Forecasting of Eddy Formations in the GoM with a Divide-and-Conquer Machine Learning Approach

Justin L. Wang, Hanqi Zhuang, Ali K. Ibrahim, Laurent Cherubin, Ali Muhamah Ali

July, 2019
  • Accurate prediction of GoM LCS is a challenge recently posed by the National Academies of Science, Engineering and Medicine

  • We propose a novel divide-and-conquer machine learning approach for forecasting of GoM LCS events

  • We use a progressively weighted smoothing function to reduce error propagation between predictions

  • Our model predicted the LCS and its eddy shedding process more than 12 weeks in advance within 60kms
Our team devised a divide-and-conquer machine learning approach to predicting the Gulf of Mexico (GoM) Loop Current System (LCS). Our methodology can experimental testing can be broken down into five main parts (1) Procuring ground truth data, (2) Data preprocessing and statistical analyses, (3) Definition and set up of prediction algorithm, (4) Simulation and testing, and (5) Performance calculations and analyses.
        We began by procuring ground truth satellite Sea Surface Height (SSH) data from HYCOM Consortium.