ºìÐÓÊÓƵ

Topic

Detection of Coccolithophore Bloom Development in the Salish Sea, Canada: Leveraging Reflectance Data from Autonomous Shipborne In Situ Radiometers and Sentinel-3A with a Random Forest Classifier

Department of Geography

Date & location

• Wednesday, April 23, 2025

• 9:00 A.M.

• Virtual Defence

Reviewers

Supervisory Committee

• Dr. Maycira Costa, Department of Geography, University of Victoria (Supervisor)

• Dr. David Atkinson, Department of Geography, UVic (Member)

• Dr. Laura Cowen, Department of Mathematics and Statistics, UVic (Outside Member) 

External Examiner

• Dr. Bing Lu, Department of Geography, Simon Fraser University 

Chair of Oral Examination

• Dr. Lijun Zhang, Department of Economics, UVic

   

Abstract

Phytoplankton are the primary producers in the ocean, forming the base of the marine food web. Among them, coccolithophores hold particular significance due to their ability to form extensive blooms and their unique role in oceanic calcium and carbonate cycling, as well as related biogeochemical processes. Current limitations in using satellite imagery to derive accurate phytoplankton data, such as chlorophyll concentrations and phytoplankton functional types stem from insufficient in situ reflectance measurements to develop models and validate satellite reflectance. To address this, we deployed a suite of hyperspectral radiometers equipped with autonomous solar tracking capability, collectively known as SAS Solar Tracker (Satlantic Inc./Sea-Bird, denoted as SAS-ST hereafter), atop a commercial ferry traversing the Salish Sea, Canada. We specified the optimal geometry for SAS-ST installation, as well as the identification and flagging of unfavourable meteorological conditions, correction for sun glint and skylight contributions, mitigation of structural interferences, and subsequent application of bidirectional reflectance distribution function (BRDF) corrections to ensure optimal data quality. Assessment of the final data quality was conducted using a quality assurance method that considers spectral shape similarity, revealing that approximately 92% of the acquired reflectance data aligned well with the global database, indicating high quality. During the data collection period of this research in the summer of 2016, an unprecedented coccolithophore bloom occurred in the Salish Sea area. Coccolithophores, a distinctive phytoplankton species, are encased in calcium carbonate plates called coccoliths, which can be shed into the water during later stages of the blooms, significantly augmenting water reflectance. Based on its unique spectral features, our research successfully identified the presence of the coccolithophore bloom and further categorized the bloom spectra into growing and decaying stages. The hyperspectral reflectance SAS-ST data were initially convolved to the Sentinel-3A OLCI 10 spectral bands spanning 400 to 709 nm. Comparison of Sentinel-3A OLCI satellite spectra with SAS-ST in situ data revealed that reflectance acquired by the OLCI satellite was underestimated, particularly in the 400-443 nm range and the decaying bloom category. Consequently, we developed an adapted machine learning algorithm based on wavelengths ranging from 490 nm to 709 nm, which increased the overall prediction accuracy for OLCI-measured coccolithophore spectra from 0.794 to 0.891, and enhanced the Kappa coefficient from 0.14 to 0.60. By leveraging data from autonomous shipborne In situ SAS-ST and Sentinel-3A OLCI, my research overcomes current limitations of coccolithophore detection algorithms in coastal waters impacted by river plumes, while also providing new insights into coccolithophore dynamics and potentially enhancing their remote sensing.