This project contains a python package that extends the functionality of the Google Earth Engine API (ee) to implement multitemporal cloud detection algorithms.

In particular it contains the code to reproduce the results of (Mateo-Garcia et al 2018) and (Gomez-Chova et al 2017).

Additional results of Mateo-Garcia et al 2018 can be browsed at http://isp.uv.es/projects/cdc/viewer_l8_GEE.html

• The Landsat-8 collection with FMask used in the articles is not longer available. We have modified the code to work with new Landsat-8 collections (LANDSAT/LC08/C01/T1_TOA/).
• We added a notebook that applies our method to Sentinel-2 images. (from collection COPERNICUS/S2/)
• Notebooks can be browsed in colab.

The following code creates a fresh conda environment with required dependencies:

 conda create -n ee python=3 numpy scipy jupyterlab matplotlib scikit-learn pillow requests luigi pandas scikit-image
pip install earthengine-api

python setup.py install

The examples folder contains several notebooks that go step by step in the proposed multitemporal cloud detection schemes.

• The notebook cloudscore_different_preds.ipynb shows ready to use examples of the proposed cloud detection scheme for Landsat-8.
• The notebook cloudscore_different_preds-S2.ipynb shows ready to use examples of the proposed cloud detection scheme for Sentinel-2.
• The notebook multitemporal_cloud_masking_sample.ipynb explains in great detail the method for background estimation proposed in (Gomez-Chova et al 2017)
• The notebook clustering_differences.ipynb explains the clustering procedure and the thresholding of the image to form the cloud mask.

The folder reproducibility contains scripts, notebooks and instructions needed to reproduce the results of Mateo-Garcia et al 2018: Multitemporal Cloud Masking in the Google Earth Engine. See reproducibility/README.md Note: due to changes in new tier Landsat-8 collections results might change.

If you use this code please cite:

@article{mateo-garcia_multitemporal_2018,
author = {Mateo-García, Gonzalo and Gómez-Chova, Luis and Amorós-López, Julia and Muñoz-Marí, Jordi and Camps-Valls, Gustau},
doi = {10.3390/rs10071079},
journal = {Remote Sensing},
language = {en},
link = {http://www.mdpi.com/2072-4292/10/7/1079},
month = {jul},
number = {7},
pages = {1079},
title = {Multitemporal {Cloud} {Masking} in the {Google} {Earth} {Engine}},
urldate = {2018-07-10},
volume = {10},
year = {2018}
} 

• Benchmarking Deep Learning Models for Cloud Detection in Landsat-8 and Sentinel-2
• Landsat-8 to Proba-V transfer learning and Domain adaptation for cloud detection

This work has been developed in the framework of the project TEC2016-77741-R (MINECO-ERDF) and the GEE Award project titled Cloud detection in the cloud granted to Luis Gómez-Chova.