A phenomenology-based algorithm for extracting submerged aquatic vegetation in lakes using Sentinel-2 time seriesImagery: a case study of <i>Potamogeton crispus</i>

Submerged aquatic vegetation (SAV), particularly <i>Potamogeton crispus</i> (P.C.), plays a critical role in lake ecosystems by supporting nutrient cycling and sediment stabilization. Accurate and large-scale monitoring of SAVs is essential for effective ecosystem management and conservation. However, traditional classification methods struggle with large-scale applications because of their reliance on extensive sample data and the influence of regional phenological variations, which limits their generalizability. This study introduces the <i>Potamogeton crispus</i> index (PCI), a novel approach developed using time-series Sentinel-2 imagery processed on the Google Earth Engine (GEE) platform. The PCI automatically captures the differences between P.C. and other aquatic vegetation during the emergence and senescence phases by utilizing maximum normalized difference vegetation index (NDVI) composites and the Otsu thresholding method, enabling large-scale extraction. Four representative lakes with significant P.C. presence were selected as study areas. The results of phenological-stage identification revealed that in Baoan Lake, the P.C. completely died at the earliest (DOY = 160), whereas in Baiyang Lake, the P.C. persisted the longest (DOY = 189). The field and high-resolution satellite image validation results demonstrated overall classification accuracies ranging from 89.9% to 95.22%, with kappa coefficients between 0.85 and 0.92. Compared with alternative methods, including the floating-leaf vegetation sensitivity index (FVSI), submerged vegetation sensitivity index (SVSI), modified normalized difference water index (MNDWI), and NDVI, the PCI achieved a 5% to 18% improvement in mapping accuracy. Further application across four additional lakes with diverse geographic and climatic conditions confirmed the robustness and adaptability of the proposed method. These findings underscore the PCI’s reliability and scalability for large-scale monitoring of P.C., demonstrating its potential for mapping other types of SAVs.