扎龙湿地土地利用信息提取方法研究

Method on Land Use Information Extraction of Zhalong Wetlands

利用2015年4~11月8景C波段Rardarsat-2影像SGX产品数据和2015年9月10日Landsat 7的7个波段的反射率数据,采用面向对象分类方法,获得8个日期的扎龙湿地淹水区,计算出淹水频次;选取的19个特征变量分别为Landsat 7影像的1波段~7波段的反射率、归一化植被指数、二阶距、对比度、相关性、差异性、能量、同质性、中值、HH和HV极化波段的后向散射系数、淹水范围和淹水频率,采用随机森林算法,提取扎龙湿地的土地利用类型信息。研究结果表明,有淹水频率参与的分类精度为91.73%,无淹水频率参与的分类精度为76.49%,精度提高主要体现在沼泽湿地与草地的分类上。本研究为准确地更新湿地基础信息提供了方法示范。

Hydrological condition changes have seriously affected the structure and functions of the wetland ecosystem, and endangered the regional ecological security, water security sustainable economic and social development. At the same time, the degradation of the wetlands is increasing in a large extent due to lack of wetland management experiences in developing countries. In order to protect wetland resources, reduce the destruction of wetlands, and formulate the scientific management measures, it is important to recognize the hydrological regime of the wetlands accurately. Monitoring hydrological regime based on remote sensing imagery can not only provide data for establishment of regional hydrological model, but also improve the classification accuracy of wetlands. Considered traditional optical image cannot distinguish the inundation of the vegetation in the wetlands, and the radar images can penetrate the vegetation through the microwave, and then detect the flooding situation under the vegetation. Eight time series Radarsat-2 images were classified to obtain the result of flood inundation area based on the object-oriented classification in this study. The inundation times were calculated by the results of flood inundation area. Two parameters of hydrological regimes（flood inundation area and inundation time） were selected as the wetland mapping predictive variables. The land use cover in the Zhalong wetlands was classified by random forest classification algorithm using 19 predictive variables including reflectivity of Landsat 1 band to 7 band; 7 texture features（angular second moment, contrast, correlation, dissimilarity, energy, homogeneity, mean）, normalized difference vegetation index（NDVI）;HH and HV polarization backscatter coefficient, flood inundation area and inundation time, and the influence of 2 hydrological regimes indicators on the classification accuracy was analyzed. The main objective of this study was to use the time series radarsat-2 images to extract the hydrological regime and improve the accuracy of classification with the new predictive variables. The results indicated that the backscattering coefficient of the flood inundation area could be distinguished with non-inundation area significantly though comparing the mean and standard deviation of backscattering coefficient values derived from Radarsat-2 SAR images; the 8 flooded inundation area were classified by the object-oriented classification, the average accuracy was more than 80%, the highest accuracy and kappa coefficient were 98.24% and 0.97 respectively on 10 September2015, while the lowest accuracy and kappa coefficient were 82.86% and 0.48 respectively on 29 April; the inundation time was obtained by overlaying the 8 inundation maps, the area in which the number of inundation times was larger, concentrated on the core area in Zhalong wetlands; the results of the wetlands＇ mapping showed that the importance score of normalized difference vegetation index was the highest, followed by that of inundation time, the flood inundation area, HH and HV backscatter coefficients respectively. Flood inundation time improved accuracy of classification of the wetlands from 76.49% to 91.37%. The commission error mainly came from the misclassification of grassland, while the omission error came from the unrecognizable wetlands.