Phenology-driven hierarchical framework for grass species classification in coastal wetlands based on Sentinel-1/2 imagery: A case study of the Jiangsu Coast, Eastern China

Coastal wetlands are among the most vulnerable and valuable ecosystems, yet they are undergoing persistent changes in land surface conditions and vegetation composition driven by climate change and intensive human activities. Accurate mapping of grass species distribution is essential for the control of invasive plants, biodiversity conservation, and vegetation carbon accounting, yet the task remains challenging due to tidal inundation, frequent cloud cover, and the high spectral similarity among coastal grass species. This study developed a hierarchical classification framework to mitigate these limitations. The intertidal zone was first delineated from multi-temporal Sentinel-2 observations by exploiting temporal differences in water coverage, which compensated for the omission of grass species affected by tidal fluctuations that conventional classification methods have overlooked. Subsequently, two phenology-based vegetation features derived from monthly NDVI were constructed to improve the discrimination of spectrally similar species in the supratidal zone and were incorporated into a random forest classifier. The proposed framework achieved an overall accuracy (OA) of 0.94, notably higher than the OA of 0.85 from the traditional single-level classification, and effectively reduced misclassifications among species with overlapping spectral signatures. The hierarchical framework increased the total detected grassland area by approximately 13657.91 ha, with Spartina alterniflora contributing the majority of the improvement due to enhanced detection in the intertidal zone. In 2023, Jiangsu coastal zones supported 177,464.8 ha of grassland, with Spartina alterniflora occupying 52490.8 ha and showing substantial dominance along the Yancheng coasts. The successful detection of large areas of S. alterniflora that were omitted by a single-level model confirmed the importance of addressing tidal interference and demonstrated the benefits of incorporating phenological differences among grass species into the classification process. These findings demonstrate that the hierarchical strategy provides a more reliable solution to large-scale grass species mapping in dynamic coastal environments and offers practical support for invasive species management and ecosystem protection.