Unveiling 3D Ocean Biogeochemical Provinces in the North Atlantic: A Systematic Comparison and Validation of Clustering Methods
Yvonne Jenniges, Maike Sonnewald, Sebastian Maneth, Are Olsen, Boris P. Koch · 2025
Read Unveiling 3D Ocean Biogeochemical Provinces in the North Atlantic: A Systematic Comparison and Validation of Clustering Methods by Yvonne Jenniges, Maike Sonnewald, Sebastian Maneth, Are Olsen, Boris P. Koch (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-18181.txt.
Introduction
Sea zones drawn by hand can mislead all work past them, so a fair draw wants a tool race, not a gut call. Yvonne Jenniges and co-authors race zone tools in a 2025 study of the north great sea. Near three hundred million reads of salt, heat, and food salts feed the race, through one frame.
Squash first, then a crowd tool won the race; two old school marks proved false friends in the race. One hundred runs of the win pair were wed to a firm map of three hundred and twenty one zones. Past zones, like the Longhurst set, read coarse next to it. Coast, deep, and cold sea cases match the known water kinds well.
Catalog record
| CREATOR | Yvonne Jenniges, Maike Sonnewald, Sebastian Maneth, Are Olsen, Boris P. Koch |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 108 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-18181
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Unveiling 3D Ocean Biogeochemical Provinces in the North Atlantic: A Systematic Comparison and Validation of Clustering Methods Yvonne Jenniges, Maike Sonnewald, Sebastian Maneth, Are Olsen, Boris P. Koch Abstract Defining ocean regions and water masses helps to understand marine processes and can serve downstream tasks such as defining marine protected areas. However, such definitions often result from subjective decisions potentially producing misleading, unreproducible outcomes. Here, the aim was to objectively define regions of the North Atlantic through systematic comparison of clustering methods within the Native Emergent Manifold Interrogation (NEMI) framework (Sonnewald, 2023). About 300 million measured salinity, temperature, and oxygen, nitrate, phosphate and silicate concentration values served as input for various clustering methods (k-Means, agglomerative Ward, and Density-Based Spatial Clustering of Applications with Noise (DBSCAN)). Uniform Manifold Approximation and Projection (UMAP) emphasised (dis-)similarities in the data while reducing dimensionality. Based on systematic validation of clustering methods and their hyperparameters using internal, external and relative validation techniques, results showed that UMAP-DBSCAN best represented the data. Strikingly, internal validation metrics proved systematically unreliable for comparing clustering methods. To address stochastic variability, 100 UMAP-DBSCAN clustering runs were conducted and aggregated following NEMI, yielding a final set of 321 clusters. Reproducibility was evaluated via ensemble overlap (88.81±1.8%) and mean grid cell-wise uncertainty (15.49±20%). Case studies of the Mediterranean Sea, deep Atlantic waters and Labrador Sea showed strong agreement with common water mass definitions. This study revealed a more detailed regionalisation compared to previous concepts such as the Longhurst provinces through systematic clustering method comparison. The applied method is objective, efficient and reproducible and will support future research on biogeochemical differences and changes in oceanic regions. [orcid=0009-0007-7933-8002] [orcid=0000-0002-5136-9604] [orcid=0000-0001-8667-5436] [orcid=0000-0003-1696-9142] [orcid=0000-0002-8453-731X] † † highlights: NEMI enabled objective, reproducible clustering of physics and biogeochemistry DBSCAN best captured natural associations within North Atlantic data UMAP embedding strengthened data associations, improving clustering of all methods Grid-cells were assigned to clusters with a mean uncertainty of about 15\text{\,}\mathrm{\%} Good agreement with known subdivisions with potential for novel insights keywords Ocean biogeochemistry ,Ocean provinces ,Machine learning ,Clustering ,Water masses ,North Atlantic † † credit: Conceptualization, data curation, funding acquisition, investigation, methodology, project administration, software, visualization, writing – original draft † † credit: Conceptualization, methodology, supervision, writing – review and editing † † credit: Conceptualization, data curation, funding acquisition, methodology, supervision, writing – review and editing † † credit: Data curation, writing – review and editing † † credit: Conceptualization, funding acquisition, methodology, supervision, visualization, writing – review and editing 1 Introduction The definition of ocean regions has offered fundamental advances in our understanding of marine (eco)systems, biodiversity distributions and their variability. Since the 19th century, efforts to delineate biogeographic patterns have evolved from early taxonomic classifications ( Forbes, 1856 ) to more comprehensive ecological and physical ocean regionalisations ( Ekman, 1935 ; Hedgpeth, 1957 ; Briggs, 1974 ; Hayden et al., 1984 ; Briggs, 1995 ; Bailey, 1998 ) . One of the most influential partitioning schemes, the ecological provinces by Longhurst (2007) , has provided a foundational framework for investigating large-scale oceanographic patterns and processes, hotspots of biodiversity and ecological relationships. For example, the Longhurst regimes helped quantify primary production ( Longhurst et al., 1995 ) , characterise tuna movements ( Logan et al., 2020 ) and trophic dynamics and food web structure ( Arnoldi et al., 2023 ) . Ocean regionalisations have been used in studies addressing fields such as biogeographic realms ( Costello et al., 2017 ) , carbon flux ( Gloege et al., 2017 ) or patterns of marine viruses ( Brum et al., 2015 ) . They also serve economic decisions in fisheries ( Juan Jordá et al., 2022 ) and policy such as the designation of marine protected areas legislation ( Spalding et al., 2007 ; Sonnewald et al., 2020 ; Zhao et al., 2020b ; Reisinger et al., 2022 ) . While previous approaches often rely on predefined thresholds and subjective criteria, emerging data-driven techniques, particularly clustering algorithms, provide a more objective framework for ocean region delineation ( Devred et al., 2007 ; Hardman-Mountford et al., 2008 ; Oliver and Irwin, 2008 ; Kavanaugh et al., 2014 ; Sayre et al., 2017 ; Sonnewald et al., 2019 ; Sonnewald et al., 2020 ) . The Native Emergent Manifold Interrogation (NEMI) method ( Sonnewald, 2023 ) addresses key challenges in analysing complex geophysical data by integrating manifold learning, clustering, stochastic ensemble methods, uncertainty quantification and intuitive validation to ensure robust and interpretable clustering outcomes. This study builds upon and enhances the NEMI method through systematic comparison of clustering algorithms applied to biogeochemical data. Fundamentally, clustering results are influenced by algorithmic biases ( Thrun, 2021 ) and, for some algorithms, the possibility of variable outcomes raising concerns about suitability, comparability and reproducibility. Two main validation strategies exist ( Rui and Wunsch, 2005 ; Ullmann et al., 2022 ) : (i) Internal validation, which assesses cohesion within clusters and separation between clusters using indices, such as the Calinski-Harabasz index ( Caliński and Harabasz, 1974 ) , Davies-Bouldin index ( Davies and Bouldin, 1979 ) and silhouette score ( Rousseeuw, 1987 ) and (ii) external validation, which utilises knowledge not seen during model training, like comparing clustering results to established ecological classifications or visual analysis in different data spaces. For example, dimensionality reduction techniques like Uniform Manifold Approximation and Projection (UMAP, McInnes et al. (2018a) ) improve the interpretability of clustering outcomes through enhancing associations between data structures and visualisation ( Allaoui et al., 2020 ) . A key challenge for clustering is quantifying uncertainty. Some clustering and dimensionality reduction algorithms are non-deterministic and a globally optimal solution is not guaranteed. For example, UMAP uses stochasticity ( McInnes et al., 2018a ) , Density-Based Spatial Clustering of Applications with Noise (DBSCAN) results may change with permuted input data ( Schubert et al., 2017 ) and k-Means is sensitive to cluster centroid initialisation ( Fränti and Sieranoja, 2019 ) . Quantifying clustering variability is crucial to assess reproducibility and reliability. One approach involves multiple runs and single-number similarity metrics, such as overlap ( Manning et al., 2008 ) . The NEMI method, which combines multiple clustering runs to form a final cluster set, has been proposed as a novel solution to represent statistical variability and quantify uncertainty. Most prior studies on ocean partitioning focus on surface waters (e.g. Devred et al. (2007) ; Longhurst (2007) ; Hardman-Mountford et al. (2008) ; Oliver and Irwin (2008) ; Vichi et al. (2011) ; Reygondeau et al. (2013) ; Fay and Mckinley (2014) ; Kavanaugh et al. (2014) ; Reygondeau et al. (2020) ; Sonnewald et al. (2020) ) despite the fact that critical processes such as particle export, upwelling or deep-water formation extend into deeper layers. Also, vertical mixing affects biodiversity leading to depth-variable community trends ( DeLong et al., 2006 ; Hörstmann et al., 2022 ) . Recent efforts to develop 3D ocean regionalisations have incorporated clustering approaches such as k-Means ( Sayre et al., 2017 ) and hybrid methods combining k-Means, CMeans, agglomerative Ward and agglomerative full linkage ( Reygondeau et al., 2017 ) . In physical oceanography, the definition of 3D water masses, mainly defined by temperature and salinity but also e.g. oxygen, has always played a