Explainable deep-learning to improve plastic-removal strategies on Galapagos beaches

Marine plastic debris is threatening the ecosystem of the Galapagos Marine Reserve and therefore, strategies are needed to effectively remove marine debris through beach cleanups. In order to clean up the beaches as efficiently as possible, our team develops a predictive model to find the most effective removal locations.

This predictive model requires a long-term plastic density prediction from far-to-near coastline, a consideration of the shoreline type and its relevance for the beaching and resuspension process of debris, and the final distribution prediction to the different shoreline sectors. This predictive model aims to unify a recursive Bayesian neural network of plastic density prediction, a neural network-based coastline classifier and a random-forest learning model for the on-shore distribution.

Our results are used by our partners from the Galapagos Conservation Trust to perform a stakeholder analysis of how intervention points of plastic arriving at the islands can best be targeted and how the model can eventually be applied to a larger domain

Researchers

  • Dr. Christian Kehl (postdoc)
  • Dr. Stefanie L. Ypma (postdoc)
  • Prof. Gerben Ruessink
  • Prof. Steven de Jong

Academic supervisors

  • Prof. Ad J. Feelders (ICS)
  • Prof. Erik van Sebille (IMAU; Freudenthal Institute)

Grant funding agency and (co-)funding non-academic partners

Galapagos Conservation Trust