基于Sentinel-2影像的天岗湖水域岸线动态监测研究

Research on dynamic monitoring Tiangang Lake shorelines based on Sentinel-2 images

摘要

【目的】准确监测河湖水域岸线的主要目标及其变化对河湖岸线管控至关重要，探索利用高时空遥感数据实现河湖水域岸线动态监测具有重要意义。【方法】以天岗湖为研究区，选取了不同时期的Sentinel-2影像，提出了分类后变化监测方法实现对天岗湖水域岸线动态变化监测。首先评估最大似然法、随机森林法和面向对象法的土地利用分类精度，然后计算土地利用变化图谱分析天岗湖水域岸线动态变化，最后对比光伏的识别精度及其时空变化。【结果】选取的三种土地利用分类方法中，面向对象的分类精度最佳，平均总体分类精度和Kappa系数为92.8%和0.91,其次为随机森林法和最大似然法。2019—2023年天岗湖江苏段的水域岸线范围内的光伏面积迅速增加，水体转入光伏的土地利用变化率最大。面向对象法对光伏识别的精度最佳，交并比IoU均值为91.4%,利用Sentinel-2影像可以准确监测不同时期光伏的动态变化。【结论】利用面向对象分类方法可以准确识别和监测水域岸线主要目标及其时空变化，可为河湖动态监管提供参考。

Abstract

[Objective] Accurately monitoring the main objective and changes of river and lake shorelines is crucial for river and lake shorelines management. Exploring the use of high spatiotemporal remote sensing data to achieve dynamic monitoring of river and lake shorelines is of great significance. [Methods] Sentinel-2 images from different periods of Tiangang Lake were selected. The post classification change detection method was proposed to monitor the dynamic changes of Tiangang lake shorelines. Firstly, the accuracy of land use classification using maximum likelihood method, random forest method, and object-based image classification method was evaluated. Then, the transitional graphic of land use was calculated to analyze the dynamic changes in the shoreline of Tiangang Lake. Finally, the accuracy and spatiotemporal changes of photovoltaics were compared. [Results] The result showed that among the three selected land use classification method, object-based image classification method had the best classification accuracy, with an average overall classification accuracy and Kappa coefficient of 92.7% and 0.91, followed by random forest method and maximum likelihood method. From 2019 to 2023, the photovoltaic area of Tiangang lake shorelines in Jiangsu Province was rapidly increased, and the land use change rate of water bodies converted to photovoltaics was the highest. The average IOU of photovoltaic recognition using object-based image classification method was 91.4%, and Sentinel-2 images can accurately monitor the dynamic changes of photovoltaics in different periods. [Conclusion] The use of object-based image classification method can accurately identify and monitor the main targets and their spatiotemporal changes along the water shoreline, providing reference for dynamic supervision of rivers and lakes.

关键词

Sentinel-2 / 面向对象分类方法 / 河湖监测 / 光伏 / 分类后变化监测

Key words

Sentinel-2 / object-based image classification method / river and lake monitoring / photovoltaic / post classification change detection

基金资助

水利部重大科技项目(SKS-2022072)

江苏省水利科技项目(2023022)

江苏省水利科学研究院自主科研项目(2023Z034)

江苏省水利科学研究院自主科研项目(2024Z007)

国家自然科学基金(41801075)

江苏省高校大学生实践创新训练计划项目(202310323098Y)