Marine Big Data-driven Machine Learning Based Monitoring of Phytoplankton Groups in the Arctic Ocean - Master Thesis

Aug 5, 2026

Upload by Ezra

The Arctic Ocean is undergoing rapid changes from climate warming, strongly affecting its physical and biological systems. Phytoplankton, as primary producers, are central to marine ecosystem functioning and biogeochemical cycles, making reliable monitoring of their distribution and abundance essential for assessing Arctic environmental health. This study develops an ensemble machine learning framework to predict Total Chlorophyll-a (TChl-a) and multiple Phytoplankton Functional Types (PFTs) using matched satellite and in-situ observations from 1997–2020, validated with PS131 expedition measurements.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages44

Reading Time1 h 51 min

AccessFree to download

Introduction

Data and Methods

Results and Discussions

Conclusions and Outlook

Master Thesis Marine Big Data-driven Machine Learning Based Monitoring of Phytoplankton Groups in the Arctic Ocean Programme: M.Sc. Space Sciences and Technologies Submitted by: Alfredo J. Bellido Rosas 6236057 Supervisors: Dr. Hongyan Xi Prof. Dr. Astrid Bracher Faculty of Physics and Electrical Engineering University of Bremen Bremen, Germany Alfred Wegener Institut Helmholtz Zentrum fu¨r Polar und Meeresforschung Bremerhaven, Germany Contents 1 Introduction 3 2 Data and Methods 6 2.1 In-situ Dataset ...................................... 6 2.2 CMEMS Dataset ..................................... 7 2.3 Matchup Extraction ................................... 9 2.4 Pre-Processing ...................................... 12 2.5 Model Selection ...................................... 13 2.5.1 Gradient Boosting Machine (GBM) ...................... 13 2.5.2 Fully Connected Neural Network (FCNN) ................... 14 2.5.3 Random Forest Regression (RFR) ....................... 15 2.5.4 Support Vector Machine (SVM) ........................ 16 2.5.5 Ridge Regression Ensemble (RRE) ....................... 17 2.6 Accuracy Assessment ................................... 18 2.6.1 Performance Metrics ............................... 18 2.6.2 Cross-validation approach ............................ 19 2.7 Validation and PFT mapping .............................. 20 2.7.1 In-situ Data .................................... 20 2.7.2 CMEMS Products ................................ 22 2.7.3 Pre-Processing .................................. 24 2.7.4 PFT Mapping .................................. 25 2.7.5 PFT Validation using in-situ Matchups .................... 25 3 Results and Discussions 26 3.1 Training Phase Performance ............................... 26 3.2 Validation Analysis .................................... 30 3.2.1 Comparison of Training and Validation Datasets ............... 30 3.2.2 Validation Performance ............................. 31 3.3 Mapping of the Arctic PFTs .............................. 33 4 Conclusions and Outlook 37 4.1 Conclusions ........................................ 37 4.2 Outlook .......................................... 37 Bibliography 40 Abstract The Arctic Ocean is experiencing rapid and signiﬁcant changes due to climate warming, profoundly impacting its physical and biological systems. Phytoplankton, as primary producers, play a crucial role in marine ecosystems and biogeochemical cycles. Monitoring their distribution and abundance is essential for understanding the health of the Arctic marine environment. This study focuses on developing an ensemble machine learning model to predict concentrations of Total Chlorophyll-a (TChl-a) and various Phytoplankton Functional Types (PFTs) in the Arctic Ocean, leveraging data from satellite observations and in-situ measurements. The ensemble model combines Gradient Boosting Machine (GBM), Fully Connected Neural Network (FCNN), Random Forest Regression (RFR), and Support Vector Machine (SVM) through a Ridge Regression Ensemble approach. The model was trained by using satellite data and model simulations outputs from Copernicus Marine Service (CMEMS) that were matched with in situ data collected during 1997-2020 and validated using in-situ measurements from the PS131 expedition [1] . The model demonstrates strong predictive capabilities, particularly for Diatoms and TChl-a, which are crucial for understanding primary production and nutrient dynamics in the Arctic. Results indicate that the ensemble model performs well in capturing the spatial and temporal distribution of TChl-a and PFTs. The model’s robust performance during the training phase and its ability to generalise to the validation dataset, regardless of its higher variability respect to the training dataset, underscore its potential for large-scale ecological monitoring. The creation of Arctic maps for PFTs and TChl-a provided val

FAQs

What is the goal of this master thesis?

To develop an ensemble machine learning model that predicts Total Chlorophyll-a and multiple phytoplankton functional types in the Arctic Ocean for ecological monitoring.

Which data sources are used for training and validation?

The model uses matched satellite observations and CMEMS model simulation outputs, matched with in-situ measurements collected from 1997–2020, and validated using in-situ measurements from the PS131 expedition.

What machine learning methods are combined in the ensemble?

The ensemble combines Gradient Boosting Machine (GBM), Fully Connected Neural Network (FCNN), Random Forest Regression (RFR), and Support Vector Machine (SVM) within a Ridge Regression Ensemble approach.

Aug 5, 2026

Upload by Ezra