Interplay between oceanic subduction and continental collision in building continental crust

Generation of continental crust in collision zones reflect the interplay between oceanic subduction and continental collision. The Gangdese continental crust in southern Tibet developed during subduction of the Neo-Tethyan oceanic slab in the Mesozoic prior to reworking during the India-Asia collision in the Cenozoic. Here we show that continental arc magmatism started with fractional crystallization to form cumulates and associated medium-K calc-alkaline suites. This was followed by a period commencing at ~70 Ma dominated by remelting of pre-existing lower crust, producing more potassic compositions. The increased importance of remelting coincides with an acceleration in the convergence rate between India and Asia leading to higher basaltic flow into the Asian lithosphere, followed by convergence deceleration due to slab breakoff, enabling high heat flow and melting of the base of the arc. This two-stage process of accumulation and remelting leads to the chemical maturation of juvenile continental crust in collision zones, strengthening crustal stratification.

Revisions are marked with Yellow Background in the clear version and with tracked changes in a separate file.

Comments from Reviewer #1
General evaluation: The manuscript entitled "Building continental crust along convergent plate boundaries" is a very interesting paper regarding the continental crustal growth on our Earth. After compiling the available geochronological and geochemical data from the igneous rocks in the Gangdese, southern Tibet, it was proposed that the arc magmatism there can be divided into two stages of evolution, i.e., early fractionation of mantle-derived primitive arc rocks, and later re-melting of the early cumulate, which made the juvenile arc crust transformed into mature continental crust. The paper is well written, and the data are of high quality, hence I recommend its publication in our journal, but some revision is still needed.
Response: Thank you for this positive evaluation.
Main comments: 1. Generally, the convergent plate boundary can be divided into the oceanic arc by oceanic subduction under another oceanic lithosphere, and continental arc by oceanic subduction under a continent. It was true that the Gangdese was mostly a continental arc (a convergent plate boundary) during the Mesozoic by subduction of the Tethyan ocean under the Asian continent, but it was also affected by collision of India to the south along the Yarlung Zangbo suture zone. Therefore, the Gangdese documented both oceanic subduction and later continental collision, and hence the title here is suggested to be modified to match the main contents of the text.
Responses: Great comment. We have revised the title to "Interplay between oceanic subduction and continental collision in building continental crust".
2. If the paper focused on the continental crustal evolution, it will be very constructive to compare the Gangdese with other arcs, especially with Cordillera and Andes along the eastern Pacific. Specific comments: 1. Line 35-37: If the setting was initiated by early oceanic subduction and then followed by continental collision, this setting is better to ascribe as collisional, but not convergent boundary.

Responses
Response: we have revised "convergent boundary" to collision zones.
2. Line 50-53: For the Gangdese, an input of Indian continental material was undoubtedly involved during the Miocene magmatism, which cannot be ignored during the magmatic evolution of a continental collisional zone.

Response:
We did not address this point in the abstract (i.e. line 46-52), but have discussed it in later sections (please see lines 191-196, 283-287, and 362-370 5. Line 75: What criteria used here to define the "post-collisional". Response: According to Zhu et al. (2015, Scientific Reports), "post-collisional" is defined as the time following complete loss of the pull force from the subducting oceanic slab due to slab breakoff. We have re-phased the sentence that now reads: "This subdivision is based on the multiple lines of evidence that constrain the timing of initial collision between India and Asia to ~60 Ma 15,16 . It is also based on post-collision being defined as the time following complete loss of the pull force from the subducting oceanic slab due to its breakoff, which started at ~45 Ma 8 ". Please see lines 78-81. 6. Line 100 (and associated figure): the Z-shaped trend for the pre-collisional is not clearly identified if it was compared with the syn-, and post-collisional rocks.

Response:
The Z-shaped trend is defined by the presence of dunite to wehrlite, pyroxenite, hornblendite samples, which are not documented by the syn-, and post-collisional rocks. This trend may be influenced by the limited number of dunite to wehrlite, pyroxenite samples. This has been clarified in lines 123-127.
7. Line 107-109: Please state the evidence that those pre-collisional mafic rocks are mantle-derived.

Response:
We have added the following two references to indicate that those pre-collisional mafic rocks are mantle-derived. Please see line 117.  9. Line 143-144: Is it possible that those ultramafic cumulates sank into mantle by delamination after its formation?
Response: We think this is less likely. We have discussed this possibility in a later section (lines 377-386).
10. Line 145-149: If four types of rock assemblages can be classified, we should discuss their origination one by one.

Response:
We have provided simple explanations for their origins as follows: (1)  11. Line 149-152: You mean that the gabbro is mantle-derived, but not a cumulate as well?
Response: Gabbro is mantle derived and either cumulate or non-cumulate. We have added the word "cumulate" to indicate the cumulitic nature of the hornblende gabbro mentioned here (in line 157).
12. Line 162: what difference between "damp" and "wet", can you give the specific values here?

Response:
We have provided the specific values and related references in line 172-173.
13. Line 164-165: Why the back-veining here is evidence of re-melting?

Response:
The dioritic dyke is younger and hotter than the wall rock (tonalite), Its emplacement resulted in the re-melting of the wall rock to form back-veining. We have re-organized this sentence to clarify this point: "Field evidence for melting is shown by back-veining of a ~48 Ma dioritic dyke by felsic magma derived from the melting of the ~79 Ma wall-rock". Please see lines 175-176.
14. Line 169-171: If the post-collisional rocks were also from re-melting of pre-existing rocks, their Sr-Nd-Hf isotopic compositions will provide strong evidence, right?
Response: Yes, this is precisely the case for felsic rocks older than 30 Ma as indicated by the similar zircon δ 18 O, but not the case for felsic rocks younger than 30 Ma that contain increased amounts of the subducted Indian slab-derived materials. These points have been addressed in later sections (lines 191-196 and 283-287).
Response: Kernel density estimation is a nonparametric technique for density estimation. It is a popular tool for visualising the distribution of data. We have added the explanation: "Colored background indicates sample density distribution as measured by bivariate kernel density, where redder background corresponds to increased sample concentration or density" in the caption of Figure 3 to explain the "bivariate kernel density". Please see line 221-223.
16. Line 216: From discussion of this sub-section, three stages of processes should be included, i.e., formation of the primitive arc, re-melting of the arc/cumulate (with in situ basement), and later input of subducted continent. The latter is the key difference between the studied Gangdese and the commonly assumed convergent plate boundaries. 18. Line 290-292: it is better to have comparisons for the major and trace elements, but not only of zircon O isotopes.

Response:
We have carefully checked the paper we cited here (Lackey et al., 2005) and references therein to find out major and trace element data of samples that have been analyzed for zircon O isotope by Lackey et al. (2005). However, the authors did not report the major and trace element data but just introduced the lithology. This makes them unavailable to compare with the Gangdese samples. To respond to this comment, therefore, we have added the lithologies (i.e. gabbros and granitoids from the Sierra Nevada batholith) to the main text. Please see lines 290.

Response
Response: Yes, the lower and upper boundaries of such doublet Moho structure are suggested to indicate the petrological and seismic Moho, respectively. To clarify this point, we added "seismic" to line 401.
25. Line 412-415: if this is correct, you have to apply delamination to keep the crust evolved toward more felsic.

Response:
We feel that there is no need to delaminate the refractory ultramafic-mafic residual layer because, as mentioned, the incoming buoyant continental crust could provide a natural barrier for delamination (Ganade et al., 2021). Instead, such residual layer may have seismically become part of the sub-arc lithospheric mantle, resulting in the vertical chemical and density stratification of continental crust in collision zones. The Gangdese crust could be the best example of such interpretation. 26. Line 426: it is too concise to cite these collisional zones without detail descriptions.

Response:
(1) The difficulty we find in providing more details is that cumulates within granitic batholiths are often not included in the dataset and their genetic relations to granitic batholiths are typically unknown.
(2) A more detailed description of these collision zones is introduced in a companion paper (Zhu et al., Under Review).
27. Line 443-445: More explanations are necessary for a better understanding.
Response: Given the limited data available for ultramafic and mafic rocks in the literature, we have added a few more explanations in terms of external forces (e.g., convergence acceleration and deceleration) and internal factors (e.g., thermal perturbations and water fluxing). Please see lines 430-433.
28. Line 445-447: as the key of the paper, it is better to specific the evolutional processes according to their chronological order, such as the arc cumulation, slab deepening and re-melting of the lower base, continental input after collision, etc.

Response:
We have re-organized this subsection along chronological order. Please see lines 425-442.

Comments from Reviewer #2
Zhu and his co-authors present a synthesis study on Gangdese magmatism and continental crust formation. The main idea of this paper is that melting of pre-existing cumulates during and after the continent collision phase is key to generating matured continental crust. Overall, this work is interesting and free of obvious pitfalls. My major criticism is that they talked too much, which makes this paper not easy to follow. The authors need to think about which observation provides the most important line of evidence instead of worrying about everything.
Response: Many Thanks! We have moved the details related to binary mixing modelling and the introduction to the rationale to the Supplementary Materials and re-organized the section of "Role of Recycled Supracrustal Component in Generating the Compositional Change". It has been reduced from 998 to 768 words.
To me, the geochemical trends shown in Fig. 3 (2) experimental studies (Dufek & Bergantz, 2005;Sisson et al., 2005) and thermodynamic modeling (Wang et al., 2022) show that the K2O contents of the partial melts are directly linked with those of the starting materials, regardless of cumulate and noncumulate origin.
We set up the conditions of simulation at 2.0 GPa and 800-1000 °C adding 2 wt% H2O, which are within the range of P-T conditions expected for the lowermost arc crust (~800-1000 °C at 35-70 km depth; Ducea et al., 2021).
The results in the figures below show that in the temperature range of ~800-1000 °C, corresponding to melt fractions of 20-30%, the produced melts have high [K2O/SiO2]c ratios >4 and high [Th/La]c ratios >2 (Figure 1 below), higher than the median [K2O/SiO2]c shown in Figure 2b or close to the median [Th/La]c in Figure 3d in the main text. Therefore, the increased K2O/SiO2 ratios since 70 Ma could be the combined result of biotite ± hornblende breakdown as they are the main K-bearing mineral phases in basaltic underplates and cumulates, whereas the increased Th/La ratios are most likely controlled by the difference of incompatibility between Th and La, i.e., Th is more incompatible than La and thus more preferentially concentrates in the melt during melting. in the temperature range of ~800-1000 °C that corresponds to 20-30% degree of melting. The grey field corresponds to temperature range of the lowermost arc crust at ~800-1000 °C and 35-70 km depth (Ducea et al., 2021).
Pertaining the second comment: (1) Dehydration melting experiments have indicated that K2O contents of partial melts are directly linked with those of the starting materials, regardless of whether they have a cumulate (Dufek & Bergantz, 2005) or noncumulate origin .
(2) Our unpublished data from dehydration melting experiments show that natural hornblende-rich cumulate samples, common in arc lower crusts, can produce shoshonitic to medium-K felsic melt compositions at 900-1000 °C and 1.0-2.0 GPa (Figure 2). Existing experimental data are shown for comparison in Figure 2. In the revision, however, (1) we did not present the results of our unpublished experimental data and the existing experimental data, to avoid complexity, and (2) we have re-organized this subsection by highlighting experimental studies and thermodynamic modeling (Wang et al., 2022). Please see line 303.