EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC-ND 4.0

A hybrid deep learning and tree-based ensemble approach for forecasting shellfish biotoxin contamination from satellite and in-situ data

Fabien Guegan, Sérgio Tavares, Susana M Rodrigues, Inês Basílio, Andreia Cruz, Jorge Cruz, Pedro R Costa, Ludwig Krippahl, Marta B Lopes

Harmful Algae · 2026-12 · 160 · 103214

DOI: 10.1016/j.hal.2026.103214 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Harmful Algae, 2026-12. Verfasst von Fabien Guegan und 8 weiteren Autor:innen. Beteiligte Einrichtung: UNIDEMI, Department of Mechanical and Industrial Engineering, NOVA School of Science and Technology (NOVA FCT), Universidade NOVA de Lisboa, Caparica…. Hauptthemen laut MEDLINE: Shellfish, Deep Learning, Marine Toxins, Environmental Monitoring. Die Arbeit nennt 33 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY-NC-ND 4.0). Gefördert von: Fundação para a Ciência e a Tecnologia.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Shellfish production is an important economic sector in several Portuguese coastal regions, yet biotoxin contamination by harmful algae poses recurrent health concerns and significant economic risks. Predicting contamination remains challenging due to the heterogeneous nature of the data and its high dimensionality. Biotoxin measurements collected by the Portuguese Institute for Sea and Atmosphere (IPMA) form irregular time series due to inconsistent sampling intervals, while satellite data from the Copernicus program provide spatially dense environmental information with thousands of measurements per time step. In this study, we evaluate the integration of satellite-derived environmental information into diarrhetic shellfish poisoning (DSP) forecasting models for the southern coast of Portugal, considering prediction horizons of up to six weeks. Satellite image sequences of sea surface temperature (SST) and chlorophyll-a concentration (chl-a) were encoded using 2D convolutional autoencoders to extract compact latent representations of environmental variability. These latent features were combined with historical DSP contamination measurements to forecast the maximum DSP concentration within future windows of one to six weeks using Random Forest and Gradient Boosting models. Results show that satellite-derived features frequently improved predictive performance during validation, although these gains were smaller and less consistent during the held-out 2024 test period and varied across stations and forecasting horizons. Nevertheless, the results highlight the potential of freely available, continuously updated satellite products to provide valuable complementary information for DSP forecasting. Their ability to capture spatially resolved environmental patterns offers a promising avenue for strengthening contamination prediction and supporting more informed shellfish production management, although further prospective evaluation across multiple years and locations is needed before operational deployment.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

ShellfishDeep LearningMarine ToxinsEnvironmental MonitoringShellfish PoisoningPortugalForecastingSatellite ImageryPrediction AlgorithmsPredictive Learning Models

Stichwörter der Autor:innen

AutoencodersHarmful algal bloomsSatellite imagingShellfish contaminationTree-based models

Substanzen

Marine Toxins

Autor:innen und Institutionen

  • UNIDEMI, Department of Mechanical and Industrial Engineering, NOVA School of Science and Technology (NOVA FCT), Universidade NOVA de Lisboa, Caparica, Portugal.
  • NOVA Laboratory for Computer Science and Informatics (NOVA LINCS), NOVA School of Science and Technology (NOVA FCT), Universidade NOVA de Lisboa, Caparica, Portugal.
  • IPMA - Instituto Português do Mar e da Atmosfera, 1495-165 Algés, Portugal.
  • Oceano Fresco S.A., Porto da Nazaré, Nazaré, Portugal.
  • UNIDEMI, Department of Mechanical and Industrial Engineering, NOVA School of Science and Technology (NOVA FCT), Universidade NOVA de Lisboa, Caparica, Portugal; Center for Mathematics and Applications (NOVA Math), NOVA…

Förderung

  • Fundação para a Ciência e a Tecnologia

Zitieren

APA

Guegan, F., Tavares, S., Rodrigues, S. M., Basílio, I., Cruz, A., Cruz, J., Costa, P. R., Krippahl, L., & Lopes, M. B. (2026). A hybrid deep learning and tree-based ensemble approach for forecasting shellfish biotoxin contamination from satellite and in-situ data. Harmful Algae, 160, 103214. https://doi.org/10.1016/j.hal.2026.103214

Vancouver

Guegan F, Tavares S, Rodrigues SM, Basílio I, Cruz A, Cruz J, et al. A hybrid deep learning and tree-based ensemble approach for forecasting shellfish biotoxin contamination from satellite and in-situ data. Harmful Algae. 2026;160:103214. doi:10.1016/j.hal.2026.103214

BibTeX
@article{Guegan2026,
  author = {Guegan, Fabien and Tavares, Sérgio and Rodrigues, Susana M and Basílio, Inês and Cruz, Andreia and Cruz, Jorge and Costa, Pedro R and Krippahl, Ludwig and Lopes, Marta B},
  title = {A hybrid deep learning and tree-based ensemble approach for forecasting shellfish biotoxin contamination from satellite and in-situ data},
  journal = {Harmful Algae},
  year = {2026},
  volume = {160},
  pages = {103214},
  doi = {10.1016/j.hal.2026.103214}
}
Kostenlos starten →