基于GEE平台的国家公园土地覆盖变化遥感检测方法构建

Developing remote sensing based methods for land cover change detection in national parks from GEE platform:a case study from the Qianjiangyuan National Park pilot area

【目的】高效、可靠地提取土地覆盖动态变化信息对于国家公园保护与管理方案制定具有基础性数据支撑作用。本研究主要目的是构建经济高效的遥感监测方法,提取钱江源国家公园体制试点区2001—2017年的土地覆盖变化信息,评估国家公园保护与管理活动的有效性。【方法】首先定义研究区的土地覆盖分类体系(耕地、森林、草地、水体、人造地表和裸地),并利用增强型目视解译方法选取历史训练样本点;其次综合Landsat多季节复合数据的光谱波段、光谱指数、纹理特征和地形特征,基于Google Earth Engine(GEE)云平台执行随机森林分类算法,生成土地覆盖数据集;最后建立土地覆盖转换规则从而识别主要的土地覆盖变化类型及其空间分布模式。【结果】独立样本集验证结果表明,钱江源国家公园2001年、2009年和2017年土地覆盖总体分类精度分别为83.12%、81.82%和87.35%。造林活动、耕地废弃和开发建设的变化信息体现出保护管理工作取得显著成效。【结论】增强型目视解译能够高效、可靠地选取历史训练样本点,从而支持历史土地覆盖数据集构建。采用基于云计算的变化检测技术可快速评估国家公园管理措施的有效性并发现生态脆弱地带,具有高效、经济及计算资源限制性少等优势,可广泛推广应用。

【Objective】Efficient and reliable extraction of land cover change information plays an important role in formulating management and protection plans for national parks.The primary objective of the current work was to develop an efficient and economical remote sensing detection method to extract land cover change information in the Qianjiangyuan National Park pilot area from 2001 to 2017,and to assess the effectiveness of the adopted protection and management measures.【Method】A land cover classification scheme,including cropland,forest,grassland,water,artificial surface,and bare land,was first defined,followed by the selection of the historical training dataset using an enhanced visual interpretation method.Next,the original spectral bands,spectral indices,and textural features derived from Landsat multi-season composite data and terrain features were incorporated as input variables for implementing the random forest classification algorithm in the Google Earth Engine(GEE)cloud platform to generate land cover datasets.Finally,the major land cover conversion types and their spatial distributions were mapped by establishing land cover conversion rules.【Result】The validation results derived from an independent sample set indicated that the overall accuracies for the 2001,2009 and 2017 classifications were 83.12%,81.82%and 87.35%,respectively.Afforestation activities,cropland abandonment,and distributions of development and construction projects demonstrated the effectiveness of the protection and management measures adopted in this park.【Conclusion】The enhanced visual interpretation process helps to identify historical sample points efficiently and reliably to support the creation of historical land cover datasets.With cloud computing-based change detection technology,it is possible to quickly assess the protection and management efficacy and locate ecologically vulnerable zones.Moreover,the method has the advantages of high efficiency,economy,and fewer limitations on computing resources;thus,the GEE-based land cover change detection method proposed in the current work is suitable for an application in other similar scenarios.