In this chapter, we explore the fascinating world of using deep learning, specifically convolutional neural networks (CNNs), to detect mesoscale ocean eddies. These eddies are small-scale vortices in the ocean that play a crucial role in climate regulation. By leveraging CNNs, which mimic the brain's visual cortex, we can identify and understand these eddies more effectively.
To detect eddies, we utilize the AVISO-SSH sea level anomaly dataset obtained from satellite altimetry measurements. This dataset provides daily global estimates of sea level anomaly. We pose the eddy detection problem as a supervised learning task and use pixel-level labels to train our deep learning model. We also discuss the challenges of overfitting and the importance of splitting the dataset into training, validation, and test sets.
One of the key challenges in machine learning is overfitting, where a model performs well on training data but fails to generalize to real-world data. To address this, we split the dataset into training, validation, and test sets. The training set is used to update the model's parameters, the validation set helps us choose the best model and hyperparameters, and the test set provides an unbiased evaluation of the final model's performance. We also highlight the significance of using py-eddy-tracker algorithm to generate segmentation masks for the data. By combining deep learning and oceanography, we can gain valuable insights into the dynamics of mesoscale ocean eddies and their impact on climate systems. This chapter provides a comprehensive overview of the process, from data preparation to training and evaluation, enabling readers to begin their own journey of understanding and contributing to this exciting field.
Please check out the source code here
The equivalent reusable Geoweaver workflow is here