Fig. 2. However, there seems to be circular reasoning in this analysis given that the SST pattern index (i.e., T1 and T2) is based on PCs. By the definition of T1, PC1 and T1 are expected to be correlated to each other, and thus I'm wondering how the result can be justified to constraint future model prediction of WNPSH.
2.The pattern of cold tongue bias stretching to the coast of Peru as shown in Fig 1c is not similar to that of the typical cold tongue bias of climate models. How would the result change if T1 is defined by equatorial SST mean in different models? And for the T2, how would be the result if T2 is defined by area-averaging the warm SST bias in the subtropics?
3.Another important issue is that physical mechanisms suggested in the study are not complete, rather the arguments are largely based on previous works and speculation. Particularly, the linkage between cold tongue SST bias in the historical run and future warming in the western Pacific is not clearly shown. The linkage between warm SST bias (or low cloud bias) and enhance land-sea thermal contrast also needs to be strengthened. Isn't there possibility that models showing stronger Arctic amplification can cause stronger low cloud and warm SST biases? The causal relationship is still not clear in the mechanism related to PC2. Overall, the mechanism section seems to overly speculative given the tenuous arguments presented.
4.The methodology used in this study has not been explained in sufficient detail. The authors explained that T1 and T2 are the projections of historical mean SST pattern onto the SST pattern associated with EOF1 and EOF2, respectively. Does the projection refer to the regression coefficient between them? Is the historical mean SST from the observations subtracted from each modeled SST to define T1 and T2? In line 210, the historical mean SST refers to one value averaged in the specified area? 5.Other minor comments are as follows.
-Line 72: EOF2 doesn't look like enhanced WNPSH. -Line 90: fewer  weaker -Line 102: What is the degree of freedom to calculate p-value? In the multi-model set used in the study, the models rooted in the same family are used, thus it seems the degree of freedom should not be simply N-2.
-Future change in circulation pattern related to EOF2 (Fig. 4) do not really indicate the enhanced WNPSH. Again, is the second inter-model mode related to WNPSH? -Authors' argument on heat waves, drought, and flood is unrelated to the topic and probably better to be avoided because there is not any analysis on these topics. -Last paragraph seems to be redundant and duplicated.
The original comments of reviewers are in italic. The revised parts are marked in red in the tracked-change version of manuscript.

Reviewers' comments:
Reviewer #1 (Remarks to the Author): The paper is well written with sound methodology and contains useful scientific information on the North Pacific Subtropical High. The paper is recommended for publication with minor suggestions.

Suggestions:
A recent study (Preethi etal 2017a)   While the scatterplots help to understand the factors that are driving the intermodel spread, a full investigation calls for further study. Since the motivation of this study is to constrain the projection by using the current state-of-the-art models instead of the   10°S-10°N).  Fig. R3 below (added in as Supplementary Fig. 7), the spread in WNPSH projection across CMIP5 models is two more times larger than that from internal variability. If measured by variance ratio ( 2 / 2 ), the contributions from internal variability is less than 20%. Considering that the emergent constraints reduce about 45% uncertainty of PCs in this study, the remaining uncertainty may partly be contributed by internal variability. We added a comment on this in discussions (L209-211) and a brief explanation in the Methods (L226-229). ( 2 / 2 ), the contributions from internal variability is less than 20%.

Specific comments:
1) L40: Change "despite of intensive" to "despite a number of intensive" Response: Done (L44). We also changed the expression during word editing.
2) L47: Remove "state-of-the-art" here, this is already stated in the paragraph above.

5) L57
: Change "75% uncertainty" to "75% of the uncertainty" Response: Done (L66). We changed the expression during word editing. Please clarify and rephrase in the text.

Response:
The expression is rephrased to explicitly show the meaning as follows, "where less than 70% of the models agree on the sign of change (Supplementary Fig.   1a) and the signal-to-noise ratio is below 0.5 ( Supplementary Fig. 1b)" (L74-76).

8) FigS1
: Add a panel showing the historical SLP over the same region to give the reader more context.

Response:
The historical SLP distribution has been plotted, with the cyan contours overlaying on the shadings (Supplementary Fig. 1a).
Response: Done (L93, L105).  (Fig. R4). The correlation coefficient between the averaged SST in the marine stratocumulus region (box in Fig. R2) and PC2 can be as high as -0.55 when the mean SST in 30°S-30°N is removed, or it is only -0.33. A statement is added in Methods to explain why a global-scale mean SST is removed for deriving the SST pattern related to the PC2 (L261-263).

Fig. R4
Historical SST pattern related to the PC2 of uncertainty in WNPSH projection.
(a) Mean SST in 30°S-30°N is subtracted from original SST in each model before regressing onto the PC2 while (b) uses the original SST.

13) L92: Add some information about the pre-industrial simulation(s) to the Methods.
Is this the PI-control scenario from each CMIP5 model?

L85-86)? Please clarify.
Response: The cold SST pattern in Fig. 1d is the intermodel spread pattern related to PC2. It only indicates the relationship of intermodel spread between current climate and future projection. If the opposite PC2 (-PC2) is used, a warm SST pattern in the stratocumulus cloud regions will be seen. In fact, most models simulate a warmer SST beneath the stratocumulus than the observations. In L98-102, we clearly point out the fact and how it could imply the observational constraint. Because the observed SST beneath the stratocumulus cloud regions is colder than the modelled, the constrained PC2 has a positive value (Fig. 7).

15) Fig 2: I am somewhat surprised to see such strong agreement between the observational datasets. I recommend creating a supplemental figure showing maps of their SST patterns over the historical period for T1 and T2.
Response: We added Supplementary Fig. 4 to show the strong agreement in climatological SST pattern in observational datasets. The strong agreement ensures a high signal-to-noise in the current climate (L126-127).

16) Calculation of T1 and T2 indices: I would like to see more detail regarding how these indices are derived. In the Fig 2 caption the authors note that "Negative T1
denotes a reduction of cold tongue bias…", but the language used is confusing. Is the index centered on the model ensemble mean value? If so, "anomaly" is more appropriate than "bias", which is often reserved when comparing models to observations.

Response:
The confusing sentence has been deleted. Detailed calculation of T1 and T2 are described in Methods (L250-260). In the revised manuscript, the indices are calculated based on Equations (2) and (3) in Methods without any postprocessing, which impacts little on the results.

17) Fig 1ab,3,4: It is difficult to see the land boundary on many of these maps. Perhaps switch the SLP contour color to grey and the land boundary to black for figs 1a and b?
Response: Revised (Figs. 1, 3 and 4).

18) L125
: I appreciate this discussion of the physical mechanisms behind the constraint, but would change the subtitle to "Physical mechanisms backing the constraint".

22) L175: "Reduce the CMIP5 intermodel spread by 75%" 75% of what? Please clarify and rephrase.
Response: Corrected. Based on the hierarchical statistical framework, reduced variance is overestimated in the original study. In the revised manuscript, it is corrected to 45% (L209).
24) L209-211: Expand on why the global-scale bias in SSTs must be subtracted.

Response:
As explained in a response above, our idea of removing a global-scale mean SST from the original historical climatology is to highlight the role of SST pattern rather than absolute values. Here we find the method does not affect the results for PC1. In Methods we now add the explanation why a global-scale mean SST is removed for deriving the SST pattern related to the PC2. More details can be found in the response (item #10) related to Fig. R4 (L261-263).

25) L212-213:
There are more robust approaches to estimating the optimal PC1 and PC2 than just using the mean of observed T1 and T2 (see main comment #2).
Response: Thanks for your suggestion. We now employ your recommended approaches to make a more robust constraint (L109-138 Response: Thank you for your comments. We have responded to your concerns one by one below.

1.The critical point is the appropriateness of the analysis used in this study. The key analysis in this paper is to find optimal principal components (PCs) from the intermodel relationship between leading PCs and sea surface temperature (SST) pattern
indexes as shown in Fig. 2 Response: Yes, the historical SST pattern related to PC1 is not identical to the typical cold tongue bias which centers in the equatorial central Pacific. Since the significant anomalies are indeed located in the central-eastern Pacific cold tongue region (180-80°W) and like the climatological SST pattern (Supplementary Fig. 4), we now call it cold-tongue-like SST pattern in the revised manuscript (L26).
As you suggested, we define a SST index in the equatorial central-eastern Pacific are robust. The CEP SST, as an important source to the PC1 uncertainty (Fig. 6a), can be used to constrain PC1 directly (Fig. R5a), as well as the ENP_NA SST to constrain PC2 (Fig. R5b). But the constraint is less effective than that constrained by the SST  For the first mode, the importance of negative shortwave-SST (SW-SST) feedback is highlighted (L165-172) which establishes the link between cold tongue SST bias in the historical run and future warming in the western Pacific (Left column in Fig. R6).
For the second mode, the associated Arctic amplification is the well-known spatial feature of global warming. There are plenty of studies addressing why the Arctic region warms more than other places under the GHG forcing. Considering the similar warming pattern related to PC2 with that related to GMST change (Supplementary Fig. 6), it is reasonable to come up with an idea to explain the spread in PC2 by global mean warming. We clearly show the physical chains from the current cold SST with more clouds in the marine stratocumulus region to future high global warming and large landsea thermal contrast (L178-205; Right Column in Fig. R6).
Based on our current knowledge, it is difficult to understand how stronger Arctic amplification, which usually means an exceptional warming in the Arctic region and surrounding continents, can cause stronger low cloud and warm SST biases.
Nevertheless, as shown in Fig. R5b, the North Atlantic and North Pacific SSTs (Fig.   1d) can be also used to constrain PC2. But the underlying mechanism remains unclear.
A possible explanation involves the Atlantic Overturning Meridional Circulation (AMOC). Warmer SST in the North Atlantic and North Pacific in the current climate is related to stronger AMOC (Zhang and Zhao 2015), implying that most CMIP5 models may underestimate the AMOC strength (Fig. R5b). A stronger AMOC, however, tends to weaken more under CO2 forcing (He et al. 2017), thus leading to weaker ocean heat uptake efficiency and higher global surface warming (Chen and Tung 2018), which can explain the PC2 uncertainty (Figs. 6f and 6h). However, a weaker AMOC could also lead to a higher global warming because it weakens less from CO2 forcing which results in smaller reduction in polarward heat transport (He et al. 2017 That is why a larger scale pattern is shown in Fig. 1b.
-Line 102: What is the degree of freedom to calculate p-value? In the multi-model set used in the study, the models rooted in the same family are used, thus it seems the degree of freedom should not be simply N-2.
Response: Strictly speaking, the degree of freedom should be less than N-2. However, it is difficult to estimate the accurate degree of freedom. The correlation coefficients in Fig. 2 are still significant (p<0.001) even if the degree of freedom is reduced to half of the original (about 15). Nevertheless, we delete the p-value in the text.
-Future change in circulation pattern related to EOF2 (Fig. 4) do not really indicate the enhanced WNPSH. Again, is the second inter-model mode related to WNPSH?
Response: The increased SLP over the western North Pacific represent the enhanced WNPSH (Fig. 5d)  Black contour lines are climatology. AMIP4xCO2 is a sensitivity run using only AGCM with prescribed present-day SST as boundary condition under 4xCO2 forcing. Hence, the experiment creates a very warm land and strong land-sea thermal contrast.
-Authors' argument on heat waves, drought, and flood is unrelated to the topic and probably better to be avoided because there is not any analysis on these topics.
-Last paragraph seems to be redundant and duplicated.