757 / 2024-09-19 11:27:49

Consistent seagrass mapping from Planet SuperDove images across various tide levels: A case study in the Li’an Lagoon

seagrass meadows,Remote sensing (RS)

Seagrass meadows are vital blue carbon ecosystems, and understanding their spatial-temporal variability is essential for their conservation and management. Remote sensing provides a cost-effective means for high-resolution, frequent, and broad-scale monitoring of seagrass distribution. Although various remote sensing algorithms have been developed for mapping seagrass distributions under low-tide scenarios, achieving accurate and consistent seagrass mapping at mid-to-high tide levels remains a challenge. In this effort, we developed a Support Vector Machine (SVM)-based Substrate Classification Model, termed SCM_SVM, for automated seagrass identification in the Li’an Lagoon using Planet SuperDove imagery, which is proved to be robust across various tidal scenarios. The inputs for SCM_SVM are Rayleigh scattering-corrected top-of-atmosphere reflectance (ρ_rc) from the eight SuperDove bands. The training dataset (~2.3 million match-ups) was constructed by matching ground-truth substrate data, visually interpreted from low-tide images, with ρ_rc from mid-to-high-tide images acquired within a two-week window. SCM_SVM was validated using independent substrate measurements from field surveys with an acoustic instrument and underwater camera, where the overall seagrass detection accuracy exceeds 89%. Remarkably, SCM_SVM provided consistent seagrass distributions for images acquired at different tidal levels in both the spatial pattern and the estimated seagrass extents. Time-series analysis from SuperDove imagery (2021-2024) revealed pronounced seasonal variations in seagrass coverage in the Li’an Lagoon, with lower extents in summer and higher extents in spring and winter that are consistent with field observations. 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.