# Actions

# Résumé des résultats à fournir

* les résultats du modèle appris sur la zone GF et testé sur la zone GF (dans la config. de test correspondant à BOOST-SWOT)
* les résultats du modèle appris sur une zone GF2 (lat : [42,52] et lon : [-45,35], patch 200*200) située le long du GF mais différente de la zone GF de test en prenant toute l’année en apprentissage et en testant sur la période «BOOST_SWOT » et la zone GF.
* les résultats d’un fine-tuning du modèle NATL (config apprentissage boost-SWOT) sur la zone GF

Les étapes suivantes seront :

* l’évaluation de modèles de référence pour l’interpolation (la méthode que tu avais identifié + transformer)
* la mise en oeuvre sur le domaine NATL60 du 4DVarNet avec exploitation de la SST

### July 5th week actions

* First experience: repeat the previous experiences with a lag of 30 days between training/test and training/validtation data.
* Second Experience: boost-swot repo experience(test: (2012,10,22) ---> (2012,12,2) ), see https://github.com/ocean-data-challenges/2020a_SSH_mapping_NATL60
* Third experience: spatial training/testing split
* training: same as previous experiences on the north atlantic region - dim_range = {'lat': slice(30.5, 45.5),'lon': slice(-67.5, -52.5)}
* validation: same as previous experiences on the north atlantic region - dim_range = {'lat': slice(30.5, 45.5),'lon': slice(-67.5, -52.5)}
* test : dim_range = {'lat': slice(33, 43),'lon': slice(-65, -55)}

### Results

### 1- First experience( with 8 nodes, 4gpus, batch_size = 4):

| Model Training/Test | SSH(MSE/%) | SSH Gradiend (MSE/%) | Resolved horizontal scale | mu 4dvar/OI | sigma 4dvar/OI | lambda_x 4dvar/OI | lambda_t 4dvar/OI |
| ------------------- | --------------- | -------------------- | ------------------------- | --- | --- | --- | --- |
| NATL/GF | 5.95e-04/56.40% | 2.11e-05/45.85% | 50km/75km | 0.95/0.93 | 0.012/0.015 | 1.13/1.34 | 6.9/8.5 |
| NATL/OSMOSIS | 1.10e-04/41.26% | 6.32e-06/39.36% | 45km/60km | 0.83/0.76 | 0.033/0.042 | 0.95/1.21 | 6.83/13.01 |
| NATL/NATL | 5.90e-04/47.66% | 1.47e-05/59.27% | 100Km/130km | | | | |



#### 2-Second Experience ( with 8 nodes, 4gpus, batch_size = 4):

| Model Training/Test | SSH(MSE/%) | SSH Gradiend (MSE/%) | Resolved horizontal scale | mu 4dvar/OI/BFN | sigma 4dvar/OI/BFN | lambda_x 4dvar/OI/BFN | lambda_t 4dvar/OI/BFN |
| ------------------- | --------------- | -------------------- | ------------------------- | --------------- | ------------------ | --------------------- | --------------------- |
| GF/GF | 8.4e-04/57.56% | 2.93e-05/45.81% | 65km/95km | 0.95/0.92/0.93 | 0.014/0.017/0.02 | 0.87/1.22/0.8 | 6.91/11.2/7.86 |
| GF2/GF |1e-03/50%|3.17e-05/41.53%|65km/95km|0.94/0.92/0.93|0.013/0.017/0.02|0.95/1.22/0.8|7.84/11.2/7.86|
| NATL FT |8.1e-04/59.48%|2.85e-05/47.45%|65km/95km|0.95/0.92/0.93|0.013/0.017/0.02|0.91/1.22/0.8|6.47/11.2/7.86|
| NATL/GF | 1.10e-03/44.6% | 3.57e-05/34.08% | 85km/95km | 0.94/0.92/0.93 | 0.014/0.017/0.02 | 1.13/1.22/0.8 | 9.4/11.2/7.86 |
| NATL/OSMOSIS | 8.78e-05/39.38% | 5.41e-06/33.43% | 45km/65km | 0.85/0.81 | 0.016/0.017 | 0.83/1.03 | 8.16/17.77 |
| NATL/NATL | 6.75e-04/36.26% | 1.52e-05/53.49% | 100km/150km | | | | |

#### 3-Third Experience ( with 8 nodes, 4gpus, batch_size = 4):

| Model Training/Test | SSH(MSE/%) | SSH Gradiend (MSE/%) | Resolved horizontal scale | mu 4dvar/OI | sigma 4dvar/OI | lambda_x 4dvar/OI | lambda_t 4dvar/OI |
| ------------------- | -------------- | -------------------- | ------------------------- | ----------- | -------------- | ----------------- | ----------------- |
| NATL/GF | 8.1e-04/40% | 2.93e-05/25% | 80km/95mk | 0.94/0.93 | 0.013/0.015 | 1.25/1.34 | 6.99/8.54 |