华北水利水电大学学报(自然科学版)

蓝藻水华的暴发会对水体生态系统造成严重影响，不同的蓝藻水华监测方法得到的结果存在较大差异。以太湖为研究对象，利用MODIS影像产品和地表分类数据，提取了太湖水域范围，并分别采用波段比值、归一化植被指数、浮游藻类指数和支持向量机、随机森林等算法提取了2010—2022年太湖蓝藻的暴发区域。通过对比分析，评价了各种方法基于MODIS影像对太湖蓝藻水华空间和时间的监测效果及适用性。5种方法的结果表明：太湖蓝藻水华2010—2014年趋于稳定，2015年开始暴发，之后规模逐渐增大，并于2021年达到顶峰，然后在2022年突然大幅度下降；支持向量机算法监测结果的一致性最好，平均Kappa系数达到0.77,结果最准确和稳定；归一化植被指数监测结果的一致性较差，平均Kappa系数仅为0.61,结果存在较大不确定性。与太湖藻华产品对比后发现，各方法在准确性和稳定性上存在差异，支持向量机算法在监测中表现最佳应优先考虑使用。但单一方法都存在不足，需进一步优化和融合，找到最适合特定场景的解决方案。

Accurate monitoring of cyanobacterial blooms is critical for aquatic ecosystem management, yet methodological discrepancies in remote sensing approaches remain unresolved. This study focuses on Lake Taihu and utilizes MODIS satellite imagery and land cover classification data to extract the spatial extent of cyanobacterial blooms in the lake. Five methods, including band ratio, normalized difference vegetation index(NDVI), floating algae index(FAI), support vector machine(SVM), and random forest(RF), are employed to estimate the outbreak areas of cyanobacterial blooms(2010—2022). Temporal analysis revealed distinct bloom dynamics: relatively stable conditions during 2010—2014 were followed by escalating outbreaks from 2015 onward, peaking in 2021 before an abrupt decline in 2022. Quantitative validation through Kappa statistics demonstrated SVM′s superior performance(mean k=0.77), significantly outperforming NDVI(k=0.61), which exhibited substantial spatial-temporal inconsistency. When compared with the Taihu algal blooms product, the monitoring results of various methods show differences in accuracy and stability, while the SVM algorithm demonstrates the best performance. Therefore, this study proposes an adaptive monitoring framework combining SVM′s classification robustness with FAI′s sensitivity to early-stage blooms. Our findings emphasize the necessity for context-specific algorithm selection and highlight the potential of hybrid approaches for operational monitoring systems.