General comments

The overall quality of the manuscript is very good. The novelty of this piece of research relies on the robustness of the model to detect regime shifts and its validation through the study case in the North Sea. The manuscript is well written and the sections are well structured. The conclusion is supported by the results and analyses that were performed.

My major concern is the issue about the scale up of the results to the ecosystem level. As it is now, the analyses performed does not enable the scale up since only the plankton community (phytoplankton and zooplankton) was assessed. I have highlighted this issue in several places throughout the manuscript.

Specific comments

Besides the specific comments embedded in the attached pdf I want to highlight the following:

Regarding the Introduction:

In my opinion, the manner of beginning this section is not helpful. See comment in the attached pdf.

The term “planktonic ecosystem” needs a better explanation. See comment in the attached pdf.

Regarding the Methods:

Abundance data is missing the units. This is crucial for any cross-comparison. See comment in the attached pdf.

Regarding the Validation of the model:

In the majority of the figures, axis labels are incorrect or missing. See comments in the attached pdf.

Regarding the Discussion:

I suggest to avoid the use of references to figures and tables. See comments in the attached pdf.

Please see comment in line 437-439 about the mismatch between the model results and changes in single plankton species.

Please see comment in line 473-474 about the detection of changes in phytoplankton and zooplankton species.

The rest of my specific comments are embedded in the attached pdf.

Technical corrections

I suggested a few number of technical corrections. These are embedded in the attached pdf.

Review of "A novel multispecies approach for the detection of ecosystem regime shifts — a case study in the North Sea" by Dees et al.

A. General Comments

This manuscript presents a novel methodology, the Regime Shift Tool (RST) model, for detecting ecosystem regime shifts, with an application to planktonic time series from the North Sea using Continuous Plankton Recorder (CPR) data. The study addresses a critical gap in the detection of regime shifts, particularly within open-ocean systems, by synthesizing multiple plankton time series into a composite likelihood metric.

The manuscript introduces an innovative and potentially valuable methodological advancement for the detection of ecosystem regime shifts. It is generally well-organized, and the research question is both timely and significant. The RST model represents a promising addition to the suite of available analytical tools for detecting complex ecological transitions.

Nevertheless, substantial revisions are required to address methodological limitations, improve the clarity of the exposition, and ensure the robustness of the interpretation of results. Several aspects concerning the methodology, results presentation, and interpretation necessitate major improvements to enhance analytical rigor and overall coherence. I therefore recommend that the manuscript be considered for publication only after major revisions have been satisfactorily completed.

To strengthen the manuscript, the authors should: (i) expand the literature review to include a detailed comparison with existing regime shift detection methods; (ii) provide comprehensive justifications for all key modeling decisions, supported by sensitivity analyses; (iii) enhance the description of the RST model to improve clarity and reproducibility; (iv) improve the quality and contextualization of figures; (v) deepen the discussion of the ecological mechanisms underlying the observed regime shifts and their management implications; and (vi) moderate claims regarding predictive capabilities unless further robust evidence is provided.

B. Specific Comments

1. Introduction

The introduction successfully contextualizes the challenges associated with detecting ecosystem regime shifts and highlights the critical role of long-term ecological datasets, such as the Continuous Plankton Recorder (CPR) records, in addressing these challenges. Nevertheless, the review of existing regime shift detection methodologies remains insufficiently comprehensive. A more detailed and critical examination of prior approaches, particularly those developed by Beaugrand et al. (2014) and others, is warranted. The authors should explicitly articulate the comparative advantages and potential limitations of the RST model relative to these established methods, thereby more clearly situating their contribution within the broader methodological landscape.

2. Methods

In the section on CPR data processing, the rationale for applying a logarithmic transformation (Equation 1) requires further elaboration. While data normalization is a standard practice, the authors should clarify the specific motivations for its use in this context and discuss any potential implications for the study’s findings. Furthermore, it would be beneficial to understand if the authors explored alternative transformations and why the logarithmic transformation was ultimately chosen. Similarly, the description of seasonality removal (Equation 2) is overly succinct. Greater detail is needed regarding the methodologies employed, including any filtering techniques or assumptions made. Importantly, the potential influence of pre-processing steps on the detection of regime shifts, which may themselves exhibit seasonal patterns, should be explicitly addressed.

The explanation of the RST model, based on Boulton and Lenton (2019) and its subsequent adaptation by Arellano-Nava et al. (2022), is currently insufficient for readers unfamiliar with these references. A more thorough, self-contained description of the algorithmic steps within the manuscript would greatly improve accessibility. In addition, it is crucial to ascertain whether the linear regressions employed in the Boulton and Lenton (2019) method were checked for underlying assumptions such as normality and homoscedasticity.

The selection of a 24-month rolling window for calculating the weighted probability of regime shifts is a critical modeling decision that requires more rigorous justification. A sensitivity analysis examining the influence of varying window lengths would strengthen the methodological foundation. It is also important to address whether the authors considered the potential impact of autocorrelation within the time series when determining the rolling window size, as autocorrelation can significantly affect the results of such analyses.

Equation 5, which integrates phytoplankton and zooplankton indicators, is predicated on an assumption of bottom-up control. Although the authors acknowledge this simplification, a more extensive discussion of its ecological plausibility, and the potential impact of alternative mechanisms such as top-down regulation or lateral environmental forcing, is necessary. It would also be valuable to discuss how the model could be adapted in future research to incorporate top-down or mixed control scenarios.

The decision to adopt a 20% threshold for regime shift identification also demands greater justification. A sensitivity analysis evaluating the effects of alternative thresholds on regime shift detection outcomes would provide essential context for interpreting the results.

Regarding model validation, the description of the simulated datasets generated using Equations 6 and 7 could be improved. The authors should explicitly discuss the extent to which these simulations reflect the complexity of real-world plankton dynamics. Moreover, the rationale for parameter choices, notably α = 0.99 and σ = 0.2, should be thoroughly explained. Furthermore, it is essential to acknowledge the inherent limitations of any model. The authors should more explicitly state the limitations of the RST model, discussing its potential applicability and the types of regime shifts it might not be able to detect. The model's sensitivity to noise is also a concern. While Type I and II errors are discussed, a more thorough noise sensitivity analysis would bolster confidence in the model's robustness.

It is also important to consider the limitations of the CPR data itself. The authors should discuss potential biases inherent in the data, such as spatial coverage, sampling frequency, and the possibility of damage to plankton during the sampling process.

3. Results

The presentation of the results is generally effective; however, the clarity and informativeness of the figures could be substantially improved. All axes should be clearly labeled with appropriate units, and figure captions should include sufficient contextual detail to aid interpretation. Additionally, annotating Figures 2–7 and 9–13 with the timing of known environmental events—such as major storms, temperature anomalies, or changes in fishing pressure—would provide valuable context and enhance the interpretability of observed patterns.

The discussion of Type I and Type II errors constitutes a notable strength of the manuscript. However, the authors should more explicitly relate these error estimates to the confidence with which regime shifts are identified. Establishing a clearer linkage between the quantified uncertainty and the interpretation of regime shift periods would enhance the robustness and transparency of the study's conclusions.

4. Discussion

The discussion provides a reasonable interpretation of the study’s results and acknowledges key limitations. However, several areas require further development to enhance the depth, rigor, and critical analysis of the findings.

First, a more detailed comparison with previous studies on regime shifts in the North Sea is necessary. Although prior research is mentioned, the authors should explicitly discuss the degree of concordance or divergence between their findings and earlier reports. Specifically, the comparison should assess how the timing and magnitude of regime shifts identified by the RST model align with or differ from those detected using established methods. Furthermore, the discussion should articulate the novel contributions of the RST model beyond merely analyzing a greater number of species, clarifying the specific insights it offers into ecosystem dynamics.

The ecological mechanisms underpinning the observed regime shifts also warrant deeper examination. While changes in plankton abundances are identified as potential drivers, a more nuanced exploration of environmental forcing factors (e.g., temperature variability, nutrient availability), biotic interactions (e.g., competition, predation), and other relevant processes is needed. In particular, the roles of climate change and anthropogenic impacts should be more thoroughly integrated into the discussion. Although these drivers are introduced earlier in the manuscript, a stronger and more explicit connection between them and the identified regime shifts is necessary to contextualize the findings.

Moreover, the discussion should more explicitly address the practical implications of applying the RST model in ecosystem management contexts. Potential applications, benefits, and limitations should be thoroughly considered. For instance, the authors could discuss how the model might inform fisheries management, conservation strategies, or adaptive management frameworks. Addressing the challenges inherent in translating model outputs into actionable management advice would also enhance the practical relevance of the study.

The influence of advection processes on plankton dynamics, and the associated difficulty of distinguishing these from genuine regime shifts, is acknowledged but remains underdeveloped. A more detailed treatment of this limitation, including suggestions for future methodological improvements, would strengthen the manuscript. For example, incorporating hydrodynamic modeling or spatial analyses into the RST framework could offer promising avenues for future research.

The manuscript’s discussion of the predictive potential of the RST model, although intriguing, is currently under-supported by the presented evidence. The authors are encouraged to temper claims regarding predictive capability or to provide more substantial empirical support. A critical evaluation of the model’s current limitations in forecasting regime shifts, including its sensitivity to noise and the risk of false positives or negatives, should be included. In addition, outlining specific future research directions to validate and enhance the model’s predictive accuracy would be valuable.

Furthermore, the limitations of the RST model itself should be discussed in greater depth. Rather than simply acknowledging that "all models are simplifications," the authors should critically evaluate the contexts in which the RST model may be less applicable, the types of regime shifts it may fail to detect, and the extent to which its performance depends on data quality. Specific limitations arising from the choice of a 24-month rolling window and the assumption of bottom-up control should be examined. Additionally, the statistical underpinnings of the model, including assumptions embedded in the Boulton and Lenton method and the potential influence of autocorrelation, should be critically assessed.

In summary, while the discussion provides a solid foundation, it requires substantial expansion to offer a more comprehensive, critical, and contextually grounded interpretation of the results. A stronger emphasis on limitations, broader ecological and management contexts, and practical applicability would significantly enhance the scientific contribution of the manuscript.