A Continuous Cloud Coverage Dataset for Robust AI in Occluded Environments

Shanika Edirisinghe; Svetlana Hensman; Bianca Schoen-Phelan

doi:10.1007/978-3-032-25809-0_20

A Continuous Cloud Coverage Dataset for Robust AI in Occluded Environments

Shanika Edirisinghe
, Svetlana Hensman
, Bianca Schoen-Phelan

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

Abstract

Cloud cover is a persistent challenge in satellite remote sensing, obscuring surface features and reducing the reliability of applications such as environmental monitoring, disaster response, and urban planning. Most existing cloud-related datasets provide binary masks, distinguishing only between cloud and clear conditions, and thus fail to capture cloud density, limiting systematic benchmarking and robustness analysis of AI models under varying occlusion. To address this gap, we introduce a continuous cloud-coverage dataset derived from SEN12MS-CR. Each Sentinel-2 tile is assigned a percentage label (0–100%) representing cloud proportion. The curated dataset includes 36,666 tiles, evenly sampled across four seasons, and is openly accessible via Zenodo. A case study using the UnCRtainTS cloud removal model shows that reconstruction quality declines progressively with higher cloud density. By supporting standardized evaluation across cloud conditions, the dataset enhances cloud-aware AI tasks, including removal, classification, and segmentation, and contributes to broader computer vision research on occlusion robustness and cognitive science perspectives on reasoning under uncertainty.

Original language	English
Title of host publication	Artificial Intelligence and Cognitive Science - 33rd Irish Conference, AICS 2025, Revised Selected Papers
Editors	Tai Tan Mai, Tomas Ward, Cathal Gurrin, Allie Ly-Duyen Tran, Lily Zhang, Luca Rossetto, Hyowon Lee, Gerard Marks, Liting Zhou, Alice Dajung Jung, Ellen Rushe
Publisher	Springer Science and Business Media Deutschland GmbH
Pages	251-262
Number of pages	12
ISBN (Print)	9783032258083
DOIs	https://doi.org/10.1007/978-3-032-25809-0_20
Publication status	Published - 2026
Event	33rd Irish Conference on Artificial Intelligence and Cognitive Science, AICS 2025 - Dublin, Ireland Duration: 1 Dec 2025 → 2 Dec 2025

Publication series

Name	Communications in Computer and Information Science
Volume	2950 CCIS
ISSN (Print)	1865-0929
ISSN (Electronic)	1865-0937

Conference

Conference	33rd Irish Conference on Artificial Intelligence and Cognitive Science, AICS 2025
Country/Territory	Ireland
City	Dublin
Period	1/12/25 → 2/12/25

UN SDGs

This output contributes to the following UN Sustainable Development Goals (SDGs)

SDG 11 Sustainable Cities and Communities

Keywords

AI evaluation
Benchmark dataset
Cloud coverage
Computer vision

Access to Document

10.1007/978-3-032-25809-0_20

Cite this

Edirisinghe, S., Hensman, S., & Schoen-Phelan, B. (2026). A Continuous Cloud Coverage Dataset for Robust AI in Occluded Environments. In T. T. Mai, T. Ward, C. Gurrin, A. L.-D. Tran, L. Zhang, L. Rossetto, H. Lee, G. Marks, L. Zhou, A. D. Jung, & E. Rushe (Eds.), Artificial Intelligence and Cognitive Science - 33rd Irish Conference, AICS 2025, Revised Selected Papers (pp. 251-262). (Communications in Computer and Information Science; Vol. 2950 CCIS). Springer Science and Business Media Deutschland GmbH. https://doi.org/10.1007/978-3-032-25809-0_20

Edirisinghe, Shanika ; Hensman, Svetlana ; Schoen-Phelan, Bianca. / A Continuous Cloud Coverage Dataset for Robust AI in Occluded Environments. Artificial Intelligence and Cognitive Science - 33rd Irish Conference, AICS 2025, Revised Selected Papers. editor / Tai Tan Mai ; Tomas Ward ; Cathal Gurrin ; Allie Ly-Duyen Tran ; Lily Zhang ; Luca Rossetto ; Hyowon Lee ; Gerard Marks ; Liting Zhou ; Alice Dajung Jung ; Ellen Rushe. Springer Science and Business Media Deutschland GmbH, 2026. pp. 251-262 (Communications in Computer and Information Science).

@inproceedings{fe57e2f93eac488a92a46fdb437f0c8e,

title = "A Continuous Cloud Coverage Dataset for Robust AI in Occluded Environments",

abstract = "Cloud cover is a persistent challenge in satellite remote sensing, obscuring surface features and reducing the reliability of applications such as environmental monitoring, disaster response, and urban planning. Most existing cloud-related datasets provide binary masks, distinguishing only between cloud and clear conditions, and thus fail to capture cloud density, limiting systematic benchmarking and robustness analysis of AI models under varying occlusion. To address this gap, we introduce a continuous cloud-coverage dataset derived from SEN12MS-CR. Each Sentinel-2 tile is assigned a percentage label (0{\textendash}100\%) representing cloud proportion. The curated dataset includes 36,666 tiles, evenly sampled across four seasons, and is openly accessible via Zenodo. A case study using the UnCRtainTS cloud removal model shows that reconstruction quality declines progressively with higher cloud density. By supporting standardized evaluation across cloud conditions, the dataset enhances cloud-aware AI tasks, including removal, classification, and segmentation, and contributes to broader computer vision research on occlusion robustness and cognitive science perspectives on reasoning under uncertainty.",

keywords = "AI evaluation, Benchmark dataset, Cloud coverage, Computer vision",

author = "Shanika Edirisinghe and Svetlana Hensman and Bianca Schoen-Phelan",

note = "Publisher Copyright: {\textcopyright} The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.; 33rd Irish Conference on Artificial Intelligence and Cognitive Science, AICS 2025 ; Conference date: 01-12-2025 Through 02-12-2025",

year = "2026",

doi = "10.1007/978-3-032-25809-0\_20",

language = "English",

isbn = "9783032258083",

series = "Communications in Computer and Information Science",

publisher = "Springer Science and Business Media Deutschland GmbH",

pages = "251--262",

editor = "Mai, \{Tai Tan\} and Tomas Ward and Cathal Gurrin and Tran, \{Allie Ly-Duyen\} and Lily Zhang and Luca Rossetto and Hyowon Lee and Gerard Marks and Liting Zhou and Jung, \{Alice Dajung\} and Ellen Rushe",

booktitle = "Artificial Intelligence and Cognitive Science - 33rd Irish Conference, AICS 2025, Revised Selected Papers",

address = "Germany",

}

Edirisinghe, S, Hensman, S & Schoen-Phelan, B 2026, A Continuous Cloud Coverage Dataset for Robust AI in Occluded Environments. in TT Mai, T Ward, C Gurrin, AL-D Tran, L Zhang, L Rossetto, H Lee, G Marks, L Zhou, AD Jung & E Rushe (eds), Artificial Intelligence and Cognitive Science - 33rd Irish Conference, AICS 2025, Revised Selected Papers. Communications in Computer and Information Science, vol. 2950 CCIS, Springer Science and Business Media Deutschland GmbH, pp. 251-262, 33rd Irish Conference on Artificial Intelligence and Cognitive Science, AICS 2025, Dublin, Ireland, 1/12/25. https://doi.org/10.1007/978-3-032-25809-0_20

A Continuous Cloud Coverage Dataset for Robust AI in Occluded Environments. / Edirisinghe, Shanika; Hensman, Svetlana ; Schoen-Phelan, Bianca.
Artificial Intelligence and Cognitive Science - 33rd Irish Conference, AICS 2025, Revised Selected Papers. ed. / Tai Tan Mai; Tomas Ward; Cathal Gurrin; Allie Ly-Duyen Tran; Lily Zhang; Luca Rossetto; Hyowon Lee; Gerard Marks; Liting Zhou; Alice Dajung Jung; Ellen Rushe. Springer Science and Business Media Deutschland GmbH, 2026. p. 251-262 (Communications in Computer and Information Science; Vol. 2950 CCIS).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution › peer-review

TY - GEN

T1 - A Continuous Cloud Coverage Dataset for Robust AI in Occluded Environments

AU - Edirisinghe, Shanika

AU - Hensman, Svetlana

AU - Schoen-Phelan, Bianca

N1 - Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2026.

PY - 2026

Y1 - 2026

N2 - Cloud cover is a persistent challenge in satellite remote sensing, obscuring surface features and reducing the reliability of applications such as environmental monitoring, disaster response, and urban planning. Most existing cloud-related datasets provide binary masks, distinguishing only between cloud and clear conditions, and thus fail to capture cloud density, limiting systematic benchmarking and robustness analysis of AI models under varying occlusion. To address this gap, we introduce a continuous cloud-coverage dataset derived from SEN12MS-CR. Each Sentinel-2 tile is assigned a percentage label (0–100%) representing cloud proportion. The curated dataset includes 36,666 tiles, evenly sampled across four seasons, and is openly accessible via Zenodo. A case study using the UnCRtainTS cloud removal model shows that reconstruction quality declines progressively with higher cloud density. By supporting standardized evaluation across cloud conditions, the dataset enhances cloud-aware AI tasks, including removal, classification, and segmentation, and contributes to broader computer vision research on occlusion robustness and cognitive science perspectives on reasoning under uncertainty.

AB - Cloud cover is a persistent challenge in satellite remote sensing, obscuring surface features and reducing the reliability of applications such as environmental monitoring, disaster response, and urban planning. Most existing cloud-related datasets provide binary masks, distinguishing only between cloud and clear conditions, and thus fail to capture cloud density, limiting systematic benchmarking and robustness analysis of AI models under varying occlusion. To address this gap, we introduce a continuous cloud-coverage dataset derived from SEN12MS-CR. Each Sentinel-2 tile is assigned a percentage label (0–100%) representing cloud proportion. The curated dataset includes 36,666 tiles, evenly sampled across four seasons, and is openly accessible via Zenodo. A case study using the UnCRtainTS cloud removal model shows that reconstruction quality declines progressively with higher cloud density. By supporting standardized evaluation across cloud conditions, the dataset enhances cloud-aware AI tasks, including removal, classification, and segmentation, and contributes to broader computer vision research on occlusion robustness and cognitive science perspectives on reasoning under uncertainty.

KW - AI evaluation

KW - Benchmark dataset

KW - Cloud coverage

KW - Computer vision

UR - https://www.scopus.com/pages/publications/105048235978

U2 - 10.1007/978-3-032-25809-0_20

DO - 10.1007/978-3-032-25809-0_20

M3 - Conference contribution

AN - SCOPUS:105048235978

SN - 9783032258083

T3 - Communications in Computer and Information Science

SP - 251

EP - 262

BT - Artificial Intelligence and Cognitive Science - 33rd Irish Conference, AICS 2025, Revised Selected Papers

A2 - Mai, Tai Tan

A2 - Ward, Tomas

A2 - Gurrin, Cathal

A2 - Tran, Allie Ly-Duyen

A2 - Zhang, Lily

A2 - Rossetto, Luca

A2 - Lee, Hyowon

A2 - Marks, Gerard

A2 - Zhou, Liting

A2 - Jung, Alice Dajung

A2 - Rushe, Ellen

PB - Springer Science and Business Media Deutschland GmbH

T2 - 33rd Irish Conference on Artificial Intelligence and Cognitive Science, AICS 2025

Y2 - 1 December 2025 through 2 December 2025

ER -

Edirisinghe S, Hensman S , Schoen-Phelan B. A Continuous Cloud Coverage Dataset for Robust AI in Occluded Environments. In Mai TT, Ward T, Gurrin C, Tran ALD, Zhang L, Rossetto L, Lee H, Marks G, Zhou L, Jung AD, Rushe E, editors, Artificial Intelligence and Cognitive Science - 33rd Irish Conference, AICS 2025, Revised Selected Papers. Springer Science and Business Media Deutschland GmbH. 2026. p. 251-262. (Communications in Computer and Information Science). doi: 10.1007/978-3-032-25809-0_20