Data-Driven Multi-Scale Channel-Aligned Transformer for Low-Carbon Autonomous Vessel Operations: Enhancing CO₂ Emission Prediction and Green Autonomous Shipping Efficiency

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This study presents an innovative and timely contribution to the field of sustainable maritime operations by proposing the MCAT model integrated with a 5G satellite IoT communication framework for accurate CO2 emissions prediction in the autonomous vessels. The combination of multi scale token reconstruction and dual level attention mechanisms to handle heterogeneous data streams is particularly noteworthy. The manuscript would benefit from a more detailed comparative analysis of the MCAT model with existing baselines beyond MAE and MSE metrics, such as R2, inference time and energy consumption during model training and inference. The integration of 5G and satellite communication is technically sound, a more in depth discussion on the deployment feasibility, particularly in remote maritime regions where infrastructure may be lacking, would enhance the practical relevance of the work.

The interpretability of the MCAT model is briefly mentioned but not sufficiently demonstrated. Given the increasing emphasis on explainable AI in critical infrastructure, the authors should provide clearer examples or visualizations of how the model explains its predictions and supports decision making for emissions reduction. Moreover, the manuscript should include a sensitivity analysis to assess the model’s robustness against missing or corrupted input data. Clarifying the scalability of the system in multi vessel scenarios and including a discussion on how this framework aligns with IMO’s long term decarbonization roadmaps (beyond 2050) could further strengthen the impact and applicability of the study.

Author Response

Dear Reviewer,

Thank you for your insightful comments and suggestions regarding our manuscript, "Data-Driven Multi-scale Channel-aligned Transformer for Low-Carbon autonomous vessel Operations: Enhancing CO2 Emission Prediction and Green autonomous shipping Efficiency" (JMSE-3662704). We appreciate the time and effort you dedicated to reviewing our work. We have carefully considered all comments and have revised the manuscript accordingly. Below, we provide a point-by-point response to your concerns.

 

The interpretability of the MCAT model is briefly mentioned but not sufficiently demonstrated. Given the increasing emphasis on explainable AI in critical infrastructure, the authors should provide clearer examples or visualizations of how the model explains its predictions and supports decision making for emissions reduction. Moreover, the manuscript should include a sensitivity analysis to assess the model’s robustness against missing or corrupted input data. Clarifying the scalability of the system in multi vessel scenarios and including a discussion on how this framework aligns with IMO’s long term decarbonization roadmaps (beyond 2050) could further strengthen the impact and applicability of the study.

 

 

Comment 1: The interpretability of the MCAT model is briefly mentioned but not sufficiently demonstrated. Given the increasing emphasis on explainable AI in critical infrastructure, the authors should provide clearer examples or visualizations of how the model explains its predictions and supports decision making for emissions reduction.

Revision 1: Modified lines 27-30 from "Furthermore, the proposed architecture supports smart autonomous shipping solutions by providing interpretable emission insights for route optimization, fuel efficiency enhancement, and compliance with CII regulations. " to: "Furthermore, the proposed architecture supports smart autonomous shipping solutions by providing demonstrably interpretable emission insights through its dual-level attention mechanism (visualized via attention maps), for route optimization, fuel efficiency enhancement, and compliance with CII regulations."

 

Revision 2: The text on lines 281-289 (Section 3.2) was changed:

From: "The dual-level attention mechanism not only captures cross-variable dependencies but also generates interpretable emission patterns. These patterns can be directly fed into autonomous autonomous shipping systems as digital twins to simulate emission outcomes under different navigation strategies (e.g., speed adjustment, route optimization). For instance, the token-level attention identifies critical temporal windows affecting emissions (e.g., acceleration phases), while the channel-level attention reveals how environmental factors (e.g., wind speed) interact with propulsion parameters-both essential for autonomous vessels to balance operational efficiency and carbon compliance."

To: "A key strength of MCAT is its inherent interpretability, derived from the dual-level attention mechanism, which not only captures cross-variable dependencies but also generates interpretable emission patterns. These patterns can be directly fed into autonomous shipping systems as digital twins to simulate emission outcomes under different navigation strategies (e.g., speed adjustment, route optimization). For instance, the token-level attention identifies critical temporal windows affecting emissions (e.g., acceleration phases), while the channel-level attention reveals how environmental factors (e.g., wind speed) interact with propulsion parameters—both essential for autonomous vessels to balance operational efficiency and carbon compliance. The visualization of these attention weights (as discussed in Section 4.8 and Figure 18) provides direct insight into the model's decision-making process."

 

 

Revision 3: Changed lines 750-753 (Section 4.7.1, description of Fig.13 results) from "Short-Term Precision (96-step, Fig.13):The predicted emission trajectory (blue) aligns closely with ground truth (red), capturing transient operational events such as acceleration-induced emission spikes (e.g., 12:00-14:00 in Fig.13)." to "Short-Term Precision (96-step, Fig.13):The predicted emission trajectory (blue) aligns closely with ground truth (red), capturing transient operational events such as acceleration-induced emission spikes (e.g., 12:00-14:00 in Fig.13). This visual alignment provides a direct interpretation of the model's capability to predict critical emission events, informing decisions like temporary speed reduction to mitigate such spikes."

 

 

Revision 4: Changed lines 764-768 (Section 4.7.1, description of Fig.15-16 results) from "By correlating these patterns with real-time weather forecasts, autonomous ships can reroute to low-carbon zones, achieving the simulated 12.3% cumulative emission reduction (Section 4.6)." to "By correlating these patterns with real-time weather forecasts, autonomous ships can reroute to low-carbon zones, achieving the simulated 12.3% cumulative emission reduction (Section 4.6). The model's ability to identify and predict these long-term patterns, visualized in the heatmaps (Fig. 8 & 9) and time-series plots (Fig. 15-16), offers strategic interpretability for voyage planning, allowing operators to understand why certain routes or operational periods are predicted to have higher emissions."

 

Revision 5: Changed lines 773-776 (Section 4.7.2, referring to Fig.17) from "Multi-Scale Feature Extraction: Token-level attention (Fig.17) isolates critical time windows (e.g., port approach phases), while channel-level attention (Fig.17) prioritizes environmental variables (e.g., wind speed), aligning with the dual attention mechanism described in Section 3.2." to "Multi-Scale Feature Extraction and Interpretability: Token-level attention (as visualized in Figure 18, top row) isolates critical time windows (e.g., port approach phases), demonstrating which historical periods the model deems most important for its current prediction. Channel-level attention (Figure 18, bottom row) prioritizes environmental variables (e.g., wind speed), showing which input features are most influential. This aligns with the dual attention mechanism described in Section 3.2, and these attention maps offer direct visual interpretability into the model's internal weighting of temporal and cross-variable features for its predictions, supporting decision-making by highlighting key emission drivers."

 

Revision 6: The text in Section 4.8 ATTENTION PATTERN MAPS (This section is crucial for interpretability), Lines 788-799, was changed from:

Figure 18 illustrates the operation of the dual attention mechanism through four subplots. These subplots visualize the attention weight distribution of two attention heads (Head1 and Head2) at the token and channel levels. The top row shows the token-level attention heatmaps, revealing how the model captures temporal dependencies by assigning attention weights at different time steps. Specifically, these visualizations demonstrate the model's ability to identify crucial temporal correlations and long-range dependencies in sequential data. In contrast, the bottom row shows the channel-level attention heatmaps, highlighting the correlations between channels and revealing the model's adaptive weighting strategy when capturing cross-variable interactions in multivariate time series."

to:"Figure 18 illustrates the operation of the dual attention mechanism through four subplots, providing visual evidence of the model's interpretability. These subplots visualize the attention weight distribution of two attention heads (Head1 and Head2) at the token and channel levels. The top row shows the token-level attention heatmaps, revealing how the model captures temporal dependencies by assigning attention weights at different time steps. Specifically, these visualizations demonstrate the model's ability to identify crucial temporal correlations and long-range dependencies in sequential data, offering an interpretable view of which past time steps are most influential for the prediction. This helps in understanding, for example, if recent operational changes or specific past events are driving the current emission forecast. In contrast, the bottom row shows the channel-level attention heatmaps, highlighting the correlations between channels (i.e., input variables) and revealing the model's adaptive weighting strategy when capturing cross-variable interactions in multivariate time series, thus making the model's reasoning about feature importance transparent (e.g., identifying if vessel speed or wind conditions are primary drivers for a given prediction). Such insights directly support emission reduction decisions by pinpointing the most impactful factors."

 

 

Revision 7: Changed lines 802-810 from:

"A prominent observation from comparing the visualizations is the significant difference in attention distribution patterns between the two levels. Token-level attention exhibits a relatively dispersed activation pattern, indicating that the model emphasizes establishing comprehensive temporal relationautonomous vessels over a longer time frame. Conversely, channel-level attention displays a more concentrated activation pattern, suggesting that the model selectively focuses on specific variable combinations with higher diagnostic significance. This structural dichotomy effectively demonstrates the dual capability of our architecture in simultaneously modeling temporal dynamics and cross-variable dependencies."

to:"A prominent observation from comparing the visualizations is the significant difference in attention distribution patterns between the two levels, which is key to MCAT's interpretability. Token-level attention exhibits a relatively dispersed activation pattern, indicating that the model emphasizes establishing comprehensive temporal relationautonomous vessels over a longer time frame. Conversely, channel-level attention displays a more concentrated activation pattern, suggesting that the model selectively focuses on specific variable combinations with higher diagnostic significance. This structural dichotomy effectively demonstrates the dual capability of our architecture in simultaneously modeling temporal dynamics and cross-variable dependencies, enhancing the interpretability of how the model arrives at its predictions by disentangling temporal and feature-wise influences and allowing users to understand the 'what' (influential variables) and 'when' (influential time periods) behind emission forecasts."

 

 

Comment 2: Sensitivity Analysis (Missing/Corrupted Data)

 

Revision 1: The title of Section 4.5 (line 617) was changed From: "Noise Robustness Analysis" to: "4.5. Sensitivity Analysis: Robustness to Missing and Corrupted Data"

 

Revision 2: The first paragraph of Section 4.5 (lines 618-621) was changed:

From: "By simulating sensor failure scenarios, a Missing Completely At Random (MCAR) pattern was applied to the dataset, filling 20% of the data with zero values. This allowed for a comparison of the prediction robustness of different models under data missing scenarios."

To: "To evaluate the model's sensitivity to missing or corrupted input data and assess its robustness, a sensitivity analysis was conducted. By simulating sensor failure scenarios, a Missing Completely At Random (MCAR) pattern was applied to the dataset, filling 20% of the data with zero values (simulating missing data) or introducing random noise to a subset of features (simulating corrupted data). This allowed for a comparison of the prediction robustness of different models under these challenging data scenarios."

 

Revision 3: The last sentence of Section 4.5 (lines 663-667) was changed:

From: "The results validate that explicit multi-scale modeling-not merely stacking attention mechanisms-is critical for robustness under data sparsity."

To: "The results validate that explicit multi-scale modeling-not merely stacking attention mechanisms-is critical for robustness under data sparsity, confirming MCAT's strong performance in this sensitivity analysis against missing input data and its suitability for real-world operational conditions where data quality can vary.”

 

 

Comment 3: Scalability (multi-vessel): Clarify the scalability of the system in multi-vessel scenarios.

 

Revision 1: The text on lines 890-895 (Section 5, Conclusion and Future Work, beginning of the "Future Work" paragraph) was changed:

From: "In future research, we will focus on three key directions to deepen the results of this study. First, we will introduce a federated learning framework to achieve secure sharing of data across fleets. Currently, the sparsity of single-autonomous vessel data limits further improvements in model performance. Through federated learning, each fleet can collaboratively train a global model without disclosing raw data. We plan to design a federated learning algorithm based on homomorphic encryption to ensure data security during transmission and computation."

To: "In future research, we will focus on three key directions to deepen the results of this study. First, to address scalability for multi-ship scenarios and enhance model generalization, we will introduce a federated learning framework. This approach allows for secure sharing of learning across fleets without centralizing sensitive raw data. Currently, the sparsity of single-autonomous vessel data limits further improvements in model performance. Through federated learning, a global MCAT model can be collaboratively trained using data from multiple vessels, potentially managed by different operators, thus improving its robustness and accuracy across diverse operational profiles. This distributed learning paradigm enhances scalability by allowing the system to learn from a wider data pool while respecting data privacy. We plan to design a federated learning algorithm based on homomorphic encryption to ensure data security during transmission and computation, a critical aspect for multi-ship and multi-stakeholder environments. The inherent architecture of the MCAT model, processing standardized data inputs, is well-suited for such distributed training, and the 5G-satellite-IoT communication architecture can support the necessary data exchange for federated updates. "

 

Comment 4: Alignment with IMO's long-term decarbonization roadmaps (beyond 2050): Discussing how this framework aligns with IMO's long-term decarbonization roadmaps (beyond 2050) could further strengthen the impact and applicability of the study.

 

Revision 1:The text at the end of the paragraph on lines 855-859 (Section 5, Conclusion and Future Work, discussing broader applications) was changed:

From (original ending of the paragraph):

"...providing comprehensive key technical support for green autonomous shipping."

To (original ending plus added text):

"...providing comprehensive key technical support for green autonomous shipping. Furthermore, the MCAT framework is designed with adaptability in mind, positioning it to align with the IMO's long-term decarbonization roadmap extending beyond 2050. As the maritime industry transitions towards alternative low-carbon and zero-carbon fuels (e.g., ammonia, methanol, hydrogen) and incorporates novel propulsion systems or energy-saving devices, the data-driven nature of MCAT allows for retraining and recalibration. The model can integrate new data streams pertinent to these future technologies (e.g., new fuel consumption patterns, alternative energy source efficiencies, carbon capture system performance). Its core capability to model complex, dynamic, and multivariate dependencies will remain crucial for optimizing vessel operations and minimizing greenhouse gas footprints, irrespective of the specific fuel or technology mix adopted post-2050. This ensures the framework's enduring relevance in supporting the industry's journey towards full decarbonization."

 

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Review-MDPI

 

Remarks on details that can be improved

1. Overall, the English is good.
2. A list of symbols and acronyms is missing (for example: IMO; MCAT, AIS, IEEE, MRV, CBAM, LTSF, URLLC, EANN, EMA, ETS, SOTA, MAE, MSE, TEU, SVM, RF, GBDT, FFN, GELU, TSN, OPNET, HFO, LFO, ECMWF, etc.).
3. I suggest at line 21, instead of “…we partition…”, I suggest “…the current work partition…”; at line 222 instead of “…we obtain…” I suggest “the authors obtained…”. In short, avoid I, we, and so on. On the same matter, see also line 226,
4. It is unclear what was the starting point: did the authors start the model from real ships?, how many?, etc. Is this question adressed at line 388?
5. Figure 8: fuel consumption/CO₂ emissions variations are directly related to the ship's geographical position. It would be very interesting to include sea and weather conditions. Is this possible?
6. Figure 9 leaves me confused. Does it refer to pollution by the hours of the day? On the left y-axis instead of “CO2 emissions” it should say "Hour of the day", and on the right y-axis it could say "CO2 emission". Isn't that right? I also suggest to replace at lines 376 and 399 “…different times…” by “…different hours…”.
7. I value the description of the experience given between lines 413-466. But it might be desirable to add an enumeration of the equipment used, how often the data was read/recorded, and exactly what data was read/recorded and from how many ships.

 

Positive remarks

1. Very interesting subject.
2. Of the 27 bibliographic references only one is before 2020.
3. Between lines 130-149, the authors unequivocally clarify the contributions of this work.
4. Figure 5 is very enlightening.
5. Figure 8 is quite interesting.
6. The results are solidly justified in sections 4.3, 4.4. The advantages of MCAT, from the current study, in relation to other systems (Rolls-Royce) in reducing emissions are well justified in Figures 17, 18 and quantified in lines 811-816.

 

What is the main question addressed by the research?

The authors propose a multi-scale channel-aligned transformer (MCAT) model that, based on 5G communications, can predict a ship's CO₂ emissions.

  What is original or relevant for the field? A system that allows reducing pollution in the naval industry is relevant.  

Author Response

Dear Reviewer,

Thank you for your insightful comments and suggestions regarding our manuscript, "Data-Driven Multi-scale Channel-aligned Transformer for Low-Carbon autonomous vessel Operations: Enhancing CO2 Emission Prediction and Green autonomous shipping Efficiency" (JMSE-3662704). We appreciate the time and effort you dedicated to reviewing our work. We have carefully considered all comments and have revised the manuscript accordingly. Below, we provide a point-by-point response to your concerns.

 

All comments from reviewers：

Overall, the English is good.

A list of symbols and acronyms is missing (for example: IMO; MCAT, AIS, IEEE, MRV, CBAM, LTSF, URLLC, EANN, EMA, ETS, SOTA, MAE, MSE, TEU, SVM, RF, GBDT, FFN, GELU, TSN, OPNET, HFO, LFO, ECMWF, etc.).

I suggest at line 21, instead of “…we partition…”, I suggest “…the current work partition…”; at line 222 instead of “…we obtain…” I suggest “the authors obtained…”. In short, avoid I, we, and so on. On the same matter, see also line 226,

It is unclear what was the starting point: did the authors start the model from real ships?, how many?, etc. Is this question adressed at line 388?

Figure 8: fuel consumption/CO2 emissions variations are directly related to the ship's geographical position. It would be very interesting to include sea and weather conditions. Is this possible?

Figure 9 leaves me confused. Does it refer to pollution by the hours of the day? On the left y-axis instead of “CO2 emissions” it should say "Hour of the day", and on the right y-axis it could say "CO2 emission". Isn't that right? I also suggest to replace at lines 376 and 399 “…different times…” by “…different hours…”.

I value the description of the experience given between lines 413-466. But it might be desirable to add an enumeration of the equipment used, how often the data was read/recorded, and exactly what data was read/recorded and from how many ships.

 

 

Comment 1: A list of symbols and acronyms is missing (for example: IMO; MCAT, AIS, IEEE, MRV, CBAM, LTSF, URLLC, EANN, EMA, ETS, SOTA, MAE, MSE, TEU, SVM, RF, GBDT, FFN, GELU, TSN, OPNET, HFO, LFO, ECMWF, etc.).

 

Revision 1: Thank you for your suggestion. An "Abbreviations" section has been added to supplement the article. The list includes:

AIS	Automatic Identification System
CBAM	Carbon Border Adjustment Mechanism
CII	Carbon Intensity Indicator
EANN	Emotional Artificial Neural Networks
ECMWF	European Centre for Medium-Range Weather Forecasts
EMA	Exponential Moving Average
ETS	Emissions Trading System
FFN	Feed-Forward Network
GBDT	Gradient Boosted Decision Trees
GELU	Gaussian Error Linear Unit
HFO	Heavy Fuel Oil
IEEE	Institute of Electrical and Electronics Engineers
IMO	International Maritime Organization
LFO	Light Fuel Oil
LSTM	Long Short-Term Memory
LTSF	Long-term Time Series Forecasting
MAE	Mean Absolute Error
MCAT	Multi-scale Channel-aligned Transformer
MRV	Monitoring, Reporting, and Verification
MSE	Mean Squared Error
OPNET	Optimized Network Engineering Tool (simulation software)
RF	Random Forests
SOTA	State-Of-The-Art
SVM	Support Vector Machine
TEU	Twenty-foot Equivalent Unit
TSN	Time-Sensitive Networking
URLLC	Ultra-Reliable Low-Latency Communication

 

Comment 2: Avoid using first person pronouns (I, We)

 

Revision 1: The relevant changes have been made. Thank you for your suggestion.

Line 225:

From: "Specifically, we partition the input time series data z along the last dimension..." to: "Specifically, the current work partitions the input time series data z along the last dimension..."

Line 228:

From: "...After partitioning, we obtain a new multi-scale token vector z', where ...." to: "...After partitioning, a new multi-scale token vector z' is obtained, where ..."

Line 242:

From: "...we append a [CLS] token to z''s T-th dimension."

To: "...a [CLS] token is appended to the T-th dimension of z'."

Action Taken: The manuscript has been searched for "we," "our," "I," and "my," and these sentences have been rephrased in the third person or passive voice.

 

Comment 3: It is unclear what was the starting point: did the authors start the model from real ships?, how many?, etc. Is this question adressed at line 388?

 

Revision 1:Thank you for pointing out the need for clarity. Line 388 indeed falls within Section 4.2 "Dataset collection and processing," which discusses the data source. To further clarify:

From: "Based on multi-source data from a container autonomous vessel in 2022, this study deeply analyzed the spatio-temporal distribution characteristics of autonomous vessel's carbon dioxide emissions. The data includes AIS..."

To: "Based on multi-source data from a single real-world container autonomous vessel operating in 2022, this study deeply analyzed the spatio-temporal distribution characteristics of autonomous vessel's carbon dioxide emissions. The data includes AIS..."

The remainder of Section 4.2 (lines 415-465, referring to the manuscript's data processing description) details the data sources (AIS, onboard sensors, ECMWF, Copernicus) and processing steps for this single vessel.

 

Comment 4: Figure 8: fuel consumption/CO2 emissions variations are directly related to the ship's geographical position. It would be very interesting to include sea and weather conditions. Is this possible?

 

Revision 1:The discussion of Figure 8 (around lines 380-383 in the original OCR, referring to the manuscript's description of Figure 8) has been augmented to address this. The revised text now reads:

"Figure 8, a hexbin plot, visually presents the distribution of CO₂ emissions across different latitudes and longitudes using intuitive color coding, revealing spatial variations and distribution patterns of emissions. The color variations in the figure illustrate the distribution of emissions across different geographical locations.While directly overlaying contemporaneous oceanographic and meteorological conditions on this specific visualization could offer further depth, these environmental factors are integral input features to the MCAT model, and their influence is implicitly captured in the predicted emission patterns discussed elsewhere (e.g., Section 4.7.1)."

 

Comment 5: Figure 9 leaves me confused. Does it refer to pollution by the hours of the day? On the left y-axis instead of “CO2 emissions” it should say "Hour of the day", and on the right y-axis it could say "CO2 emission". Isn't that right? I also suggest to replace at lines 376 and 399 “…different times…” by “…different hours…”.

 

Revision 1: (Figure 9 Axis Labels and Legend):

Thank you for highlighting the confusion and for the suggestions regarding Figure 9.

The figure itself (Figure 9) has been revised:

The left Y-axis, which was previously labeled "CO2 Emissions" (with a range 0-23 noted in the reviewer's comment), has been changed to "Hour of the Day".

The color bar legend, previously labeled "CO2", now correctly represents "CO2 Emission (units, e.g., kg or ppm-m³)".

 

Revision 2: (Figure 9 Caption):

The caption for Figure 9 (line 385, original OCR number) was changed:

From: "Figure 9. Time series plot of CO₂ emissions."

To: "Figure 9. Heatmap of CO2 emission concentrations by hour of day."

 

Comment 6: Modification of text mentioning "different times"

 

Revision 1: Merge line 386: "Figure 9 visually presents the CO₂ emission concentrations of the autonomous vessel at different times of the day in the form of a heatmap." with lines 408–410 (discussion of Figure 9): "Figure 9 shows carbon dioxide emission concentration at various times of the day, revealing the significant impact of time on emissions." into a single paragraph: "Figure 9 shows carbon dioxide emission concentration at different times of the day (diurnal variations), revealing the significant impact of these daily patterns on emissions."

 

Comment 7: I value the description of the experience given between lines 413-466. But it might be desirable to add an enumeration of the equipment used, how often the data was read/recorded, and exactly what data was read/recorded and from how many ships.

 

Revision 1: Thank you for this valuable suggestion. Section 4.2 ("Dataset collection and processing," corresponding to lines 413-466 of the original manuscript) has been revised and enhanced to provide clearer details on the aspects you mentioned:

Number of Ships:

The text in the paragraph starting "Based on multi-source data..." (around line 398 of the original manuscript, now in the revised Section 4.2) was changed:

From: "Based on multi-source data from a container autonomous vessel in 2022..."

To: "Based on multi-source data from a single real-world container autonomous vessel operating in 2022, this study deeply analyzed the spatio-temporal distribution characteristics of autonomous vessel's carbon dioxide emissions."

This explicitly clarifies that the study utilizes data from one vessel.

Equipment Used:

The text describing onboard sensor data collection (around line 425 of the original manuscript, now in the revised Section 4.2) was enhanced:

From: "For the collection and processing of onboard sensor data, this study utilized multiple sensors installed on autonomous vessels to collect real-time data. Among these, mass flow meters recorded fuel consumption data for the main engine, auxiliary engine, and boiler."

To: "For the collection and processing of onboard sensor data, this study utilized data from multiple sensors typically installed on modern autonomous vessels. Key among these were Coriolis mass flow meters recording fuel consumption data (HFO and LFO) for the main engine, auxiliary engine, and boiler. Other sensor data included engine RPM, shaft power, GPS for speed."

Data Recording Frequency and Specific Data Recorded:

Throughout the revised Section 4.2, we have ensured that the recording frequencies and specific data parameters are clearly stated for each data source:

AIS Data: Update frequencies vary but data were ultimately aggregated on an hourly basis. Recorded data include dynamic (longitude, latitude, speed, heading, ROT) and static components, and navigation distance.

Onboard Sensor Data: Fuel consumption (HFO, LFO for main, auxiliary, boiler) was initially recorded in kg/h and then processed to hourly estimates, along with speed and pitch angle. Engine RPM, shaft power, and GPS speed are also included.

Meteorological and Sea Condition Data (ECMWF & Copernicus): Data such as 10-meter-height wind components, temperature, humidity, significant wave height, peak wave frequency, sea surface temperature, and current velocity components were collected or processed to an hourly resolution.

We believe these revisions to Section 4.2 now provide a more comprehensive and clear enumeration of the data foundation as requested.

 

Please see the attachment.

Author Response File: Author Response.pdf