Abstract/Description/Artist Statement
Particle settling velocity serves as an essential component in ocean biological pump, as it determines particle retention time in the water column. Stokes’ law has been widely used to predict particle settling velocities by particle size and excess density in aquatic environments. However, an increasing number of studies suggest that Stokes’ law fits poorly in the size-velocity relationship of observations on small oceanic particles. Here, we present a series of novel approaches to investigate the relative contribution of settling velocities by the particle shape and optical densities using machine learning (ML) models and principal component analysis (PCA), based on 3906 measurements derived from an in-situ sediment trap camera deployed in Bermuda. We observe a substantial increase on R2 score compared to simple linear regression by implementing ML models, while the morphological features only explain no more than 50% of the variance.
Faculty Advisor/Mentor
Alexander Bochdansky
Faculty Advisor/Mentor Email
ABochdan@odu.edu
Faculty Advisor/Mentor Department
Department of Ocean and Earth Sciences
College/School Affiliation
College of Sciences
Student Level Group
Graduate/Professional
Presentation Type
Poster
Included in
Artificial Intelligence and Robotics Commons, Biogeochemistry Commons, Oceanography Commons
How Much Does Shape Matter: Investigating the Impact of Marine Particle Morphological Features on in-situ Settling Velocities Using PCA and Various ML Models
Particle settling velocity serves as an essential component in ocean biological pump, as it determines particle retention time in the water column. Stokes’ law has been widely used to predict particle settling velocities by particle size and excess density in aquatic environments. However, an increasing number of studies suggest that Stokes’ law fits poorly in the size-velocity relationship of observations on small oceanic particles. Here, we present a series of novel approaches to investigate the relative contribution of settling velocities by the particle shape and optical densities using machine learning (ML) models and principal component analysis (PCA), based on 3906 measurements derived from an in-situ sediment trap camera deployed in Bermuda. We observe a substantial increase on R2 score compared to simple linear regression by implementing ML models, while the morphological features only explain no more than 50% of the variance.