Unveiling Seaward Expansion Pattern in Mangrove Forests Using Uav Remote Sensing and Deep Learning

Abstract

Mangrove wetlands, as ecologically critical transition zones along land-sea interfaces, are naturally expanding seaward in China due to rapid sediment accretion. However, the spatiotemporal dynamics of species competition and stand structure during this expansion remain poorly understood. In this study, we developed an integrated remote sensing framework combining satellite imagery, low-altitude unmanned aerial vehicle (UAV) data, and deep learning techniques to investigate China's largest contiguous mangrove. A novel algorithm based on Mask R-CNN was implemented, achieving high accuracy in automated species identification (Aegiceras corniculatum and Avicennia marina, overall accuracy: 87.7%) and canopy parameter extraction (precision >70%), enabling high-resolution quantification of forest structural dynamics. Results show that A. corniculatum acts as the initial pioneer species in colonizing newly accreted mudflats under extreme tidal stress. As environmental conditions stabilize, A. marina becomes established and eventually outcompetes A. corniculatum, leading to monospecific A. marina stands in older zones. Interspecific competition intensity follows a hump-shaped trend: peaking during the early successional phase (1–7 years), then declining as A. marina dominance increases. Correspondingly, stand spatial distribution shifts from uniform to random with decreasing stand age. These findings reveal a distinct pattern of sequential species replacement and competition-driven community succession during mangrove seaward expansion. The methodological framework and ecological insights presented here provide valuable guidance for near-natural mangrove afforestation and the restoration of degraded coastal ecosystems, contributing to sustainable and scalable wetland conservation strategies.

Keywords: Mangrove, UAV remote sensing, Deep learning, Seaward expansion pattern, Species composition, Stand structure