时间序列植被覆盖度断点检测方法研究

断点检测技术能够提取时间序列数据中发生变化的信息,对时序变化特征的研究和分析具有重要意义。在植被变化研究中,它有助于发掘时序植被参数数据中潜在的连续变化信息（如火烧、砍伐和病虫害等）,具有较广阔的应用前景。本文以2000-2015年空间分辨率为250 m的福建省长汀县MODIS NDVI遥感时序产品计算的植被覆盖度（Vegetation Fractional Coverage, VFC）为数据源,利用DBEST（Detecting Breakpoints and Estimating Segments in Trend）模型开展植被变化断点检测与分析,并讨论了模型参数对实验结果的影响。经研究发现,使用DBEST推荐的第一、第二水平变化阈值（即θ₁、θ₂分别为0.1和0.2）可较好地界定植被覆盖度的变化级别;变化持续时长φ可根据所用时序数据的类型和当地植被变化特点进行调整,表示断点变化级别的阈值β也可根据研究目标自定义;本文φ取3,β取0.2,实验得到断点位置和断点类型的精确度分别为92%和80%,表明DBEST模型能够很好地提取时间序列VFC数据中的重要变化信息,与当地实际情况比较吻合。

The Breakpoints Detection Method Using Time Series of Vegetation Fractional Coverage

Detecting breakpoints plays an important role in plotting and analyzing time series of the changing characteristics such as firing, logging, diseases and insect pests in vegetation. It is a useful technique of extracting the significant information in time series data. We focused on the method of Detecting Breakpoints and Estimating Segments in Trend (DBEST). We studied the detection of vegetation breakpoints by using vegetation fractional coverage (VFC) data which is derived from MODIS NDVI remote sensing images ( 250 m) from 2000 to 2015 in Changting County of Fujian Province. In order to determine if the results of breakpoints detection are reasonable, the primary experiment is to test the applicability of DBEST method by using the VFC data of various changing types in time series. We select several samples of time series data which covered the key water and soil erosion conservation area. The vegetation changes more frequently in this area for conducting the break-points detection experiments. We make an accuracy evaluation of changing time and changing types by using the temporal trajectories and Landsat remote sensing images of every point. We find that DBEST is suitable for VFC time series data of Changting, by using the default first and second level-shift-thresholds (θ₁ = 0.1, θ₂ = 0.2) which indicated that DBEST could define the changing level of VFC, but the duration-thresholdφ should be adjusted according to the study area and the type of time series data (we setφ=3). Those parameters have weak influences on the accuracy of breakpoints positions, but have more effects on the changing types of breakpoints. On the whole, the excessive intervention is not necessary for detecting vegetation in DBEST. However, through a lot of experiments we believe that the threshold of the changing magnitude can be modified by our own need to gain a satisfying results. Finally, we set β = 0.2 to fit our own research targets. The precision of the changing time is 92%, greater than the changing types (80%), indicating that DBEST method works well in extracting the important changing information for VFC time series. Meanwhile, the experimental results are broadly consistent with the varying conditions of the local vegetation.