摘要： 红树林是维系海岸带生态安全与生物多样性的关键生态系统, 明确其时空动态及驱动机制, 对湿地保护与适应性管理具有重要意义。以广西北海金海湾国际湿地红树林为研究对象, 基于Google Earth Engine(GEE)平台, 筛选 2019—2025 年云量低于 30% 的 Sentinel-2 卫星影像, 获取有效影像 487景, 通过NDVI-P75算法逐年合成低潮位影像, 通过ResNet34-UNet构建AI模型实现红树林高精度提取, 并结合Google高清历史影像、实地调查及大疆无人机航拍对结果进行分析, 结果表明: (1)2019—2025年间, 研究区红树林面积总体呈扩张趋势, 从146.86ha增加至167.23ha, 具有“整体扩张、局部波动”的特征; (2)空间分异明显, 东部核心保育区以年均2.05ha·yr⁻¹的速率持续扩张, 中区以年均1.19ha·yr⁻¹稳步增长, 西区则呈现“波动恢复型”, 年均增长仅 0.16ha·yr⁻¹; (3)2022年为研究期间红树林面积的阶段性低谷,主要归因于外来速生种拉关木(Laguncularia racemosa)的人工清理工程, 导致2021—2022年间红树林面积净减少6.13ha。此后, 随着“退塘还林”生态修复工程的实施, 以及金海湾湿地2022年入选国际重要湿地名录以来开展的外来物种系统管控与差异化功能分区管理, 红树林面积逐步恢复, 并进入持续增长阶段。金海湾红树林这一 “局部波动、总体扩张” 的动态变化表明, 在外来种清除、湿地修复与分区分类管理的协同作用下, 人为干扰可得到有效缓解, 这为同类型滨海湿地的可持续治理提供科学依据。

关键词: 红树林, 深度学习, 北海金海湾, 国际重要湿地, 遥感

Abstract: Mangroves are key coastal ecosystems that maintain ecological security and biodiversity. Understanding their spatiotemporal dynamics and driving mechanisms is essential for wetland conservation and adaptive management. Taking the mangroves of the Jinhai Bay Ramsar Wetland in Beihai, Guangxi, China, as the study area, this study employed the Google Earth Engine (GEE) platform to acquire 487 valid Sentinel-2 images with cloud cover below 30% during 2019–2025. Annual low-tide composite images were generated using the NDVI-P75 algorithm, and a ResNet34-UNet-based artificial intelligence (AI) model was developed for high-accuracy mangrove mapping. The mapping results were further interpreted and validated using Google high-resolution historical imagery, field surveys, and DJI unmanned aerial vehicle (UAV) observations. The results showed that: (1) from 2019 to 2025, the mangrove area exhibited an overall expansion trend, increasing from 146.86ha to 167.23ha, characterized by overall expansion with local fluctuations; (2) pronounced spatial heterogeneity was observed among different functional zones. The eastern core conservation zone expanded continuously at an average rate of 2.05ha·yr⁻¹, the central zone increased steadily at 1.19ha·yr⁻¹, whereas the western zone exhibited a fluctuation–recovery pattern with an average annual increase of only 0.16ha·yr⁻¹; and (3) the mangrove area reached a temporary low point in 2022, mainly due to the artificial removal of the invasive fast-growing species Laguncularia racemosa, resulting in a net loss of 6.13ha between 2021 and 2022. Subsequently, with the implementation of the pond-to-mangrove ecological restoration project, together with systematic invasive species control and differentiated functional zoning management following the designation of the Jinhai Bay Wetland as a Ramsar Wetland in 2022, the mangrove area gradually recovered and entered a phase of sustained growth. The observed dynamic pattern of local fluctuations but overall expansion indicates that the synergistic implementation of invasive species removal, wetland restoration, and zoned management can effectively mitigate anthropogenic disturbances, providing a scientific basis for the sustainable management of similar coastal wetlands.

Key words: Mangrove Forest, Deep Learning, Beihai Jinhai Bay, Ramsar Wetland, Remote Sensing