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

被引：0
作者：
Huang, Haiyan [1 ]
Roy, David P. [1 ,2 ]
De Lemos, Hugo [1 ]
Qiu, Yuean [1 ]
Zhang, Hankui K. [3 ]
机构：
[1] Michigan State Univ, Ctr Global Change & Earth Observat, E Lansing, MI 48824 USA
[2] Michigan State Univ, Dept Geog Environm & Spatial Sci, E Lansing, MI 48824 USA
[3] South Dakota State Univ, Geospatial Sci Ctr Excellence, Dept Geog & Geospatial Sci, Brookings, SD 57007 USA
来源：
SCIENCE OF REMOTE SENSING | 2025年 / 11卷
关键词：
Cloud; Cloud shadow; Deep learning; HLS; Sentinel-2; Swin-Unet; ATMOSPHERIC CORRECTION; IMAGERY; AEROSOL;
D O I：
10.1016/j.srs.2025.100213
中图分类号：
X [环境科学、安全科学];
学科分类号：
08 ; 0830 ;
摘要：
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-ofatmosphere (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 x 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.
引用
收藏
页数：15
相关论文
共 68 条
  • [1] Alguacil A., Pinto W.G., Bauerheim M., Jacob M.C., Moreau S., Effects of boundary conditions in fully convolutional networks for learning spatio-temporal dynamics, Joint European Conference On Machine Learning And Knowledge Discovery In Databases
  • [2] Sep 13-17, pp. 102-117, (2021)
  • [3] Alvera-Azcarate A., Van der Zande D., Barth A., dos Santos J.F.C., Troupin C., Beckers J.M., Detection of shadows in high spatial resolution ocean satellite data using DINEOF, Rem. Sens. Environ., 253, (2021)
  • [4] Aybar C., Ysuhuaylas L., Loja J., Gonzales K., Herrera F., Bautista L., Yali R., Flores A., Diaz L., Cuenca N., Espinoza W., Prudencio F., Llactayo V., Montero D., Sudmanns M., Tiede D., Mateo-Garcia G., Gomez-Chova L., CloudSEN12, a global dataset for semantic understanding of cloud and cloud shadow in Sentinel-2, Sci. Data, 9, 1, (2022)
  • [5] Aybar C., Bautista L., Montero D., Contreras J., Ayala D., Prudencio F., Loja J., Ysuhuaylas L., Herrera F., Gonzales K., Valladares J., Flores L.A., Mamani E., Quinonez M., Fajardo R., Espinoza W., Limas A., Yali R., Alcantara A., Leyva M., Loayza-Muro R., Willems B., Mateo-Garcia G., Gomez-Chova L., CloudSEN12+: the largest dataset of expert-labeled pixels for cloud and cloud shadow detection in Sentinel-2, Data Brief, (2024)
  • [6] Bauer-Marschallinger B., Falkner K., Wasting petabytes: a survey of the Sentinel-2 UTM tiling grid and its spatial overhead, ISPRS J. Photogrammetry Remote Sens., 202, pp. 682-690, (2023)
  • [7] Bolton D.K., Gray J.M., Melaas E.K., Moon M., Eklundh L., Friedl M.A., Continental-scale land surface phenology from harmonized Landsat 8 and Sentinel-2 imagery, Rem. Sens. Environ., 240, (2020)
  • [8] Cahalan R.F., Oreopoulos L., Wen G., Marshak A., Tsay S.C., DeFelice T., Cloud characterization and clear-sky correction from Landsat-7, Rem. Sens. Environ., 78, 1-2, pp. 83-98, (2001)
  • [9] Cao H., Wang Y., Chen J., Jiang D., Zhang X., Tian Q., Wang M., Swin-unet: unet-like pure transformer for medical image segmentation, Computer Vision – ECCV 2022 Workshops, Lecture Notes in Computer Science, 13803, (2023)
  • [10] Caron M., Touvron H., Misra I., Jegou H., Mairal J., Bojanowski P., Joulin A., Emerging properties in self-supervised vision transformers, Proceedings of the IEEE/CVF International Conference on Computer Vision, pp. 9650-9660, (2021)
1234567