pith. sign in

arxiv: 2606.27168 · v1 · pith:DYS2JQTFnew · submitted 2026-06-25 · 🧬 q-bio.QM · nlin.CD

Deep learning model emulators for marine biogeochemistry forecasting from days to decades

Jozef Skakala , Ieuan Higgs , David Moffat This is my paper

Pith reviewed 2026-06-26 01:26 UTC · model grok-4.3

classification 🧬 q-bio.QM nlin.CD
keywords marine biogeochemistrydeep learning emulatorsLSTM1D CNNocean forecastingphytoplankton bloomsclimate projectionsmodel emulation
0
0 comments X

The pith

Deep learning emulators match or beat a marine biogeochemistry model in both 10-day forecasts and multi-decadal projections while remaining stable.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tests whether two deep-learning architectures can stand in for a complex one-dimensional marine biogeochemistry model. LSTM networks emulate selected variables at daily steps, while physics-informed 1D CNNs emulate the full water-column system. Both stay stable across multi-decadal runs, reproduce the parent model's decadal climate projections, and forecast 10-day trajectories, including the timing of spring phytoplankton blooms years ahead. When the emulators are instead trained on reanalysis data they improve forecast skill scores on phytoplankton, zooplankton and other variables relative to the original model. The work is framed as a step toward cheaper, higher-skill ocean predictions once the approach is tested in three-dimensional regional models.

Core claim

Both the LSTM and 1D-CNN emulators remain largely stable over multi-decadal timescales and accurately reproduce the parent model in decadal climate projections and short-range (10-day) forecasting applications, including the ability to predict the timing of phytoplankton Spring blooms several years in advance; when trained on reanalysis data the emulators substantially outperform the parent model's forecast skill score for several key ecosystem variables.

What carries the argument

LSTM networks and physics-informed 1D CNNs that take ocean physics simulator inputs and output daily-resolution biogeochemical states.

If this is right

  • Operational marine forecasting could run at higher resolution or with larger ensembles at the same computational budget.
  • Climate-scale simulations of marine ecosystems become feasible at lower cost while preserving daily variability and extreme-event statistics.
  • Explainability methods applied to the emulators can reveal key physical drivers of emergent ecosystem behaviour.
  • Marine autonomous observing systems could use the emulators for on-board short-range prediction.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If the 1D-to-3D translation holds, the same architectures could be coupled to existing ocean physics models to produce hybrid Earth-system forecasts.
  • Training directly on reanalysis rather than model output may allow the emulators to correct systematic biases present in the parent biogeochemistry model.
  • Daily-resolution emulation opens the possibility of ingesting real-time satellite or in-situ data streams for continuous correction of the forecast state.

Load-bearing premise

That success inside a one-dimensional water-column model will carry over once horizontal advection, mesoscale features and three-dimensional boundary conditions are added.

What would settle it

A side-by-side run of the emulators inside a full three-dimensional regional model that shows whether forecast skill scores on phytoplankton and zooplankton remain higher than those of the parent model.

Figures

Figures reproduced from arXiv: 2606.27168 by David Moffat, Ieuan Higgs, Jozef Skakala.

Figure 1
Figure 1. Figure 1: The ocean surface time series for four selected biogeochemistry variables from the 1D GOTM-FABM-ERSEM simulation spanning the 2002-2025 period. The different data (train￾ing/validation/test) are clearly marked by the vertical lines [PITH_FULL_IMAGE:figures/full_fig_p010_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Six selected ocean biogeochemistry and physics variables from the GOTM-FABM￾ERSEM simulation, shown in Hovm¨oller plots. As in Fig.1, the training, validation and test periods are clearly marked. pened throughout the Spring-Summer period nearly every day, with slightly bigger gaps in the Autumn-Winter period. The assimilation was done for the 1st January 2004 - 31st December 2025 period, after the initial … view at source ↗
Figure 3
Figure 3. Figure 3: The LSTM prediction of nine selected variables (see Tab.3) across the validation (2016-2018) and test (2019-2025) data period. The predicted LSTM ensemble median value is in blue with two quartiles around the median in aqua color. The simulator validation and test data are in red. The magnitude of the inter- and intra-annual variability simulated by the emula￾tors is broadly comparable to that of the GOTM-… view at source ↗
Figure 4
Figure 4. Figure 4: Similar to Fig.3, but simulated by the physics-informed 1D CNN emulator. The plotted variables are defined in Tab.3. of the Appendix demonstrate how well the size of anomalies can be forecast by the em￾ulators on a decadal scale as a function of lead time (in years). The Figures compare the prediction (RMSE) skill with the scale of the anomalies (see the normalized RMSE in Eq.4), which amounts to the RMSE … view at source ↗
Figure 5
Figure 5. Figure 5: Hovm¨oller plots compared for eight selected variables across the test period (2019- 2025) between the test data (left-hand panels) and the 1D CNN emulator (middle column). The difference between those two (emulator minus test data) is shown in the right-hand panels. Of particular interest is the ability to forecast specific events that have a major im￾pact on marine biogeochemistry in temperate seas, such… view at source ↗
Figure 6
Figure 6. Figure 6: The upper-row bar-plots show the values of mean absolute differences between the emulated seasonal climatology and the seasonal climatology from the training data. The val￾ues were normalized by the mean climatology value from the training data and are shown for a range of selected variables. The bottom-row plots show the ratio of mean absolute values of daily anomalies from the emulator runs and from the … view at source ↗
Figure 7
Figure 7. Figure 7: The Pearson correlation calculated for the LSTM ensemble median and validation￾test data separately for each year throughout the validation and test period (2016-2025). The Pearson correlation is calculated for the selected biogeochemistry variables. We investigated the stability of the decadal LSTM forecast relative to a red, skewed, multiplicative noise applied to its inputs. The multiplicative noise was… view at source ↗
Figure 8
Figure 8. Figure 8: The LSTM ensemble median RMSE relative to the reference ERSEM simulation for each year of the validation-test data period. The same panels show LSTM RMSE normalized to the RMSE of a zero-anomaly prediction model (no knowledge of anomalies, all future prediction is climatology). rably smaller, as shown in Fig.A8:B of the Appendix. This means that the LSTM em￾ulator, once trained, is highly stable with respe… view at source ↗
Figure 9
Figure 9. Figure 9: The timing of bloom onset, expressed as the day of year, is shown for the Spring bloom (upper panel) and the second bloom occurring in late Summer to early Autumn (lower panel). The Spring bloom onset was defined as the date when surface total chlorophyll-a first exceeded a threshold of 0.5 mg.m−3 . The onset of the second bloom was defined as the date when surface total chlorophyll-a exceeded the same thr… view at source ↗
Figure 10
Figure 10. Figure 10: The RMSE skill of short-range (up to 10-day) prediction of LSTM emulator (en￾semble median) relative to the persistence RMSE skill. The skill is compared across the 1-10 day forecast lead time. The skill was averaged across 3 separate years (from the 2022-2024 period). The 2025 year was left out due to the degradation of model skill in that year. produce much more skilled forecast than the free run simula… view at source ↗
Figure 11
Figure 11. Figure 11: The upper two panels (one for LSTM, another for 1D CNN) compare the surface total chlorophyll-a concentrations for the reanalysis (red), the emulator (blue), the GOTM￾FABM-ERSEM simulator forecast (green) and the assimilated OC-CCI observations (pink dots). For LSTM the blue represents ensemble median and the aqua colour the two quartiles around median. Prediction of anomalies is shown for the emulators i… view at source ↗
Figure 12
Figure 12. Figure 12: Left-hand panels A, C show the MAE skill of the emulators to predict the reanal￾ysis data across the test period (2019-2025) relative to the same MAE skill of the free run. The right-hand panels B, D show the same for low-pass filtered data with a 21 day window. This is to demonstrate the impact of hard-to-predict assimilation cycle on the emulator skill to predict anomalies. The upper panels A-B show the… view at source ↗
Figure 13
Figure 13. Figure 13: The top panel shows the daily mean ±1 standard deviation of surface diatom chlorophyll over the 2017-2025 validation and test period; shaded regions denote key seasonal regimes of phytoplankton behaviour. The remaining panels show, for each seasonal regime, the mean ±1 standard deviation across years of the maximum absolute sensitivity of surface diatom chlorophyll to each input. Seasonal regimes windows … view at source ↗
Figure 14
Figure 14. Figure 14: Mean lag at which sensitivity of surface diatoms chlorophyll-a (A), vertically av￾eraged diatoms chlorophyll-a (B), and sea bottom dissolved oxygen concentration (C) to each forcing input first drops below 50% (filled circle), 10% (circle with cross), and 5% (small open circle) of its peak mean value, summarising the effective memory timescales across the inputs. –38– [PITH_FULL_IMAGE:figures/full_fig_p0… view at source ↗
Figure 15
Figure 15. Figure 15: Top panel shows predicted and target value of surface diatoms chlorophyll over 2024-2025. Remaining panels show temporal evolution of absolute sensitivity of surface diatoms chlorophyll to each biological/forcing input, taking the maximum sensitivity within the 30-day lookback window for historic variables. Darker colours indicate periods of stronger sensitivity. Zoomed into 2024-2025. –39– [PITH_FULL_IM… view at source ↗
read the original abstract

Deep-learning emulators have emerged as a promising approach for reducing the computational cost of Earth System Models while potentially improving forecasting skill. Here, we demonstrate the successful emulation of a high-complexity marine biogeochemistry model within a simplified one-dimensional water-column framework. We explore two emulator architectures: Long Short-Term Memory (LSTM) neural networks that emulate a selected subset of variables at daily resolution, and physics-informed one-dimensional Convolutional Neural Networks (1D CNNs) that emulate the full pelagic system throughout the water column also at daily resolution. Using ocean physics simulator inputs, both emulators remain largely stable over multi-decadal timescales and accurately reproduce the parent model in both decadal climate projections and short-range (10-day) forecasting applications. The former includes the ability to predict the timing of phytoplankton Spring blooms several years in advance. When trained on reanalysis data, the emulators substantially outperform the parent model's forecast skill score for several key ecosystem variables, including phytoplankton and zooplankton. If similar performance can be achieved in three-dimensional regional applications, these emulators could provide substantially higher-quality predictions at a fraction of the computational cost. We further apply novel explainability techniques to identify key drivers of emulator behaviour and gain insights into emergent ecosystem dynamics. Performance is evaluated using a range of metrics, including the reproduction of daily variability and extreme events. These approaches have considerable potential for future applications in operational forecasting, climate-scale simulations, and marine autonomous systems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript develops and tests two deep-learning emulators (LSTM networks for selected variables at daily resolution and physics-informed 1D CNNs for the full pelagic system) of a high-complexity marine biogeochemistry model inside a strictly one-dimensional water-column framework. It reports that both emulators remain stable over multi-decadal timescales, reproduce the parent model in decadal climate projections and 10-day forecasts (including multi-year advance prediction of phytoplankton spring-bloom timing), and, when trained on reanalysis data, substantially outperform the parent model on several key variables. Novel explainability methods are applied to identify drivers, and the work is positioned as a step toward lower-cost operational and climate-scale forecasting if the 1D results generalize to three-dimensional settings.

Significance. If the reported 1D stability, bloom-timing skill, and reanalysis outperformance hold under rigorous validation, the work would demonstrate a viable route to computationally inexpensive marine biogeochemistry emulators with potential value for operational forecasting and long-term simulations. The application of explainability techniques to emergent ecosystem dynamics is a positive addition. However, the absence of quantitative error bars, explicit training/validation splits, drift diagnostics, and any 2D/3D sensitivity tests limits the immediate impact.

major comments (2)
  1. [Abstract/Methods] Abstract and Methods: the claims of 'successful reproduction,' 'largely stable' multi-decadal behavior, and 'substantially outperform' on reanalysis data are presented without reported quantitative error bars, explicit training/validation/test splits, or drift diagnostics over decades; these omissions make it impossible to assess whether the central performance assertions are robust.
  2. [Abstract/Discussion] Abstract and Discussion: all quantitative results (multi-decadal stability, spring-bloom prediction years ahead, reanalysis outperformance) are obtained inside a 1-D column where horizontal advection, mesoscale features, and lateral boundary conditions are absent by construction. No ablation study, 2-D schematic experiment, or sensitivity test is described that would indicate whether the LSTM or 1D-CNN architectures retain skill or stability once these processes must be emulated or supplied as inputs; this directly bears on the manuscript's claims for operational forecasting and climate-scale use.
minor comments (1)
  1. [Abstract] The abstract states that performance is evaluated with 'a range of metrics, including the reproduction of daily variability and extreme events,' but does not list the specific metrics or show example time series; adding these would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and detailed review. The comments highlight important aspects of robustness and scope that we address point-by-point below. Where the suggestions strengthen the manuscript without altering its core 1D focus, we have incorporated revisions.

read point-by-point responses

  1. Referee: [Abstract/Methods] Abstract and Methods: the claims of 'successful reproduction,' 'largely stable' multi-decadal behavior, and 'substantially outperform' on reanalysis data are presented without reported quantitative error bars, explicit training/validation/test splits, or drift diagnostics over decades; these omissions make it impossible to assess whether the central performance assertions are robust.

    Authors: We agree that explicit quantitative support improves clarity. The revised manuscript now includes (i) error bars on all reported skill scores and stability metrics, (ii) a clear statement of the training/validation/test split (70/15/15 with temporal blocking to avoid leakage), and (iii) drift diagnostics (mean and maximum absolute drift per variable over 20-year integrations). These additions appear in the updated Methods and Results sections and in new supplementary figures. revision: yes

  2. Referee: [Abstract/Discussion] Abstract and Discussion: all quantitative results (multi-decadal stability, spring-bloom prediction years ahead, reanalysis outperformance) are obtained inside a 1-D column where horizontal advection, mesoscale features, and lateral boundary conditions are absent by construction. No ablation study, 2-D schematic experiment, or sensitivity test is described that would indicate whether the LSTM or 1D-CNN architectures retain skill or stability once these processes must be emulated or supplied as inputs; this directly bears on the manuscript's claims for operational forecasting and climate-scale use.

    Authors: The study is deliberately restricted to a 1D water-column setting, as stated in the title, abstract, and introduction; the text already qualifies all forecasting and climate claims with the conditional 'if similar performance can be achieved in three-dimensional regional applications.' We have expanded the Discussion to reiterate this scope limitation and to outline the additional inputs (horizontal velocities, lateral boundaries) that would be required for 3D extension. Performing 2D/3D ablation or sensitivity experiments lies outside the present scope, which is to establish baseline emulator viability in 1D before such extensions. revision: partial

Circularity Check

0 steps flagged

No circularity: performance claims rest on held-out test data and reanalysis, not self-referential definitions

full rationale

The paper trains LSTM and 1D-CNN emulators on parent-model output and reanalysis data, then evaluates reproduction of variables, multi-decadal stability, bloom timing, and forecast skill on held-out periods. These are standard supervised-learning metrics on independent splits; no equation or claim reduces by construction to a fitted parameter or self-citation. The 1D-to-3D extrapolation is explicitly flagged as untested future work rather than assumed. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claims rest on standard supervised-learning assumptions (i.i.d. training/test splits, sufficient data coverage of extremes) plus the domain assumption that 1D vertical dynamics dominate the variables of interest. No new physical entities or ad-hoc constants are introduced in the abstract.

axioms (1)
  • domain assumption The 1D water-column framework captures the dominant vertical processes that determine the target biogeochemical variables.
    Invoked when the authors extrapolate 1D results to potential 3D use.

pith-pipeline@v0.9.1-grok · 5793 in / 1471 out tokens · 20691 ms · 2026-06-26T01:26:12.718094+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

174 extracted references · 12 canonical work pages · 2 internal anchors

  1. [1]

    org/abs/2406.01465

    AIFS--ECMWF's data-driven forecasting system , author=. arXiv preprint arXiv:2406.01465 , year=

    work page arXiv

  2. [2]

    Science Advances , volume=

    Generative emulation of weather forecast ensembles with diffusion models , author=. Science Advances , volume=. 2024 , publisher=

    2024

  3. [3]

    Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

    Deep residual learning for image recognition , author=. Proceedings of the IEEE conference on computer vision and pattern recognition , pages=

  4. [4]

    Science , volume=

    Learning skillful medium-range global weather forecasting , author=. Science , volume=. 2023 , publisher=

    2023

  5. [5]

    Nature , volume=

    Accurate medium-range global weather forecasting with 3D neural networks , author=. Nature , volume=. 2023 , publisher=

    2023

  6. [6]

    FourCastNet: A Global Data-driven High-resolution Weather Model using Adaptive Fourier Neural Operators

    Fourcastnet: A global data-driven high-resolution weather model using adaptive fourier neural operators , author=. arXiv preprint arXiv:2202.11214 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  7. [7]

    Journal of Advances in Modeling Earth Systems , volume=

    A reliable generative adversarial network approach for climate downscaling and weather generation , author=. Journal of Advances in Modeling Earth Systems , volume=. 2025 , publisher=

    2025

  8. [8]

    Earth's Future , volume=

    AI-driven weather forecasts to accelerate climate change attribution of heatwaves , author=. Earth's Future , volume=. 2025 , publisher=

    2025

  9. [9]

    Nature Communications , volume=

    Artificial intelligence for modeling and understanding extreme weather and climate events , author=. Nature Communications , volume=. 2025 , publisher=

    2025

  10. [10]

    Geophysical Research Letters , volume=

    Samudra: An AI global ocean emulator for climate , author=. Geophysical Research Letters , volume=. 2025 , publisher=

    2025

  11. [11]

    Ocean Science , volume=

    Long-term prediction of the Gulf Stream meander using OceanNet: a principled neural-operator-based digital twin , author=. Ocean Science , volume=. 2025 , publisher=

    2025

  12. [12]

    arXiv preprint arXiv:2509.12490 , year=

    SamudrACE: Fast and accurate coupled climate modeling with 3D ocean and atmosphere emulators , author=. arXiv preprint arXiv:2509.12490 , year=

    work page arXiv

  13. [13]

    arXiv preprint arXiv:2402.04342 , year=

    Building ocean climate emulators , author=. arXiv preprint arXiv:2402.04342 , year=

    work page arXiv

  14. [14]

    Geophysical Research Letters , volume=

    On some limitations of current machine learning weather prediction models , author=. Geophysical Research Letters , volume=. 2024 , publisher=

    2024

  15. [15]

    arXiv preprint arXiv:2405.15412 , volume=

    ORCA: A global ocean emulator for multi-year to decadal predictions , author=. arXiv preprint arXiv:2405.15412 , volume=

    work page arXiv

  16. [16]

    arXiv preprint arXiv:2410.11807 , year=

    Regional ocean forecasting with hierarchical graph neural networks , author=. arXiv preprint arXiv:2410.11807 , year=

    work page arXiv

  17. [17]

    arXiv preprint arXiv:2308.03152 , year=

    Ai-goms: Large ai-driven global ocean modeling system , author=. arXiv preprint arXiv:2308.03152 , year=

    work page arXiv

  18. [18]

    Journal of Marine Systems , volume=

    Estimating time-dependent parameters for a biological ocean model using an emulator approach , author=. Journal of Marine Systems , volume=. 2012 , publisher=

    2012

  19. [19]

    Biogeosciences Discussions , volume=

    Reviews and syntheses: parameter identification in marine planktonic ecosystem modelling , author=. Biogeosciences Discussions , volume=. 2016 , publisher=

    2016

  20. [20]

    Frontiers in Marine Science , volume=

    Identifying drivers and dynamics of phytoplankton in the Black Sea: application of a neural emulator , author=. Frontiers in Marine Science , volume=. 2026 , publisher=

    2026

  21. [21]

    Environmental Modelling & Software , pages=

    Physically-informed Deep Neural Network for the joint prediction of chlorophyll-a and hydrographic 3D structure of the Mediterranean Sea , author=. Environmental Modelling & Software , pages=. 2025 , publisher=

    2025

  22. [22]

    Science of the Total Environment , volume=

    Sub-monthly time scale forecasting of harmful algal blooms intensity in Lake Erie using remote sensing and machine learning , author=. Science of the Total Environment , volume=. 2023 , publisher=

    2023

  23. [23]

    Salva Rühling Cachay, Miika Aittala, Hailey James, and Rose Yu

    Fourcastnet 3: A geometric approach to probabilistic machine-learning weather forecasting at scale , author=. arXiv preprint arXiv:2507.12144 , year=

    work page arXiv

  24. [24]

    Biogeosciences , volume=

    iMarNet: an ocean biogeochemistry model intercomparison project within a common physical ocean modelling framework , author=. Biogeosciences , volume=. 2014 , publisher=

    2014

  25. [25]

    EGUsphere , volume=

    Equation discovery for climate impact: emulating impact models for unexplored climate scenario with interpretable symbolic regression , author=. EGUsphere , volume=. 2026 , publisher=

    2026

  26. [26]

    ICES Journal of Marine Science , volume=

    On conditions for the vernal blooming of phytoplankton , author=. ICES Journal of Marine Science , volume=. 1953 , publisher=

    1953

  27. [27]

    Limnology and oceanography , volume=

    Critical depth and critical turbulence: two different mechanisms for the development of phytoplankton blooms , author=. Limnology and oceanography , volume=. 1999 , publisher=

    1999

  28. [28]

    Ecology , volume=

    Abandoning Sverdrup's critical depth hypothesis on phytoplankton blooms , author=. Ecology , volume=. 2010 , publisher=

    2010

  29. [29]

    Journal of Geophysical Research: Machine Learning and Computation , year=

    Estimating carbon pools in the European Shelf sea environment: replacing reanalysis by model-informed machine learning? , author=. Journal of Geophysical Research: Machine Learning and Computation , year=

  30. [30]

    Nature Reviews Methods Primers , volume=

    Ocean biogeochemical modelling , author=. Nature Reviews Methods Primers , volume=. 2022 , publisher=

    2022

  31. [31]

    Frontiers in Marine Science , volume=

    Advancing marine biogeochemical and ecosystem reanalyses and forecasts as tools for monitoring and managing ecosystem health , author=. Frontiers in Marine Science , volume=. 2019 , publisher=

    2019

  32. [32]

    Quarterly Journal of the Royal Meteorological Society , volume=

    Validating and improving the uncertainty assumptions for the assimilation of ocean-colour-derived chlorophyll a into a marine biogeochemistry model of the northwest European shelf seas , author=. Quarterly Journal of the Royal Meteorological Society , volume=. 2023 , publisher=

    2023

  33. [33]

    Quarterly Journal of the Royal Meteorological Society , pages=

    Assimilation of machine-learning-predicted nitrate to improve the quality of phytoplankton forecasting in the shelf-sea environment , author=. Quarterly Journal of the Royal Meteorological Society , pages=. 2026 , publisher=

    2026

  34. [34]

    Biogeosciences , volume=

    Hybrid machine learning data assimilation for marine biogeochemistry , author=. Biogeosciences , volume=. 2026 , publisher=

    2026

  35. [35]

    Journal of Geophysical Research: Oceans , volume=

    Sensitivity of shelf sea marine ecosystems to temporal resolution of meteorological forcing , author=. Journal of Geophysical Research: Oceans , volume=. 2020 , publisher=

    2020

  36. [36]

    Journal of Geophysical Research: Oceans , volume=

    Assessment of skill and portability in regional marine biogeochemical models: Role of multiple planktonic groups , author=. Journal of Geophysical Research: Oceans , volume=. 2007 , publisher=

    2007

  37. [37]

    Journal of Marine Systems , volume=

    Gaussian anamorphosis extension of the DEnKF for combined state parameter estimation: Application to a 1D ocean ecosystem model , author=. Journal of Marine Systems , volume=. 2012 , publisher=

    2012

  38. [38]

    Method and parameter estimates , author=

    Simultaneous data-based optimisation of a 1D-ecosystem model at three locations in the North Atlantic Ocean: Part 1. Method and parameter estimates , author=. Journal of Marine Research , volume=. 2003 , publisher=

    2003

  39. [39]

    Ocean Modelling (submitted) , year=

    Improving the prediction of ecosystem indicators by assimilating biogeochemical data from multiplatform observing systems , author=. Ocean Modelling (submitted) , year=

  40. [40]

    Progress in Oceanography , volume=

    Control of simulated ocean ecosystem indicators by biogeochemical observations , author=. Progress in Oceanography , volume=. 2025 , publisher=

    2025

  41. [41]

    Handbook of Brain Theory and Neural Networks , editor =

    LeCun, Yann and Bengio, Yoshua , title =. Handbook of Brain Theory and Neural Networks , editor =

  42. [42]

    Neural Computation , volume=

    Long Short-Term Memory , author=. Neural Computation , volume=

  43. [43]

    Journal of Marine systems , volume=

    Description of a flexible and extendable physical--biogeochemical model system for the water column , author=. Journal of Marine systems , volume=. 2006 , publisher=

    2006

  44. [44]

    Environmental modelling & software , volume=

    A general framework for aquatic biogeochemical models , author=. Environmental modelling & software , volume=. 2014 , publisher=

    2014

  45. [45]

    Data assimilation methodology , author=

    Physical and biological modeling in the Gulf Stream region:: I. Data assimilation methodology , author=. Deep Sea Research Part I: Oceanographic Research Papers , volume=. 2000 , publisher=

    2000

  46. [46]

    Journal of Marine Systems , volume=

    The carbonate system in the North Sea: Sensitivity and model validation , author=. Journal of Marine Systems , volume=. 2012 , publisher=

    2012

  47. [47]

    Biogeosciences , volume=

    Relationships between the surface concentration of particulate organic carbon and optical properties in the eastern South Pacific and eastern Atlantic Oceans , author=. Biogeosciences , volume=. 2008 , publisher=

    2008

  48. [48]

    Science of the Total Environment , volume=

    Modelling terrigenous DOC across the north west European Shelf: Fate of riverine input and impact on air-sea CO2 fluxes , author=. Science of the Total Environment , volume=. 2024 , publisher=

    2024

  49. [49]

    Zenodo , year=

    NECCTON: Technical specification of the pelagic lower trophic level products , author=. Zenodo , year=

  50. [50]

    Journal of Marine Systems , volume=

    Modelling the combined impacts of climate change and direct anthropogenic drivers on the ecosystem of the northwest European continental shelf , author=. Journal of Marine Systems , volume=. 2015 , publisher=

    2015

  51. [51]

    Assimilation of machine learning-predicted nitrate to improve the quality of phytoplankton forecasting in the shelf sea environment

    Combining machine learning with data assimilation to improve the quality of phytoplankton forecasting in a shelf sea environment , author=. submitted to QJRMS, arXiv:2508.02400, https://arxiv.org/abs/2508.02400 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  52. [52]

    Journal of Geophysical Research: Oceans , volume=

    Algorithm development and validation for satellite-derived distributions of DOC and CDOM in the US Middle Atlantic Bight , author=. Journal of Geophysical Research: Oceans , volume=. 2008 , publisher=

    2008

  53. [53]

    Frontiers in Marine Science , pages=

    Uncertainty in ocean-color estimates of chlorophyll for phytoplankton groups , author=. Frontiers in Marine Science , pages=

  54. [54]

    Frontiers in Applied Mathematics and Statistics , volume=

    Joint interpolation and representation learning for irregularly sampled satellite-derived geophysical fields , author=. Frontiers in Applied Mathematics and Statistics , volume=

  55. [55]

    Journal of Geophysical Research: Oceans , volume=

    Modeling the carbon fluxes of the northwest European continental shelf: Validation and budgets , author=. Journal of Geophysical Research: Oceans , volume=. 2012 , publisher=

    2012

  56. [56]

    Biogeosciences , volume=

    Investigating ecosystem connections in the shelf sea environment using complex networks , author=. Biogeosciences , volume=. 2024 , publisher=

    2024

  57. [57]

    NEMO-FABM-ERSEM free run 2016-2020 simulations , year =

    Sk. NEMO-FABM-ERSEM free run 2016-2020 simulations , year =

    2016

  58. [58]

    NEMO-FABM-ERSEM reanalysis 2016-2020 simulations , year =

    Sk. NEMO-FABM-ERSEM reanalysis 2016-2020 simulations , year =

    2016

  59. [59]

    NN code and weights , year =

    Sk. NN code and weights , year =

  60. [60]

    Earth System Science Data Discussions , volume=

    Global carbon budget 2024 , author=. Earth System Science Data Discussions , volume=. 2024 , publisher=

    2024

  61. [61]

    Estuarine, coastal and shelf science , volume=

    Carbon dioxide in European coastal waters , author=. Estuarine, coastal and shelf science , volume=. 2006 , publisher=

    2006

  62. [62]

    Carbon and nutrient fluxes in continental margins: A global synthesis , pages=

    Global synthesis1 , author=. Carbon and nutrient fluxes in continental margins: A global synthesis , pages=. 2010 , publisher=

    2010

  63. [63]

    2008 , publisher=

    Chemical oceanography and the marine carbon cycle , author=. 2008 , publisher=

    2008

  64. [64]

    The carbon cycle and atmospheric CO2: Natural variations Archean to present , volume=

    Ocean carbon pumps: Analysis of relative strengths and efficiencies in ocean-driven atmospheric CO2 changes , author=. The carbon cycle and atmospheric CO2: Natural variations Archean to present , volume=. 1985 , publisher=

    1985

  65. [65]

    Frontiers in Marine Science , volume=

    Satellite ocean colour: current status and future perspective , author=. Frontiers in Marine Science , volume=. 2019 , publisher=

    2019

  66. [66]

    Earth-Science Reviews , volume=

    Sensing the ocean biological carbon pump from space: A review of capabilities, concepts, research gaps and future developments , author=. Earth-Science Reviews , volume=. 2021 , publisher=

    2021

  67. [67]

    Frontiers in Marine Science , volume=

    Validation and intercomparison of ocean color algorithms for estimating particulate organic carbon in the oceans , author=. Frontiers in Marine Science , volume=. 2017 , publisher=

    2017

  68. [68]

    Journal of Geophysical Research: Oceans , volume=

    A color-index-based empirical algorithm for determining particulate organic carbon concentration in the ocean from satellite observations , author=. Journal of Geophysical Research: Oceans , volume=. 2018 , publisher=

    2018

  69. [69]

    Remote Sensing of Environment , volume=

    Size-partitioned phytoplankton carbon and carbon-to-chlorophyll ratio from ocean colour by an absorption-based bio-optical algorithm , author=. Remote Sensing of Environment , volume=. 2017 , publisher=

    2017

  70. [70]

    Applied Optics , volume=

    Reconciling models of primary production and photoacclimation , author=. Applied Optics , volume=. 2020 , publisher=

    2020

  71. [71]

    Advances in neural information processing systems , volume=

    A unified approach to interpreting model predictions , author=. Advances in neural information processing systems , volume=

  72. [72]

    Proceedings of the National Institute of Sciences of India , volume=

    On the generalised distance in statistics , author=. Proceedings of the National Institute of Sciences of India , volume=

  73. [73]

    Facta Universitatis, Series: Mathematics and Informatics , pages=

    Mahalanobis distance and its application for detecting multivariate outliers , author=. Facta Universitatis, Series: Mathematics and Informatics , pages=

  74. [74]

    Neural computation , volume=

    A review of recurrent neural networks: LSTM cells and network architectures , author=. Neural computation , volume=

  75. [75]

    IEEE transactions on neural networks and learning systems , volume=

    A survey of convolutional neural networks: analysis, applications, and prospects , author=. IEEE transactions on neural networks and learning systems , volume=. 2021 , publisher=

    2021

  76. [76]

    Remote sensing of Environment , volume=

    Pan-Arctic optical characteristics of colored dissolved organic matter: Tracing dissolved organic carbon in changing Arctic waters using satellite ocean color data , author=. Remote sensing of Environment , volume=. 2017 , publisher=

    2017

  77. [77]

    Frontiers in Marine Science , volume=

    A machine learning model-based satellite data record of dissolved organic carbon concentration in surface waters of the global open ocean , author=. Frontiers in Marine Science , volume=. 2024 , publisher=

    2024

  78. [78]

    Quarterly Journal of the Royal Meteorological Society , volume=

    Global seasonal forecast system version 5 (GloSea5): A high‐resolution seasonal forecast system , author=. Quarterly Journal of the Royal Meteorological Society , volume=

  79. [79]

    Geoscientific Model Development , volume=

    The CO5 configuration of the 7 km Atlantic Margin Model: large-scale biases and sensitivity to forcing, physics options and vertical resolution , author=. Geoscientific Model Development , volume=

  80. [80]

    IEEE transactions on neural networks , volume=

    The graph neural network model , author=. IEEE transactions on neural networks , volume=

Showing first 80 references.