基于云平台和BFAST算法的地表变化检测方法

准确识别地表变化的时空信息,有助于探究地表自然环境和生态系统发展演变的规律,支撑相关的科研与行政管理工作。本文以河南某生态保护修复工程部分实施范围为研究区域,基于Google Earth Engine(GEE)云平台,以2013—2020年的98景Landsat8/OLI遥感影像作为数据源,应用Breaks for additive season and trend(BFAST)算法对地表变化进行了信息提取和制图。首先基于GEE云平台对Landsat8/OLI地表反射率数据集进行调用和预处理,基于CFMask算法对遥感数据集进行云影掩膜,开展光谱指数(植被指数NDVI)的计算以及时间序列数据集的构建。其次基于时序数据集与BFAST算法构建由趋势项、季节项和残差项组成的广义线性回归模型,通过最小二乘法求解模型中的未知参数集,以此进一步构建时序拟合模型,而后基于残差的Moving sums(MOSUM)方法对时序结构变化进行检测。最后从检测结果中抽取像元样点,通过与Google Earth高分辨率影像数据叠置和目视解译,开展结果验证和精度评价。结果表明,本文提出的方法在研究区的时序地表变化...

Land Surface Change Detection Method Based on Cloud Platform and BFAST Algorithm

Accurately identifying the spatio-temporal information of surface changes will help to explore the law of development and evolution of surface natural environment and ecosystems, and support related scientific research and administrative management. Taking part of the implementation area of an ecological protection and restoration project in Henan Province as the study area, based on the Google Earth Engine (GEE) cloud platform, using 98 view Landsat8/OLI remote sensing images from 2013 to 2020 as the data source, and the Breaks for additive season and trend (BFAST) algorithm theory was applied to extract and map information on land surface changes. The methodological experimental process included: firstly, the Landsat8/OLI land surface reflectance dataset was called and pre-processed based on GEE, the cloud shadow masking of the remote sensing dataset based on CFMask algorithm, the calculation of the spectral index (vegetation index NDVI) and the construction of the time series dataset. Secondly, based on the time series data set and the BFAST algorithm theory, a generalized linear regression model consisted of trend terms, seasonal terms and residual terms was constructed, and the unknown parameter set in the model was solved by the least square method, so as to further construct a time series fitting model and detect time-series structure changes in near real-time based on the Moving sums of the residuals (MOSUM) method. Finally, image element sample points were extracted from the detection results, and the results were validated and evaluated in terms of accuracy by overlaying with Google Earth high-resolution image data and visual interpretation. The analysis of the results showed that the method proposed had high detection accuracy in the detection of time-series land surface ecological changes in the study area (83.7% overall accuracy, 86.5%, 80.7% and 87.7% accuracy of the detection results in the sub-years from 2018 to 2020, respectively) in the detection of time-series land surface changes in the study area. Overall, the method proposed was a basic method for remote sensing big database construction, near real-time disturbance identification and monitoring of land surface ecological information and other technologies, which can provide technical support and decision-making support for the investigation and monitoring of ecological protection and restoration in national land space and assessment and early warning.