A global Swin-Unet Sentinel-2 surface reflectance-based cloud and cloud shadow detection algorithm for the NASA Harmonized Landsat Sentinel-2 (HLS) dataset

The NASA Harmonized Landsat Sentinel-2 (HLS) data provides global coverage atmospherically corrected surface reflectance with a 30m cloud and cloud shadow mask derived using the Fmask algorithm applied to top-of-atmosphere (TOA) reflectance. In this study we demonstrate, as have other researchers, l...

Full description

Saved in:
Bibliographic Details
Main Authors: Haiyan Huang, David P. Roy, Hugo De Lemos, Yuean Qiu, Hankui K. Zhang
Format: Article
Language:English
Published: Elsevier 2025-06-01
Series:Science of Remote Sensing
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S2666017225000197
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849329203633192960
author Haiyan Huang
David P. Roy
Hugo De Lemos
Yuean Qiu
Hankui K. Zhang
author_facet Haiyan Huang
David P. Roy
Hugo De Lemos
Yuean Qiu
Hankui K. Zhang
author_sort Haiyan Huang
collection DOAJ
description The NASA Harmonized Landsat Sentinel-2 (HLS) data provides global coverage atmospherically corrected surface reflectance with a 30m cloud and cloud shadow mask derived using the Fmask algorithm applied to top-of-atmosphere (TOA) reflectance. In this study we demonstrate, as have other researchers, low Sentinel-2 Fmask performance, and present a solution that applies a deep learning Swin-Unet model to the HLS surface reflectance to provide unambiguously improved cloud and cloud shadow detection. The model was trained and assessed using 30m HLS surface reflectance for the 13 Sentinel-2 bands and corresponding CloudSEN12+ annotations, that define cloud, thin cloud, clear, and cloud shadow, and is the largest publicly available expert annotation set. All the CloudSEN12 annotations with coincident HLS Sentinel-2 data were considered. A total of 8672 globally distributed 5 × 5 km data sets were used, 7362 to train the model, 464 for internal model validation, and 846 to independently assess the classification accuracy. The HLS Sentinel-2 Fmask had F1-scores of 0.832 (cloud), 0.546 (cloud shadow), and 0.873 (clear), and the Swin-Unet model had higher performance with F1-scores of 0.891 (cloud and thin cloud combined), 0.710 (cloud shadow), and 0.923 (clear) despite the use of surface and not TOA reflectance. The Swin-Unet thin cloud class had low accuracy (0.604 F1-score) likely due to atmospheric correction issues and thin cloud variability that are discussed. The comprehensively trained model provides a solution for users who wish to improve the HLS Sentinel-2 cloud and cloud shadow masking using the available HLS Sentinel-2 surface reflectance data.
format Article
id doaj-art-097bbeb74c29490dbe17c05cde85f604
institution Kabale University
issn 2666-0172
language English
publishDate 2025-06-01
publisher Elsevier
record_format Article
series Science of Remote Sensing
spelling doaj-art-097bbeb74c29490dbe17c05cde85f6042025-08-20T03:47:20ZengElsevierScience of Remote Sensing2666-01722025-06-011110021310.1016/j.srs.2025.100213A global Swin-Unet Sentinel-2 surface reflectance-based cloud and cloud shadow detection algorithm for the NASA Harmonized Landsat Sentinel-2 (HLS) datasetHaiyan Huang0David P. Roy1Hugo De Lemos2Yuean Qiu3Hankui K. Zhang4Center for Global Change and Earth Observations, Michigan State University, East Lansing, MI, 48824, USACenter for Global Change and Earth Observations, Michigan State University, East Lansing, MI, 48824, USA; Department of Geography, Environment, & Spatial Sciences, Michigan State University, East Lansing, MI, 48824, USA; Corresponding author. Center for Global Change and Earth Observations, Michigan State University, East Lansing, MI, 48824, USA.Center for Global Change and Earth Observations, Michigan State University, East Lansing, MI, 48824, USACenter for Global Change and Earth Observations, Michigan State University, East Lansing, MI, 48824, USAGeospatial Sciences Center of Excellence, Department of Geography and Geospatial Sciences, South Dakota State University, Brookings, SD, 57007, USAThe NASA Harmonized Landsat Sentinel-2 (HLS) data provides global coverage atmospherically corrected surface reflectance with a 30m cloud and cloud shadow mask derived using the Fmask algorithm applied to top-of-atmosphere (TOA) reflectance. In this study we demonstrate, as have other researchers, low Sentinel-2 Fmask performance, and present a solution that applies a deep learning Swin-Unet model to the HLS surface reflectance to provide unambiguously improved cloud and cloud shadow detection. The model was trained and assessed using 30m HLS surface reflectance for the 13 Sentinel-2 bands and corresponding CloudSEN12+ annotations, that define cloud, thin cloud, clear, and cloud shadow, and is the largest publicly available expert annotation set. All the CloudSEN12 annotations with coincident HLS Sentinel-2 data were considered. A total of 8672 globally distributed 5 × 5 km data sets were used, 7362 to train the model, 464 for internal model validation, and 846 to independently assess the classification accuracy. The HLS Sentinel-2 Fmask had F1-scores of 0.832 (cloud), 0.546 (cloud shadow), and 0.873 (clear), and the Swin-Unet model had higher performance with F1-scores of 0.891 (cloud and thin cloud combined), 0.710 (cloud shadow), and 0.923 (clear) despite the use of surface and not TOA reflectance. The Swin-Unet thin cloud class had low accuracy (0.604 F1-score) likely due to atmospheric correction issues and thin cloud variability that are discussed. The comprehensively trained model provides a solution for users who wish to improve the HLS Sentinel-2 cloud and cloud shadow masking using the available HLS Sentinel-2 surface reflectance data.http://www.sciencedirect.com/science/article/pii/S2666017225000197CloudCloud shadowDeep learningHLSSentinel-2Swin-Unet
spellingShingle Haiyan Huang
David P. Roy
Hugo De Lemos
Yuean Qiu
Hankui K. Zhang
A global Swin-Unet Sentinel-2 surface reflectance-based cloud and cloud shadow detection algorithm for the NASA Harmonized Landsat Sentinel-2 (HLS) dataset
Science of Remote Sensing
Cloud
Cloud shadow
Deep learning
HLS
Sentinel-2
Swin-Unet
title A global Swin-Unet Sentinel-2 surface reflectance-based cloud and cloud shadow detection algorithm for the NASA Harmonized Landsat Sentinel-2 (HLS) dataset
title_full A global Swin-Unet Sentinel-2 surface reflectance-based cloud and cloud shadow detection algorithm for the NASA Harmonized Landsat Sentinel-2 (HLS) dataset
title_fullStr A global Swin-Unet Sentinel-2 surface reflectance-based cloud and cloud shadow detection algorithm for the NASA Harmonized Landsat Sentinel-2 (HLS) dataset
title_full_unstemmed A global Swin-Unet Sentinel-2 surface reflectance-based cloud and cloud shadow detection algorithm for the NASA Harmonized Landsat Sentinel-2 (HLS) dataset
title_short A global Swin-Unet Sentinel-2 surface reflectance-based cloud and cloud shadow detection algorithm for the NASA Harmonized Landsat Sentinel-2 (HLS) dataset
title_sort global swin unet sentinel 2 surface reflectance based cloud and cloud shadow detection algorithm for the nasa harmonized landsat sentinel 2 hls dataset
topic Cloud
Cloud shadow
Deep learning
HLS
Sentinel-2
Swin-Unet
url http://www.sciencedirect.com/science/article/pii/S2666017225000197
work_keys_str_mv AT haiyanhuang aglobalswinunetsentinel2surfacereflectancebasedcloudandcloudshadowdetectionalgorithmforthenasaharmonizedlandsatsentinel2hlsdataset
AT davidproy aglobalswinunetsentinel2surfacereflectancebasedcloudandcloudshadowdetectionalgorithmforthenasaharmonizedlandsatsentinel2hlsdataset
AT hugodelemos aglobalswinunetsentinel2surfacereflectancebasedcloudandcloudshadowdetectionalgorithmforthenasaharmonizedlandsatsentinel2hlsdataset
AT yueanqiu aglobalswinunetsentinel2surfacereflectancebasedcloudandcloudshadowdetectionalgorithmforthenasaharmonizedlandsatsentinel2hlsdataset
AT hankuikzhang aglobalswinunetsentinel2surfacereflectancebasedcloudandcloudshadowdetectionalgorithmforthenasaharmonizedlandsatsentinel2hlsdataset
AT haiyanhuang globalswinunetsentinel2surfacereflectancebasedcloudandcloudshadowdetectionalgorithmforthenasaharmonizedlandsatsentinel2hlsdataset
AT davidproy globalswinunetsentinel2surfacereflectancebasedcloudandcloudshadowdetectionalgorithmforthenasaharmonizedlandsatsentinel2hlsdataset
AT hugodelemos globalswinunetsentinel2surfacereflectancebasedcloudandcloudshadowdetectionalgorithmforthenasaharmonizedlandsatsentinel2hlsdataset
AT yueanqiu globalswinunetsentinel2surfacereflectancebasedcloudandcloudshadowdetectionalgorithmforthenasaharmonizedlandsatsentinel2hlsdataset
AT hankuikzhang globalswinunetsentinel2surfacereflectancebasedcloudandcloudshadowdetectionalgorithmforthenasaharmonizedlandsatsentinel2hlsdataset