A Deep Learning Model for Cloud Detection and Classification in Sentinel-3 OLCI L1B Imagery

Exploring deep learning cloud segmentation on Sentinel-3 images.

Requirements

  • M.Sc. in Data Science, Computer Science, Artificial Intelligence, Mathematics, or similar
  • Strong knowledge of Python and deep learning frameworks (PyTorch, Lightning)
  • Basic background in computer vision, preferably semantic segmentation or domain adaptation
  • Basic knowledge of remote sensing and multispectral satellite imagery

Description

This thesis addresses a practical gap in the Sentinel-3 OLCI data processing workflow: unlike L2 products, L1B data does not include a cloud mask, requiring users to separately download and process the corresponding L2 product solely to obtain cloud flagging information. This dependency introduces overhead in terms of storage, bandwidth, and processing complexity — particularly in operational or large-scale analysis pipelines where only L1B radiance data may be readily available. To bridge this gap, this work should propose a supervised deep learning model capable of generating cloud masks directly from OLCI L1B top-of-atmosphere reflectance data across its 21 spectral bands, eliminating the need to access L2 products for cloud screening purposes. Ground truth labels are derived from co-registered L2 cloud mask products, which serve as a reliable proxy for supervised training and evaluation. Beyond binary cloud detection, the model should be extended to perform cloud type classification, distinguishing between categories such as optically thick clouds, thin cirrus, fractional or sub-pixel cloud cover, and clear sky.

Main Activities

  • Literature review on satellite cloud detection models, domain adaptation for remote sensing, and multi-sensor fusion
  • Characterization of spectral band correspondences and distribution shifts between Sentinel-3 OLCI and other satellites commonly used for such task (i.e. Sentinel-2)
  • Data cleaning, validation and preprocessing
  • Development of a cloud detection model
  • [Enhancement] Development of a cloud classification model

Contacts