High-resolution land cover classification for Stordalen, Abisko region, northern Sweden

An object-based approach was used to generate a detailed land cover classification covering the Stordalen area near Abisko, northern Sweden. First, an orthophoto was combined with the DEM resampled to 1 m spatial resolution. A segmentation layer was generated by grouping pixels into homogeneous areas with a minimum region size of 130 m². From this a water mask was classified using the red band of the orthophoto and a slope layer. A land cover training set was created by combining field survey information with visual interpretation of the orthophoto and topography. The following layers were used as input for the classification algorithm: an orthophoto, elevation and slope; a SPOT5 4-band satellite image, and NDVI, SAVI and NIR/SWIR as SPOT5 derivatives. The segments were then classified using a support vector machine algorithm. Artificial surfaces were hand digitized and masked out. The individual thematic classes are described in TableA1.
The quality of the classification was assessed using a set of 108 ground control points. The accuracy assessment results in a Kappa value of 0.71 and an overall accuracy of 74% for all land covers excluding water and artificial areas see TableA2.

本研究采用面向对象方法，对瑞典北部阿比斯科（Abisko）附近的斯托达伦（Stordalen）区域开展精细化土地覆盖分类制图。首先，将正射影像（orthophoto）与重采样至1米空间分辨率的数字高程模型（DEM）进行融合；通过将像素聚类为同质性区域（最小区域面积为130平方米）生成分割图层。基于上述结果，利用正射影像的红色波段与坡度图层完成水体掩膜的分类。通过整合野外调查数据、正射影像目视解译结果与地形信息，构建土地覆盖训练数据集。分类算法的输入图层包括：正射影像、高程与坡度图层；SPOT5四波段卫星影像，以及SPOT5衍生的归一化差分植被指数（NDVI）、土壤调整植被指数（SAVI）与近红外/短波红外（NIR/SWIR）指标。随后采用支持向量机（Support Vector Machine）算法对分割单元进行分类。人工地表通过手动数字化方式提取并予以掩膜剔除。各专题分类类别详见表A1。
本研究采用108个地面控制点对分类结果开展质量评估。精度评估结果显示：排除水体与人工地表的所有土地覆盖类型的Kappa系数为0.71，总体精度达74%，具体详见表A2。