Home

Awesome

Open Satellite Image Cloud Detection Resources (OpenSICDR)

We collect the latest open-source tools and datasets for cloud and cloud shadow detection, and launch this online project (Open Satellite Image Cloud Detection Resources, i.e., OpenSICDR) to promote the sharing of the latest research outputs of the field. If you would like to provide new resources, please kindly contact <a href='https://zhiweili.net/'>Dr. Zhiwei Li</a> at <a href="mailto:dr.lizhiwei@gmail.com">dr.lizhiwei(AT)gmail.com</a> or submit an update request.

Source:

Zhiwei Li, Huanfeng Shen, Qihao Weng, Yuzhuo Zhang, Peng Dou, Liangpei Zhang. Cloud and Cloud Shadow Detection for Optical Satellite Imagery: Features, Algorithms, Validation, and Prospects. ISPRS Journal of Photogrammetry and Remote Sensing, vol. 188, pp. 89-108, 2022. (Link, PDF)

<br />

Contributors:

<br />

Update logs:

Feb 28, 2024: Added two new cloud detection datasets, GF1MS-WHU and GF2MS-WHU.

June 5, 2024: 1) Added one new cloud detection dataset, CloudSEN12; Added parts of cloud mask products in Google Earth Engine;

<br />

Open-Source Datasets for Cloud and Cloud Shadow Detection

NameImage SourceReferencesDescriptionsLink
L7_IrishLandsat-7 (30 m)Scaramuzza et al., 2012; USGS., 2016aContains 206 Landsat-7 scenes from nine global latitude zones with manually generated masks, of which only 45 scenes are labeled for cloud shadows.Link
L8_SPARCSLandsat-8 (30 m)Hughes and Hayes, 2014; USGS., 2016cContains 80 subsets of Landsat-8 scenes with a size of 1000×1000 pixels that are labeled for both clouds and cloud shadows.Link
L8_BiomeLandsat-8 (30 m)Foga et al., 2017; USGS., 2016bContains 96 Landsat-8 scenes from eight global biomes with manually generated cloud masks, of which 32 scenes are labeled for cloud shadows.Link
95-CloudLandsat-8 (30 m)Mohajerani and Saeedi, 2019Contains 95 Landsat-8 images and associated pixel-level cloud labels that is an extension of the previously established 38-Cloud dataset.Link
Snow-Cloud Validation MasksLandsat-8 (30 m)Stillinger and Collar, 2019Contains 13 Landsat-8 images and corresponding clouds and snow labels at mid-latitude mountainous regions.Link
RICE_datasetLandsat-8 (30 m)Lin et al., 2019Contains 450 Landsat-8 images and corresponding cloud-free images and cloud labels with a size of 512×512 pixels in one of two subsets of the dataset.Link
WHU Cloud DatasetLandsat-8 (30 m)Ji et al., 2021Contains 7 Landsat-8 images and corresponding cloud-free historical images and cloud and shadow masks in six different regions.Link
S2-HollsteinSentinel-2 (10 m)Hollstein et al., 2016Consists 5,647,725 pixels based on images acquired over the entire globe with cloud, cirrus, snow, shadow, and water labels.Link
S2-BaetensHagolleSentinel-2 (10 m)Baetens et al., 2018, 2019Provides cloud masks for 38 Sentinel-2 scenes selected in 2017 or 2018, each with cloud and cloud shadow labels.Link
T-S2/T-PSSentinel-2 (10 m)<br />PlanetScope (3 m)Shendryk et al., 2019Contains 4,993 Sentinel-2 and 4,943 PlanetScope subscenes with a size of 512×512 pixels and only RGB and NIR bands over the Wet Tropics of Australia, each is labeled at the block level.Link
Sentinel-2 Cloud Mask CatalogueSentinel-2 (10 m)Francis et al., 2020Comprises 20 m resolution cloud masks for 513 subscenes, of which 424 subscenes are labeled for cloud shadows.Link
Sentinel-2 KappaZetaSentinel-2 (10 m)Domnich et al., 2021Contains 4403 labeled image blocks with a size of 512×512 pixels from 155 Sentinel-2 images over the Northern European terrestrial area.Link
WHUS2-CDSentinel-2 (10 m)Li et al., 2021Contains 32 Sentinel-2 images distributed in Mainland China and its reference cloud masks labeled at 10 m resolution.Link
CloudSEN12Sentinel-2 (10 m)Aybar et al., 2022Contains 49,400 Sentinel-2 image patches, each sized 509×509 pixels, evenly distributed across all continents except Antarctica.Link
GF1_WHUGaofen-1 WFV (16 m)Li et al., 2017Contains 108 globally distributed GF-1 WFV scenes and their manually labeled cloud and cloud shadow masks.Link
Levir_CSGaofen-1 WFV (16 m)Wu et al., 2021Contains 4,168 globally distributed Gaofen-1 WFV scenes (down sampled to 160 m resolution) and the corresponding geographical data, cloud, and snow labels.Link
GF1MS-WHU<br />GF2MS-WHUGaofen-1 PMS (8 m)<br />Gaofen-2 PMS (8 m)Zhu et al., 2024Contains 141 unlabeled and 33 well-annotated 8-m Gaofen-1 PMS multispectral images;<br />Contains 163 unlabeled and 29 well-annotated 4-m Gaofen-2 multispectral images.Link
WDCD datasetGaofen-1 PMS (8 m)<br />Ziyuan-3 MUX (5.8 m)Li et al., 2020Contains over 200,000 globally distributed Gaofen-1 image blocks labeled at the block level for training and 30 Gaofen-1 and Ziyuan-3 scenes labeled at the pixel level for validation and testing.Link
N/AGaofen series (N/A)Sun et al., 2020Contains 745 paired NIR-R-G composited images and corresponding pixel-level labels with a size of 256×256 pixels.Link
AIR-CDGaofen-2 PMS (4 m)He et al., 2021Contains 34 Gaofen-2 full images and the corresponding cloud labels distributed at different regions of China.Link
HRC_WHUGoogle Earth (0.5 m to 15 m)Li et al., 2019Comprises 150 globally distributed high-resolution images (0.5 m to 15 m resolution, three RGB channels) and the corresponding cloud masks.Link
<br />

Open-Source Tools for Cloud and Cloud Shadow Detection

NameApplicable Images (Primarily)ReferencesDescriptions (Data and Method)Link
LandsatFmaskLandsat 4-8<br />Sentinel-2Zhu et al., 2012 & 2015Mono-temporal<br />Physical rule basedLink
TmaskLandsat 4-8Zhu and Woodcock, 2014Multi-temporal<br />Temporal change basedLink
MSScvmLandsat MSSBraaten et al., 2015Multi-source<br />Physical rule basedLink
MFmaskLandsat 4-8Qiu et al., 2017Multi-source<br />Physical rule basedLink
MCM-GEELandsat-8Mateo-García et al., 2018Multi-temporal<br />Temporal change basedLink
Cloud-NetLandsat-8Mohajerani and Saeedi, 2019Mono-temporal<br />DL basedLink
CmaskLandsat-8Qiu et al., 2020Multi-temporal<br />Temporal change basedLink
DAGANSLandsat-8<br />Proba-VMateo-Garcia et al., 2020Mono-temporal<br />DL basedLink
FCNNLandsats-8<br />Sentinel-2López-Puigdollers et al., 2021Mono-temporal<br />DL basedLink
Sentinel-2MAJASentinel-2<br />VENμS<br />Landsat-8Hagolle et al., 2010Multi-temporal<br />Temporal change basedLink
cB4S2Sentinel-2Hollstein et al., 2016Mono-temporal<br />Machine learning basedLink
Sen2CorSentinel-2Main-Knorn et al., 2017Mono-temporal<br />Physical rule basedLink
s2cloudlessSentinel-2Zupanc, 2017Mono-temporal<br />Machine learning basedLink
FORCESentinel-2<br />Landsat 4-8Frantz et al., 2018Mono-temporal<br />Physical rule basedLink
KappaMaskSentinel-2Domnich et al., 2021Mono-temporal<br />DL basedLink
CD-FM3SFSentinel-2Li et al., 2021Mono-temporal<br />DL basedLink
GaofenMFCGaofen-1 WFVLi et al., 2017Mono-temporal<br />Physical rule basedLink
GeoInfoNetGaofen-1 WFVWu et al., 2021Mono-temporal<br />DL basedLink
OthersN/AHR imagesXie et al., 2017Mono-temporal<br />DL basedLink
<br />

Open-Source Cloud and Cloud Shadow Mask Products in Google Earth Engine

[1] Sentinel-2: Cloud Probability. [Link]

[2] Sentinel-2: Cloud Score+. [Link]

<br />

References