Tracking the spatial and temporal evolution of salt marsh vegetation based on UAV sampling and seasonal phenology from Landsat data.

Abstract

Salt marshes, valued for their ecological importance, have been increasingly degraded in recent decades. Preserving salt marshes necessitates a critical approach that involves monitoring vegetation distribution and species composition. This study presents a high-precision salt marsh mapping framework for the Yellow River Delta (YRD), integrating Unmanned Aerial Vehicle (UAV), machine learning and seasonal phenological features from Landsat data. UAV data facilitate sampling efficiency, while seasonal phenology improves species differentiation in classification models. Among the tested algorithms, the Random Forest algorithm achieved the highest overall accuracy (89 %), outperforming support vector machines, gradient-boosted decision trees and deep neural network, particularly in identifying mixed-vegetation zones. Autumn phenological features emerged as critical discriminators for vegetation type classification. From 1991 to 2022, the salt marsh area exhibited an initial decline, followed by stabilization, and subsequent expansion, reaching 259.15 km² in 2022. Notably, the invasive species Spartina alterniflora expanded significantly after 2009, reaching 61.4 km² before its eradication in 2021. This research demonstrates that integrating UAV and seasonal phenological data provides a scalable, high-precision approach for long-term salt marsh monitoring. The framework provides robust tools and actionable insights for conservation, invasive species management, and ecosystem restoration.