Two-Stage CNN-LSTM

Learning the spatio-temporal link between wind and significant wave height (Hs)

This repository provides the Python implementation for predicting significant wave height from wind fields with a two-stage deep-learning pipeline that combines Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) units. The full approach is described in:

Obakrim, S. et al. “Learning the spatiotemporal relationship between wind and significant wave height using deep learning,” Environmental Data Science, 2022. doi:10.1017/eds.2022.35

Data sources

Variable	Archive	Link
Significant wave height (Hs)	HOMERE hindcast (Ifremer)	Link
Wind fields	Climate Forecast System Reanalysis (CFSR)	Link

Pre-processed datasets used in the paper can be downloaded directly here.

Method overview

1. CNN stage — extracts spatial features from consecutive wind-field snapshots.
2. LSTM stage — learns temporal dependencies and predicts the corresponding Hs time series.

Citation

@article{obakrim2022twostage,
  title  = {Two-stage CNN–LSTM for learning the spatio-temporal relationship between wind and significant wave height},
  author = {Obakrim, S. and others},
  journal= {Environmental Data Science},
  year   = {2022},
  doi    = {10.1017/eds.2022.35}
}