A Chromaticity Baseline Height Index and Sliding Window Iterative Thresholding for Automatic Harmful Algal Bloom Extraction

Frequent harmful algal blooms (HABs) have become an important ecological issue affecting lakes worldwide. Existing HABs indices are easily affected by complex water backgrounds, while global threshold segmentation methods cannot effectively adapt to the spatial heterogeneity of HABs, resulting in false detection over turbid waters and omission of HABs. To address these issues, this study developed a chromaticity baseline height index (CBHI) based on the spectral structural differences between HABs and background waters in chromaticity space and proposed a sliding window iterative thresholding (SWIT) method for adaptive and stable HABs extraction. The results show that CBHI effectively distinguishes HABs from clear and turbid waters. CBHI combined with SWIT achieved the highest accuracy on the validation dataset, with mean F1-score (F1) and Intersection of Union (IoU) values of 0.950 and 0.906, respectively. The extracted HABs area was highly consistent with the reference data (R2 = 0.964, mean absolute percentage error = 11.796%, and NRMSE = 4.961%). Its performance was better than that of CBHI combined with global thresholding (F1 = 0.816–0.859 and IoU = 0.696–0.761) and the alternative floating algal index combined with the global iterative thresholding method (F1 = 0.726 and IoU = 0.576). Comparison with the random forest classifier further showed that CBHI+SWIT achieved better overall performance than the supervised learning method. Validation across multiple lakes demonstrated that the proposed method can adapt to severe, moderate, and mild HABs outbreaks (F1 = 0.869–0.993), indicating good generalization capability. This method provides technical support for remote sensing monitoring of HABs in complex waters and aquatic environmental management.