Hou, Siyuan, Xiaolong Yu, Wendian Lai, Fei Zhang, Hanyang Qiao, Peng Cheng, and Zhongping Lee. 2025. “Consistent Seagrass Mapping Across Various Tide Levels From Planet SuperDove Images.” IEEE Transactions on Geoscience and Remote Sensing 63:1–14. doi:10.1109/TGRS.2025.3574821.
Abstract
Seagrass meadows are vital blue carbon ecosystems, and remote sensing provides a cost-effective means of monitoring their changes at high spatiotemporal resolution. While existing algorithms excel in low-tide mapping, accurately and consistently identifying seagrass at mid-to-high tide levels remains challenging. This study presents a support vector machine (SVM)-based substrate classification model (SCM_SVM) for automated seagrass identification across various tidal conditions using Planet SuperDove imagery, with a demonstration provided for the Li’An Lagoon (LAL). By training with ~1.8 million matched ground-truth substrate data and Rayleigh scattering-corrected top-of-atmosphere reflectance ( ρrc ), SCM_SVM could robustly identify seagrass from SuperDove ρrc measurements across low-to-high tide levels. Validation against independent field measurements indicated a detection accuracy of seagrass exceeding 85%. Notably, SCM_SVM provided consistent seagrass distributions in spatial patterns and extents for images acquired under different tidal levels. Time-series analysis from 2021 to 2023 revealed a significant decline of −0.15 km2/yr in the area. These results underscore the potential of SuperDove for high spatiotemporal resolution monitoring of seagrass dynamics. Future work will focus on enhancing the global applicability of SCM_SVM and extending it to detect other submerged vegetation.
Fig. 3.
Proposed framework for seagrass mapping in this study. (a) Construction of the training dataset using matchups between satellite radiometric measurements and groundtruth substrate data. (b) Model development based on machine learning. (c) Model validation and application.
