A large-scale mangrove species classification method using time-series data with phenological information and gaussian mixture model

Keywords: mangrove forest, species classification, phenology, gaussian mixture model, large scale
Abstract Type: Paper Abstract

Abstract

Mangrove forests play a critical role in ecological balance and socioeconomic stability, particularly in tropical and subtropical regions. This study introduces a novel approach to map three dominant mangrove species (Rhizophora mangle, Avicennia germinans, and Laguncularia racemosa) by leveraging phenology from high-temporal-resolution HLS imagery. Using harmonic regression to model phenological changes in time-series vegetation indices (NDVI, EVI, MVI, and RVI), we achieved an 86% overall classification accuracy across the species. A Gaussian mixture model, incorporating key phenological features including average vegetation cover and seasonal variation magnitude, successfully distinguished among three mangrove species. Our findings demonstrate the efficacy of combining harmonic regression with GMM-based unsupervised classification to capitalize on phenological variations, presenting a scalable solution for large-scale mangrove species mapping. This research offers valuable insights for enhancing mangrove ecosystem monitoring and management, supporting biodiversity conservation, carbon storage, and coastal resilience.