由全国1：10万比例尺资源环境遥感数据库生成的1km格网土 利用／土地覆盖成分数据，分析了依据面积最大原则生成1km单一性值栅格数据与原始成分类型数据之间的误差分布，研究了按照面积最大和中心属性值法从1km栅格数据粗化为4，8和10km栅格数据过程中的误差分布特征，并探讨了地貌特征对这种尺度效应的影响，结果显示，1km单一生栅格数据压缩的城镇及居民点的实际面积（减少多于60％），而增加了林地和旱地的面积，对于单元栅格而言，各类的面积误差（成分面积与尺度转换后保留的单一类型面积之间的差值）与相应尺度栅格范围内的土 地利用复杂程度是对应的，由最大面积法的数据尺度转换引起的误差是随着尺度的增大而增加，而中心属性值法转换过程的误差呈现近似随机分布，应用面积最大法转换空间数据尺度中的误差分布受地貌类型的影响是显著的，水田与旱地面积在数据粗化过程中产生的误差在山地和丘陵区明显高于在其地貌类型上的误差，中心属性值法数据化受地貌类型的影响并不显著。

Based on the national large-scale land-use/land-cover vector data（1 : 100000），we produced 1 km resolution grid cell data set in which each type of land-use/land-cover is represented as area percentage. From the 1 km grid cell data set，1 km grid data with one unique attribute in each grid is built and then a series of grid data sets with 4 km， 8 km and 10 km spatial resolution are derived using maximum area attribution decision law and the centric attribute decision law separately. As to the derived data sets with different scale and scaling up method，we analyzed the distribution of error in area of each type of land-use/land-cover type. In addition，we explored the relationship between the error in area and physiognomy types. The result shows that：1）1 km grid data decrease the real residential area and constructive land（more than 60% of the total real area is lost），while the forestry land and dry farming land are increased；2）the difference between real area and the computed area from grid data is compatible with the land-use / land-cover complicacy； 3）the errors from scaling up data with maximum area law increase when scale increases，while the errors with centric attribute law randomly distributes when scale changes；4）physiognomy types can impact on the errors distribution significantly when using the maximum attribution law in scaling up. For instance，the area of paddy land and dry farming land in mountain area and hill area will decrease obviously when it is scaled up，which differs significantly from that in other area with other types of physiognomy. However，physiognomy has little effect on the error distribution with centric attribution law when the grid data is scaled up.