Skip to content
Source

Overview

I present some of my findings for working on machine learning applied to ocean applications.

API

You can find the Documentation for the Codes here


ARGO Project

Summary

In this project, we are trying to predict profiles of Bbp. This is a multi-output ML problem with high dimensional inputs and high dimensional outputs.

Details

You can find the repository with all of the reproducible code, specific functions and notebooks at github.com/IPL-UV/ml4ocean.


Subset Dataset

We look at the North Atlantic and SubTropical Gyre regions to try and get an idea about how some ML methods might perform in this scenarios. As a first pass, we did minimal preprocessing with some standard PCA components reduction and we were ambitious and attempted to predict all 273 depths for the profiles. We had some success as we were able to do a decent job with Random forests and MultiLayer Perceptrons.

Relevant Materials

  • ISPRS 2020 Publication - Pending
Todo
  • 0.0 - Data Exploration Notebook
  • 1.0 - Data Preprocessing Notebook
  • 2.0 - Loading the Preprocessed Data
  • 3.0 - Baseline ML Algorithm
  • 4.0 - Visualization & Summary Statistics

Global Dataset

After we found success in the subset data, we decided to go for the global dataset and see how we do. We did reduce the problem difficulty by predicting only 19 depth levels instead of \sim273.

Todo
  • 0.0 - Data Exploration Notebook
  • 1.0 - Data Preprocessing Notebook
  • 4.0 - Visualization & Summary Statistics