the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 16 Jul 2025
A Digital Twin Ocean: Can we improve Coastal Ocean Forecasts using targeted Marine Autonomy?
Abstract. This study outlines the development and testing of a Digital Twin Ocean (DTO) framework, aimed at improving coastal ocean forecasts through the use of autonomous underwater gliders. A fleet of gliders were deployed in the western English Channel during August–September 2024 to collect measurements of temperature, salinity, chlorophyll and oxygen, aiming to track the movement of the harmful algal bloom Karenia mikimotoi. Measurements were assimilated into a very high resolution (1.5 km) numerical forecast model, with an implementation of biogeochemistry data assimilation for this purpose. The model forecast was then used by a probabilistic uncertainty model to plan a series of waypoints to navigate the glider fleet towards features of interest. By utilising a continuous feedback loop of measurement, prediction, guidance, and refinement a system with real time coupling between the real ocean environment and its digital counterpart has been established.
Building upon a prior pilot study of Ford et al. (2022), this work improves every element of the system to addresses several limitations of the prior configuration. Whilst a bloom was present in the wider area, measurements and modeling suggest it didn't enter the glider operation zone. Despite this and other operational challenges the mission clearly demonstrates the benefits of such a system. The ability to simultaneously track multiple features of interest, namely chlorophyll and oxygen, would not have been possible with a single glider resulting in significant benefits to the system. Furthermore, the improvement to biogeochemical forecasting has been demonstrated through a series of post mission experiments, highlighting the advantages of high temporal resolution observations and increased spatial resolution of the model.
- Preprint
(2912 KB) - Metadata XML
- BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2025-3346', Anonymous Referee #1, 27 Aug 2025
-
RC2: 'Comment on egusphere-2025-3346', Anonymous Referee #2, 24 Jan 2026
This manuscript presents the use of ocean gliders to improve coastal forecasting in the western English Channel during a period of two months. It is an important regional effort to advance coastal ocean forecasting that couples multi-glider adaptive sampling, high-resolution biogeochemical modeling (1.5 km), and data assimilation with probabilistic path planning.
General comments
The authors indicate the first implementation of the AMM15 (1.5 km) with biogeochemical data assimilation, which is a non-trivial advance, and it is a substantial contribution for the regional coastal forecasting. They have also demonstrated that 7 km shelf models can fundamentally miss hypoxia-relevant variability that has an important impact on the marine ecosystem. The latter demonstrates that the multi-glider coordination is not optional but necessary when tracking coupled biogeochemical features. The results are relevant to coastal forecasting, ecosystem monitoring, and early warning systems. That said, the manuscript sometimes oversells the DTO concept rhetorically, rather than letting the results speak, so a more restrained tone would improve the overall readiness of the manuscript and demonstrate better the importance of the results. The modeling and assimilation framework is technically sound and appropriate for answering the questions in this study. However, some points require revisiting. The assumptions of the data assimilations are reasonable but weakly justified. Given that conclusions hinge on biogeochemical improvements, this deserves explicit discussion of limitations and representativeness error, not only references. Furthermore, treating AMM15-DT as the best possible ocean state is pragmatic, but the manuscript should state clearly that it is not validation, only relative consistency. Without independent oxygen or chlorophyll validation, some conclusions should be framed more cautiously. Moreover, the change in the path-planning strategy during the mid-mission from uncertainty-driven to gradient-driven sampling is understandable from an operational perspective, but it has not been analyzed sufficiently from a scientific standpoint. The author missed the opportunity to explicitly discuss the significance of the failure of uncertainty in targeting near boundaries, which is a crucial negative result that needs to be addressed.
With all the above in mind, I would recommend the manuscript with a major revision before publishing.
Specific comments:
Lines 18-21: This is correct but very generic, and we miss the scientific motivation of this. It will be useful to mention here why DTOs are needed specifically for coastal forecasting.
Line 26: Here you have to justify why it is essential. Could you explain why autonomy without DTO feedback is insufficient for the coastal system?
Line 26: This is a secondary aspect, and it is not related to the work. You either make it shorter or remove it. The authors should focus more on the scientific return and not on the cost efficiency.
Line 28: This statement is too broad and risks overstating the maturity of the field.
Line 43: The limitations have been listed but biogeochemistry is not explained, especially why the 7 km resolution model and single glider mission fail to capture chlorophyll and oxygen features.
Lines 51-53 and 61-63: The choice of Karenia mikimotoi is scientifically justified, but the chlorophyll from the glider is only a proxy, and the species-specific detection remains unresolved, which is why the adaptive and multivariable sampling is needed. However, the author needs to clarify here if the oxygen is treated as a direct indicator for the bloom or as an independent parameter.
Line 66: This sentence needs to be rephrased as "overselling outcomes," given that the bloom did not enter the study domain, as it is not guaranteed the bloom tracking, but surely indicates the system testing capability.
Line 74: It shifts the reader's focus away.
Line 78: It will be beneficial if the authors specify what the manuscript evaluates here.
Line 84: This statement is vague. It will be important for the authors to describe the frequency of the DTO system. Is it aligned with the real-time observations or performed daily? Furthermore, the authors should specify the quality level of the data incorporated into the model. Is quality control performed in real-time observations?
Line 89: “AI derived” is vague and somewhat fashionable. I would recommend the author here to use accurate terminology.
Lines: 94-95. Further information is needed for the horizontal and vertical resolution? Furthermore, at which depth range were the gliders operated during the experiment? Are shallow Slocum gliders or deep?
Line 97: The authors well stated the issue with the oxygen sensor, which is something that can happen during an experiment. However, the paper should discuss the impact of data assimilation if the oxygen variable was used as an indication of the bloom.
Line 99: Did you focus on the oxygen hysteresis between downcast and upcast profiles? Please describe the handling of the oxygen data in NRT and delay mode.
Line 100: A table that describes each sensor type, frequency, model, and glider is needed for clarity.
Line 110: Could you provide us more details regarding the manual calibration that was performed in the glider chlorophyll observation? Is it about dark counts, correction factor, or something else?
Line 115: Could you please specify the correction methods that were performed?
Line 134: Is there any specific reason why the SLSTR SST was not assimilated, particularly given the value of these observations near the coast?
Line 183: Could you shortly justify the use of median values rather than mean values, especially on the biogeochemical data that are skewed?
Line 209: Why are 6 and 7 days key for chlorophyll-A and dissolved oxygen? Do you have the examination for forecasting the decorrelation scales?
Line 211: I assume you refer to the maximum of chlorophyll and minimum of oxygen.
Line 221: Please link this to the classic adaptive sampling strategy that was used before the DTOs.
Line 260: The trajectories of gliders should be shown in Figure 1 as part of a zoomed-in map.
Line 279: How do you verify that the Karenia bloom that was detected in the Celtic Sea occurred? Please confirm if you carried out independent measurements.
Line 281: Please indicate and report the periods that the system was fully operational. This sentence is very vague.
Line 285: The reader until now has seen an extensive discussion of the DTO architecture and path planning, but with the scope in this manuscript, it is too late. I restructure the manuscript suggested for clarity.
Lines 295-301: The names of experiments, AMM15-NRT, AMM15-DT, etc., are introduced without any name convention.
Lines 303-304: This is an assumption of the reference, not a validation. Please indicate on figure 3 the period of the time-average RMSD.
Line 315. I assume the spatial-average RMSD in Figure 4 is for the whole domain, but what about the results on the small red box where the gliders are operating? Did you notice any influence on assimilation from the number of vertical profiles that assimilated in the red box?
Line: 321-322: The gliders mentioned here target ‘event states; it will be useful for the author to mention here the glider limitations regarding this. The glider is progressing ~21 km/day. The thresholds are reasonable, but what kind of criteria have you used to define them for the study area?
Lines 331-335: It is indicated that the variances are caused only by the varying resolutions of the models, rather than any variations in configuration? In addition, it is mentioned that the difference is significant due to the assimilation, yet you do not provide a numerical value for that difference. Is the satellite chlorophyll assimilation behaving differently across resolutions?
Line 336: It has been mentioned that in the bottom oxygen the difference is large, but it has been discussed how the models are so sensitive with the resolution. Are there other mechanisms that can affect such things as the timescale of ventilation?
Lines 344-345: That shows only the variability between the models and does not validate the models. What is the variability in the glider oxygen observations in the red box compared to the simulations for the same area?
Lines 362-364: In the section, the authors should compare the post-mission paths with the actual operation paths flown during the mission, making clear how assimilation alters their trajectory with what the glider should have done in practice.
Lines 370-372. The authors should consider that typical glider progress is ~21 km/day and explicitly interpret path divergences of similar magnitude as effectively a one-day navigation error, with direct consequences for missed or delayed event sampling.
In section 4: The conclusion would benefit from a focused summary that synthesizes the main scientific findings rather than restating the framework or future potential. In particular, the authors should explicitly summarize the demonstrated impacts of multi-glider assimilation and model resolution.
Citation: https://doi.org/10.5194/egusphere-2025-3346-RC2
Viewed
| HTML | XML | Total | BibTeX | EndNote | |
|---|---|---|---|---|---|
| 1,068 | 116 | 29 | 1,213 | 53 | 51 |
- HTML: 1,068
- PDF: 116
- XML: 29
- Total: 1,213
- BibTeX: 53
- EndNote: 51
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
General comments
The article describes the use of glider BGC and physical targeted observations to improve coastal ocean forecasts in the western English Channel. The gliders are driven in real time based on an automatic procedure to optimize the sampling of chlorophyl maximum and oxygen minimum associated with spring bloom.
Targeted observations from autonomous platforms to improve monitoring and forecast of the coastal ocean is a very promising approach for different real time applications. This paper addresses its implementation in real time with a coupled physical and BGC ocean monitoring and forecasting system to track specific events and drive the gliders in the area of interest. It follows a previous paper from Ford et al. (2022) which was the proof of concept with some identified limitations. Here, the model resolution is increased and three gliders instead of one are sampling the area and are assimilated in real time.
This paper focus on the sensitivity of the BGC analysis, forecasts and glider automated path to the model increased resolution from 7 km to 1.5 km and the assimilation of better qualified observations.
Unfortunately, the paper in its current status lacks some validation of the sensitivity experiments to support the model estimate improvements mentioned in the conclusion. There is no diagnostic to evaluate the realism of the different simulations between each others. The impact of a higher resolution model and assimilation of DT versus NRT observations is analysed by comparing the different simulations with the simulation at high resolution with the DT observations assimilated, including the glider ones. There is a need to show that the changes between the simulations are associated to more realistic analysis and forecasts, at least for the oxygen minimum and chlorophyl maximum. This could be achieved by comparing the analysis and forecasts to the assimilated observations and if possible, to other independent observations, as done in Ford et al. (2022).
I would recommend major revision before publication.
Specific comments:
l.13: “features of interest, namely chlorophyll and oxygen”: the later ones are variables not features. From my understanding, the feature of interest in this paper is the spring bloom associated with oxygen minimum and chlorophyll-a maximum.
L.27: making our observations more efficient and through the use of low-carbon autonomous platforms, such as ocean gliders (Testor et al. (2019)): “and” can be removed.
l.49: Three issues in previous study by Ford et al. are mentionned as improved in this paper but I did not find where the reduction of “biases in the observational source” is addressed in this paper.
l.140: …from satellite the physics observations… -> …from satellite, the physical observations…
l.175: Does NEMO and ERSEM shares the same spatial resolution at 1.5 km in the AMM15 configuration?
l.223: I would add for clarity some information in the sentence, even if mentioned later: …transect based sampling of the feature of interest … disregarding forecast uncertainty …
l.299: Does the other in-situ observations than the glider ones are also DT quality in AMM15-DT?
l.303-312: In this section the impact of glider data assimilation is discussed but the simulations that are compared to infer it are AMM15-DT and AMM15-NoG-“NRT”. Those simulations differ also in processing level of the assimilated data sets, in addition to the glider assimilation. The comparison should be computed against the AMM15-NRT so the differences could solely be attributed to glider data assimilation.
Figure 4: “depth average”: Can you specify over which range of depth the average is computed?
l.351: the impact of model differences on the glider: Do you mean on the glider path?
l.369-372: Can you also interpret the different paths of the gliders in AMM15-NRT and AMM15-NoG by looking at the Chl-a maximum and Oxygen minimum locations in the 2 simulations? Which of the two path is the better sampling the different extrema?
l.387: The improved realism of the simulation with the increased resolution is not shown in the paper. Only differences between the simulations are analysed.
l.396: I do not find where the impact on glider path of using DT mode observations instead of NRT ones is shown. Figure 10 shows the different paths for the simulation with/without glider NRT observations assimilated.
Technical corrections
l.10: to addresses
I would suggest checking the format of references in the text. For example, line 24: “… within marine autonomy Ford et al. (2022).” could be changed to ” … within marine autonomy (Ford et al., 2022).”
l.68: assoiated -> associated