伊犁新垦区土壤盐碱化遥感信息的提取

土壤盐碱化是土地资源利用的主要障碍因子之一，利用遥感数据快速准确地提取盐碱地信息及其空间分布，可以为土地资源开发利用提供重要的科学依据，有着重要的现实意义。本文以新疆伊犁新垦区为研究区域，利用专家知识建立决策树，使用2008年8月的ETM数据，结合基础地理信息数据，提取不同程度的盐碱地信息。研究表明，利用基于知识的决策树方法，结合多源数据，能够有效提取盐碱地信息，并且分类结果准确，精度达到89.3% 。新垦区盐碱地占全区面积的10%，主要分布在伊犁河南岸大灌区，各级盐碱地比例为：重盐碱地占1%、中盐碱地占17%、轻盐碱地占82%。

Extraction of Soil Salinization from Remote Sensing Information Based on Knowledge Discovery over the Arid Area

Soil salinization, a manifestation of land desertification and degradation, is one of major constraints on land resource utilization and a primary restriction for local agricultural development. It is also an import factor affecting stability of local ecological environment. Saline land information and its distributions derived from remotely sensed information can therefore provide important and meaningful reference for land resource exploration and development. In the present work, we investigated soil salinization distributions from remotely sensed data over Yili nearly reclaimed area in Xinjiang Uygur Autonomous Region, a typical arid area of Northwest China. Given that the accuracy of soil salinization information extracted from satellite imagery is generally lower than that of classification, the decision tree method with the aid of expert knowledge in conjunction with Landsat ETM+ imagery and basic geographic information data was utilized to extract saline land information of different degrees. The decision tree method is to employ a multistage or sequential hierarchical decision model. The basic rationale involved in any multistage approach is to break up a complex decision into a combination of several simpler decisions. After fully comparing all possible combinations of several spectral indices and components, the authors selected four characteristic variables, i.e., normalized difference vegetation index (NDVI), brightness value of K-T transform, the third band of principal component transform, DEM and ground water depth. The authors explored and realized automatic extraction of soil salinization information. It was found that NDVI shows the potential to differentiate vegetation from background, DEM can provide geomorphic conditions, and ground water is closely related to the soil salinization. Results demonstrated that saline land information can be successfully extracted using the decision tree method with the aid of expert knowledge and a variety of data sources. Based on field sampling and error matrix analysis, results indicated that the accuracy of soil salinization information is justified, showing overall classification accuracy of 89.3% and Kappa value of 0.87. Saline land area accounted for 10 percent of the entire study area, primarily distributed within the irrigated area in the south of the Yili River. The percentage of different degree saline land is strong salination 1%, moderate salination 17.1%, and light salination 82%. This study would offer reference for management and recovery of soil salinization and regional agricultural sustainable development, also providing reference for future research on land salination over arid areas.