Skip to content

A PyTorch implementation of the Multi-Mode CNN to reconstruct Chlorophyll-a time series in the global ocean from oceanic and atmospheric physical drivers

License

Notifications You must be signed in to change notification settings

JoanaR/multi-mode-CNN-pytorch

Repository files navigation

A Multi-Mode Convolutional Neural Network (CNNMM) to reconstruct satellite-derived Chlorophyll-a time series in the global ocean from physical drivers

This repository contains the code of the model presented in the paper [A Multi-Mode Convolutional Neural Network to reconstruct Chlorophyll-a time series in the global ocean from physical drivers](Frontiers in Marine Science, 2023).

Contents

This repository contains the following PyTorch code:

  • Implementation of the multi-mode CNNMM8 Chl time series regression from oceanic and atmospheric predictors :

Results

Our model achieves the following performance on INDIGO Benchmark dataset :

Model name r2 RMSE Slope Seas Inter N param Time computation Km travelled by car
CNNMM8 0.87 0.28 0.90 1.00 0.96 803 920 39 h 8.9

See the paper for more details.

Requirements

Python Libraries :

  • torch==1.4.0
  • torchvision==0.5.0
  • numpy==1.18.1
  • carbontracker==1.1.5

INDIGO dataset download :

The benchmark formated dataset is available for download here. You can also find the required files to run the code in this repository. This dataset was built from the following source of data :

Proxy used as predictors Acronyme Products Initial spatio-temporal resolutions
Sea Surface Temperature SST Reyn_SmithOIv2 SST dataset Monthly on a 1◦ × 1◦ spatial grid
Sea Level Anomaly SLA Ssalto/Duacs merged product of CNES/SALP project Weekly on a 1/3◦ × 1/3◦ spatial grid
Zonal and Meridional surface winds Uera, Vera Atmospheric model reanalysis ERA interim 4 Every 5-days on a 0.25◦ × 0.25◦ spatial grid
Zonal and Meridional surface total currents u,v OSCAR unfiltered satellite product Every 5-days on a 0.25◦ × 0.25◦ spatial grid
Short-wave radiations SW NCEP/NCAR Numerical reanalysis Daily on a 2◦ grid
Binary continental mask mask
Bathymetry bathy GEBCO 15 arc seconds

Animation

Chl recontructed data over [2012-2015] :

animation_Chl_reconstructed_2012_2015.mov

Chl reference satellite data over [2012-2015] :

animation_Chl_satellite_reference_2012_2015.mov

References

Roussillon Joana, Fablet Ronan, Gorgues Thomas, Drumetz Lucas, Littaye Jean, Martinez Elodie (2023). A Multi-Mode Convolutional Neural Network to reconstruct satellite-derived Chlorophyll-a time series in the global ocean from physical drivers. Frontiers in marine science. doi: 10.3389/fmars.2023.1077623

Roussillon Joana, Fablet Ronan, Gorgues Thomas, Drumetz Lucas, Littaye Jean, Martinez Elodie (2022). satellIte phytoplaNkton Drivers In the Global Ocean over 1998-2015 (INDIGO Benchmark dataset). SEANOE. https://doi.org/10.17882/91910

DOI