|
|
--- |
|
|
license: mit |
|
|
--- |
|
|
|
|
|
# Dataset for Accurate Mediterranean Sea Forecasting via Graph-based Deep Learning |
|
|
|
|
|
This repo contains the dataset to get started with SeaCast and reproduce published results. |
|
|
|
|
|
The file structure is as follows: |
|
|
|
|
|
``` |
|
|
model_weights/ |
|
|
graphs/ |
|
|
data/ |
|
|
└── mediterranean/ |
|
|
└── samples/ |
|
|
├── test/ |
|
|
├── train/ (*202310* files) |
|
|
└── val/ (*202403* files) |
|
|
├── observations/ |
|
|
├── metrics/ |
|
|
└── static/ |
|
|
``` |
|
|
|
|
|
where the folders contain: |
|
|
|
|
|
- `model_weights`: SeaCast model weights |
|
|
- `graphs`: hierarchical graphs for the Mediterranean Sea |
|
|
- `test`: test set from July to December 2024 |
|
|
- `train`: training subset from October 2023 |
|
|
- `val`: validation subset from March 2024 |
|
|
- `observations`: satellite and in-situ measurements for evaluation |
|
|
- `metrics`: precalculated metrics |
|
|
- `static`: static data to run SeaCast |
