遥感影像软分类在区域水土流失评估中的应用

Application of Remote Sensing Image Soft Classification to Regional Soil Erosion Susceptibility Assessment

以深圳市宝安区为例 ,通过最大似然法对遥感影像进行光谱软分类 ,得到了每个像元分属于不同土地利用 /覆盖类型的似然度 ;基于土地利用 /覆盖和水土流失的内在关系 ,利用层次分析法确定了每种土地利用 /覆盖类型影响水土流失的权重 ,在此基础上 ,使用土地利用特征指标来定量反映每个像元对于水土流失的综合影响 .结果表明 ,这种方法具有有效和快速的特点 ,不仅能够反映水土流失潜在性的相对强弱 ,而且可在像元层次上反映这种强弱的空间分布特征 ,在地形复杂、土地利用 /覆盖类型分布破碎的区域具有优越性 ,其结果可为水土流失的治理和详细调查选点提供依据 .

The land use/cover structure has causal-effect relationships to soil erosion, it therefore can be used to access the susceptibility of regions to erosion by water. The Analytical Hierarchy Process (AHP) is used to obtain the weight of each land use/cover category to soil loss. And the soil loss potentiality can be calculated by a characteristic index that is defined as the sum of products between the weigh of each category and the area percentage of each category for the concerned region. As remotely sensed image classification is more and more used to extract the land use/cover information, the characteristic index above mentioned has two shortcomings. Due to the presence of mixed pixels in images, each pixel can only be labelled as the dominating category in the land use/cover classification result image, which may lead to un-accurate if not false indexes especially in the rugged areas. On the other hand, the index cannot be obtained for every pixel, thus its spatial distribution information is limited. To overcome the two disadvantages, this paper applied a soft classification to remote sensing images, and possibilities of each pixel to every land use/cover category are obtained. And an adjusted index is set forth which uses the possibilities instead of the area percentage. Hence the index can be calculated for each pixel in the study region and its spatial distribution can be revealed. The adjusted index is applied to Bao′an, a rugged district in Shenzhen city. The statistical results of the whole district and its 9 towns reveal that the adjusted indexes are nearly the same to the original indexes in term of mean but they have smaller standard deviations and different spatial distributions.