Beyond chlorophyll: machine learning estimates of diagnostic phytoplankton pigments from multispectral ocean colour data
Authors: David Moffat, Angus Laurenson, Victor Martinez-Vicente, Gemma Kulk, Xuerong Sun, Robert J. W. Brewin, Shubha Sathyendranath
Organizations: Plymouth Marine Laboratory, Plymouth, UK · National Center for Earth Observation, Plymouth Marine Laboratory, UK · Department of Earth and Environmental Sciences, Centre for Geography and Environmental Science, University of Exeter, Cornwall, United Kingdom
Phytoplankton play a central role in marine ecosystems and the global carbon cycle, with different groups contributing differently to ocean biogeochemical processes. While standard techniques exist for monitoring phytoplankton concentration from ocean-colour data, their community composition remains difficult to observe at large scales. Chlorophyll-a, widely available from satellite ocean-colour observations, is commonly used as a measure of phytoplankton biomass but provides limited information on taxonomic composition. Accessory pigments, some of which are diagnostic of important phytoplankton groups, offer additional information on community structure, but their retrieval from ocean-colour data is challenging because of limited spectral resolution and strong covariance with chlorophyll-a. In this study, we evaluate machine learning methods for estimating diagnostic pigment concentrations from multispectral satellite observations. Using a global dataset of 33,640 High Performance Liquid Chromatography (HPLC) measurements matched with ESA Ocean Colour Climate Change Initiative (OC-CCI) reflectance data, we compare Random Forest and TabPFN models trained on multispectral reflectance with baseline models using chlorophyll-a alone. A temporally stratified validation scheme is employed to reduce the effects of autocorrelation. Results show that multispectral models consistently outperform approaches based solely on satellite-derived chlorophyll-a, demonstrating that ocean-colour reflectance contains additional information relevant to pigment discrimination. Improvements vary by pigment, with those strongly correlated with chlorophyll-a showing limited gains, while others exhibit substantial improvement. These findings highlight the potential of machine learning to extract ecologically relevant information from satellite data beyond conventional chlorophyll-based approaches.
Figures & tables
Pigment
Marker type
Taxonomic affiliation
19’-butanoyloxyfucoxanthin
Ambiguous
Haptophytes, Pelagophytes
19’-hexanoyloxyfucoxanthin
Ambiguous
Chrysophytes, Haptophytes
Alloxanthin
Unambiguous
Cryptophytes
Monovinyl Chlorophyll- a
Ambiguous
All phytoplankton, except Prochlorococcus
Divinyl chlorophyll- a
Unambiguous
Prochlorococcus
Chlorophyll-b
Ambiguous
Chlorophytes, Euglenophytes, Prasinophytes
Table 1: Overview of pigments estimated by the machine learning models and their taxonomic affiliation with phytoplankton groups. Information taken from 31 , with the taxonomic affiliation identified for those phytoplankton groups in which the specific pigment is always or often present. Some pigments are unique markers to certain phytoplankton groups (i.e., unambiguous) and others appear across multiple phytoplankton groups (i.e., ambiguous).
Figure 1: Relationships between chlorophyll -a and other pigments in the in situ dataset, including the Pearson correlation in log space. The bottom right figure shows the Pearson correlation in log space of each pigment compared with each other pigment.
Figure 2: Global map of match-up dataset. Blue points show locations of training data, whilst red points show data held back for validation, from years 2002, 2007, 2012, and 2017
Model
Method
Predictor variables
Max N (Training)
Max N (Validation)
TabPFN
Tabular Foundation Model
Rrs , Kd(490) , Chl- a
6,625
1,701
RF
Random Forest model
Rrs , Kd(490) , Chl- a
6,625
1,701
Baseline
Random Forest model
Chl- a
6,625
1,701
Table 2: Overview of the different machine learning models developed to predict eight phytoplankton pigments using ocean colour satellite observations. Details of the models are provided in Section 3.4 . The maximum total number of observations ( N ) used for training and validation of the models is also provided, but specific numbers vary per pigment, due to not all pigments being captured for all samples.
Pigment
Model
n
r
bias
RMSD
CP-RMSD
R 2
TChlb
predFM
6656
0.76
3.22
3.41
1.14
0.52
TChlb
predRF chl
6656
0.71
3.21
3.41
1.15
0.13
TChlb
predRF
6656
0.76
3.21
3.41
1.15
0.47
Allo
predFM
4631
0.65
3.68
3.86
1.18
0.36
Allo
predRF chl
4631
0.58
3.68
3.86
1.18
-0.25
Allo
predRF
4631
0.62
3.68
3.86
1.18
0.29
Table 3: Comparison of phytoplankton pigment retrieval performance for the foundation model (predFM), chlorophyll-only random forest (predRF chl), and full random forest (predRF). Statistics shown are sample size (n), Pearson correlation (r), bias, root mean square difference (RMSD), Centre-pattern root mean square difference (CP-RMSD), and coefficient of determination (R 2 ).
Figure 3: Scatter plots of predicted versus measured pigment concentrations for 19’-butanoyloxyfucoxanthin (a), 19’-butanoyloxyfucoxanthin (b), alloxanthin (c). The x-axis shows HPLC-derived measurements and the y-axis shows model predictions of each pigment in mg m -3 . Results for the training (blue) and validation (orange) datasets are presented for each type of model, i.e., the TabPFN (a,d,g) and RF (b,e,h) models that incorporate spectral information and the baseline model (c,f,i) that is based on chlorophyll-a as a predictor.
Figure 4: As in Figure 3 , for chlorophyll -b (a), fucoxanthin (b), peridinin (c), and zeaxanthin (d).
Figure 5: Comparison of model performance on the validation dataset. Panel (a) shows the validation R2 values obtained for each pigment and model configuration. Panel (b) shows the improvement in R2 achieved by the Random Forest model using multispectral reflectance relative to the chlorophyll-only Random Forest baseline, illustrating the additional predictive information contained within the reflectance observations.
Figure 6: SHAP Explainability plots showing the impact of each input variable on pigment concentrations. The x-axis represents the change in predicted pigment concentrations on a logarithmic scale, while the y-axis lists the input features used in the RF models. Features are ordered by importance, with the top feature having the greatest influence. The colours indicate the distribution of feature values and their contribution to the predicted pigment concentrations. Panels (a–g) correspond to the different pigments.
Figure 7: Global maps of predicted pigment concentrations (left column) and pigment-to-chlorophyll- a ratios (right column) for May 2010 generated using the Random Forest model that incorporates spectral information. Panels (a–h) show 19’-butanoyloxyfucoxanthin, 19’-hexanoyloxyfucoxanthin, alloxanthin and total chlorophyll- b .
Figure 8: Global maps of predicted pigment concentrations (left column) and pigment-to-chlorophyll- a ratios (right column) for May 2010 generated using the Random Forest model. Panels (a–f) show fucoxanthin, peridinin and zeaxanthin.
Marine ecosystems are increasingly impacted by climate change, necessitating tools to identify and predict spatial habitat information. To build such tools, ecological marine provinces, "eco-provinces", ecologically meaningful regions in the global ocean can be used. We use unsupervised machine learning (ML) to identify eco-provinces with corresponding uncertainty measures based on output of a global simulation of phytoplankton functional types. Our work aims to create a proof of concept to predict eco-provinces based on satellite ocean color data. To do so, we develop a hierarchy of explainable dense ensemble networks to infer how well the eco-provinces can be detected from modeled ocean color fields. Key results include that the delineated eco-provinces are both ecologically meaningful and can be inferred with high skill. However, no straightforward relationship was found where adding more input data to the network consistently improves inference skill, and there is an intricate tradeoff between inputs and prediction fidelity. Our work is a case for optimism and a cautionary tale of needing uncertainty quantification and careful validation of prediction fidelity.
University of California Davis, Davis, CA, 95616, USA · University of Washington, Seattle, WA, 98195, USA · NOAA Geophysical Fluid Dynamics Laboratory, Princeton, NJ, 08540, USA +1
Machine learning (ML) is widely used for the development of ocean colour algorithms, but most studies focus on model parameter training and hyperparameter tuning. The optimisation of the data that feeds the models - i.e., Feature Engineering (FE) - is not fully explored. We assess the impact of FE in ocean colour machine learning models and we propose an optimisation framework that includes seven sequenced levels of data transformation: i. band choice, ii. log scaling, iii. spectral shape normalisation, iv. index extraction, v. principal component analysis, vi. feature scaling, and vii. zero-to-one scaling. We demonstrate the application for Multi-layer perceptron, Support Vector Machines, and eXtreme Gradient Boosting Trees on Sentinel-3 OLCI observations in the Norwegian coastal waters. The models are trained to estimate Chlorophyll-a concentration [Chl-a] and Secchi disk depth (Zsd). Results show that accuracy is highly variable among FE found in six studies using Sentinel-3 OLCI and the ones that we optimise. The R range from 0.01 to 0.55 for [Chl-a] and from 0.15 to 0.68 for Zsd, where the optimised FE shows the top results. The ML models with optimised FE could also improve by two times the R and reduce up to 63% of the mean absolute error when compared to CHL_OC4ME and CHL_NN standard algorithms. Nevertheless, no common optimised FE is found for all target variables and ML models, suggesting that FE optimisation is necessary for each application. Therefore, our proposed framework can be key for improving the accuracy of water quality monitoring in coastal waters.
Edson Silva, Julien Brajard, Simon Cappe +2
Nansen Environmental and Remote Sensing Center, and Bjerknes Centre for Climate Research, Bergen, Vestland, Norway · Nansen Environmental and Remote Sensing Center, Bergen, Vestland, Norway
Marine plankton underpin aquatic food webs and play a key role in global CO2 sequestration, making reliable species identification critical for understanding ocean health and climate feedbacks. Existing classification models perform well on individual collections but fail to generalize across instruments and environments due to isolated training datasets and inconsistent labels. To address this, we introduce Planktonzilla-17M, a unified dataset consolidating publicly available plankton image collections spanning thirteen imaging systems. It comprises 17.4 million images with standardized taxonomy and geo-environmental metadata, including 3.74 million plankton images spanning over 602 taxonomic classes, of which 201 are identified at the species level, making it the largest and most comprehensive plankton image dataset to date. Using this large-scale dataset, we perform a controlled comparison between supervised and CLIP-style image--text training on a shared ViT backbone. We find that a supervised classifier matches or exceeds CLIP-style training when trained using taxonomic lineage as text. We further observe that BioCLIP and BioCLIP2 perform poorly on plankton in zero-shot and few-shot settings. Leveraging Planktonzilla-17M improves plankton classification performance, highlighting the limitations of current biological foundation models in marine imaging domains.
Alan Gerson Contreras Montanares, Luis Valenzuela, Luis Martí +1
Inria Chile Research Center · Inria — Institut national de recherche en sciences et technologies du numérique · Av. Apoquindo 2827, Las Condes 7550268 Chile.