Abstract: The Pinglu Canal estuary, located in the northern part of Maowei Sea, represents a typical mangrove tidal flat area subject to strong human disturbances. In this study, we employed remote sensing imagery and a GeoAI deep learning framework, using a U-Net model to quantitatively extract the spatiotemporal dynamics of mangroves in the estuary. We systematically analyzed the combined impacts of waves, oyster aquaculture, vessel traffic, and silt-promoting wave-dissipation engineering on mangrove evolution. The results show that between 2016 and 2025, the overall mangrove area in the Pinglu Canal estuary declined at an average rate of 6.7 ha/yr. The erosion rate accelerated markedly from 2021 to 2023, with the area reduction increasing from 3.4 ha/yr before 2021 to 16.3 ha/yr, and the shoreline retreating at a rate of -4.89 m/yr. Subsequently, from 2023 to 2025, the rate of area loss slowed to 5.75 ha/yr, with shoreline retreat decreasing to -3.69 m/yr; erosion primarily occurred along both sides of the river channel. The study indicates that increased vessel traffic and the removal of oyster rafts are major drivers of mangrove degradation, whereas ecological silt-promoting wave-dissipation structures effectively enhance shoreline stability and mitigate mangrove loss. These findings provide critical support for the coordinated development of canal navigation and mangrove ecosystem protection.

Key words: U-Net, deep learning, mangrove, Pinglu Canal estuary