基于决策树的Landsat TM/ETM+图像中太湖蓝藻水华信息提取

以9期Landsat TM/ETM+影像为数据源,基于K-T变换和归一化植被指数(NDVI),建立了湖泊蓝藻水华信息提取的决策树模型.基于大气顶面反射率图像,选用2005年10月17日太湖图像进行了对比验证,表明决策树模型比单波段阈值法、多波段阈值法(RVI、DVI、NDVI)能够更有效地提取蓝藻水华信息,区分陆生植被、水生植物和水华,省去了水体掩膜的过程.使用太湖2002年10月25日和2011年7月22日图像、巢湖2005年8月12日的图像,验证决策树模型方法和工作流程的有效性.使用多期TM图像确定了阈值的取值范围,其中,亮度、绿度、NDVI的下限值依次为0.191、-0.007、-0.054,湿度下限范围为0.07~0.15;亮度阈值上限范围为0.3~0.7、绿度为0.2~0.5、湿度为0.1~0.3,这些结果可作为湖泊蓝藻水华遥感监测的参考.

Decision tree model of extracting blue-green algal blooms information based on Landsat TM/ETM+imagery in Lake Taihu

A decision tree model (KTNI) based on the K-T transformation and Normalized Difference Vegetation Index (NDVI) was developed to extract the blue-green algal blooms information in the lakes using 9 Landsat TM/ETM+images. This paper compared the KTNI with the single-band threshold value method and the multi-band threshold value methods (RVI, DVI, NDVI) in accuracy of extracting the blue-green algal blooms information, and then validated the KTNI using two observed images of the Lake Taihu and one image of the Lake Chaohu. The results showed that the KTNI can mask the water body automatically and separate the blue-green algal blooms from the land vegetation and aquatic plants effectively. Model thresholds and their ranges were determined based on the TM images;the lower limits of the Brightness, Greenness and NDVI were 0.191, -0.007 and -0.054, respectively, and Wetness ranged from 0.07 to 0.15. For the upper limits, the Brightness was from 0.3 to 0.7, Greenness was from 0.2 to 0.5, and Wetness was from 0.1 to 0.3. The decision tree model and threshold values can be used as the reference for remote sensing monitoring on the blue-green algal blooms in lakes.