| 摘要: |
| 砂质基底海岛地形受潮流水动力影响显著，地形演变迅速，及时掌握其周边地形动态变化规律对海岛保护与管理具有重要意义。文章以某砂质基底海岛周边海域为研究区，基于 2013 年环境减灾卫星（HJ-1B）遥感数据，构建了 4 种能够削弱悬浮泥沙影响的水深遥感反演模型（支持向量机、随机森林、全连接神经网络和卷积神经网络）。通过系统对比各模型精度，发现卷积神经网络（CNN）模型表现最优， 其反演结果与实测水深的决定系数（R 2）达 0.98，均方根误差（RMSE）为 0.74 m，平均相对误差（MRE）为 5.97%。为进一步验证模型适用性并揭示地形演变规律，将 CNN 模型迁移应用于 2017 年环境卫星影像，实现了 2013—2017 年间地形变化的定量分析。结果表明：CNN 模型在迁移应用中仍保持良好性能，能够准确重构区域地形宏观格局与微地貌细节；2013—2017 年，研究区整体地形以淤积为主，淤积区面积占比 62.93%，平均淤积厚度为 2.66 m（年均约 0.53 m），冲刷区面积占比 37.07%，平均冲刷厚度为 1.16 m（年均约0.23 m），整体呈现向东淤积趋势，需重点关注海岛北侧约 1 km 处出现的局部冲刷现象。文章构建的高精度反演模型与多时相动态监测方法，可为砂质基底海岛的地形稳定性评估与海岸带管理提供可靠技术支撑。 |
| 关键词: 砂质基底海岛 水深反演 HJ-1B 环境卫星 卷积神经网络 动态监测 冲淤演变 |
| DOI：10.20016/j.cnki.hykfygl.2026.03.011 |
| 投稿时间：2025-10-28修订日期：2026-02-20 |
| 基金项目:2021 年江苏省自然资源科技项目（2021040）；2021 年江苏省海洋科技创新项目（JSZRHYKJ202101）． |
|
| Satellite-Based Dynamic Monitoring of Topographic Changes in Sandy Substrate Islands |
| LI Jing,XIAO Yizhe,CHEN Hao |
| Jiangsu province surveying & mapping engineering institute |
| Abstract: |
| Sandy substrate islands，significantly influenced by tidal hydrodynamics，experience rapid geomorphological changes，making it crucial to monitor the dynamic evolution of their surrounding topography for effective island conservation and management．This study focuses on the coastal waters surrounding a sandy substrate island．Utilizing 2013 HJ-1B satellite imagery，four water depth inversion models-Support Vector Machine（SVM）， Random Forest（RF），Fully Connected Neural Network（FNN），and Convolutional Neural Network（CNN）-were developed to mitigate the impact of suspended sediments．A systematic comparison revealed that the CNN model achieved optimal performance，with a coefficient of determination（R 2）of 0.98，a Root Mean Square Error（RMSE） of 0.74 m，and a Mean Relative Error（MRE）of 5.97% against measured water depth data．To further validate the model’s applicability and investigate topographic evolution，the CNN model was transferred and applied to 2017 HJ- 1A/B imagery，enabling quantitative analysis of topographic changes between 2013 and 2017．Results indicate that the CNN model maintained robust performance in the transfer applications，accurately reconstructing both the macroscale terrain and microscale topographic details．During 2013—2017，the study area was dominated by siltation，with accretion zones accounting for approximately 62.93% of the total area and an average siltation thickness of 2.66 m（approximately 0.53 m annually）．Erosion zones covered about 37.07% of the area，with an average erosion depth of 1.16 m（approximately 0.23 m annually）．An overall eastward siltation trend was observed，while localized scour near approximately 1 km north of the island requires particular attention．The high-precision inversion model and multi- temporal dynamic monitoring methodology developed in this study provide a reliable technical framework for assessing topographic stability and supporting coastal zone management of sandy substrate islands． |
| Key words: Sandy substrate islands Bathymetry inversion HJ-1B satellite Convolutional neural network（CNN） Dynamic monitoring Erosion and deposition evolution |
