基于时序遥感数据轨迹分析的干旱/半干旱区土地覆被变化检测研究

Ecological environment in worldwide arid zones have been seriously damaged in the context of global climate change and increasing economic activities. It is extremely urgent to detect changes in arid ecological environment. Trajectory-based feature mining of time-series remote sensing datasets has been commonly used for land cover conversion (LCC) area detection. Due to the high temporal-spatial variability of land cover change in arid zones, existing methods not only lack ecological diagnosis but also hard to establish robust means that can discriminate “change signals” from high “noise” in arid land cover. This project aims to develop a novel time-series trajectory analysis method for LCC area detection in arid area, which integrates land cover classification method to define land cover type before and after change in LCC areas. Thus, an automatic framework that efficiently detects the location (change/no change), time, type (from-to) and process (change trajectory) of land cover conversion in large-scale arid area can be established. The research content includes: ①establish a key input variable system for the detection of LCC in arid zones; ②detect LCC area (change/no change) in time-series trajectories using the key variables through feature mining algorithms; ③define land cover change type (from-to) by designing land cover classification method that is suitable for large-scale arid area. This project not only has great theoretical significance for revealing the evolution process of ecological environment and human-land relationship in arid area, but also provides a technical framework for dynamic monitoring of land cover in typical arid areas, such as the core area of “Silk Road Economic Zone” in Central Asia and other typical arid areas around the world.

在全球气候变化背景下，经济活动日益频繁，世界范围内干旱区的生态环境均已遭到严重破坏，研究干旱区生态环境的变化十分迫切。基于时序遥感数据的轨迹特征挖掘技术能快速识别土地覆被类型转换(LCC)区域，但干旱区土地覆被有高时空变异性，现有方法在干旱区应用普遍缺乏生态意义诊断且不能有效克服高“噪声”对检测结果的影响。基于此，本项目拟研发适合干旱区LCC检测的时序轨迹特征挖掘方法，并结合土地覆被分类算法确定转换前后的类型，以此构建一套适用于大范围干旱区的，快速自动检测LCC位置、时间、类型、过程的方法体系。主要研究内容包括：①适用于干旱区LCC检测的输入变量体系构建；②基于多变量时序轨迹特征挖掘的LCC区域识别；③面向大范围干旱区的土地覆被分类算法识别转换类型。本项目对揭示干旱区生态环境及人地关系演化有重要理论意义，且能为地处“丝绸之路经济带”核心区的中亚乃至全球干旱区的土地覆被动态监测提供技术基础。

干旱区土地覆被变化具有高时空变异性，现有土地覆被变化检测算法在干旱区应用普遍缺乏生态意义诊断且不能有效克服高“噪声”对检测结果的影响。为解决这一问题，本项目拟研发适合干旱区的快速自动检测LCC位置、时间、类型、过程的方法体系。本项目首先构建了适用于干旱区LCC检测的输入变量体系构建，然后开发了基于多变量时序轨迹特征挖掘的LCC区域识别算法，最后基于改进的C5.0决策树算法实现了对变化前后土地覆被类型的识别。本项目构建的方法体系在中国北方内蒙古地区科尔沁实验区和乌兰察布实验区应用都有较好的表现，该方法体系的提出对揭示干旱区生态环境及人地关系演化有重要理论意义，同时为大范围准确监测干旱区土地覆被变化提供了有效手段。