the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 24 Aug 2026
The Stratified Microbial Carbon Pump: Thermal Stratification Enhances Refractory Dissolved Organic Carbon Production and Stabilizes Carbon in Alkaline Karst Waters through Keystone Microbial Interaction Networks
Abstract. Recalcitrant dissolved organic carbon (RDOC) represents a persistent fraction of the carbon pool and contributes substantially to regional carbon budgets in terrestrial aquatic ecosystems. Given the substantial carbon sequestration potential of karst waters, elucidating the dynamics and persistence of RDOC is essential for understanding carbon retention in these geologically distinctive environments. Karst reservoirs represent important yet highly complex components of regional carbon cycling. Here, RDOC accumulation and its underlying mechanisms were examined in the Dalongdong (DLD) Reservoir, a dissolved-carbon-rich karst water body, over three contrasting thermal phases: incubating thermal stratification (ITS), obvious thermal stratification (OTS), and mixing (MX). Our results revealed that RDOC dynamics were predominantly regulated by microbial processes rather than benthic carbon inputs. The establishment of thermal stratification generated pronounced physicochemical gradients that facilitated vertical niche partitioning among keystone taxa, thereby regulating DOM bioavailability through taxon-specific metabolic pathways. Bacterial network analysis further indicated that facilitative interactions favored the generation of labile carbon, whereas competitive interactions under environmental stress promoted RDOC accumulation. Notably, the distinctive geochemical conditions of the karst system facilitated Ca–P co-precipitation, resulting in persistent phosphorus limitation and an elevated C:P ratio in the water column. This nutrient imbalance reduced microbial carbon use efficiency and suppressed extracellular enzyme activities, consequently favoring the conversion of autochthonous labile carbon into more persistent RDOC. Collectively, these findings suggest that the synergistic coupling of the biological carbon pump (BCP) and microbial carbon pump (MCP), reinforced by geochemical phosphorus sequestration, represents an important mechanism underlying long-term carbon retention in alkaline, calcium-rich aquatic systems. Our findings further demonstrate that karst-specific geochemical conditions interact with thermal stratification to regulate microbial processes governing carbon persistence. In particular, MCP efficiency appears to be modulated by carbonate-weathering-derived dissolved inorganic carbon (DIC), providing mechanistic insights into the enhanced carbon sequestration capacity of geologically distinctive inland waters. Protecting these highly efficient carbon-sequestering ecosystems may therefore contribute to atmospheric CO2 removal and provide an additional pathway for advancing global climate mitigation.
- Preprint
(2002 KB) - Metadata XML
-
Supplement
(1354 KB) - BibTeX
- EndNote
Status: final response (author comments only)
-
RC1: 'Comment on egusphere-2026-4596', Anonymous Referee #1, 04 Sep 2026
-
AC1: 'Reply on RC1', Junbing Pu, 14 Oct 2026
Response to reviewers’ comments
Ms. NO.: egusphere-2026-4596
Title: The Stratified Microbial Carbon Pump: Thermal Stratification Enhances Refractory Dissolved Organic Carbon Production and Stabilizes Carbon in Alkaline Karst Waters through Keystone Microbial Interaction Networks
Journal: Biogeosciences
Reviewer #1:
This study takes the Dalongdong karst reservoir in Guangxi as the research object, and divides the thermal regime into three phases: incubating thermal stratification, obvious thermal stratification, and water‑column mixing. It explores the regulatory mechanisms of thermal stratification on the production of recalcitrant dissolved organic carbon (RDOC). Thermal stratification creates physicochemical gradients that shape the ecological niches of keystone bacteria. Facilitative bacterial interactions promote the production of labile carbon, while competitive interactions drive RDOC accumulation. Calcium‑phosphorus co‑precipitation in karst water induces phosphorus limitation, reduces microbial carbon‑use efficiency, and triggers the coupling of the biological carbon pump (BCP) and the microbial carbon pump (MCP). This work reveals the carbon sequestration mechanism of karst reservoirs and provides theoretical support for inland aquatic carbon‑sink assessment and climate‑change mitigation. My main comments are listed below:
Reviewer comment 1:
The abstract states: “Bacterial network analysis further indicated that facilitative interactions favored the generation of labile carbon, whereas competitive interactions under environmental stress promoted RDOC accumulation.” How is this conclusion demonstrated?
Response 1:
We thank the reviewer for raising this important point. We agree that the original wording overstated what can be inferred from co-occurrence network analysis. In particular, positive and negative associations in a microbial co-occurrence network cannot be equated directly with facilitation and competition, respectively, nor can they demonstrate that microbial interactions themselves caused the production of labile carbon or the accumulation of RDOC.
Our original interpretation was based on the covariation among bacterial network structure, DOM composition, and environmental conditions across thermal phases and water-column layers. Positive microbial associations tended to occur under conditions with greater availability of relatively labile DOM fractions, whereas stronger negative associations were observed under some environmentally constrained conditions that also exhibited higher RDOC levels. However, the revised layer- and period-specific analyses showed that these relationships were not uniform. After Benjamini-Hochberg false discovery rate correction and, where applicable, adjustment for major environmental variables, most cohesion-DOM relationships were no longer statistically significant. This indicates that microbial network organization should be interpreted primarily as a community-level characteristic that covaries with environmental conditions and DOM composition, rather than as an independent causal driver of RDOC accumulation.
We have therefore removed the terms “facilitative interactions” and “competitive interactions” and revised the causal wording in the Abstract. The revised sentence now reads: “Bacterial network analysis further suggested that positive microbial associations were linked to greater labile carbon availability, whereas stronger negative associations under environmental stress coincided with higher RDOC levels.” (L37-39)
This revised wording distinguishes statistical network associations from demonstrated ecological interactions and reflects more accurately the evidence provided by our data. Consistent with this revision, the concluding statement of the Abstract now emphasizes that thermal stratification, microbial community organization, and nutrient limitation jointly covary with RDOC dynamics, rather than implying that particular microbial interaction types directly promote RDOC production or accumulation.
Reviewer comment 2:
Please elaborate in detail on the synergistic coupling between the biological carbon pump (BCP) and microbial carbon pump (MCP) and its environmental implications.
Response 2:
We sincerely thank the reviewer for this insightful comment. We agree that the coupling between the biological carbon pump (BCP) and microbial carbon pump (MCP) was insufficiently explained in the original manuscript, although it represents an important conceptual framework for interpreting carbon persistence in the Dalongdong Reservoir. We have therefore expanded the relevant discussion and revised the Abstract to explicitly describe the carbon-transfer pathway linking these two processes.
Conceptually, the BCP and MCP represent interconnected rather than independent carbon-sequestration pathways (Jiao et al., 2010; Legendre, 2024). In the BCP, dissolved inorganic carbon (DIC) is initially incorporated into phytoplankton and other autotrophic biomass through photosynthetic carbon fixation, generating autochthonous organic matter (Jiao et al., 2010; Legendre, 2024). A substantial fraction of this freshly produced organic carbon can subsequently enter the dissolved organic matter (DOM) pool through extracellular release, cell lysis, grazing-related release, and solubilization of particulate organic matter (Azam and Malfatti, 2007; Thornton, 2014). This BCP-derived DOM therefore provides an important substrate pool for subsequent microbial processing (Azam and Malfatti, 2007; Jiao et al., 2010). The MCP acts downstream by selectively utilizing and transforming labile and semi-labile DOM through microbial uptake, metabolic processing, biomass turnover, and release of transformation products, thereby generating DOM components with reduced biological availability and promoting the formation and persistence of recalcitrant dissolved organic carbon (RDOC) (Ogawa et al., 2001; Jiao et al., 2010; Jiao et al., 2024).
In the Dalongdong Reservoir, our results further suggest that this BCP-MCP linkage is strongly modulated by karst-specific geochemical and hydrodynamic conditions. Carbonate weathering supplies abundant DIC and Ca²⁺, potentially supporting autochthonous carbon production while simultaneously facilitating Ca-P co-precipitation. The resulting phosphorus limitation alters microbial resource allocation, extracellular enzyme activity, community interactions, and DOM utilization. Thermal stratification further establishes vertically differentiated physicochemical environments and microbial niches, leading to depth-specific processing of BCP-derived organic substrates. Together, these processes connect the production of autochthonous organic carbon with its subsequent microbial transformation and preservation as RDOC.
We have also clarified the environmental implications of this coupling. Rather than simply increasing organic carbon production, BCP-MCP coupling determines the fate of fixed carbon by regulating whether autochthonous organic matter is rapidly remineralized to CO₂ or transferred into a more persistent dissolved organic carbon pool. In the alkaline and Ca-rich karst reservoir studied here, geochemically induced phosphorus limitation and stratification-mediated microbial niche differentiation can strengthen carbon persistence by both modifying microbial transformation pathways and constraining rapid DOM recycling. Consequently, the coupling between DIC fixation, autochthonous organic matter production, and MCP-mediated RDOC formation provides a mechanistic pathway through which carbonate-rich inland waters may enhance long-term aquatic carbon retention.
Accordingly, we have revised the Abstract and expanded the Discussion (Section 4.4) to more clearly describe (i) the substrate connection between the BCP and MCP, (ii) the regulatory role of thermal stratification and karst geochemistry, and (iii) the implications of this coupling for RDOC persistence and regional aquatic carbon sequestration.
Reviewer comment 3:
Please provide relevant references to support the statement that BCP processes can induce phosphorus limitation in lake–reservoir systems.
Response 3:
We sincerely thank the reviewer for this valuable suggestion. We agree that the relationship between BCP activity and phosphorus limitation in lake–reservoir systems require stronger literature support and a clearer mechanistic explanation. We have therefore revised this section and added relevant references.
Previous studies have demonstrated that enhanced photosynthetic activity can promote phosphorus removal from the water column, particularly in calcium-rich and alkaline aquatic systems. During intensive primary production, phytoplankton uptake of CO₂ and/or HCO₃⁻ increases water-column pH and promotes CaCO₃ supersaturation and precipitation. Dissolved phosphate can subsequently be incorporated into or co-precipitated with newly formed CaCO₃, thereby decreasing the pool of bioavailable phosphorus. This photosynthetically induced CaCO₃-P co-precipitation has long been documented in hard-water lakes and has been proposed as negative feedback on primary production (Otsuki and Wetzel, 1972; Hamilton et al., 2009).
More importantly, this mechanism has also been demonstrated in karst river–reservoir systems. (Sun et al., 2022)Sun et al. (2022) showed that, following damming, nutrient limitation of BCP shifted from predominantly carbon limitation in riverine reaches toward nitrogen or phosphorus limitation in reservoirs, while BCP-associated phosphorus removal reduced total phosphorus concentrations and increased water-column N:P ratios. Recent studies of karst reservoirs further demonstrated that enhanced BCP activity can stimulate CaCO₃ precipitation and associated phosphorus co-precipitation, thereby accelerating phosphorus removal from surface waters (Zhang et al., 2025; Han et al., 2025).
Accordingly, we have revised the manuscript to clarify that BCP does not directly “cause” phosphorus limitation through carbon fixation alone. Rather, enhanced biological carbon fixation can indirectly intensify phosphorus limitation through biological uptake of phosphorus and, particularly in Ca-rich alkaline waters, through photosynthetically induced CaCO₃-P co-precipitation and subsequent particle settling. Relevant references have now been added to support this interpretation (L93-103).
Reviewer comment 4:
Section 4.1 is suggested to include more evidence on RDOC quantities; currently this section contains very limited supporting data.
Response 4:
We sincerely thank the reviewer for this important suggestion. We agree that the original Section 4.1 focused primarily on microbial community organization and potential functional differentiation, whereas quantitative evidence for the RDOC pool itself was insufficiently integrated into the discussion. We have therefore substantially revised this section to incorporate both the temporal and vertical variations in RDOC concentration and relative contribution, together with the statistically supported environmental and microbial associations.
Specifically, we now report the vertical distribution of RDOC during obvious thermal stratification (OTS). RDOC concentrations were 1.33 mg L⁻¹ in the epilimnion, 1.25 mg L⁻¹ in the thermocline, and 1.11 mg L⁻¹ in the hypolimnion, demonstrating clear vertical differentiation of the persistent DOM pool. At the whole-reservoir scale, RDOC accounted for 23.0% of the resolved DOM pool during OTS, corresponding to an estimated concentration of 1.21 mg L⁻¹, while LDOC remained relatively abundant at 3.65 mg L⁻¹. During mixing (MX), RDOC remained quantitatively important, with an estimated concentration of 1.26 mg L⁻¹ and an increased relative contribution of 46.2%, whereas LDOC decreased by more than 90%, from 3.65 to 0.30 mg L⁻¹. We therefore interpret the high RDOC proportion during MX cautiously as being consistent with persistence and redistribution of RDOC accumulated during the preceding stratified period, together with continued microbial processing during overturn, rather than as evidence of enhanced de novo RDOC production during mixing (L737-740, L769-773).
We also strengthened the quantitative interpretation by incorporating the revised layer-specific correlation analyses. After BH-FDR correction, RDOC remained positively associated with water temperature in the thermocline (ρ = 0.886, FDR = 0.033) and negatively associated with FI (ρ = -0.850, FDR = 0.048). In the hypolimnion, the negative association between pH and RDOC remained significant (ρ = -0.905, FDR = 0.040). These results indicate that statistically robust RDOC associations were layer-specific rather than uniform throughout the water column. By contrast, most cohesion-DOM relationships did not remain significant after multiple-testing correction or environmental adjustment, and we have therefore revised the Discussion to avoid interpreting microbial cohesion as an independent driver of RDOC accumulation.
In addition, we retained comparison with previous experimental work conducted in the same DLD Reservoir. That degradation experiment showed that microbial RDOC conversion efficiency during thermal stratification was approximately fourfold higher than during mixing and reached its maximum within the thermocline during OTS. This previous process-based evidence complements the present field observations, in which the thermocline also exhibited the most robust environment-RDOC relationships after multiple-testing correction (L740-748).
Accordingly, the revised Section 4.1 now integrates direct quantitative evidence for RDOC concentration and relative contribution, vertical and temporal variation in the RDOC pool, statistically robust layer-specific associations, and independent process-based evidence from previous experiments. We believe that these additions provide substantially stronger quantitative support for the discussion of RDOC dynamics while avoiding overinterpretation of microbial network associations as direct causal effects.
Reviewer comment 5:
Additional data-supported evidence is recommended for this section.
Response 5:
We thank the reviewer for this constructive suggestion. We agree that the original discussion in Section 4.1 relied too heavily on ecological interpretation and did not sufficiently integrate quantitative evidence from our dataset. We have therefore revised this section to incorporate representative hydrochemical, RDOC, and statistically supported microbial-environment relationships.
Specifically, we now provide quantitative evidence for the vertical physicochemical gradient established during obvious thermal stratification (OTS). Dissolved oxygen (DO) decreased from 8.62 mg L⁻¹ in the epilimnion to 1.42 mg L⁻¹ in the hypolimnion, demonstrating pronounced vertical hydrochemical stratification. Correspondingly, RDOC concentrations were 1.33 mg L⁻¹ in the epilimnion, 1.25 mg L⁻¹ in the thermocline, and 1.11 mg L⁻¹ in the hypolimnion, providing direct evidence for vertical differentiation of the persistent carbon pool (L663-668).
We also incorporated the results of the revised layer-specific correlation analysis. After global BH-FDR correction, RDOC remained significantly associated with water temperature in the thermocline (ρ = 0.886, FDR = 0.033) and negatively associated with FI (ρ = -0.850, FDR = 0.048), while a significant negative association between pH and RDOC persisted in the hypolimnion (ρ = -0.905, FDR = 0.040). In contrast, several nominal associations detected in the epilimnion did not remain significant after multiple-testing correction. We have therefore revised the discussion to distinguish statistically robust relationships from uncorrected correlations and to avoid overinterpreting microbial or network associations as direct causal effects.
In addition, the revised text now integrates the observed associations between keystone-taxon abundance and hydrochemical conditions, together with their positions within microbial interaction networks and predicted functional profiles. These data are used to support the interpretation that stratification-associated environmental gradients covary with microbial community organization and RDOC dynamics, while explicitly recognizing that network position and functional prediction do not constitute direct evidence of metabolic activity or causal regulation.
These additions substantially strengthen the data-supported linkage among thermal stratification, hydrochemical heterogeneity, microbial reorganization, and RDOC variability in Section 4.1, while keeping the interpretation consistent with the statistical strength and limitations of the available evidence.
Reviewer comment 6:
Lines 639: two redundant spaces appear before “RDOC”.
Response 6:
We sincerely thank the reviewer for the careful reading and pointing out this typographical error. We have removed the two extra spaces before “RDOC” in L639 (now L726 in the revised manuscript). The relevant sentence has been corrected accordingly.
Reviewer comment 7:
The font size of text elements in Figure 6 could be enlarged.
Response 7:
We sincerely thank the reviewer for this valuable suggestion to improve the readability of our figures. In accordance with your comment, we have significantly enlarged all text elements in Fig.6 (Now it has been changed to Fig. 7), including the axis titles, tick labels, legend entries, and any annotations within the figure.
Reviewer comment 8:
In Section 4.2, the authors argue that benthic release is not the dominant driver of RDOC accumulation, based on three observations: first, RDOC concentrations in the hypolimnion are not markedly higher than in the epilimnion, contradicting the vertical pattern expected for sediment‑dominated inputs; second, optical indices show low HIX values in hypolimnetic waters, whereas sediment‑derived dissolved organic matter typically exhibits strong humic signatures; third, FI shows no significant correlation with RDOC in most water layers and is even negatively correlated within the thermocline, suggesting that the RDOC pool is not dominated by sediment‑borne allochthonous signals. This is an interesting finding. Could the authors comment on whether better approaches could be applied to constrain carbon sources and further confirm that RDOC is predominantly autochthonously produced?
Response 8:
We sincerely thank the reviewer for this insightful suggestion. We agree that the vertical distribution of RDOC and fluorescence indices provide only indirect constraints on carbon sources and cannot, by themselves, definitively determine whether the RDOC pool is predominantly autochthonous. We have therefore revised Section 4.2 to make this limitation explicit and to incorporate the additional carbon-pool and stable-isotope relationships available in our dataset.
Several coordinated relationships provide additional support for active autochthonous organic-carbon production within the water column. DIC was strongly negatively correlated with DOC, indicating that lower inorganic-carbon concentrations coincided with greater accumulation of dissolved organic carbon. DIC was also strongly negatively correlated with δ¹³CDIC. Because autotrophic carbon fixation preferentially assimilates ¹²C, enhanced biological utilization of DIC can reduce DIC concentrations while leaving the residual DIC pool relatively enriched in ¹³C (Jiang et al., 2020; Farmer et al., 2021). In addition, δ¹³CDIC was positively correlated with LDOC, indicating that conditions associated with stronger biological DIC utilization also coincided with greater availability of labile dissolved organic carbon. Together, these relationships are consistent with biological utilization of inorganic carbon and concurrent production of autochthonous organic substrates.
However, we now emphasize that these natural-abundance isotope relationships should be interpreted cautiously. δ¹³CDIC can also be influenced by carbonate weathering, respiration, air-water CO₂ exchange, carbonate speciation, and hydrodynamic mixing. Therefore, the observed DIC-DOC, DIC-δ¹³CDIC, and δ¹³CDIC-LDOC relationships support the plausibility of autochthonous carbon production, but they do not constitute direct isotope tracing of carbon transfer from DIC into RDOC.
We have also revised the interpretation of the microbial-network evidence. Positive cohesion was associated with greater LDOC availability, whereas stronger negative cohesion coincided with LDOC depletion and higher RDOC levels in the whole-dataset patterns. Nevertheless, these cohesion relationships were period dependent and were not uniformly retained after multiple-testing correction or environmental adjustment. Accordingly, we now interpret bacterial cohesion as a community-level descriptor that covaries with DOM composition and environmental conditions, rather than as direct evidence that microbial interactions quantitatively convert LDOC into MCP-derived RDOC.
The revised source-attribution argument therefore rests on several mutually consistent, but indirect, observations: RDOC did not show pronounced enrichment in hypolimnetic waters relative to the epilimnion; HIX remained relatively low in the hypolimnion; FI showed no significant relationship with RDOC in most water layers and was negatively associated with RDOC in the thermocline; carbon-pool and stable-isotope relationships were consistent with biological DIC utilization and autochthonous substrate production; and microbial community organization covaried with changes in the relative distribution of labile and more persistent DOM fractions. These observations argue against a simple benthic-dominance scenario, but they do not exclude sedimentary or terrestrial contributions.
We agree with the reviewer that more direct approaches would be required to constrain RDOC sources more rigorously in the future. In particular, controlled ¹³C-labeling experiments could directly trace the transfer of inorganic carbon into phytoplankton-derived DOC and subsequently into more persistent DOM fractions. Molecular-level DOM characterization, such as high-resolution mass spectrometry, could further distinguish autochthonous and terrestrially or sediment-derived molecular signatures. Direct measurements of sediment porewater chemistry, benthic diffusive fluxes, sediment resuspension, and water-column particle fluxes would also help quantify possible benthic contributions. Combining these approaches with controlled microbial degradation experiments would provide a much stronger basis for distinguishing between newly produced autochthonous RDOC, redistributed pre-existing RDOC, and sediment-derived organic carbon. The revised content can be found in Section 4.2.
Reviewer comment 9:
Could you further clarify the difference between the biological carbon pump (BCP) and the microbial carbon pump (MCP)? In addition, is the occurrence of RDOC strongly influenced by its degradability?
Response 9:
We sincerely thank the reviewer for this important comment. We agree that the distinction between the biological carbon pump (BCP) and microbial carbon pump (MCP), as well as the meaning of recalcitrance in RDOC, was not sufficiently clarified in the original manuscript. We have therefore revised the relevant sections (Section 4.4) to explicitly distinguish these two mechanisms and to clarify the controls on RDOC persistence.
The BCP and MCP represent distinct but interconnected mechanisms of aquatic carbon sequestration. The BCP primarily describes the fixation of inorganic carbon by phytoplankton and other autotrophs and the subsequent transfer of biologically fixed carbon through particulate settling and, to some extent, dissolved organic carbon export. Thus, the BCP is principally associated with the production and physical redistribution of biologically fixed carbon. In contrast, the MCP primarily describes microbial transformation of dissolved organic matter. Through selective uptake, metabolic transformation, biomass turnover, and release of microbial products, microorganisms convert a fraction of labile or semi-labile DOM into compounds that exhibit reduced biological availability and longer environmental persistence. Therefore, unlike the BCP, the MCP does not necessarily require vertical particle transport; instead, it promotes carbon storage by increasing the persistence of carbon within the dissolved organic carbon pool. These two processes can nevertheless be closely coupled. Organic carbon initially produced through autotrophic fixation provides an important substrate source for microbial metabolism. Phytoplankton exudation, cell lysis, grazing, and solubilization of particulate organic matter transfer part of this biologically fixed carbon into the DOM pool, which is subsequently processed by heterotrophic microorganisms. Thus, the BCP can provide organic substrates for the MCP, whereas the MCP influences the ultimate fate of this carbon by determining whether it is rapidly remineralized or retained as relatively persistent RDOC.
We also agree with the reviewer that RDOC persistence is closely related to its degradability. However, we would like to emphasize that “recalcitrance” should not necessarily be interpreted as an absolute and immutable molecular property. Current understanding indicates that DOC persistence is controlled jointly by intrinsic molecular characteristics and extrinsic environmental conditions. Some RDOC compounds may persist because their molecular structures are intrinsically resistant to microbial degradation, whereas others may be potentially degradable but remain unused because their concentrations are below microbial uptake thresholds or because appropriate microbial degraders, nutrients, enzymatic capabilities, or favorable physicochemical conditions are absent. Consequently, the apparent recalcitrance of DOM is partly context dependent. This distinction is particularly relevant to the Dalongdong Reservoir. In our study, persistent phosphorus limitation associated with Ca–P co-precipitation, together with thermal-stratification-induced gradients in nutrient availability and microbial community composition, may constrain microbial access to and utilization of certain DOM components. Meanwhile, microbial transformation can generate or selectively preserve compounds with relatively low bioavailability. Therefore, the observed RDOC accumulation should be interpreted as the combined outcome of microbial transformation, molecular properties, and environmental constraints on further degradation, rather than as evidence that all RDOC molecules are intrinsically non-degradable.
References
Azam F., Malfatti F., 2007. Microbial structuring of marine ecosystems. Nature Reviews Microbiology. 5 (10): 782-791.
Farmer J. R., Hertzberg J. E., Cardinal D., Fietz S., Hendry K., Jaccard S. L., Paytan A., Rafter P. A., Ren H., Somes C. J., Sutton J. N., 2021. Assessment of C, N, and Si Isotopes as Tracers of Past Ocean Nutrient and Carbon Cycling. Global Biogeochemical Cycles. 35 (7): 1-28
Hamilton S. K., Bruesewitz D. A., Horst G. P., Weed D. B., Sarnelle O., 2009. Biogenic calcite–phosphorus precipitation as a negative feedback to lake eutrophication. Canadian Journal of Fisheries and Aquatic Sciences. 66 (2): 343-350.
Han Y., He H., Liu Z., Lai C., Ma Z., Liu X., Li D., Shao M., Cao W., Li H., Hao P., Zhao Y., Xu H., Li Y., Yin L., 2025. Attributes of karst lakes in sustaining net autotrophy and carbon sink effects. Journal of Hydrology. 647: 132404.
Jiang Y., Lei J., Hu L., Xiao Q., Wang J., Zhang C., Ali H., 2020. Biogeochemical and physical controls on the evolution of dissolved inorganic carbon (DIC) and δ13CDIC in karst spring-waters exposed to atmospheric CO2(g): Insights from laboratory experiments. Journal of Hydrology. 583: 124294.
Jiao N., Herndl G. J., Hansell D. A., Benner R., Kattner G., Wilhelm S. W., Kirchman D. L., Weinbauer M. G., Luo T., Chen F., Azam F., 2010. Microbial production of recalcitrant dissolved organic matter: long-term carbon storage in the global ocean. Nature Reviews Microbiology. 8: 593-599.
Jiao N., Luo T., Chen Q., Zhao Z., Xiao X., Liu J., Jian Z., Xie S., Thomas H., Herndl G. J., Benner R., Gonsior M., Chen F., Cai W.-J., Robinson C., 2024. The microbial carbon pump and climate change. Nature Reviews Microbiology. 22 (7): 408-419.
Legendre L., 2024. Jigsaw puzzle of the interwoven biologically-driven ocean carbon pumps. Progress in Oceanography. 229: 103338.
Ogawa H., Amagai Y., Koike I., Kaiser K., Benner R., 2001. Production of Refractory Dissolved Organic Matter by Bacteria. Science. 292 (5518): 917-920.
Otsuki A., Wetzel R. G., 1972. Coprecipitation of phosphate with carbonates in a marl lake. . Limnology and Oceanography. 17 (5): 763-767.
Sun H., Han C., Liu Z., Wei Y., Ma S., Bao Q., Zhang Y., Yan H., 2022. Nutrient limitations on primary productivity and phosphorus removal by biological carbon pumps in dammed karst rivers: Implications for eutrophication control. Journal of Hydrology. 607: 1-13.
Thornton D. C. O., 2014. Dissolved organic matter (DOM) release by phytoplankton in the contemporary and future ocean. European Journal of Phycology. 49 (1): 20-46.
Zhang W., Wang W., Xu S., Sun Q., Shi W., Man J., Yu S., Yang Y., Wu W., Hu X., Wu Q., Wu P., Li S.-L., 2025. Effectively mitigated eutrophication risk by strong biological carbon pump (BCP) effect in karst reservoirs. Water Research. 278: 123395.
Citation: https://doi.org/10.5194/egusphere-2026-4596-AC1
-
AC1: 'Reply on RC1', Junbing Pu, 14 Oct 2026
-
RC2: 'Comment on egusphere-2026-4596', Anonymous Referee #2, 30 Sep 2026
This manuscript describes the effects of thermal stratification on the spatiotemporal dynamics of dissolved organic matter (DOM) components, bacterial community structure and interactions, and recalcitrant/refractory dissolved organic carbon (RDOC) in the DLD Reservoir, with emphasis on the microbial carbon pump (MCP), biological carbon pump (BCP), and Ca–P co-precipitation in a karst reservoir. The topic is potentially interesting and relevant to understanding carbon sequestration in inland waters. However, the manuscript currently has the following issues:
- The terms “recalcitrant dissolved organic carbon (RDOC)” and “refractory dissolved organic carbon” should be unified throughout the manuscript. The title and main text use different expressions. In general, the latter is preferred for inland waters, whereas the former is used for soils and sediments.
- On what basis were the “incubating thermal stratification (ITS)” and “obvious thermal stratification (OTS)” periods classified? If thermal stratification is classified dichotomously, it should be mixing versus stratification. If classified by degree, light, moderate, and strong stratification need to be defined quantitatively; which category does “obvious” belong to? The classification principle for “incubating” differs from that for degree. Why are they compared together, and which degree category would it fall into? If the classification criteria differ, how can they be compared?
- The first two sentences of the Abstract should reflect the importance of temperature or thermal stratification for RDOC; the transition to thermal stratification is abrupt.
- Line 38: The full name of DOM is not provided at first mention. Line 42: The full name of Ca–P is not provided at first mention.
- The Abstract is too long and should be condensed.
- Many statements in the Abstract are discussion content rather than research results, and the evidence for causal mechanisms is insufficient. Please remove or soften overly strong statements.
- Terms such as CO2, DIC, and BCP are not spelled out at first mention. Please check the entire manuscript.
- The BCP mechanism is not clearly explained.
- Lines 80–82 do not clearly explain how the mechanisms differ between karst and non-karst regions.
- Line 86: Is “autotrophic organic carbon” a recognized technical term? Please use a more conventional term.
- Line 97: “dissolved organic carbon” needs an abbreviation. Please check the use of abbreviations and full names throughout the manuscript.
- Lines 102–112: The many references are redundant and are not organized around the same question; they appear to be a disorganized collection of MCP papers. What common issue do these references address? Are they intended to show that the microbial species responsible for MCP differ in karst waters, or that MCP is affected by environmental differences? A paragraph should describe one unresolved issue, and environmental effects and microorganisms should be separated. In addition, the scientific question addressed by this study is not clearly stated.
- The environmental, climatic, and nutrient characteristics of karst systems are not described.
- Lines 121–123: Is 38% of global inland waters a large contribution? It seems that rivers may contribute more than lakes/reservoirs.
- Lines 127–134 contain no references. References should be added after the cited content, not only at the end of a paragraph.
- Lines 130–131: It should be clearly stated how karst and non-karst areas differ.
- Line 139: “particularly elevated alkalinity and Ca–P co-precipitation” is mentioned abruptly and has not been introduced previously. Background information on this aspect should be added.
- Lines 137–143: The scientific question and hypotheses are not adequately motivated. It should be clearly explained why the combined effects of elevated alkalinity, Ca–P co-precipitation, and physical stratification on MCP need to be studied. In addition, how the functions of BCP and MCP are defined and how they are coupled is unclear. These descriptions are qualitative; they need to be made quantitative and specific. If the study only analyzes relationships, this content may be better placed in the Discussion.
- Lines 143–148: The research tasks of this study cannot fully address the scientific question or test the scientific hypotheses.
- Lines 149–162: The detailed description of reservoir characteristics should be moved to the Methods section.
- Lines 163–172: The work conducted in the specific reservoir can be condensed together with Lines 143–148 in the Introduction, and some content can be moved to the Methods section.
- Are there only three monitoring stations? Is this representative enough of the reservoir? The names are D3, D5, and D8. If there are only three, why not call them D1, D2, and D3?
- Line 221: On what basis was the 2.5 m interval chosen?
- Line 223: The manufacturer and country of the instruments should be provided.
- Insufficient justification for thermal stratification staging and layer division may bias inter-layer comparisons. The authors divided the DLD Reservoir into ITS, OTS, and MX periods and used fixed layers of 0–5 m, 2.5–7.5 m, and ≥7.5 m as the surface, thermocline, and bottom layers. However, the thermocline ranged from 1.58–6.02 m in April and 1.53–10 m in August, indicating large seasonal differences in thermocline position and thickness. Fixed-depth division may mix thermocline samples with hypolimnion samples and affect subsequent comparisons of DOM and bacterial communities. It is recommended to stratify samples according to the actual thermocline boundaries of each profile, or at least provide validation of the fixed-layer approach and state the number of samples and replicates for each period and water layer.
- DOM component classification, PARAFAC validation, and concentration conversion assumptions need clarification. The authors divided components into LDOC, SLDOC, and RDOC, but the classification criteria are not given in the Results. Different combinations of C1–C5 were resolved by PARAFAC in different periods. It should be explained whether these components are consistent across periods, whether split-half validation was performed, and how Ex/Em peaks correspond to known DOM components. More importantly, calculating LDOC/RDOC/SLDOC concentrations as “fluorescence intensity contribution × DOC concentration” assumes equal fluorescence quantum yields among components, that all DOC is fluorescent DOM, and that fluorescence contribution represents carbon mass proportion. The results should be expressed as “fluorescence-estimated concentrations” or “relative concentrations,” and the formula and uncertainty should be clearly stated.
- The sequencing data volume and diversity description need verification, and figure labels are inconsistent with the main text. The statement that “411,458 high-quality bacterial OTUs” were obtained is unusually high; in amplicon studies, the number of OTUs/ASVs is typically much lower, and this may actually refer to reads or features. Please verify the units and provide sequencing depth, quality-control workflow, number of replicates, and rarefaction strategy. In addition, the Fig. 2f/g caption refers to “top 10/top 20 bacterial genera and species levels,” whereas Section 3.3.1 discusses the phylum level and Section 3.3.2 discusses the genus level; the figures and text should be made consistent.
- Network analysis and keystone taxon identification lack key parameters and robustness tests. The authors classified nodes into Module hubs, Connectors, Network hubs, and Peripherals based on Zi–Pi, but the Results do not explain how the networks were constructed, whether they were based on OTU tables, correlation thresholds, module detection methods, random network comparisons, or p-value correction. The number of network nodes differs greatly among periods, and direct comparison of keystone taxon numbers may be affected by network size. Please provide network construction parameters, numbers of nodes/edges, number of modules, randomization tests, and sensitivity analyses of keystone taxon identification to demonstrate robustness.
- Correlation analyses are overinterpreted as causal mechanisms. The manuscript extensively uses Spearman correlations and cohesion regressions and interprets the results as “promoting RDOC production,” “inhibiting LDOC formation,” and “promoting RDOC accumulation through the MCP effect.” However, correlation is not causation, and confounding factors such as temperature, nutrients, dissolved oxygen, and Chl-a were not controlled. The MCP effect is also not defined or directly measured. It is recommended to change the wording to “correlated/associated,” add multiple-comparison correction (e.g., FDR), use stratified regression or mixed-effects models, and if necessary use structural equation modeling to distinguish direct and indirect effects.
- The U-shaped/linear relationships between cohesion and DOM may be confounded by period. The U-shaped relationship between positive cohesion and RDOC has its left side mainly from MX and right side mainly from OTS, which likely reflects period differences rather than continuous variation in the same mechanism. Network structures differ among periods, and cohesion values may not be directly comparable. It is recommended to perform stratified regression by period or use mixed-effects models with period as a random effect, and to report sample size, confidence intervals, and outlier influence. The simultaneous roles of positive and negative cohesion as promoting factors for RDOC also require explanation of collinearity and ecological meaning.
- Environmental factors and Bayesian regression models need diagnostics and justification for variable selection. Fig. 4 involves many environmental, spectral, and diversity indices, which are strongly collinear (e.g., WT, DO, pH, pCO₂, HCO₃⁻, Chl-a, FI, HIX, β:α, Simpson). It should be explained which variables were included in the Bayesian linear regression, whether they were standardized, how priors were specified, whether VIF was checked, and how posterior distributions and model fit were evaluated. R² values of 0.530 and 0.655 are reported, but overfitting risk and cross-validation are not. Please add model diagnostics, variable selection procedures, and uncertainty of effect sizes.
- Conclusions about “DOM bioavailability” should be cautious because direct degradation experiments are lacking. The title and Discussion address DOM bioavailability, but the Results are mainly based on fluorescence component classification and concentration estimation, without direct BDOC measurements or DOM degradation incubations. Whether LDOC/SLDOC/RDOC are equivalent to bioavailable fractions needs justification. The current conclusion that “keystone taxa are more closely associated with RDOC production and persistence than with LDOC/SLDOC generation and degradation” is premature. It is recommended to restrict the conclusions to associations at the fluorescence-component level and discuss the limitations of extrapolation from a single reservoir and three seasons.
- Writing, terminology, and figures/tables need comprehensive revision. There are spelling errors such as “Incabating thermal stractification” and “DLD Reservoirwas”; terms such as TS, ITS, OTS, MX, and thermal stratification period are used inconsistently; full names should be given when RDOC/LDOC/SLDOC first appear. The captions of Fig. 2, Fig. 3, and Table 1 need to be consistent with the classification levels in the main text. Please standardize terminology, proofread the text, and ensure that each conclusion corresponds to specific figures/tables and statistical results.
- Mechanistic inferences are too strong; correlation analyses are directly elevated to causal mechanisms. Sections 4.1–3 frequently use causal expressions such as “drives,” “regulates,” “contributes directly,” “governed by,” and “determines,” but the supporting evidence is mainly network analysis, Spearman correlations, RDA, and Bayesian regression. For example, “keystone taxa directly contribute to RDOC formation,” “positive cohesion reflects the MCP mechanism,” and “Ca-P co-precipitation drives BCP to MCP transition” all lack direct experimental or process evidence. It is recommended to add structural equation modeling, laboratory degradation experiments, isotope labeling, or metagenomic/metatranscriptomic evidence to distinguish direct effects from indirect associations.
- The concepts and quantitative indicators of MCP/BCP are unclear, and whether RDOC is equivalent to MCP products needs clarification. The manuscript takes MCP as a core mechanism but does not clearly define how it is measured: Are RDOC fluorescence components sufficient to represent MCP products? How is MCP efficiency quantified? Are there independent indicators for the conversion of BCP to MCP? In Section 4.2, “about 20% of biomass can be converted to persistent RDOC” is directly cited from the literature and not verified in this study. It is recommended to clarify the definitions and boundaries of MCP and BCP, add RDOC molecular composition, degradation indices, incubation experiments, or stable isotope evidence, and avoid equating “fluorescence-estimated RDOC” directly with “refractory carbon produced by the microbial carbon pump.”
- Statistical models have risks of period confounding and overfitting of quadratic relationships. In Section 4.2, the U-shaped relationship between positive cohesion and RDOC has its left side mainly from MX and right side mainly from OTS, which likely reflects period differences rather than continuous variation in the same mechanism; network structures and cohesion values may not be directly comparable among periods. In Section 4.3, the quadratic relationships of Ca²⁺–N:P and C:P–RDOC have R² of only about 0.395, and the quadratic terms may be driven by a few points. It is recommended to perform stratified regression by period or use mixed-effects models with period as a random effect; report sample size, confidence intervals, outlier sensitivity, and multiple-comparison correction; and explain whether the quadratic terms are ecologically meaningful.
- The P-limitation mechanism in the karst reservoir lacks direct geochemical and physiological evidence. Section 4.3 proposes that Ca-P co-precipitation reduces SRP and increases C:P, thereby inhibiting CUE and promoting RDOC accumulation. However, the manuscript does not provide direct evidence for SRP, Ca-P mineral saturation indices, or HAP precipitation, nor does it measure CUE, extracellular enzyme activity, or microbial P-limitation status. There is tension between TP as a positive predictor of RDOC and the claim that P limitation promotes RDOC accumulation; the difference between total P and bioavailable P needs to be explained. It is recommended to add SRP, Ca2+/SRP relationships, enzyme activities, CUE, or P-addition experiments; otherwise this mechanism can only be described as a correlation, not proven.
- Functional prediction results are overinterpreted as actual metabolic functions. Fig. 6 is based on 16S functional prediction and Spearman correlations and cannot prove that Limnobacter fixes CO₂, Nodosilinea uses CCMs, or Fluviicola participates in RDOC formation. Correlations between taxon abundance and predicted functions only indicate potential functions, not expression or activity. It is recommended to change statements such as “fixes CO₂” and “participates in methane oxidation” to “potential functions” and to add metagenomic, metatranscriptomic, cultivation, or isotope-tracing evidence, or at least clearly state the limitations of functional prediction.
- Extrapolated conclusions are too strong; sampling from a single reservoir in three periods is insufficient to generalize to “karst reservoirs as regional carbon sinks.” Section 4.3 concludes that “karst reservoirs as regional carbon sinks,” but the study is based only on the DLD Reservoir in three periods and lacks multi-reservoir, multi-year, and spatial replication. Network structures, cohesion, and DOM components are highly sensitive to sampling period and reservoir hydrological conditions. It is recommended to limit the conclusions to the DLD Reservoir and discuss uncertainties; generalizing to karst reservoirs would require multi-system comparisons and carbon mass-balance estimates.
Citation: https://doi.org/10.5194/egusphere-2026-4596-RC2 -
AC2: 'Reply on RC2', Junbing Pu, 14 Oct 2026
Response to reviewers’ comments
Ms. NO.: egusphere-2026-4596
Title: The Stratified Microbial Carbon Pump: Thermal Stratification Enhances Refractory Dissolved Organic Carbon Production and Stabilizes Carbon in Alkaline Karst Waters through Keystone Microbial Interaction Networks
Journal: Biogeosciences
Reviewer #2:
Reviewer comment 1:
The terms “recalcitrant dissolved organic carbon (RDOC)” and “refractory dissolved organic carbon” should be unified throughout the manuscript. The title and main text use different expressions. In general, the latter is preferred for inland waters, whereas the former is used for soils and sediments.
Response 1:
We thank the reviewer for pointing out this inconsistency. We have unified the terminology throughout the manuscript to “refractory dissolved organic carbon (RDOC)”, which is more appropriate for inland water systems. We have carefully checked the title, abstract, and main text to ensure consistent usage (L28).
Reviewer comment 2:
On what basis were the “incubating thermal stratification (ITS)” and “obvious thermal stratification (OTS)” periods classified? If thermal stratification is classified dichotomously, it should be mixing versus stratification. If classified by degree, light, moderate, and strong stratification need to be defined quantitatively; which category does “obvious” belong to? The classification principle for “incubating” differs from that for degree. Why are they compared together, and which degree category would it fall into? If the classification criteria differ, how can they be compared?
Response 2:
Thank you for this important comment. We agree that the terminology used in the original manuscript may have caused confusion because “incubating thermal stratification (ITS)” and “obvious thermal stratification (OTS)” could be interpreted as representing different classification principles. In this study, ITS, OTS, and MX were not defined as categorical levels of stratification intensity (e.g., weak, moderate, and strong stratification). Rather, they represent successive seasonal evolutionary phases of the reservoir thermal regime, identified from the temporal development of the vertical temperature structure together with the Schmidt Stability Index (SSI) (Idso and Oceanography, 1973) and the relative water-column stability index (RWCS/H) (Shi et al., 2016).
Specifically, ITS represents the period during which thermal stratification begins to develop. During this phase, vertical temperature gradients become detectable, but the water column has not yet reached a fully developed and persistent stratified state. Correspondingly, SSI and RWCS/H begin to increase as the vertical density structure progressively strengthens. OTS represents the subsequent period when thermal stratification is well established and clearly expressed in the temperature profile, accompanied by relatively high SSI and RWCS/H values. MX represents the period after stratification weakens or collapses, when vertical temperature differences are substantially reduced and water-column mixing is re-established.
Therefore, our classification is based on the seasonal progression of thermal structure from formation, through establishment, to breakdown and mixing, rather than on fixed thresholds of stratification intensity. Although stratification intensity generally increases from ITS to OTS, the terms ITS and OTS describe different stages of the same temporal process rather than independent categories such as “mild” and “strong” stratification. The three phases were compared because they represent sequential thermal states of the same reservoir and were identified using the same set of thermal-profile and stability indicators. This provides a consistent framework for evaluating how changes in thermal structure across the seasonal cycle are associated with variations in physicochemical conditions, microbial communities, and RDOC dynamics. To avoid further ambiguity, we have revised the Methods section to explicitly state that ITS, OTS, and MX are thermal-regime evolutionary phases rather than stratification-intensity classes, and we have clarified the criteria used to identify each phase. The detailed description of the method can be found in Text.S4 of the Supplementary Materials.
Reviewer comment 3:
The first two sentences of the Abstract should reflect the importance of temperature or thermal stratification for RDOC; the transition to thermal stratification is abrupt.
Response 3:
Thank you for this helpful comment. We agree that the role of thermal stratification was introduced too abruptly in the original Abstract. We have therefore revised the opening sentences to explicitly emphasize the importance of thermal stratification in regulating dissolved organic matter dynamics and RDOC persistence in reservoirs. This revision provides a clearer transition from the general background of RDOC to the specific focus of our study on different thermal phases in Dalongdong Reservoir (L27-30).
Reviewer comment 4:
Line 38: The full name of DOM is not provided at first mention. Line 42: The full name of Ca–P is not provided at first mention.
Response 4:
Thank you for pointing out these errors. After the revision of the abstract, “dissolved organic matter (DOM)” appeared for the first time in the abstract. Therefore, we did not make any changes to DOM in the original text at L27 (the revised article is at L35). Additionally, at L41, we added the full name of Ca-P – “calcium-phosphorus (Ca-P)”, so that it appeared in the text in its full form for the first time, enhancing the readability of the article.
Reviewer comment 5:
The Abstract is too long and should be condensed.
Response 5:
Thank you for this suggestion. We have substantially shortened the Abstract by removing redundant background information, reducing extended mechanistic interpretation, and retaining only the essential study objectives, major results, and principal implications. The revised Abstract is therefore more concise and focused on the core findings of the study (L27-45).
Reviewer comment 6:
Many statements in the Abstract belong to the Discussion rather than the Results, and the evidence for causal mechanisms may differ between karst and non-karst regions.
Response 6:
We appreciate this important comment. We agree that several statements in the previous version of the Abstract were overly interpretive and extended beyond the direct evidence provided by our dataset. We have therefore revised the Abstract to distinguish more clearly between observed results and mechanistic interpretation. In particular, causal expressions such as “resulting in,” “regulating,” and “driving” were replaced where appropriate with more evidence-based wording such as “was associated with,” “coincided with,” and “covaried with.” We also removed broader claims implying that the observed mechanisms are necessarily specific to karst systems. Instead, the revised Abstract now states that the relationships observed in Dalongdong Reservoir provide evidence for understanding carbon persistence in carbonate-rich inland waters, while emphasizing that further comparative studies are needed to determine whether these relationships differ between karst and non-karst aquatic systems (L27-45).
Reviewer comment 7:
Terms such as CO2, DIC, and BCP are not spelled out at first mention. Please check the entire manuscript.
Response 7:
Thank you for raising this question. I apologize for omitting the full names of the three abbreviations, namely CO2, DIC and BCP, in my article. I have now corrected this and added the full names to the places where these abbreviations first appeared. Please refer to L50, L51 and L80 of the article for details.
Reviewer comment 8:
The BCP mechanism is not clearly explained.
Response 8:
Thank you for this important comment. We agree that the BCP mechanism was not sufficiently explained in the original manuscript. We have therefore revised this section to clarify the complete pathway of the BCP, from photosynthetic fixation of dissolved inorganic carbon (DIC) to the production of autochthonous organic carbon (AOC), its subsequent partitioning into dissolved and particulate organic carbon pools, and its retention or export through processes such as particle settling and organic matter preservation.
We have also clarified how this pathway may be modified in karst waters by carbonate chemistry and nutrient availability. Specifically, photosynthetic uptake of CO₂ and HCO₃⁻ can increase pH and promote CaCO₃ precipitation, which may enhance phosphorus removal through CaCO₃-associated co-precipitation and thereby alter nutrient limitation of the BCP. These revisions provide a clearer mechanistic link among DIC utilization, AOC production, organic carbon retention, and nutrient regulation in karst aquatic systems. Please refer to pages L65-89 (Karst aquatic systems……BCP-mediated carbon fixation and retention) for details.
Reviewer comment 9:
Lines 80–82 do not clearly explain how the mechanisms differ between karst and non-karst regions.
Response 9:
Thank you for this important comment. We agree that the original text did not clearly distinguish the fundamental BCP mechanism from the environmental factors that regulate it in karst versus non-karst aquatic systems.
In the revised manuscript, we now clarify that the core BCP mechanism itself is not unique to karst waters: in both karst and non-karst systems, inorganic carbon is fixed by primary producers and transferred into autochthonous organic carbon (AOC), which may subsequently enter dissolved or particulate organic carbon pools and contribute to carbon retention or export. The key difference lies in the geochemical environment regulating this process. Karst waters are commonly characterized by high DIC, Ca²⁺, alkalinity, and pH. Under these conditions, photosynthetic uptake of CO₂ and HCO₃⁻ can further increase pH and promote CaCO₃ precipitation. Phosphate may then be removed through CaCO₃-associated co-precipitation and particle settling, potentially intensifying phosphorus limitation and altering the efficiency of the BCP.
We have revised L 80–82 (Now referring to L 83 to L89: In karst waters, abundant DIC……BCP-mediated carbon fixation and retention of the text) accordingly and also expanded the explanation of the BCP pathway to clarify both the common mechanism and the distinctive geochemical regulation occurring in karst waters.
Reviewer comment 10:
Line 86: Is “autotrophic organic carbon” a recognized technical term? Please use a more conventional term.
Response 10:
Thank you for pointing this out. We agree that, in this context, "autotrophic organic carbon" is not an appropriate standard term. The intended term should be "autochthonous organic carbon (AOC)", i.e. organic carbon produced through in situ primary production within the aquatic system. This term has been widely adopted in research on the carbon cycle in aquatic systems, including previous studies on karst water bodies and reservoirs. Following revisions throughout the paper, AOC now appears earlier in the text. Consequently, in the revised manuscript (L84-85), we have replaced "autotrophic organic carbon" with "autochthonous organic carbon".
Reviewer comment 11:
Line 97: “dissolved organic carbon” needs an abbreviation. Please check the use of abbreviations and full names throughout the manuscript.
Respense 11:
Thank you for your correction. The term “dissolved organic carbon” at L115 was an oversight on our part—we forgot to include the abbreviation, and we have now added DOC.
Reviewer comment 12:
Lines 102–112: The many references are redundant and are not organized around the same question; they appear to be a disorganized collection of MCP papers. What common issue do these references address? Are they intended to show that the microbial species responsible for MCP differ in karst waters, or that MCP is affected by environmental differences? A paragraph should describe one unresolved issue, and environmental effects and microorganisms should be separated. In addition, the scientific question addressed by this study is not clearly stated.
Response 12:
Thank you for this insightful comment. We agree that the original paragraph contained too many references and did not clearly organize them around a common scientific question. In the revised manuscript, we have substantially reorganized this section and reduced the number of examples.
Specifically, we now separate the existing literature into two distinct unresolved issues. The first concerns how environmental conditions, such as hydrological variability, nutrient availability, and thermal stratification, influence MCP-related carbon transformation and RDOC accumulation. The second concerns the microbial mechanisms involved, including the roles of specific taxa, community composition, and taxon–taxon interactions in RDOC formation.
We have also clarified the central scientific question addressed by this study. The revised text now focuses on how environmental stratification and nutrient constraints interact with microbial community organization to regulate RDOC dynamics in a karst reservoir. Accordingly, the literature cited in this section is now used to support these two specific knowledge gaps rather than being presented as a general collection of MCP-related studies (L121-143: Recent studies have demonstrated that microbial processes are……in karst reservoirs).
Reviewer comment 13:
The environmental, climatic, and nutrient characteristics of karst systems are not described.
Response 13:
Thank you for this helpful comment. We agree that the previous revision focused primarily on carbonate chemistry and did not sufficiently describe the broader environmental, climatic, and nutrient context of karst aquatic systems.
We have therefore expanded this section to clarify these characteristics. Specifically, we now describe the high DIC, Ca²⁺, alkalinity, and pH that commonly characterize carbonate-rich waters; the pronounced seasonal variability in temperature and precipitation typical of subtropical karst regions, which can influence hydrological connectivity, thermal stratification and mixing, and external material inputs; and the distinctive nutrient dynamics of karst waters, particularly the strong geochemical control on phosphorus through CaCO₃-associated adsorption and co-precipitation.
These additions clarify that BCP efficiency in karst systems is not controlled by carbonate chemistry alone, but by the combined effects of hydrochemical conditions, climatic seasonality, hydrological variability, and nutrient availability. We have revised the paragraph accordingly to provide a more complete environmental context for interpreting BCP regulation in karst reservoirs (L65-75: Karst aquatic systems……with CaCO₃ and subsequently removed through particle settling).
Reviewer comment 14:
Lines 121–123: Is 38% of global inland waters a large contribution? It seems that rivers may contribute more than lakes/reservoirs.
Response 14:
Thank you for this important comment. We agree that the original wording may have overstated the relative contribution of lakes and reservoirs to global inland-water carbon emissions. In particular, more recent global estimates indicate that CO₂ evasion from streams and rivers exceeds that from lakes and reservoirs (Raymond et al., 2013). We have therefore removed the statement that lakes and reservoirs account for approximately 38% of aquatic carbon emissions and no longer describe them as major contributors on the basis of this percentage.
Instead, we now emphasize their broader roles in carbon transformation, storage, burial, and atmospheric exchange. The revised sentence reads: “Lakes and reservoirs are important components of inland aquatic ecosystems and play substantial roles in carbon transformation, storage, burial, and atmospheric exchange, although global CO₂ evasion from streams and rivers is estimated to exceed that from lakes and reservoirs (Tranvik et al., 2009; Raymond et al., 2013).” For further details, see L144–147 of the text.
Reviewer comment 15:
Lines 127–134 contain no references. References should be added after the cited content, not only at the end of a paragraph.
Response 15:
Thank you for pointing this out. We agree that the references in the original version were placed too generally at the end of the paragraph, which made it unclear which specific statements they supported.
In the revised manuscript, we have redistributed the references so that each citation now follows the specific statement it supports. References concerning carbonate-rich hydrochemistry and in situ primary production are placed immediately after those statements, while references related to thermal stratification, vertical physicochemical gradients, microbial community restructuring, and carbon transformation are cited separately at the corresponding locations (L150-173).
Reviewer comment 16:
Lines 130–131: It should be clearly stated how karst and non-karst areas differ.
Response 16:
Thank you for this important comment. We agree that the original statement was too general and did not specify how carbon cycling in karst aquatic systems differs from that in non-karst systems. We have therefore revised this part to describe the differences more explicitly.
In the revised manuscript, we clarify that karst reservoirs are commonly characterized by carbonate-rich hydrochemistry, high concentrations of dissolved inorganic carbon and Ca²⁺, and strong interactions between carbonate precipitation, nutrient availability, and biological carbon fixation. These conditions can enhance the contribution of in situ primary production to organic carbon formation and can also affect phosphorus availability through CaCO₃-associated removal. In contrast, many non-karst lakes and reservoirs are less strongly influenced by carbonate weathering and may depend more heavily on catchment-derived organic matter and external nutrient inputs. We have therefore replaced the previous general statement with a more specific description of these hydrochemical and biogeochemical differences, while avoiding the implication that all karst and non-karst systems follow the same contrasting pattern (L150-166: “Unlike many non-karst lakes……by restricting vertical water-column mixing (Pu et al., 2020; Li et al., 2021)”).
Reviewer comment 17:
Line 139: “particularly elevated alkalinity and Ca–P co-precipitation” is mentioned abruptly and has not been introduced previously. Background information on this aspect should be added.
Response 17:
Thank you for this helpful comment. We agree that the discussion of increased alkalinity and Ca–P co-precipitation appeared too abruptly in the original manuscript because the relevant geochemical background had not been sufficiently introduced beforehand.
To address this concern, we have added background information in the Introduction describing the coupling between the biological carbon pump (BCP), carbonate precipitation, and phosphorus removal in karst waters. Specifically, we now explain that intensive photosynthetic uptake of CO₂ and HCO₃⁻ can increase pH and promote CaCO₃ precipitation, during which phosphate may co-precipitate with CaCO₃ and subsequently be removed through particle settling, thereby reducing bioavailable phosphorus and potentially intensifying P limitation (Otsuki and Wetzel, 1972; Hamilton et al., 2009). We also added evidence that impoundment in karst river–reservoir systems can shift BCP limitation from carbon toward nitrogen or phosphorus (Sun et al., 2022), and that enhanced BCP activity may further promote CaCO₃-associated phosphorus removal in karst reservoirs (Zhang et al., 2025; Han et al., 2025). Detailed revisions can be found from L93 to L103: “Intensive photosynthetic uptake of CO₂ and HCO₃⁻ further increases……than acting independently in carbonate-rich waters” in the text.
Reviewer comment 18:
Lines 137–143: The scientific question and hypotheses are not adequately motivated. It should be clearly explained why the combined effects of elevated alkalinity, Ca–P co-precipitation, and physical stratification on MCP need to be studied. In addition, how the functions of BCP and MCP are defined and how they are coupled is unclear. These descriptions are qualitative; they need to be made quantitative and specific. If the study only analyzes relationships, this content may be better placed in the Discussion.
Response 18:
Thank you for this important comment. We agree that the original hypothesis was stated too strongly relative to the evidence provided by our dataset. In particular, our study does not directly quantify the functional coupling between the BCP and MCP, nor does it experimentally demonstrate a causal pathway linking alkalinity, Ca–P co-precipitation, thermal stratification, and MCP activity.
We have therefore revised section 1 (L174-187) substantially. First, we now provide a clearer motivation for examining these factors together. Seasonal thermal stratification generates vertical gradients in temperature, oxygen, and nutrient availability, whereas the carbonate-rich geochemical environment of karst reservoirs can influence phosphorus availability and C stoichiometry through Ca-associated phosphorus removal. These processes may jointly modify the physicochemical conditions under which microbial carbon transformation occurs.
Second, we have replaced the original causal hypothesis with two explicit research questions focusing on (1) temporal variation in RDOC across thermal phases and (2) the relationships between RDOC dynamics and physicochemical gradients, phosphorus availability, C stoichiometry, microbial community structure, and microbial interactions.
Third, because our current data do not allow direct quantitative assessment of BCP–MCP functional coupling, we no longer present such coupling as a tested hypothesis in the Introduction. Instead, its potential mechanistic significance is discussed separately in Section 4.4, where we distinguish the roles of the BCP and MCP and evaluate their possible interaction based on the observed relationships and previous studies. We believe this revision better aligns the stated objectives with the scope and strength of the available evidence.
Reviewer comment 19:
Lines 143–148: The research tasks of this study cannot fully address the scientific question or test the scientific hypotheses.
Response 19:
Thank you for this important comment. We agree that the original research tasks were not fully aligned with the scientific questions and implied stronger mechanistic testing than supported by our data. We have therefore revised the study objectives to focus on RDOC variations across thermal phases and their associations with physicochemical conditions, phosphorus availability, C stoichiometry, and microbial community organization. The corresponding research tasks were also simplified to match these questions directly. Potential BCP–MCP coupling is no longer presented as a directly tested causal mechanism but is discussed separately in Section 4.4 based on the observed relationships. Please refer to L 184 to 187 (We further evaluated whether keystone taxa……carbon processing and carbon persistence in inland waters) in the text for details.
Reviewer comment 20:
Lines 149–162: The detailed description of reservoir characteristics should be moved to the Methods section.
Response 20:
Thank you for this helpful suggestion. We agree that the detailed characteristics of the DLD Reservoir are more appropriate in the methodological description than in the Introduction. We have therefore moved the relevant information to the Supplementary Materials (Text S4) and integrated it with the description of the thermal stratification indices and thermal-regime classification.
In the revised Text S4, the hydrological and hydrochemical characteristics of the reservoir are now presented together with the vertical temperature profiles, Schmidt Stability Index (SSI), RWCS/H, and the criteria used to distinguish the ITS, OTS, and MX phases. This integration provides a more coherent methodological context for understanding how the specific characteristics of the DLD Reservoir relate to the seasonal development and breakdown of thermal stratification.
Reviewer comment 21:
Lines 163–172: The work conducted in the specific reservoir can be condensed together with Lines 143–148 in the Introduction, and some content can be moved to the Methods section.
Response 21:
Thank you for this helpful suggestion. We agree that the original Introduction contained overlapping descriptions of the study objectives and analytical approaches. Therefore, we condensed the content of the L143 to 148 and L163 to 172 of the previous text into a single paragraph, which is presented in Supplementary Materials (Text S5, L280–303), with a focus on the core research questions and objectives.
Methodological details, including high-frequency thermal monitoring and microbial co-occurrence network analysis, have been moved to the Methods section. We also removed overly broad mechanistic and climate-warming statements that were not directly tested by the present dataset. The revised Introduction now more clearly distinguishes the scientific questions from the analytical methods and avoids repetition.
Reviewer comment 22:
Are there only three monitoring stations? Is this representative enough of the reservoir? The names are D3, D5, and D8. If there are only three, why not call them D1, D2, and D3?
Response 22:
Thank you for this important comment. We apologize that the original description may have caused confusion regarding the number and naming of the sampling stations.
D3, D5, and D8 do not indicate that only three stations were originally established in the reservoir. These labels are retained from the existing sampling-site numbering system used for Dalongdong Reservoir. In the present study, three representative stations (D3, D5, and D8) were selected from the broader reservoir monitoring network rather than renumbered as D1, D2, and D3, in order to maintain consistency with previous sampling records and facilitate comparison among different monitoring campaigns.
The three stations were selected to represent the major spatial characteristics of the reservoir, including differences in hydrological position and water-column conditions. In addition, our study focused primarily on vertical variations in dissolved organic matter, microbial communities, and related biogeochemical processes during different sampling periods. Therefore, intensive vertical sampling was conducted at each station, providing substantially greater resolution along the water column than would be indicated by the number of surface sampling locations alone.
We agree that three stations cannot capture all possible small-scale spatial heterogeneity within the reservoir. The station names D3, D5, and D8 have been retained because they correspond to the established site identifiers used in the reservoir monitoring system.
Reviewer comment 23:
Line 221: On what basis was the 2.5 m interval chosen?
Response 23:
Thank you for this important comment. The 2.5 m vertical sampling interval was selected based on both previous investigations of the thermal structure and carbon biogeochemistry of the DLD Reservoir and the need to resolve major vertical gradients in the present study. Previous studies in the same reservoir have demonstrated pronounced depth-dependent changes in water temperature, pH, conductivity, DO, Chl-a, DIC, and δ¹³CDIC during thermal stratification, with several physicochemical and biological features changing markedly within the upper several meters of the water column (Li et al., 2021; Wu et al., 2017). More recent work on carbon transformation in the DLD Reservoir further showed that, during summer stratification, the epilimnion generally occupies the upper approximately 2.5 m, while the thermocline extends through roughly the 2.5–7.5 m depth range, with clear differences in DIC, DOC, POC, and associated biogeochemical processes among the epilimnion, thermocline, and hypolimnion (Su et al., 2026).
On this basis, a 2.5 m sampling interval was considered sufficiently fine to capture the major hydrochemical and carbon-isotope gradients associated with thermal stratification while maintaining consistent vertical coverage among sampling sites and periods. The interval therefore was not chosen solely for logistical convenience, but was informed by the vertical structure previously documented for the DLD Reservoir. At the same time, we acknowledge that the exact position and thickness of the thermocline varied among sampling periods. Accordingly, the fixed 2.5 m sampling resolution should be regarded as an approximation for resolving the principal water-column gradients rather than as a precise delineation of instantaneous thermocline boundaries. We have clarified this rationale and limitation in the revised Methods.
Reviewer comment 24:
Line 223: The manufacturer and country of the instruments should be provided.
Response 24:
We thank the reviewer for this request. The PONSEL brand of water quality analysis instruments is manufactured by PONSEL MESURE, which is part of the AQUALABO Group. The company is based in Caudan, France.
The full manufacturer details are as follows:
Manufacturer: PONSEL MESURE – Groupe AQUALABO
Country of Origin: France
We have added this information to the revised manuscript (see L236-240).
Reviewer comment 25:
Insufficient justification for thermal stratification staging and layer division may bias inter-layer comparisons. The authors divided the DLD Reservoir into ITS, OTS, and MX periods and used fixed layers of 0–5 m, 2.5–7.5 m, and ≥7.5 m as the surface, thermocline, and bottom layers. However, the thermocline ranged from 1.58–6.02 m in April and 1.53–10 m in August, indicating large seasonal differences in thermocline position and thickness. Fixed-depth division may mix thermocline samples with hypolimnion samples and affect subsequent comparisons of DOM and bacterial communities. It is recommended to stratify samples according to the actual thermocline boundaries of each profile, or at least provide validation of the fixed-layer approach and state the number of samples and replicates for each period and water layer.
Response 25:
Thank you for this important comment. We agree that the thermocline position and thickness varied seasonally and that the fixed depth intervals used in our study do not exactly coincide with the thermocline boundaries at every sampling period.
Our vertical classification should therefore be regarded as an operational approximation based on the 2.5 m vertical resolution of the water-sampling design, rather than as an exact representation of the instantaneous thermocline boundaries. Water samples for DOM and microbial analyses were collected at discrete 2.5 m intervals, whereas thermocline boundaries were derived from the continuous temperature profiles. Consequently, it was not possible to assign every biological or chemical sample precisely to the dynamically changing thermocline boundaries. To maintain a consistent basis for comparison among sampling periods, we therefore used fixed depth intervals to represent the upper, transition, and lower water-column zones. These depth classes were selected with reference to the observed temperature profiles and were intended to capture the major vertical structure of the reservoir rather than to define the thermocline with high spatial precision. The seasonal development of stratification itself was evaluated independently using the vertical temperature profiles, Schmidt Stability Index (SSI), RWCS/H, and thermocline boundaries.
We acknowledge that this approximation may introduce some overlap between transition-layer and lower-layer samples, particularly when the thermocline expands or shifts vertically. We have now clarified this limitation in the Text.S5 (L331-337) and emphasize that the fixed-depth classification was used primarily to provide a consistent comparative framework across thermal phases, rather than to represent exact thermocline positions.
Reviewer comment 26:
DOM component classification, PARAFAC validation, and concentration conversion assumptions need clarification. The authors divided components into LDOC, SLDOC, and RDOC, but the classification criteria are not given in the Results. Different combinations of C1–C5 were resolved by PARAFAC in different periods. It should be explained whether these components are consistent across periods, whether split-half validation was performed, and how Ex/Em peaks correspond to known DOM components. More importantly, calculating LDOC/RDOC/SLDOC concentrations as “fluorescence intensity contribution × DOC concentration” assumes equal fluorescence quantum yields among components, that all DOC is fluorescent DOM, and that fluorescence contribution represents carbon mass proportion. The results should be expressed as “fluorescence-estimated concentrations” or “relative concentrations,” and the formula and uncertainty should be clearly stated.
Response 26:
Thank you for your suggestion.
(1) The classification criteria for LDOC, SLDOC, and RDOC during each sampling period are described in detail in Text S3 and Table S1 of the Supplementary Materials. A corresponding cross-reference has also been added in Section 3.2 of the main text (L431, 440).
(2) The identified DOM components differed among sampling periods. Detailed descriptions and visualizations are provided in Fig. S2 and Text S3. Three fluorescent DOM components (C1, C2, and C3) were identified during the ITS period; four components (C1, C3, C4, and C5) were identified during the OTS period; and three components (C1, C2, and C4) were identified during the MX period. Based on their classification, C1 was assigned to the SLDOC fraction, C2, C3, and C5 to the LDOC fraction, and C4 to the RDOC fraction. All fluorescent components identified by parallel factor analysis (PARAFAC) were validated using split-half analysis. Because split-half validation is a critical step in ensuring the reliability of the PARAFAC results, this procedure has been specifically emphasized in Section 3.2 of the main text (L419).
(3) The assignment of Ex/Em peaks to known DOM components was based on comparison with the OpenFluor database together with the classification of the corresponding Ex/Em peak characteristics. This point has been clarified in Section 3.2 of the revised manuscript (L424–427).
(4) We agree that, owing to instrumental limitations, three-dimensional excitation–emission matrix fluorescence spectroscopy alone cannot identify all dissolved organic matter (DOM) components in water samples. However, previous studies have shown that dissolved organic carbon (DOC) constitutes a substantial proportion of DOM, with approximately 50% of DOM represented by DOC (Moody and Worrall, 2017). Consequently, the terms DOM and DOC are often used interchangeably in related studies, and in this context, fluorescent DOM components were used as proxies for DOC fractions. More detailed characterization of DOM at the molecular level will require techniques such as Fourier transform ion cyclotron resonance mass spectrometry (FT-ICR MS) in future studies. Accordingly, the concentrations of LDOC, SLDOC, and RDOC are reported as “fluorescence-estimated concentrations”, and the uncertainty associated with this estimation approach has been added to Section 3.2 of the main text (L432–440) and Text S3.
Reviewer comment 27:
The sequencing data volume and diversity description need verification, and figure labels are inconsistent with the main text. The statement that “411,458 high-quality bacterial OTUs” were obtained is unusually high; in amplicon studies, the number of OTUs/ASVs is typically much lower, and this may actually refer to reads or features. Please verify the units and provide sequencing depth, quality-control workflow, number of replicates, and rarefaction strategy. In addition, the Fig. 2f/g caption refers to “top 10/top 20 bacterial genera and species levels,” whereas Section 3.3.1 discusses the phylum level and Section 3.3.2 discusses the genus level; the figures and text should be made consistent.
Response 27:
Thank you for this careful and important comment. We rechecked the original sequencing and bioinformatic workflow and found that the previous statement of “411,458 high-quality bacterial OTUs” was incorrect. The value referred to sequence reads rather than taxonomic features. Moreover, because the sequencing data were processed in QIIME2 using the Deblur algorithm, the resulting feature table represents amplicon sequence variants (ASVs), rather than conventional OTUs generated by similarity-based clustering. We have therefore corrected the terminology throughout the manuscript.
The sequencing and quality-control workflow have now been described in greater detail. Briefly, the V4 region of bacterial 16S rRNA genes was amplified using primers 515F and 806R and sequenced on an Illumina NovaSeq 6000 platform using PE250 chemistry. Paired-end reads were merged using FLASH (v1.2.11), demultiplexed according to barcode sequences using Sabre, and processed in QIIME2 (v2020.2). Sequences with an average quality score below Q30, sequences shorter than 200 bp, and sequences containing ambiguous bases were removed. Deblur was then used for sequence denoising and chimera removal, followed by construction of the ASV feature table. Taxonomy was assigned using a Naïve Bayes classifier trained on the SILVA 138 database. Each sample was amplified in three technical PCR replicates. The PCR products were pooled in equal amounts before library construction to reduce amplification-related variability. Before diversity analyses, sequencing libraries were normalized by rarefaction to the sequencing depth of the sample with the lowest retained read count, ensuring comparable sequencing effort among samples. Alpha- and beta-diversity analyses were subsequently conducted using the rarefied feature table. After rechecking the feature tables, 4,210, 6,531, and 28,066 unique ASVs were detected during the ITS (April), OTS (August), and MX (November) periods, respectively, after excluding control samples and considering only samples included in the present study. These totals represent the union of ASVs detected within each sampling period and should not be interpreted as direct measures of between-period alpha diversity because the number and vertical coverage of samples differed among periods.
The larger number of ASVs detected during the MX period may also partly reflect the ecological consequences of complete water-column mixing. During stratification, bacterial taxa can occupy distinct vertical niches, whereas overturn redistributes microorganisms among previously separated water layers. As a result, individual samples collected during MX may contain taxa originating from a broader range of former depth-specific niches, potentially increasing the number of detectable ASVs. Consequently, the revisions in the main text are at L444-445 in 3.3.1 and in the methods section 2.2.3 (L332-358).
In addition, we rechecked Fig. 2f/g and the corresponding Sections 3.3.1 and 3.3.2. The taxonomic labels and captions have been revised so that the phylum- and genus-level descriptions are fully consistent between the figures and the main text (L412-413).
Reviewer comment 28:
Network analysis and keystone taxon identification lack key parameters and robustness tests. The authors classified nodes into Module hubs, Connectors, Network hubs, and Peripherals based on Zi–Pi, but the Results do not explain how the networks were constructed, whether they were based on OTU tables, correlation thresholds, module detection methods, random network comparisons, or p-value correction. The number of network nodes differs greatly among periods, and direct comparison of keystone taxon numbers may be affected by network size. Please provide network construction parameters, numbers of nodes/edges, number of modules, randomization tests, and sensitivity analyses of keystone taxon identification to demonstrate robustness.
Response 28:
Thank you for this valuable suggestion. We agree that the original manuscript did not provide sufficient methodological detail regarding the network construction procedure and did not explicitly report the results of robustness tests. We would like to clarify that, in the revised manuscript, we retained the same network construction procedure and the same Zi–Pi-based method for identifying keystone taxa as used in the original manuscript, without modifying the primary analytical framework. The additional analyses described below were conducted solely to improve methodological transparency and to evaluate the robustness of the original network results, rather than to redefine the original networks or keystone taxa.
Microbial co-occurrence networks were constructed at the ASV level. Prior to network construction, low-abundance and low-frequency ASVs were removed based on their mean relative abundance and occurrence frequency. Pairwise associations among the retained ASVs were calculated using Spearman’s rank correlation with pairwise complete observations. P values were corrected for multiple comparisons using the Benjamini–Hochberg procedure, and only associations with |ρ| ≥ 0.60 and adjusted p ≤ 0.001 were retained as network edges. The resulting networks were treated as undirected weighted co-occurrence networks. Network modules were identified using the fast-greedy community detection algorithm implemented in the igraph package. Node topological roles were classified according to their Zi and Pi values using the same criteria as in the original manuscript: module hubs (Zi > 2.5 and Pi < 0.62), connectors (Zi < 2.5 and Pi > 0.62), network hubs (Zi > 2.5 and Pi > 0.62), and peripherals. Module hubs, connectors, and network hubs were regarded as putative keystone taxa. Importantly, neither these classification thresholds nor the original Zi–Pi calculation procedure was changed during revision.
As requested, we further supplemented the major network topological parameters for the three sampling periods, including the numbers of nodes and edges, numbers of positive and negative associations, mean degree, average path length, network diameter, network density, clustering coefficient, number of modules, and modularity. These results are now provided in Supplementary Table.S3. The three networks differed markedly in size and connectivity. The April, August, and November 2023 networks contained 352, 549, and 432 nodes and 6,403, 34,564, and 6,558 edges, respectively. A total of 12, 3, and 15 modules were identified in the three periods, with modularity values of 0.257, 0.249, and 0.250, respectively. Thus, although network size and connectivity differed substantially among sampling periods, all three networks exhibited clear modular organization.
We also agree with the reviewer that direct comparison of the absolute numbers of keystone taxa among networks of different sizes may be affected by network size. Therefore, while retaining the keystone taxa originally identified by the Zi–Pi analysis for each period, we additionally calculated the proportion of keystone taxa relative to the total number of nodes in each network:
Keystone proportion (%) = number of keystone taxa / total number of network nodes × 100.
The results showed that 23, 16, and 11 putative keystone taxa were identified in April, August, and November 2023, corresponding to 6.53%, 2.91%, and 2.55% of the total network nodes, respectively. Notably, the August network was the largest network, containing 549 nodes, yet both the absolute number and the relative proportion of keystone taxa were lower than those in April. The November network contained 432 nodes, but the number of keystone taxa further decreased to 11. Therefore, the temporal differences in keystone-taxon numbers cannot be explained simply by differences in network size. In the revised manuscript, we now report both the absolute number and the normalized proportion of keystone taxa and avoid interpreting temporal differences solely on the basis of absolute counts.
To further evaluate whether the observed network structures could have arisen by chance, we additionally generated 999 randomized networks for each empirical network while preserving the node degree distribution and compared their major topological properties with those of the empirical networks. The results showed that all three empirical networks had markedly higher clustering coefficients and modularity than the corresponding randomized networks. For example, the clustering coefficient of the April network was 0.567, compared with a mean value of 0.365 for the randomized networks, while the empirical modularity was 0.257 compared with a randomized mean of 0.093. For the August network, the corresponding clustering coefficients were 0.670 and 0.477, and the modularity values were 0.249 and 0.045. For the November network, the corresponding values were 0.555 and 0.341 for the clustering coefficient and 0.250 and 0.103 for modularity. These differences were significant in all three periods (empirical P = 0.002). Thus, the empirical networks consistently exhibited significantly greater clustering and modular organization than expected under randomized network structure, indicating that the observed network architecture was not simply generated by random connectivity.
In addition, we performed supplementary sensitivity analyses using alternative correlation and significance thresholds. The original network-filtering criterion (|ρ| ≥ 0.60 and adjusted P ≤ 0.001) was retained as the primary analytical criterion, while alternative thresholds were used only to evaluate whether the overall network structure and Zi–Pi-based keystone-taxon identification were overly dependent on a single filtering criterion. Robustness was evaluated by comparing network size, module structure, keystone-taxon proportion, Jaccard similarity among keystone-taxon sets, and the retention rate of keystone taxa relative to the primary network.
The sensitivity analyses showed that the major network-level patterns were generally stable, whereas the exact identities of Zi–Pi-defined keystone taxa displayed some period-dependent sensitivity to increasingly stringent correlation thresholds. In the April network, increasing the correlation threshold from |ρ| ≥ 0.60 to 0.65 and 0.70 did not change the network size or keystone-taxon composition; both the Jaccard similarity and keystone-taxon retention rate remained 1.00, indicating high stability under moderate changes in the correlation threshold. In contrast, in the August and November networks, increasing the correlation threshold removed part of the network edges and nodes and resulted in some changes in the composition of Zi–Pi-defined keystone taxa. When the threshold was further increased to |ρ| ≥ 0.80, all three networks became substantially sparser and changes in keystone-taxon composition became more pronounced. We therefore do not interpret the sensitivity analyses as evidence that all keystone taxa are completely stable under all threshold settings. Rather, the results indicate that the major network-level organizational features are generally robust, whereas the identities of specific Zi–Pi-defined keystone taxa may vary under more stringent filtering conditions, particularly in the August and November networks.
In summary, we retained the original network construction procedure and Zi–Pi-based keystone-taxon identification method in the revised manuscript, while adding the complete network construction parameters, network topological properties for all three sampling periods, network-size-normalized keystone-taxon proportions, 999 random-network tests, and threshold-sensitivity analyses. These additional analyses showed that all three empirical networks exhibited significant non-random clustering and modular organization, and that the temporal differences in keystone-taxon abundance were not simply determined by network size. Additionally, we have also replaced Fig. 5 and Table. S3 in the original text. And the method section 2.2.4 in the text has been revised (from L361 to L382).
Reviewer comment 29:
Correlation analyses are overinterpreted as causal mechanisms. The manuscript extensively uses Spearman correlations and cohesion regressions and interprets the results as “promoting RDOC production,” “inhibiting LDOC formation,” and “promoting RDOC accumulation through the MCP effect.” However, correlation is not causation, and confounding factors such as temperature, nutrients, dissolved oxygen, and Chl-a were not controlled. The MCP effect is also not defined or directly measured. It is recommended to change the wording to “correlated/associated,” add multiple-comparison correction (e.g., FDR), use stratified regression or mixed-effects models, and if necessary use structural equation modeling to distinguish direct and indirect effects.
Response 29:
Thank you for this important comment. We agree that the original manuscript overinterpreted several correlation- and cohesion-based relationships as causal mechanisms. Because this study is observational and did not directly manipulate microbial interactions or quantify MCP activity or efficiency, we have revised the manuscript to consistently use association-based terms such as “associated with”, “correlated with”, and “covaried with”, rather than causal expressions such as “promoted” or “inhibited”. The MCP is now discussed only as a possible interpretive framework.
To address the statistical concerns, we re-analysed the relationships among environmental variables, DOM fractions, microbial cohesion, and key bacterial taxa. First, all Spearman correlation analyses were subjected to Benjamini–Hochberg FDR correction, and only associations remaining significant after correction were treated as robust. Second, where sample size permitted, cohesion–DOM relationships were further evaluated after controlling for WT, TN, TP, DO, and Chl-a. For subsets with insufficient observations, we avoided overparameterized models and restricted interpretation to FDR-corrected associations.
Third, we added stratified and mixed-model analyses for key bacterial taxa and DOM fractions across the epilimnion, thermocline, hypolimnion, and mixed period. These results showed that microbial–DOM relationships were not consistent across all strata and periods, supporting a condition-dependent rather than universal causal interpretation.
We also considered structural equation modelling (SEM). However, the present sample size and replication were insufficient for a statistically stable and identifiable SEM relative to the number of environmental, microbial, and DOM variables. We therefore did not force an SEM, but instead used the more conservative combination of FDR-corrected correlation analysis, environmentally adjusted regression where feasible, and stratified/mixed-model analyses. Accordingly, causal wording has been removed, multiple-comparison correction and additional hierarchical analyses have been added, and MCP is now described as a possible mechanism rather than a directly measured process. See section 4.1 in the main text for detailed revisions.
Reviewer comment 30:
The U-shaped/linear relationships between cohesion and DOM may be confounded by period. The U-shaped relationship between positive cohesion and RDOC has its left side mainly from MX and right side mainly from OTS, which likely reflects period differences rather than continuous variation in the same mechanism. Network structures differ among periods, and cohesion values may not be directly comparable. It is recommended to perform stratified regression by period or use mixed-effects models with period as a random effect, and to report sample size, confidence intervals, and outlier influence. The simultaneous roles of positive and negative cohesion as promoting factors for RDOC also require explanation of collinearity and ecological meaning.
Response 30:
Thank you for this important comment. We agree that the apparent U-shaped relationship between positive cohesion and RDOC in the pooled data could be confounded by differences among sampling periods, because MX samples mainly occupied the lower cohesion range whereas OTS samples mainly occupied the higher range.
We therefore re-analysed the relationship using period-stratified regressions and a period-adjusted model in which positive cohesion was centered within each period and Period was included as a fixed effect. The analysis included 40 observations (MX, n = 15; ITS, n = 12; OTS, n = 13). After accounting for period effects, the within-period linear association was modest (β = 4.657, 95% CI: 0.072–9.242, P = 0.0467) and was no longer significant after BH correction (adjusted P = 0.187). The quadratic term was not significant (P = 0.425; 95% CI: −195.32 to 84.12) and did not improve model fit. Period-stratified and leave-one-out analyses likewise did not support a consistent nonlinear response. We therefore removed the previous interpretation of a general U-shaped mechanism and now describe the positive cohesion–RDOC relationship as period dependent. We also clarified that positive and negative cohesion are network-level descriptors of microbial association structure rather than independent causal drivers of RDOC. Their relationships with RDOC are therefore interpreted as reflecting different modes of community organization under contrasting environmental conditions, rather than as simultaneous “promoting” effects. Potential collinearity between cohesion metrics was also considered in the revised interpretation.
Sample sizes, confidence intervals, influence diagnostics, and multiple-testing corrections have been added to the revised manuscript and Supplementary Information (Section 3.3.3 and Text. S2).
Reviewer comment 31:
Environmental factors and Bayesian regression models need diagnostics and justification for variable selection. Fig. 4 involves many environmental, spectral, and diversity indices, which are strongly collinear (e.g., WT, DO, pH, pCO₂, HCO₃⁻, Chl-a, FI, HIX, β:α, Simpson). It should be explained which variables were included in the Bayesian linear regression, whether they were standardized, how priors were specified, whether VIF was checked, and how posterior distributions and model fit were evaluated. R² values of 0.530 and 0.655 are reported, but overfitting risk and cross-validation are not. Please add model diagnostics, variable selection procedures, and uncertainty of effect sizes.
Response 31:
Thank you for this important comment. We agree that the original manuscript did not provide sufficient information regarding predictor selection, multicollinearity, prior specification, model diagnostics, effect uncertainty, or the potential risk of overfitting. We therefore reanalyzed both the LDOC and RDOC Bayesian regression models using a more explicit hypothesis-driven variable-selection and validation procedure.
Positive and negative bacterial cohesion (Pos and Neg) were specified a priori as core variables because they represent the microbial interaction metrics directly related to our study hypothesis. Environmental, optical, and diversity variables were treated as candidate covariates. Before model fitting, continuous predictors were standardized to zero mean and unit variance, allowing posterior coefficients to be compared on a common scale. Pairwise Spearman correlations and variance inflation factors (VIFs) were then examined to evaluate multicollinearity. Rather than allowing an automatic selection procedure to remove the hypothesis-driven cohesion variables, Pos and Neg were retained as prespecified core variables, whereas highly collinear auxiliary covariates were removed iteratively. This procedure excluded pH, DO, PD, and DIC from the final models because of high multicollinearity. The final full models therefore included Pos, Neg, WT, Chl-a, TP, TN, Simpson, HIX, FI, β:α, and SR.
Because Pos and Neg remained strongly collinear with the broader predictor set (final VIFs ≈18.94 and 20.89, respectively), we now explicitly acknowledge that their individual coefficients should be interpreted cautiously. For LDOC, the posterior estimates for Pos and Neg were 0.197 (95% CrI: −1.510 to 1.896) and −0.589 (95% CrI: −2.392 to 1.231), respectively. For RDOC, the corresponding estimates were 0.260 (95% CrI: −0.287 to 0.823) and 0.377 (95% CrI: −0.209 to 0.961). Thus, although the directions of some posterior associations were consistent with the pairwise correlation patterns, the 95% credible intervals overlapped zero. We therefore no longer interpret Pos or Neg as statistically well-resolved independent drivers of LDOC or RDOC.
Both Bayesian models were fitted using stan_lm with an R² prior specified as R2(location = 0.5, what = "mean"), providing moderate model-level regularization. MCMC convergence was evaluated using R-hat statistics, effective sample sizes, and trace plots. For the RDOC model, all R-hat values were approximately 1.00 and effective sample sizes were large, indicating satisfactory convergence. Posterior predictive checks were also conducted to evaluate agreement between observed and posterior-predicted distributions.
Model explanatory power and predictive robustness are now reported separately. The revised LDOC model had a Bayesian R² of 0.525 (95% CrI: 0.331–0.684), whereas its leave-one-out cross-validated R² (LOO-R²) was 0.402 (95% CrI: 0.145–0.583). The revised RDOC model had a Bayesian R² of 0.539 (95% CrI: 0.339–0.698) and a LOO-R² of 0.415 (95% CrI: 0.176–0.610). The reduction from in-sample Bayesian R² to LOO-R² indicates appreciable predictive uncertainty and is now explicitly considered when interpreting model performance.
We additionally compared the full models with reduced core models containing only Pos and Neg. The core models explained 43.0% and 37.9% of the posterior variance in LDOC and RDOC, respectively. However, LOO comparisons showed that the predictive differences between the full and core models were small relative to their uncertainty (LDOC: ΔELPD ≈ 0.4, SE = 4.6; RDOC: ΔELPD ≈ 2.6, SE = 4.5). Therefore, we do not interpret the additional covariates or cohesion metrics as providing unequivocal independent predictive effects.
We have revised Fig. 5 and the corresponding Results and Discussion accordingly. Posterior effect estimates are now presented together with their 95% credible intervals, and terms such as “promotive predictor,” “inhibitory predictor,” and “driving effect” have been replaced by descriptions of posterior association direction and uncertainty where appropriate. The previously reported R² values of 0.530 and 0.655 have also been replaced with the re-estimated Bayesian R² and LOO-R² values from the revised models. The revised content can be found in pages L633-654 of the article. Additionally, the titles of the two corresponding figures, Fig. 5e and Fig. 5f, have also been modified (L626-632). Finally, corresponding modifications were also made in the method section (Text. S2).
Reviewer comment 32:
Conclusions about “DOM bioavailability” should be cautious because direct degradation experiments are lacking. The title and Discussion address DOM bioavailability, but the Results are mainly based on fluorescence component classification and concentration estimation, without direct BDOC measurements or DOM degradation incubations. Whether LDOC/SLDOC/RDOC are equivalent to bioavailable fractions needs justification. The current conclusion that “keystone taxa are more closely associated with RDOC production and persistence than with LDOC/SLDOC generation and degradation” is premature. It is recommended to restrict the conclusions to associations at the fluorescence-component level and discuss the limitations of extrapolation from a single reservoir and three seasons.
Response 32:
Thank you for this important comment. We agree that the original manuscript used the term “DOM bioavailability” too broadly relative to the measurements available in the present study. Our Results are primarily based on EEM-PARAFAC fluorescence-component classification and concentration estimation, and we did not directly quantify BDOC or perform degradation incubations as part of the present dataset. Therefore, the fluorescence-defined LDOC, SLDOC, and RDOC fractions should not be treated as direct measurements of DOM bioavailability.
We have revised the manuscript accordingly. Throughout the Abstract, Results, and Discussion, expressions such as “DOM bioavailability,” “regulation of DOM bioavailability,” and “driving effects on DOM bioavailability” have been replaced with more precise terms such as “fluorescence-defined DOM fractions,” “relative distribution of LDOC, SLDOC, and RDOC fractions,” and “apparent lability and persistence inferred from fluorescence characteristics.”
We have also revised the section titles to avoid implying direct regulation or causal control. Specifically, Sections 3.3.2–3.3.4 now describe statistical associations between keystone taxa, bacterial cohesion, environmental factors, and fluorescence-defined DOM fractions rather than direct effects on DOM bioavailability.
In addition, we have clarified that LDOC, SLDOC, and RDOC are operationally defined fluorescence-based fractions and are not equivalent to directly measured bioavailable, semi-labile, or refractory DOC pools. Direct validation would require BDOC measurements, controlled degradation incubations, or molecular-level DOM characterization. So, we have also reorganized the discussion and conclusion sections.
Finally, we added an explicit limitation that the present conclusions are based on a single reservoir sampled during three thermal phases. Therefore, the observed relationship should be understood as a specific association within a particular location and period in the DLD reservoir, and more extensive inferences will require future verification across multiple reservoirs, over multiple years, and based on process-based approaches. Detailed revisions can be found in Sections 3.3.2, 3.3.3, 3.3.4, Sections Discussion and Conclusion.
Reviewer comment 33:
Writing, terminology, and figures/tables need comprehensive revision. There are spelling errors such as “Incabating thermal stractification” and “DLD Reservoirwas”; terms such as TS, ITS, OTS, MX, and thermal stratification period are used inconsistently; full names should be given when RDOC/LDOC/SLDOC first appear. The captions of Fig. 2, Fig. 3, and Table 1 need to be consistent with the classification levels in the main text. Please standardize terminology, proofread the text, and ensure that each conclusion
Response 33:
Thank you for this careful review. We have thoroughly proofread and revised the manuscript to correct typographical errors and standardize terminology throughout. The full names of RDOC, LDOC, and SLDOC are now provided at first mention, and the use of TS, ITS, OTS, and MX has been made consistent. The captions of Figures 2–3 and Table 1 have also been revised to match the classifications used in the main text. In addition, we checked the Results section to ensure that each major conclusion is supported by the corresponding figure, table, and statistical result.
Reviewer comment 34:
Mechanistic inferences are too strong; correlation analyses are directly elevated to causal mechanisms. Sections 4.1-4.3 frequently use causal expressions such as “drives,” “regulates,” “contributes directly,” “governed by,” and “determines,” but the supporting evidence is mainly network analysis, Spearman correlations, RDA, and Bayesian regression. For example, “keystone taxa directly contribute to RDOC formation,” “positive cohesion reflects the MCP mechanism,” and “Ca-P co-precipitation drives BCP to MCP transition” all lack direct experimental or process evidence. It is recommended to add structural equation modeling, laboratory degradation experiments, isotope labeling, or metagenomic/metatranscriptomic evidence to distinguish direct effects from indirect associations.
Response 34:
Thank you for this important comment. We agree that several expressions in Sections 4.1–4.3 overstated the mechanistic strength of the evidence. The present study is based mainly on co-occurrence networks, Spearman correlations, RDA, and Bayesian regression, which identify statistical associations but cannot by themselves establish direct causal pathways.
We have therefore systematically revised Sections 4.1–4.3 by replacing causal terms such as “drives,” “regulates,” “determines,” and “directly contributes” with more appropriate expressions such as “is associated with,” “covaries with,” “is consistent with,” and “may contribute to.” We also removed statements that treated microbial network properties or Ca–P-related patterns as direct evidence of MCP activity or BCP–MCP transition.
Although structural equation modeling was not appropriate for the current dataset and isotope-labeling or meta-omics measurements were not available from the original sampling campaign, an independent laboratory degradation experiment using samples from the same reservoir has previously provided process-based evidence for microbial transformation and RDOC conversion during thermal stratification (“High carbon fixation during thermal stratification period in a subtropical periodic stratified reservoir: Evidences from RDOC conversion efficiency driven by MCP”). We now cite this study only as complementary experimental support for the plausibility of the proposed mechanisms, rather than as direct causal evidence generated by the present dataset.
We have also clarified that the present manuscript analyzes the environmental and microbial characteristics of the initial reservoir samples before incubation, whereas the previously published study focused on subsequent degradation dynamics and RDOC conversion during incubation. Accordingly, the mechanistic interpretations in the revised Discussion are now explicitly framed as plausible explanations supported by both the observed associations and previous experimental evidence.
Reviewer comment 35:
The concepts and quantitative indicators of MCP/BCP are unclear, and whether RDOC is equivalent to MCP products needs clarification. The manuscript takes MCP as a core mechanism but does not clearly define how it is measured: Are RDOC fluorescence components sufficient to represent MCP products? How is MCP efficiency quantified? Are there independent indicators for the conversion of BCP to MCP? In Section 4.2, “about 20% of biomass can be converted to persistent RDOC” is directly cited from the literature and not verified in this study. It is recommended to clarify the definitions and boundaries of MCP and BCP, add RDOC molecular composition, degradation indices, incubation experiments, or stable isotope evidence, and avoid equating “fluorescence-estimated RDOC” directly with “refractory carbon produced by the microbial carbon pump.”
Response 35:
Thank you for this important and constructive comment. We agree that the original manuscript did not clearly distinguish the conceptual and quantitative boundaries among the biological carbon pump (BCP), the microbial carbon pump (MCP), and refractory dissolved organic carbon (RDOC), and that several statements could be interpreted as equating fluorescence-estimated RDOC directly with MCP products. We have therefore substantially revised Section 4.2. In the revised manuscript, the BCP is defined as autotrophic fixation of DIC and the subsequent transfer of newly fixed autochthonous carbon into dissolved and particulate organic carbon pools. In contrast, the MCP refers to subsequent microbial transformation of organic matter through selective uptake, metabolism, biomass turnover, and release of microbial products, through which a fraction of relatively bioavailable carbon may enter more persistent DOC pools. We now explicitly state that these two processes are conceptually distinct, although they may be ecologically connected.
We also clarify that RDOC is treated in this study as an operationally defined persistent carbon pool rather than as being equivalent to MCP-derived carbon. Accordingly, the fluorescence-estimated RDOC fractions are interpreted as optical proxies of relatively refractory DOM and are not considered direct measurements of MCP products. Bacterial cohesion is likewise treated as an indicator of microbial community organization rather than a direct measure of MCP activity.
The revised Section 4.2 also clarifies the evidence supporting a potential BCP–MCP linkage. The relationships among DIC, δ¹³CDIC, DOC, and LDOC are interpreted as being consistent with biological DIC utilization and autochthonous organic-carbon production, but not as direct isotope tracing of carbon transfer from DIC into RDOC. Similarly, the contrasting relationships of positive and negative bacterial cohesion with LDOC and RDOC are interpreted as association evidence for shifts in carbon bioavailability and persistence rather than as direct proof of LDOC-to-RDOC conversion.
Importantly, MCP efficiency was not directly quantified in the present field study. We therefore removed any implication that the present dataset provides a quantitative measurement of MCP efficiency or direct BCP-to-MCP carbon transfer. A previously published laboratory degradation experiment conducted using samples from the same DLD Reservoir showed that microbial RDOC conversion efficiency during thermal stratification was substantially higher than during mixing and reached its maximum in the thermocline during OTS (Jia et al., 2025). We now use this result only as independent process-based evidence supporting the plausibility of enhanced microbial RDOC conversion under stratified conditions, rather than as proof that the fluorescence-estimated RDOC observed in the present study was exclusively produced by the MCP.
We have also deleted the statement that approximately 20% of microbial biomass can be converted into persistent RDOC. This value is now identified explicitly as a literature-derived estimate rather than a value measured in the present study and is no longer used as quantitative evidence for MCP efficiency in the DLD Reservoir.
Finally, we added an explicit limitation stating that direct verification of MCP-derived RDOC and BCP-to-MCP carbon transfer would require molecular-level DOM characterization, controlled degradation experiments, and stable-isotope tracing approaches. These measurements were not available in the present study. The revised text therefore presents the observed patterns as indirect ecological evidence for microbial involvement in carbon persistence rather than direct proof of MCP-derived RDOC production.
Reviewer comment 36:
Statistical models have risks of period confounding and overfitting of quadratic relationships. In Section 4.2, the U-shaped relationship between positive cohesion and RDOC has its left side mainly from MX and right side mainly from OTS, which likely reflects period differences rather than continuous variation in the same mechanism; network structures and cohesion values may not be directly comparable among periods. In Section 4.3, the quadratic relationships of Ca²⁺–N:P and C:P–RDOC have R² of only about 0.395, and the quadratic terms may be driven by a few points. It is recommended to perform stratified regression by period or use mixed-effects models with period as a random effect; report sample size, confidence intervals, outlier sensitivity, and multiple-comparison correction; and explain whether the quadratic terms are ecologically meaningful.
Response 36:
Thank you for these important comments. We agree that the originally reported nonlinear relationships may have been affected by sampling-period differences and influential observations. We therefore re-analysed the relevant relationships using period-adjusted and period-stratified regressions, together with influence diagnostics, leave-one-out sensitivity tests, and multiple-testing correction.
(1) For positive cohesion–RDOC, the apparent pooled U-shaped pattern was largely associated with differences among periods, with MX and OTS occupying different cohesion ranges. After within-period centering and adjustment for Period, the quadratic term was no longer significant (P = 0.425), whereas only a modest within-period linear association remained (β = 4.657, 95% CI: 0.072–9.242, P = 0.0467; BH-adjusted P = 0.187). Period-stratified and sensitivity analyses likewise did not support a consistent nonlinear response. We therefore now interpret this relationship as period dependent rather than as a universal U-shaped mechanism. (2) Similarly, for Ca²⁺–N:P, the period-adjusted relationship remained significantly positive (n = 38, β = 0.256, 95% CI: 0.161–0.352, P < 0.001; R² = 0.465), supporting a linear rather than quadratic relationship. For C:P–RDOC, the period-adjusted linear association was also significant (n = 27, β = 0.00885, 95% CI: 0.00312–0.01459, P = 0.004; R² = 0.299), whereas the quadratic term was not significant (P = 0.169) and was sensitive to individual observations.
Accordingly, we have revised Sections 3.3.3, 4.2 and 4.3 to emphasize period-adjusted linear or period-dependent associations, removed the previous interpretation of the quadratic terms as general ecological thresholds or universal nonlinear mechanisms, and added sample sizes, confidence intervals, influence diagnostics, leave-one-out results, and multiple-testing corrections.
Reviewer comment 37:
The P-limitation mechanism in the karst reservoir lacks direct geochemical and physiological evidence. Section 4.3 proposes that Ca-P co-precipitation reduces SRP and increases C:P, thereby inhibiting CUE and promoting RDOC accumulation. However, the manuscript does not provide direct evidence for SRP, Ca-P mineral saturation indices, or HAP precipitation, nor does it measure CUE, extracellular enzyme activity, or microbial P-limitation status. There is tension between TP as a positive predictor of RDOC and the claim that P limitation promotes RDOC accumulation; the difference between total P and bioavailable P needs to be explained. It is recommended to add SRP, Ca2+/SRP relationships, enzyme activities, CUE, or P-addition experiments; otherwise this mechanism can only be described as a correlation, not proven.
Response 37:
Thank you for this important comment. We agree that the original version overstated the mechanistic evidence for phosphorus limitation and Ca–P precipitation. Our dataset does not include SRP, Ca–P mineral saturation indices, direct evidence of HAP precipitation, microbial carbon use efficiency, extracellular enzyme activities, or P-addition experiments. Therefore, these processes cannot be considered directly demonstrated by the present study.
We have substantially revised Section 4.3 to distinguish observed relationships from mechanistic interpretation. The Ca²⁺–N:P relationship is now described as being consistent with possible geochemical phosphorus immobilization rather than as direct evidence of Ca–P precipitation. Likewise, the relationships between C:P stoichiometry and RDOC are presented as correlations, while reduced CUE, altered enzyme investment, and microbial P limitation are discussed only as mechanisms proposed by previous studies.
We have also clarified the apparent discrepancy between TP and phosphorus limitation. TP represents the total phosphorus pool and does not distinguish directly bioavailable P from particulate or mineral-associated forms. Because SRP was not measured, the positive TP–RDOC relationship cannot be interpreted as evidence of increased microbial P availability.
Accordingly, we have removed statements implying that Ca–P precipitation, P limitation, reduced CUE, or enzyme suppression were directly proven in the DLD Reservoir. We now present these processes as plausible mechanisms that require further verification using SRP measurements, mineral saturation calculations, enzyme assays, CUE measurements, and nutrient-addition experiments.
Reviewer comment 38:
Functional prediction results are overinterpreted as actual metabolic functions. Fig. 6 is based on 16S functional prediction and Spearman correlations and cannot prove that Limnobacter fixes CO₂, Nodosilinea uses CCMs, or Fluviicola participates in RDOC formation. Correlations between taxon abundance and predicted functions only indicate potential functions, not expression or activity. It is recommended to change statements such as “fixes CO₂” and “participates in methane oxidation” to “potential functions” and to add metagenomic, metatranscriptomic, cultivation, or isotope-tracing evidence, or at least clearly state the limitations of functional prediction.
Response 38:
Thank you for this important comment. We agree that the original manuscript overinterpreted several 16S-based functional predictions as actual metabolic activities. Because Fig. 6 (Now it has been changed to Fig. 7) is based on predicted functions and Spearman correlations, these analyses can only indicate potential functional associations and cannot demonstrate gene expression, metabolic activity, or direct taxon-specific contributions to RDOC formation.
We have therefore systematically revised the relevant text by replacing expressions such as “fixes CO₂,” “participates in methane oxidation,” and “contributes to RDOC formation” with more cautious terms such as “was associated with predicted carbon-fixation functions,” “showed potential methane-oxidation functions,” and “may potentially be involved in RDOC-related carbon transformation.” We have also explicitly stated in the revised Discussion that PICRUSt2/FAPROTAX predictions and taxon–function correlations represent functional potential rather than direct physiological evidence.
The samples used in this study were collected approximately three years ago, and sufficient archived material is no longer available for additional metagenomic, metatranscriptomic, or isotope-tracing analyses. We therefore cannot retrospectively provide these additional lines of evidence. We acknowledge this limitation explicitly in the revised manuscript. However, an independent laboratory degradation experiment using samples from the same DLD Reservoir was previously reported in our study entitled “High carbon fixation during thermal stratification period in a subtropical periodic stratified reservoir: Evidences from RDOC conversion efficiency driven by MCP” (Jia et al., 2025). That experiment demonstrated that microbial RDOC conversion efficiency during thermal stratification was approximately fourfold higher than during mixing and reached its maximum in the thermocline during OTS. We have incorporated this result only as complementary process-based evidence supporting the plausibility of enhanced microbial carbon transformation during stratification. We have not used it to claim that the specific taxa identified by 16S functional prediction directly expressed the predicted pathways.
Accordingly, the revised manuscript now clearly distinguishes between (i) potential functions inferred from 16S-based prediction; (ii) statistical taxon–function associations, and (iii) independent process-based evidence from the previously published degradation experiment. Specific modifications can be found in Section 4.1 of the text.
Reviewer comment 39:
Extrapolated conclusions are too strong; sampling from a single reservoir in three periods is insufficient to generalize to “karst reservoirs as regional carbon sinks.” Section 4.3 concludes that “karst reservoirs as regional carbon sinks,” but the study is based only on the DLD Reservoir in three periods and lacks multi-reservoir, multi-year, and spatial replication. Network structures, cohesion, and DOM components are highly sensitive to sampling period and reservoir hydrological conditions. It is recommended to limit the conclusions to the DLD Reservoir and discuss uncertainties; generalizing to karst reservoirs would require multi-system comparisons and carbon mass-balance estimates.
Response 39:
Thank you for this important comment. We agree that the original conclusion was too broad relative to the spatial and temporal scope of the present study. Our data were obtained from a single karst reservoir during three thermal phases and therefore cannot support a general conclusion that karst reservoirs as a whole function as regional carbon sinks.
We have revised Section 4.3 and the Conclusions accordingly. The revised text now limits the interpretation to the DLD Reservoir and describes the observed patterns as site-specific evidence that carbonate-rich geochemistry, nutrient stoichiometry, and microbial community organization may favor RDOC persistence under particular seasonal conditions.
We have also added an explicit statement that broader extrapolation to karst reservoirs will require multi-reservoir and multi-year comparisons, together with whole-system carbon mass-balance assessments. The references in Section 4.3 have been retained and are now used to support the plausibility of the proposed mechanisms rather than to imply that the present study demonstrates a universal regional carbon-sink function.
References
Hamilton S. K., Bruesewitz D. A., Horst G. P., Weed D. B., Sarnelle O., 2009. Biogenic calcite–phosphorus precipitation as a negative feedback to lake eutrophication. Canadian Journal of Fisheries and Aquatic Sciences. 66 (2): 343-350.
Han C., Wei Y., Sun H., Ma S., Yan H., Bao Q., Zhang Y., Xue Y., He H., 2025. Mechanisms of phosphorus removal by biological carbon pump in karst reservoirs and their role in eutrophication mitigation: Insights from two subtropical karst reservoirs in Guizhou Province, China. Journal of Hydrology. 663: 134130.
Idso S. B. J. L., Oceanography, 1973. On the concept of lake stability1. 18: 681-683.
Jia Y., Su X., Zhang T., He Q., Li J., Pu J., 2025. High carbon fixation during thermal stratification period in a subtropical periodic stratified reservoir: Evidences from RDOC conversion efficiency driven by MCP. Water Research. 278: 123316.
Li J., Pu J., Zhang T., 2021. Transport and transformation of dissolved inorganic carbon in a subtropical groundwater‑fed reservoir, south China. Water Research. 209: 117905.
Moody C. S., Worrall F., 2017. Modeling rates of DOC degradation using DOM composition and hydroclimatic variables. Journal of Geophysical Research: Biogeosciences. 122 (5): 1175-1191.
Otsuki A., Wetzel R. G., 1972. Coprecipitation of phosphate with carbonates in a marl lake. . Limnology and Oceanography. 17 (5): 763-767.
Raymond P. A., Hartmann J., Lauerwald R., Sobek S., McDonald C., Hoover M., Butman D., Striegl R., Mayorga E., Humborg C., Kortelainen P., Dürr H., Meybeck M., Ciais P., Guth P., 2013. Global carbon dioxide emissions from inland waters. Nature. 503 (7476): 355-359.
Shi J.-C., Huang T.-L., Wen G., Liu F., Qiu X.-P., Wang B.-S., 2016. The Variation Characteristic of Sulfides and VOSc in a Source Water Reservoir and Its Control Using a Water-Lifting Aerator. International Journal of Environmental Research and Public Health. 13 (4): 427.
Su X., Qin J., Lai W., Jia Y., Chen P., Zhang T., Li J., Pu J., 2026. Sources and transformations of inorganic and organic carbon in a stratified groundwater-fed reservoir, south China. Journal of Hydrology. 666: 134800.
Sun H., Han C., Liu Z., Wei Y., Ma S., Bao Q., Zhang Y., Yan H., 2022. Nutrient limitations on primary productivity and phosphorus removal by biological carbon pumps in dammed karst rivers: Implications for eutrophication control. Journal of Hydrology. 607: 127480.
Tranvik L. J., Downing J. A., Cotner J. B., Loiselle S. A., Striegl R. G., Ballatore T. J., et al., 2009. Lakes and reservoirs as regulators of carbon cycling and climate. Limnol. Oceanogr. 54: 2298-2314.
Wu F., Pu J., Li J., Zhang T., Li L., Huang S., 2017. Impacts of thermal stratification on the hydrochemistry and dissolved inorganic carbon in a typical karst reservoir in summer. Impacts of thermal stratification on the hydrochemistry and dissolved inorganic carbon in a typical karst reservoir in summer. 38: (8): 3209-3217.
Zhang W., Wang W., Xu S., Sun Q., Shi W., Man J., Yu S., Yang Y., Wu W., Hu X., Wu Q., Wu P., Li S.-L., 2025. Effectively mitigated eutrophication risk by strong biological carbon pump (BCP) effect in karst reservoirs. Water Research. 278: 123395.
Citation: https://doi.org/10.5194/egusphere-2026-4596-AC2
-
RC3: 'Comment on egusphere-2026-4596', Anonymous Referee #3, 04 Oct 2026
General comments
This manuscript investigated how thermal stratification in a karst reservoir influences RDOC accumulation through microbial interactions and geochemical phosphorus limitation. The authors have collected a substantial amount of field data across three thermal phases and combined DOM optical characterization with 16S rRNA-based network analysis. However, the central conclusions, especially the causal role of thermal stratification, the dominance of microbial over benthic processes, and the Ca–P co-precipitation mechanism, are not sufficiently supported by the current data and analyses. The distinction between observed associations and demonstrated mechanisms should also be strengthened throughout the manuscript. I recommend the following major revisions to the manuscript before it can be considered for publication.
1. The main claim that thermal stratification enhances RDOC production and stabilization is not fully convincing. RDOC concentration was slightly higher during the mixing period (1.26 mg/L) than during the obvious thermal stratification period (1.21 mg/L). No RDOC component was resolved in ITS. The proportion is even higher in MX (46.2% vs 23.0%). Moreover, stratification strength differs considerably between April and August (Fig. 2b), yet RDOC is only resolved in August; no relationship between stratification intensity and RDOC can be derived. The authors need to clarify this in the manuscript. More importantly, the manuscript should clarify whether “RDOC” here refers to a fluorescence-defined component or to experimentally demonstrated refractory carbon. These two concepts should not be treated as equivalent.
2. The three sampling periods correspond to April, August, and November, and are simultaneously associated with different thermal structures, temperatures, biological conditions, and seasonal environmental forcing. The observed differences between ITS, OTS, and MX cannot necessarily be attributed specifically to thermal stratification. How did the authors distinguish the effect of thermal stratification from the broader seasonal effect? Please clarify.
3. The manuscript interprets positive cohesion as facilitative interactions and negative cohesion as competitive or inhibitory interactions, and subsequently suggests that competitive interactions promote RDOC accumulation. However, these cohesion metrics are derived from correlations among bacterial taxa and do not directly measure biological interactions. Likewise, a positive or negative association between cohesion and RDOC concentration does not establish that the interaction itself caused RDOC production. For statements such as “competitive interactions promote RDOC production”, please provide more evidence.
Specific comments
L36-37: “predominantly regulated by microbial processes rather than benthic carbon inputs” is too strong. Rephrase as an inference.
L39-41: The network and cohesion analyses mainly indicate associations among taxa. The conclusion that competitive or cooperative interactions promote RDOC formation requires more cautious wording.
L43-45: “resulting in persistent phosphorus limitation” implies causation. Use “may have contributed to” or similar wording.
L98-100: This sentence conflates humic-like fluorescence with recalcitrance. Please revise this statement and provide supporting references.
L348: The identification of keystone taxa appears to be mainly based on network topology. The authors should clarify whether independent functional evidence supports their designation as keystone taxa.
L411-421: RDOC and LDOC concentrations are calculated from fluorescence contribution × DOC. Please give the exact equation and assumptions for converting fluorescence contributions into RDOC/LDOC concentrations. How were different fluorescence efficiencies handled? Please discuss the limitations of this proxy.
L585-587: This is misleading. In the Bayesian models, alpha diversity does not represent bacterial activity and pH does not represent alkalinity.
L694: “The succession of thermal stratification drives shifts in key environmental parameters.” I don’t think “drives” is appropriate here. I don’t see evidence for this in the data presented.
L709-710: RDOC was not detected in ITS, so how can rising positive cohesion be “accompanied by an increase in RDOC”?
L724-727: What is the DIC-enhanced MCP mechanism?
L750-751: RDOC accumulation is independent of benthic input is not sufficiently supported. Sediment porewater, benthic flux, and resuspension were not measured. Please clarify or remove.
L770-774: This sentence is kind of confusing. What does “rate-coupling signature” mean here? I don’t follow how a positive correlation between δ13C-DIC and RDOC can confirm that RDOC originates from water-column internal carbon processing, rather than simply reflecting concurrent changes in the DIC and organic carbon pools.
L813: “RDOC increased significantly along the C:P gradient” is not supported by the presented analysis. The quadratic fit has R2 = 0.385 (Fig. 8), leaving most of the variance unexplained. I don’t see any support for this. I don’t follow this line of reasoning at all, please clarify or remove.
Figure captions should also be revised for clarity. For example, the caption of Fig. 3 is not very clear. Please clarify the effective sample size, i.e., n = 3.
Technical corrections
L214-215, L258, L315, L340: Section numbering is inconsistent (2.2 followed by 2.3.1–2.3.4).
L378: “DLD Reservoirwas” should be “DLD Reservoir was”.
Table 1: “Incabating” should be “Incubating”.
L896: “Origian 2021” should be “Origin 2021”.Citation: https://doi.org/10.5194/egusphere-2026-4596-RC3 -
AC3: 'Reply on RC3', Junbing Pu, 14 Oct 2026
Response to reviewers’ comments
Ms. NO.: egusphere-2026-4596
Title: The Stratified Microbial Carbon Pump: Thermal Stratification Enhances Refractory Dissolved Organic Carbon Production and Stabilizes Carbon in Alkaline Karst Waters through Keystone Microbial Interaction Networks
Journal: Biogeosciences
Reviewer #3:
General comments
This manuscript investigated how thermal stratification in a karst reservoir influences RDOC accumulation through microbial interactions and geochemical phosphorus limitation. The authors have collected a substantial amount of field data across three thermal phases and combined DOM optical characterization with 16S rRNA-based network analysis. However, the central conclusions, especially the causal role of thermal stratification, the dominance of microbial over benthic processes, and the Ca–P co-precipitation mechanism, are not sufficiently supported by the current data and analyses. The distinction between observed associations and demonstrated mechanisms should also be strengthened throughout the manuscript. I recommend the following major revisions to the manuscript before it can be considered for publication.
Reviewer comment 1:
The main claim that thermal stratification enhances RDOC production and stabilization is not fully convincing. RDOC concentration was slightly higher during the mixing period (1.26 mg/L) than during the obvious thermal stratification period (1.21 mg/L). No RDOC component was resolved in ITS. The proportion is even higher in MX (46.2% vs 23.0%). Moreover, stratification strength differs considerably between April and August (Fig. 2b), yet RDOC is only resolved in August; no relationship between stratification intensity and RDOC can be derived. The authors need to clarify this in the manuscript. More importantly, the manuscript should clarify whether “RDOC” here refers to a fluorescence-defined component or to experimentally demonstrated refractory carbon. These two concepts should not be treated as equivalent.Response 1:
Thank you for this important comment. We agree that the original interpretation overstated the evidence linking thermal stratification to RDOC production and stabilization. We have therefore revised the relevant sections to clarify the observed relationships and distinguish field-based associations from experimentally supported mechanisms.
First, we now distinguish between RDOC concentration and relative proportion. The fluorescence-estimated RDOC-equivalent concentrations were 1.21 mg L⁻¹ during OTS and 1.26 mg L⁻¹ during MX, whereas the relative fluorescence contributions were 23.0% and 46.2%, respectively. The higher RDOC proportion during MX therefore does not necessarily indicate enhanced RDOC production. Instead, it may reflect the persistence of previously accumulated organic matter, vertical redistribution during water-column mixing, and continued microbial processing. We have revised the Discussion to clarify this interpretation.
Second, no RDOC-related PARAFAC component was resolved during ITS. We now explicitly acknowledge this observation and clarify that the absence of a resolved fluorescence component does not imply the complete absence of chemically refractory DOC.
Third, we agree that the present observations do not establish a direct quantitative relationship between stratification intensity and RDOC production. Accordingly, we have revised statements implying direct causation and now describe thermal stratification as a seasonal physical structure associated with changes in environmental conditions, microbial community organization, and DOM composition. Importantly, a previous degradation experiment conducted in the same reservoir (Jia et al., 2025) provides independent process-based evidence that microbial RDOC conversion efficiency was higher during stratification than during mixing, particularly in the thermocline during OTS. This experimental evidence supports the potential role of thermal stratification in facilitating microbial carbon transformation, although the present field observations do not independently establish this mechanism.
Finally, we have clarified that RDOC in the present field dataset refers to a fluorescence-defined RDOC-related fraction and its semi-quantitative DOC-equivalent estimate, rather than experimentally verified refractory carbon. We now explicitly acknowledge the limitations of fluorescence-based classification and avoid treating these estimates as direct evidence of molecular recalcitrance or MCP-derived RDOC.
Overall, these revisions clarify that the present field observations provide evidence of seasonal associations among thermal structure, microbial community organization, and RDOC-related DOM fractions, while the previous experimental findings provide complementary support for the proposed role of thermal stratification in microbial carbon transformation. We have therefore retained the original title to reflect the broader mechanistic framework, while refining the manuscript to avoid overstating the causal evidence provided by the present field dataset.
Reviewer comment 2:
The three sampling periods correspond to April, August, and November, and are simultaneously associated with different thermal structures, temperatures, biological conditions, and seasonal environmental forcing. The observed differences between ITS, OTS, and MX cannot necessarily be attributed specifically to thermal stratification. How did the authors distinguish the effect of thermal stratification from the broader seasonal effect? Please clarify.Response 2:
Thank you for this important comment. We agree that the three sampling periods (April, August, and November) differ not only in thermal structure, but also in temperature, biological activity, and other seasonally varying environmental conditions. Therefore, the observed differences among ITS, OTS, and MX cannot be attributed exclusively to thermal stratification.
In this study, ITS, OTS, and MX were defined as successive seasonal thermal-regime phases of the same reservoir, based on vertical temperature profiles together with the Schmidt Stability Index (SSI) (Idso and Oceanography, 1973) and the relative water-column stability index (RWCS/H) (Shi et al., 2016), rather than as independent classes of stratification intensity. ITS represents the developing stratification phase, OTS the well-established stratified phase, and MX the post-stratification mixing phase. The detailed criteria are now clarified in Text S5 of the Supplementary Materials.
We have revised the manuscript accordingly to avoid attributing all observed biological and biogeochemical differences directly to thermal stratification. Throughout the revised text, we now describe these patterns as being associated with seasonal changes in thermal structure and concurrent environmental conditions, rather than as effects caused solely by stratification. We also explicitly acknowledge that, because thermal regime and season covary in this three-period field design, their independent effects cannot be fully disentangled. Resolving this issue would require higher-frequency temporal sampling or multi-year observations that separate thermal structure from broader seasonal forcing.
Reviewer comment 3:
The manuscript interprets positive cohesion as facilitative interactions and negative cohesion as competitive or inhibitory interactions, and subsequently suggests that competitive interactions promote RDOC accumulation. However, these cohesion metrics are derived from correlations among bacterial taxa and do not directly measure biological interactions. Likewise, a positive or negative association between cohesion and RDOC concentration does not establish that the interaction itself caused RDOC production. For statements such as “competitive interactions promote RDOC production”, please provide more evidence.
Response 3:Thank you for this important comment. We agree that positive and negative cohesion are derived from taxon co-occurrence patterns and should not be interpreted as direct measurements of facilitative or competitive biological interactions. We have therefore revised the manuscript throughout to replace terms such as “facilitative interactions” and “competitive interactions” with “positive microbial associations” and “negative microbial associations.”
We have also removed statements implying that negative cohesion directly promotes RDOC production. The revised text now states that positive associations were linked to greater LDOC availability, whereas stronger negative associations coincided with higher RDOC levels. These relationships are interpreted as community-level association patterns rather than causal effects on RDOC formation. We therefore now describe cohesion as covarying with DOM composition and carbon persistence rather than as a direct driver of RDOC production.
Specific comments
Reviewer comment 4:L36-37: “predominantly regulated by microbial processes rather than benthic carbon inputs” is too strong. Rephrase as an inference.
Response 4:
Thank you for this comment. We agree that the original wording was too strong because benthic carbon inputs were not directly measured. We have therefore revised the statement to: “RDOC variation was more closely associated with microbial community characteristics than with the available indicators of benthic carbon input.” We also added a limitation noting that sedimentary contributions cannot be excluded without direct measurements of porewater chemistry, benthic fluxes, and resuspension. Detailed revisions can be found in L32-34 of the text.
Reviewer comment 5:
L39-41: The network and cohesion analyses mainly indicate associations among taxa. The conclusion that competitive or cooperative interactions promote RDOC formation requires more cautious wording.
Response 5:Thank you for this comment. We agree that the original wording overstated the ecological meaning of the network and cohesion metrics. Because these metrics are derived from co-occurrence patterns, they indicate statistical associations among taxa rather than directly measured facilitative or competitive interactions. We have therefore revised the sentence to: “Bacterial network analysis further suggested that positive microbial associations were linked to greater labile carbon availability, whereas stronger negative associations under environmental stress coincided with higher RDOC levels.” This revision avoids inferring direct causal effects of microbial interactions on RDOC formation. Detailed revisions can be found in L36-39 of the text.
Reviewer comment 6:
L43-45: “resulting in persistent phosphorus limitation” implies causation. Use “may have contributed to” or similar wording.Response 6:
Thank you for this comment. We agree that the original wording, “resulting in persistent phosphorus limitation,” implied a causal relationship that was not directly demonstrated by our data. We have therefore revised the sentence to avoid causal interpretation. The revised text now reads: “The reservoir also exhibited persistent phosphorus limitation and high C:P ratios, together with geochemical conditions consistent with potential calcium–phosphorus (Ca–P) co-precipitation.”
This revision distinguishes the observed phosphorus limitation and C:P pattern from the inferred geochemical mechanism and avoids implying that Ca–P co-precipitation was directly responsible for the phosphorus limitation. Detailed revisions can be found in L39-41 of the text.
Reviewer comment 7:
L98-100: This sentence conflates humic-like fluorescence with recalcitrance. Please revise this statement and provide supporting references.
Response 7:
Thank you for this important comment. We agree that the original sentence conflated humic-like fluorescence with molecular recalcitrance. Humic-like fluorescence reflects optical characteristics of DOM and may be associated with more processed or aromatic material, but it does not by itself demonstrate resistance to biodegradation. We have therefore revised the text to distinguish optical characteristics from experimentally demonstrated recalcitrance.
We have also added an additional reference discussing the use and limitations of optical DOM indices as proxies for biodegradability. Detailed revisions can be found in L116-121 of the text.
Reviewer comment 8:
L348: The identification of keystone taxa appears to be mainly based on network topology. The authors should clarify whether independent functional evidence supports their designation as keystone taxa.
Response 8:
Thank you for this valuable comment. We agree that the identification of keystone taxa was primarily based on network topology rather than independent functional validation. Accordingly, we have revised the Methods to use the term “putative keystone taxa” and explicitly defined their identification based on within-module connectivity (Zi) and among-module connectivity (Pi). Specifically, module hubs, connectors, and network hubs were classified as putative keystone taxa. We further clarified that predicted functional profiles and taxon–function associations provide only supplementary evidence and do not constitute direct functional validation. In addition, network robustness was assessed using degree-preserving randomization and threshold-sensitivity analyses. These clarifications have been incorporated into the revised Methods section. (L361-382).Reviewer comment 9:
L411-421: RDOC and LDOC concentrations are calculated from fluorescence contribution × DOC. Please give the exact equation and assumptions for converting fluorescence contributions into RDOC/LDOC concentrations. How were different fluorescence efficiencies handled? Please discuss the limitations of this proxy.
Response 9:
Thank you for this comment. We have clarified the calculation of fluorescence-estimated LDOC, SLDOC, and RDOC concentrations by providing the exact equation and underlying assumptions. The relative fluorescence contribution of each PARAFAC component was multiplied by the measured bulk DOC concentration, and components assigned to the same category were summed before calculation. We now explicitly state that component-specific fluorescence efficiencies were not independently corrected and that non-fluorescent DOM is not represented by EEM-PARAFAC. Accordingly, these values are now described as semi-quantitative fluorescence-estimated DOC-equivalent concentrations, rather than direct measurements of chemically defined LDOC, SLDOC, or RDOC pools, and the associated limitations have been added to the manuscript. Detailed revisions can be found in Section 3.2 and Text S3 of the text.Reviewer comment 10:
L585-587: This is misleading. In the Bayesian models, alpha diversity does not represent bacterial activity and pH does not represent alkalinity.
Response 10:
Thank you for your valuable feedback. The terms “bacterial activity” and “alkalinity” in the original text are indeed incorrect descriptions of the α diversity index (Simpson) and pH. After further refining the Bayesian linear model, the indicators that have a positive predictive effect on the increase of LDOC in water bodies were re-screened and ranked. We also checked and corrected the incorrect descriptions of the effects of these indicators. For details, please refer to pages L634-655.Reviewer comment 11:
L694: “The succession of thermal stratification drives shifts in key environmental parameters.” I don’t think “drives” is appropriate here. I don’t see evidence for this in the data presented.
Response 11:
Thank you for this comment. We agree that the term “drives” implied a causal relationship that was not directly demonstrated by our data. We have therefore revised the sentence to: “The seasonal evolution of thermal stratification alters the physicochemical environment of the water column and is accompanied by changes in bacterial association patterns across depths and thermal phases.” (L814-816) This wording reflects the observed temporal association without implying direct causation.Reviewer comment 12:
L709-710: RDOC was not detected in ITS, so how can rising positive cohesion be “accompanied by an increase in RDOC”?
Response 12:
Thank you for pointing this out. We agree that the original wording was inaccurate because RDOC was not detected during ITS. We have revised the sentence to avoid implying an increase in RDOC during this period. Please refer to Sections 3.3.3 and 4.2 for details.
Reviewer comment 13:
L724-727: What is the DIC-enhanced MCP mechanism?
Response 13:Thank you for pointing this out. We agree that the term “DIC-enhanced MCP mechanism” was insufficiently defined and could imply a mechanistic pathway that was not directly demonstrated in the present study. We have therefore removed this expression from the revised manuscript. Instead, we now describe the observed relationships more explicitly: biological DIC utilization may contribute to the production of autochthonous organic substrates, while subsequent microbial processing may be associated with the persistence of fluorescence-defined RDOC fractions. We also clarify that our stable-isotope and fluorescence data provide only indirect association evidence and do not directly demonstrate carbon transfer from DIC into MCP-derived RDOC. Direct verification would require isotope-tracing or other process-based experiments. The revised content is as shown in Section 4.2 of the text.
Reviewer comment 14:
L750-751: RDOC accumulation is independent of benthic input is not sufficiently supported. Sediment porewater, benthic flux, and resuspension were not measured. Please clarify or remove.
Response 14:Thank you for this important comment. We agree that the original wording was too strong because sediment porewater chemistry, benthic fluxes, and resuspension were not directly measured in this study. We have therefore revised Section 4.2 to avoid concluding that RDOC accumulation was independent of benthic inputs.
The revised text now states that benthic release was unlikely to be the dominant source of RDOC under the observed conditions, based on the absence of pronounced RDOC enrichment in deep waters, relatively low hypolimnetic HIX, and weak or negative relationships between FI and RDOC. However, we now explicitly acknowledge that these observations provide only indirect source constraints and cannot exclude sedimentary contributions. Accordingly, the manuscript now states that a benthic contribution cannot be ruled out without direct measurements of porewater chemistry, benthic fluxes, resuspension, or other source-tracing approaches (L898-906).
Reviewer comment 15:
L770-774: This sentence is kind of confusing. What does “rate-coupling signature” mean here? I don’t follow how a positive correlation between δ13C-DIC and RDOC can confirm that RDOC originates from water-column internal carbon processing, rather than simply reflecting concurrent changes in the DIC and organic carbon pools.Response 15:
Thank you for this comment. We agree that the term “rate-coupling signature” was unclear and that the original wording overstated what could be inferred from the isotope relationships. We have therefore removed this expression and revised Section 4.2 to distinguish concurrent isotope–carbon-pool associations from direct source tracing. The revised text now states that the relationships among DIC, δ¹³C-DIC, DOC, and LDOC are consistent with enhanced biological DIC utilization and autochthonous organic-carbon production. We also now explicitly acknowledge that δ¹³C-DIC can be influenced by carbonate weathering, respiration, air–water CO₂ exchange, carbonate speciation, and mixing, and that direct verification of DIC-to-RDOC transfer would require isotope-tracing or other process-based experiments. The specific modifications can be found in Section 4.2, L845-856 and 867-870.
Reviewer comment 16:
L813: “RDOC increased significantly along the C:P gradient” is not supported by the presented analysis. The quadratic fit has R2 = 0.385 (Fig. 8), leaving most of the variance unexplained. I don’t see any support for this. I don’t follow this line of reasoning at all, please clarify or remove.
Figure captions should also be revised for clarity. For example, the caption of Fig. 3 is not very clear. Please clarify the effective sample size, i.e., n = 3.Response 16:
Thank you for this valuable comment. We agree that the original interpretation of the C–RDOC relationship was insufficiently supported. We have therefore reanalyzed this relationship using a period-adjusted linear model, which revealed a significant positive association (n = 27, β = 0.00885, 95% CI: 0.00312–0.01459, p = 0.004; R² = 0.299). In contrast, the quadratic term was not significant (P = 0.169), and additional period-stratified, influence, and leave-one-out analyses did not support a consistent nonlinear response. Accordingly, we have removed the previous quadratic interpretation and revised the text to emphasize that C is associated with RDOC variation but does not establish an ecological threshold or a dominant controlling mechanism. We have also revised Fig. 9 and its caption to reflect the updated analysis and clarified the sample sizes and their definitions in the figure captions.
Technical corrections
Reviewer comment 17:
L214-215, L258, L315, L340: Section numbering is inconsistent (2.2 followed by 2.3.1–2.3.4).Response 17:
Thank you for your valuable suggestions. Such errors have now been corrected, as detailed in L230, 274, 331, and 360 of the text. We have also conducted a thorough review of the entire manuscript to address similar issues.
Reviewer comment 18:
L378: “DLD Reservoirwas” should be “DLD Reservoir was”.Response 18:
Thank you for your valuable feedback. The error here was due to the author failing to insert a space while typing, which has now been corrected. See L399 in the text.
Reviewer comment 19:
Table 1: “Incabating” should be “Incubating”.Response 19:
Thank you for your suggestion. The error here was caused by a typo during typing and has now been corrected; see Table 1 for details.
Reviewer comment 20:
L896: “Origian 2021” should be “Origin 2021”.Response 20:
Thank you for your suggestion. The error here was caused by a typo during typing and has now been corrected; see L1075 for details.
References
Idso S. B. J. L., Oceanography, 1973. On the concept of lake stability1. 18: 681-683.
Jia Y., Su X., Zhang T., He Q., Li J., Pu J., 2025. High carbon fixation during thermal stratification period in a subtropical periodic stratified reservoir: Evidences from RDOC conversion efficiency driven by MCP. Water Research. 278: 123316.
Shi J.-C., Huang T.-L., Wen G., Liu F., Qiu X.-P., Wang B.-S., 2016. The Variation Characteristic of Sulfides and VOSc in a Source Water Reservoir and Its Control Using a Water-Lifting Aerator. International Journal of Environmental Research and Public Health. 13 (4): 427.
Citation: https://doi.org/10.5194/egusphere-2026-4596-AC3
-
AC3: 'Reply on RC3', Junbing Pu, 14 Oct 2026
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 275 | 121 | 75 | 471 | 82 | 70 | 77 |
- HTML: 275
- PDF: 121
- XML: 75
- Total: 471
- Supplement: 82
- BibTeX: 70
- EndNote: 77
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This study takes the Dalongdong karst reservoir in Guangxi as the research object, and divides the thermal regime into three phases: incubating thermal stratification, obvious thermal stratification, and water‑column mixing. It explores the regulatory mechanisms of thermal stratification on the production of recalcitrant dissolved organic carbon (RDOC). Thermal stratification creates physicochemical gradients that shape the ecological niches of keystone bacteria. Facilitative bacterial interactions promote the production of labile carbon, while competitive interactions drive RDOC accumulation. Calcium‑phosphorus co‑precipitation in karst water induces phosphorus limitation, reduces microbial carbon‑use efficiency, and triggers the coupling of the biological carbon pump (BCP) and the microbial carbon pump (MCP). This work reveals the carbon sequestration mechanism of karst reservoirs and provides theoretical support for inland aquatic carbon‑sink assessment and climate‑change mitigation. My main comments are listed below: