Published July 11, 2023 | Version 1.0
Model Open

METAR cloud cover over weather stations in Sentinel-2 satellite images - Machine Learning models

Description

This upload contains the Machine Learning models trained on Sentinel-2 satellite images of level L1C, of the project MSCC (also known as AIMLSSE) for the ordinal classification of cloud cover in the METAR format as commonly used by weather stations.

The dataset for these models can be found here: https://doi.org/10.5281/zenodo.14691474

The model itself is a Convolutional Neural Network (CNN), built on the EfficientNetV2 Model and adapted to the new task using Transfer Learning.
There are two models provided:

efficientnet_v2_s_final_16km_300_ndsi_ord_regr_automatic_labels.pt Model trained on the automatically gathered observation data from available weather stations in the METAR cloud cover format.
efficientnet_v2_s_final_16km_300_ndsi_ord_regr_manual_labels.pt Model trained on manually labeled images in the METAR cloud cover format.

Files

Files (163.8 MB)

Additional details

Related works

Is part of
Thesis: 10.5445/IR/1000161948 (DOI)

Dates

Submitted
2023-07-11

Software

Repository URL
https://github.com/ErikWessel/MSCC
Programming language
Python
Development Status
Inactive