Elucidating ecological complexity: Unsupervised learning determines global marine eco-provinces

Area of Science:

• Marine ecology
• Ecological modeling
• Unsupervised machine learning

Background:

• Marine ecosystems exhibit complex community structures influenced by nutrient dynamics.
• Existing methods for defining marine ecological provinces often lack objectivity and scalability.
• Understanding spatial patterns in plankton communities is crucial for ecosystem assessment.

Purpose of the Study:

• To develop and apply an unsupervised learning method for identifying global marine ecological provinces.
• To define robust aggregated eco-provinces (AEPs) for improved ecosystem model interpretation.
• To explore the influence of nutrient supply on marine community structure using the defined AEPs.

Main Methods:

• The Systematic Aggregated Eco-province (SAGE) method was employed, utilizing plankton community structure and nutrient flux data.
• t-stochastic neighbor embedding (t-SNE) was used for dimensionality reduction to handle non-Gaussian data covariance.
• Density-based spatial clustering of applications with noise (DBSCAN) identified over a hundred initial eco-provinces.
• A connectivity graph with ecological dissimilarity defined robust aggregated eco-provinces (AEPs).

Main Results:

• Over a hundred distinct eco-provinces were identified globally.
• Robust aggregated eco-provinces (AEPs) were objectively defined by nesting the initial eco-provinces.
• The study explored the control of nutrient supply rates on community structure within the defined AEPs.

Conclusions:

• The developed SAGE method provides unique and interpretable marine ecological provinces and AEPs.
• These novel classifications aid in the interpretation of complex ecosystem models.
• The findings facilitate marine ecosystem model intercomparison and enhance monitoring capabilities.