Image-to-Image
fastai
climate
File size: 1,266 Bytes
b1a06a9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
---
license: mit
datasets:
- CEMAC/synthetic_lee_waves
metrics:
- mse
pipeline_tag: image-to-image
library_name: fastai
tags:
- climate
---

# Model Card for LeeWaveNet

<!-- Provide a quick summary of the model. -->

This repository contains four neural-network models, trained using [fastai](https://docs.fast.ai/), for detecting and determining characteristics of trapped lee waves using maps of 700 hPa vertical velocity as input.

* The base model [segmodel.pkl](https://huggingface.co/CEMAC/LeeWaveNet/blob/main/segmodel.pkl) generates a segmentation mask indicating where trapped lee waves are present. This model uses a U-Net architecture with Resnet-34 (pre-trained on ImageNet) as the encoder model.
* Three alternative model heads have been trained on synthetic data: [amplitude_0.0625.pkl](https://huggingface.co/CEMAC/LeeWaveNet/blob/main/amplitude_0.0625.pkl), [wavelength_0.125.pkl](https://huggingface.co/CEMAC/LeeWaveNet/blob/main/wavelength_0.125.pkl) and [orientation_0.25.pkl](https://huggingface.co/CEMAC/LeeWaveNet/blob/main/orientation_0.25.pkl). These predict the amplitude, wavelength and orientation of detected waves respectively.

For full details, please see the article by [Coney et al. (2023)](https://doi.org/10.1002/qj.4592).