TLDR: Researchers developed a new method using Sentinel 2 satellite images and AI to predict chlorophyll-a (Chl-a) levels, an indicator of algal blooms, throughout the Mar Menor lagoon’s water column. This approach, which combines atmospheric correction and various machine learning models, provides more detailed and depth-specific insights than previous surface-only monitoring, helping to anticipate and manage eutrophication crises in this vital Spanish lagoon.
The Mar Menor, Europe’s largest hypersaline coastal lagoon located in southeastern Spain, has faced severe environmental challenges, particularly eutrophication crises that have devastated its biodiversity and water quality. A new research paper titled “Chlorophyll-a Mapping and Prediction in the Mar Menor Lagoon Using C2RCC-Processed Sentinel 2 Imagery” by Antonio Mart´ınez-Ibarra, Aurora Gonz ´alez-Vidal, Adri ´an C ´anovas-Rodr´ıguez, and Antonio F. Skarmeta, introduces a robust methodology to enhance the monitoring of this vital ecosystem.
Monitoring chlorophyll-a (Chl-a), a key indicator of phytoplankton biomass, is crucial for anticipating harmful algal blooms and guiding mitigation efforts. Traditionally, in-situ measurements, while precise, are limited in their spatial and temporal coverage. This study addresses these limitations by developing a reliable method to predict and map Chl-a concentrations across the entire water column of the Mar Menor, offering a more comprehensive and scalable view.
The researchers integrated nearly a decade of Sentinel 2 satellite imagery with ground truth data from buoys. The satellite images underwent atmospheric correction using C2RCC processors, which are specifically designed for complex coastal waters. Buoy data were aggregated by depth into four layers: 0–1 meter, 1–2 meters, 2–3 meters, and 3–4 meters. A variety of machine learning and deep learning algorithms, including Random Forest, XGBoost, CatBoost, Multilayer Perceptron Networks, and ensemble models, were then trained and validated using cross-validation techniques.
The study’s findings reveal that the performance of the predictive models is dependent on depth. At the surface (0–1 meter), XGBoost and ensemble models, utilizing C2X-Complex processed data with a 9×9 pixel aggregation window, achieved a high R² value of 0.89. For the 1–2 meter depth, CatBoost and ensemble models performed best, reaching an R² of 0.87. Interestingly, at 2–3 meters, unprocessed Top of Atmosphere (TOA) reflectances combined with the K-Nearest Neighbors (KNN) algorithm yielded the best results with an R² of 0.81. As expected, performance declined at the deepest layer (3–4 meters), where Random Forest achieved an R² of 0.66.
The generated maps successfully reproduced known eutrophication events, such as the 2016 crisis and a surge detected in 2025, confirming the robustness of the methodology. This end-to-end, validated approach for depth-specific chlorophyll mapping goes beyond previous efforts that primarily focused on surface estimates. Its integration of multispectral band combinations, buoy calibration, and advanced machine learning models provides a transferable framework that could be applied to other turbid coastal systems facing similar environmental challenges.
The study also highlights the importance of selecting appropriate processing methods and spatial aggregation strategies based on depth. For instance, larger aggregation windows (e.g., 5×5, 9×9, 15×15 pixels) generally provided more accurate results by smoothing out local fluctuations in reflectance values. While the C2X-Complex processor consistently delivered strong performance across most depths, the use of unprocessed TOA reflectances at 2-3 meters suggests that different depths may require tailored processing approaches.
Also Read:
- Advanced Satellite Monitoring and AI System Boost Wildfire and Air Quality Tracking
- MSCloudCAM: A New Approach to Cloud Detection in Multispectral Satellite Data
Looking ahead, the researchers propose several avenues for future work, including automating the chlorophyll map generation pipeline, studying historical Chl-a concentrations over the last decades, incorporating additional satellite-derived features like turbidity, and extending the methodology to higher-resolution satellite platforms. This research represents a significant step forward in leveraging remote sensing and artificial intelligence for comprehensive environmental monitoring of vulnerable aquatic ecosystems like the Mar Menor. For more details, you can read the full research paper here.
