基于SI-Albedo特征空间的干旱区盐渍化土壤信息提取研究——以克里雅河流域绿洲为例

选取塔里木南缘克里雅河流域绿洲为研究靶区,利用Landsat-ETM+卫星图像数据和野外调查数据分析盐渍化土壤与地表反照率(Albedo)、土壤盐分指数(SI)之间的关系。回归分析发现,盐渍化土壤在SI-Albedo特征空间分布具有显著规律,即非盐渍化土壤呈团状分布;轻、中度盐渍化土壤具有线性分布特征;非盐渍化土壤与轻度盐渍化土壤分异明显。结合分异规律,编制分类算法模型,得到研究区盐渍化土壤信息提取结果,并与传统监督最大似然分类法结果进行对比分析。结果表明,在SI-Albedo特征空间中定量快速提取盐渍化土壤信息的总体效果较好,对准确且自动提取干旱区盐渍化土壤信息以及区域尺度盐渍化遥感监测研究具有重要意义。

SI-ALbedo space based extraction of salinization information in arid area

Soil salinization is getting more and more attention the world over for its adverse impact on the social economy, the environment, and the agricultural eco-system. The total area of salinized soil in Xinjiang reaches 8.476×10⁶hm², accounting for 31.1% of the total cultivated land. It is, therefore, necessary and important to study soil salinization in arid regions for solution to this problem. Remote sensing (RS) technology demonstrates a number of advantages in this field. But how to extract salinization information accurately from RS images is the basis of the study. In this paper a case study of Yutian County monitoring soil salinization by means of remote sensing, is carried out. Yutian County was selected for this study because of its importance as a significant site for agricultural development. Located in the south of the Keriya oasis, it has recently been exposed to severe soil salinization. Seven-spectrum-band Enhanced Thematic Mapper-plus (ETM+) images dated October 7, 2002 were used against the data of soil features obtained from field investigation and analysis of typical soil information, to extract Salinization Index (SI) and land surface albedo, which are very important biophysical parameters of land surface. In this paper the relationship between salinization index (SI) and albedo was analyzed quantitatively. Through experiment and theoretical reasoning, the authors proposed a conception of SI-Albedo space and discussed its biophysical characteristics. Analysis revealed that location could be used to improve the current strategies for salinization in the SI-Albedo space, and hence the strategies for salinization mapping, by defining measurements in this feature space. Therefore, the authors present a methodology to monitor severity of salinization. Field data, available data in the literature, and ancillary data were linked with land cover characteristics (salinization index, land surface albedo) derived from Landsat ETM+ multispectral images. An information extraction model, using the decision tree classification method, was established and applied to classification of RS images. Results indicate that the classification based on SI-Albedo space has a higher classification accuracy than the one based on maximum likelihood. Its highest overall accuracy is about 0.92% higher than the maximum likelihood. Although both techniques show some mix-class phenomena in the classification result, but the classification based on SI-Albedo space has less than the maximum likelihood, and thus a higher separability. Salinity soil distribution maps show that the soil salinization of this study area is relatively severe and varying in degree and type; The area is dominated with light salinization and moderate salinization. The former is distributed mainly in farmland, while the latter around the Bostan swamp. And based on the salinized soil map, the salinity soil early-warning line was derived for anticipating further soil degradation. Such contrasting and complementary behavior suggests a potential synergism between salinization index and land surface albedo for mapping and monitoring of a complex soil salinization environment such as Keriya oasis.