The comparison map profile method:: A strategy for multiscale comparison of quantitative and qualitative images
The comparison map profile method:: A strategy for multiscale comparison of quantitative and qualitative images
复制标题
DOI：
10.1109/tgrs.2008.919379
复制
发表时间：
2008-09-01
影响因子：
8.2
通讯作者：
Hely, Christelle
中科院分区：
文献类型：
--
作者：
Gaucherel, Cedric;Alleaume, Samuel;Hely, Christelle
比较地图剖面（CMP）方法比较两个空间明确的数据集（原始图像）在每个点，并通过几个空间尺度同时。CMP将移动窗口概念与定量或定性数据的相似性指数相结合，以可视化和量化输出：在剖面上报告平均相似性值的变化及其随尺度的变异性，在单尺度地图上估计区域之间的相似性，并在平均多尺度地图上评估其持续性。CMP方法首先使用在方格图案中具有轻微差异的两个图像来说明。第二，使用CMP方法对非洲植被进行了两组比较。第一组涉及叶面积指数（LAI）的定量数据：将AVHRR-NVDI产品提取的遥感LAI图像与动态全球植被模型（DGVM）输出的模拟LAI进行比较，分别使用距离和互相关系数进行数值和结构模式的定量比较。第二组图像处理的定性数据：遥感产品的土地覆盖类型的IGBP-NIODIS进行比较DGVM分类叶面积指数输出到土地覆盖类型，使用Kappa统计相似性指数。结果表明，考虑到空间模式使用CMP方法减少了50%的平均相关性，并增加了50%的距离相比，全球像素到像素的索引。同样，土地覆盖图的比较成本仅为全球Kappa值的35%。从森林到草地的植被赤道梯度是两种类型数据集之间最持久的相似区域。潜在的限制和CMP方法的优势进行了讨论。
The comparison map profile (CMP) method compares two spatially explicit data sets (original images) at each point and through several spatial scales simultaneously. The CMP combines the moving window concept with similarity indices for quantitative or qualitative data to visualize and quantify outputs: Changes in mean similarity value and its variability through scales are reported on a profile, similarities between regions are estimated on monoscale maps, and their persistence through scales assessed on a mean multiscale map. The CMP method is first illustrated using two images with slight difference in the checkered pattern. Second, two sets of comparisons related to African vegetation are conducted using the CMP method. The first set deals with quantitative data of leaf area index (LAI): Remote-sensed LAI images extracted from the AVHRR-NVDI product are compared to simulated LAI output from a dynamic global vegetation model DGVM) using the distance and the cross-correlational coefficient for quantitative comparison of values and structure patterns, respectively,. The second set of images deals with qualitative data: the remote-sensed product of land cover type by IGBP-NIODIS is compared to the DGVM classified LAI output into land cover types using the Kappa statistics as similarity index. Results show that taking spatial patterns into account using the CMP method decreases the mean correlation by 50%, and increases the distance by 50% as compared to the global pixel-to-pixel indices. Similarly, comparison of land cover maps costs only 35% of the global Kappa value. Equatorial gradients of vegetation from forests to grassland are the most persistent similar regions between both types of data sets. Potential limits and strengths of the CMP method are discussed.