New Deep Learning System Automatically Spots Sea Snot Using Satellite Data

Jim Crocker
26th September, 2025

New Deep Learning System Automatically Spots Sea Snot Using Satellite Data

Aerial photo of marine mucilage ("sea snot")

Photo: Ceylan Yüceoral

Key Findings

  • This study, focused on the Marmara Sea, aimed to automatically detect mucilage formations (“sea snot”) using Sentinel-1 satellite data to aid timely intervention
  • Researchers successfully used deep learning and machine learning models, achieving 96% to 100% accuracy in identifying mucilage-covered areas from clear water
  • VH polarization data from Sentinel-1 proved more effective than VV polarization for mucilage detection, likely due to its ability to capture the widespread nature of the blooms
Mucilage events, often described as ‘sea snot’, are large accumulations of organic matter in the water column and pose a growing threat to marine ecosystems globally. These blooms can smother marine life, disrupt the food chain, and impact tourism and fisheries. Recent, particularly severe outbreaks in the Sea of Marmara have highlighted the urgency for effective monitoring and rapid detection methods.[1] University of Piri Reis, University of Batman, University of California at Riverside, and University of Khuzestan researchers undertook a study to improve the automated identification of mucilage formations using satellite imagery and advanced computational techniques. The study focused on a region of the Marmara Sea affected by a significant mucilage event in May 2021, specifically the area between Armutlu and Zeytinbağı. The core problem addressed was the need for a reliable, scalable method to differentiate mucilage-covered areas from clear water, enabling timely intervention and mitigation efforts. Traditional monitoring relies heavily on visual inspection and in-situ sampling, which is both time-consuming and limited in scope. To address this, the researchers collected 1300 GPS-tagged samples from areas confirmed to be covered in mucilage (May 17-22) and 2600 samples from clear water areas (June 21-22). These samples served as ground truth data for training and validating the image analysis models. The key innovation lies in the use of time-series analysis of Sentinel-1 satellite images. Sentinel-1 is a radar satellite, meaning it uses microwave signals instead of visible light, allowing it to penetrate clouds and provide data regardless of weather conditions – a significant advantage over optical sensors. The researchers specifically utilized two polarization bands of the Sentinel-1 data: VV and VH. These bands represent the backscattering of the radar signal from the water surface, providing information about the roughness and composition of the area. Different materials reflect radar signals differently, creating a unique “signature” that can be used for identification. By recording the numerical backscattering values from these bands, a unique dataset was created to train machine learning and deep learning algorithms. The study builds upon earlier work demonstrating the potential of satellite remote sensing for mucilage detection[2]. Previous research using Sentinel-2 data established the spectral characteristics of different mucilage types – white (freshly formed), yellow (wind/current transported), and brown (aged) – and correlated these with sea surface temperatures, suggesting a link between warming waters and bloom formation. However, these optical methods are susceptible to cloud cover.[3] showed the correlation of MODIS reflectance profiles with field measurements, but also acknowledged limitations in accuracy, particularly near the shoreline. ’s study goes further by employing radar data, overcoming the limitations of cloud interference. The researchers tested a range of algorithms, including deep learning models like Recurrent Neural Networks (RNNs) and Convolutional Neural Networks (CNNs), alongside more traditional machine learning methods such as Decision Trees, Naive Bayes, and Support Vector Machines (SVM). A hybrid approach combining the Transformer Method with Logistic Regression was also implemented. The results were highly promising, with accuracies ranging from 96% to 100% across all models. The success of these models indicates that Sentinel-1 data, combined with advanced computational techniques, can effectively and automatically detect mucilage formations. This has significant implications for environmental monitoring. Early detection allows for targeted cleaning efforts, reducing the impact on marine ecosystems and associated industries. The high accuracy achieved suggests that this methodology can be deployed at scale, providing a cost-effective and reliable solution for monitoring mucilage blooms in the Sea of Marmara and potentially other vulnerable regions. Furthermore, the study indirectly supports findings from[4], which highlighted the growing importance of deep learning in machine learning and its ability to handle large datasets. The substantial amount of satellite data required for this type of analysis makes deep learning an ideal choice, as it excels at identifying patterns and making predictions from complex information.[5] also demonstrates the power of integrating multi-source data, a concept that could be further expanded upon in future research by combining Sentinel-1 data with optical data from Sentinel-2 and other sources to provide a more comprehensive understanding of mucilage bloom dynamics.

EnvironmentEcologyMarine Biology

References

Main Study

1) A hybrid deep learning framework combining transformer and logistic regression models for automatic marine mucilage detection using sentinel-1 SAR data: A case study in Armutlu-Zeytinbağı, Marmara Sea

Published 25th September, 2025

https://doi.org/10.1371/journal.pone.0330721

Related Studies

2) Detection of mucilage phenomenon in the Sea of Marmara by using multi-scale satellite data.

https://doi.org/10.1007/s10661-022-10267-6

3) Daily monitoring of marine mucilage using the MODIS products: a case study of 2021 mucilage bloom in the Sea of Marmara, Turkey.

https://doi.org/10.1007/s10661-022-09831-x

4) Review of deep learning: concepts, CNN architectures, challenges, applications, future directions.

https://doi.org/10.1186/s40537-021-00444-8

5) Monitoring phycocyanin in global inland waters by remote sensing: Progress and future developments.

https://doi.org/10.1016/j.watres.2025.123176


Related Articles

An unhandled error has occurred. Reload 🗙