Improving Location and Prediction Region Accuracy in Marine Animal Movement with Temporal Fusion Transformer.

Annual Control Talk by PhD student Jorge Medina.

Abstract: 

Marine animal movement refers to the changes in the locations of marine animals over time. Accurately predicting these movements is important for applications such as marine spatial planning for fishing and shipping, disease control, and understanding the impacts of climate change on migration routes. In this talk, we will explore the potential of Temporal Fusion Transformer (TFT) to improve the forecasting accuracy in two ways: more accurate point predictions and enhanced quality of bidimensional prediction regions, i.e. predicted areas where the animal is likely to be. The model surpasses traditional state-space models (SSMs) like Random Walk, Move-Persistence, and Correlated Random Walk. We will also examine TFT adaptations that allows it to outperform SSMs at missing data imputation, a necessary preprocess step in marine animal movement research due to the intermittent nature of tracking data.