用哨兵2号影像+浮标数据，实现对马尔梅诺湖叶绿素a的深度精准预测。
Chlorophyll-a Mapping and Prediction in the Mar Menor Lagoon Using C2RCC-Processed Sentinel 2 Imagery
- 结合哨兵2号遥感与浮标数据，用机器学习建模预测水体不同深度的叶绿素a浓度。
- 表面到3-4米深度的预测误差（RMSLE）在0.34至0.39之间，决定系数最高达0.76。
- 方法可复用于其他浑浊近岸水域，助力藻华预警和生态管理。
西班牙东南部欧洲最大的高盐度海岸潟湖马尔梅诺近年来饱受富营养化危机，严重威胁生物多样性和水质。监测叶绿素a（反映浮游植物生物量）对提前预警有害藻华、指导治理措施至关重要。传统现场测量精度高但时空覆盖有限。卫星遥感提供更全面的视角，支持规模化长期监测。本研究旨在突破以往叶绿素监测多限于表层或时间覆盖不足的局限，建立可靠的水柱全深度叶绿素a预测与制图方法。研究整合哨兵2号影像与浮标实测数据，构建能实现高分辨率、深度分层监测的模型，提升富营养化早期预警能力。哨兵2影像经C2RCC大气校正处理，浮标数据按深度聚合。采用CatBoost、XGBoost、SVM及MLP等多种机器学习算法，通过交叉验证与多目标优化进行训练与验证。测试了不同波段组合与空间聚合策略以优化预测性能。结果显示预测效果随深度变化：表面至3-4米处的均方根对数误差（RMSLE）为0.34至0.39，决定系数（R²）分别为0.76、0.76、0.70和0.60。生成的地图成功再现已知富营养化事件。研究提出了一套端到端、经过验证的叶绿素制图方法，其融合多光谱波段组合、浮标校准与建模，可推广至其他浑浊近岸系统。
原文摘要 · Abstract (English)
The Mar Menor, Europe's largest hypersaline coastal lagoon, located in southeastern Spain, has undergone severe eutrophication crises, with devastating impacts on biodiversity and water quality. Monitoring chlorophyll-a, a proxy for phytoplankton biomass, is essential to anticipate harmful algal blooms and guide mitigation strategies. Traditional in situ measurements, while precise, are spatially and temporally limited. Satellite-based approaches provide a more comprehensive view, enabling scalable and long-term monitoring. This study aims to overcome limitations of chlorophyll monitoring, often restricted to surface estimates or limited temporal coverage, by developing a reliable methodology to predict and map chlorophyll-a concentrations across the water column of the Mar Menor. This work integrates Sentinel 2 imagery with buoy-based ground truth to create models capable of high-resolution, depth-specific monitoring, enhancing early-warning capabilities for eutrophication. Sentinel 2 images were atmospherically corrected using C2RCC processors. Buoy data were aggregated by depth. Multiple ML algorithms, including CatBoost, XGBoost, SVMs, and MLPs, were trained and validated using a cross-validation scheme with multi-objective optimization functions. Band-combination experiments and spatial aggregation strategies were tested to optimize prediction. The results show depth-dependent performance. The Root Mean Squared Logarithmic Error (RMSLE) obtained ranges from 0.34 at the surface to 0.39 at 3-4 m, while the R2 value was 0.76 at the surface, 0.76 at 1-2 m, 0.70 at 2-3 m, and 0.60 at 3-4 m. Generated maps successfully reproduced known eutrophication events. The study delivers an end-to-end, validated methodology chlorophyll mapping. Its integration of multispectral band combinations, buoy calibration, and modeling offers a transferable framework for other turbid coastal systems.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。