基于Kappa系数的景观变化测度——以辽宁省中部城市群为例

分析了辽宁省中部城市群1988～1998年间的景观变化,结果表明沼泽地从景观中完全消失,沙地的变化弹性最大,耕地的变化最小。优势景观类型(耕地)的转化方向较多,非优势景观类型(沙地)转化的方向少。某些景观类型之间较稳定地相互转化(林地和耕地) ,但某些景观类型之间转化极不稳定(耕地和沼泽地)。总体上,景观的变化趋势是:居住用地面积增加,占居了周围的耕地,促进了耕地向菜地的转化。面积大的景观类型变化小(林地和耕地) ,相反,面积小的类型变化大。Kappa系数分析结果表明,位置、数量、随机和标准Kappa系数都降低,即景观在数量上丢失3.86 %的信息,位置上丢失5 .95 %的信息,丢失的综合信息为6 .89%。而且建议进行综合信息评价时用随机Kappa系数,因为它既不考虑数量,又不考虑位置的影响。虽然Kappa系数从位置、数量和综合信息方面揭示了景观变化,但是这些分析是单方面的,而且没有说明在整个图的一致率中由于空间位置、数量和偶然因子所引起的一致率和变化率。景观变化过程中,不考虑空间位置和数量的情况下,景观在空间上随机分布,某个象元属于某个类型的概率等于1/ J(J=类型总数) (8% )。随景观中类型数量的减少,偶然一致率所占的比重增加,因此建议对单个类型不要进行一致性分析。位置一致率在整个一致率中

Measuring spatial information changes using Kappa coefficients: a case study of the city groups in central Liaoning Province

To analyze the landscape change of city groups in central Liaoning Province, we classified the remote sensing images of 1988 and 1998 into 12 cover types., The transformation matrix between each two stages was created in GIS (Geographical Information System). Marsh disappeared from the landscape and reed fields showed a declining trend (only remained 6.13%) in this period and could disappear in future. The most change was seen from sandy land which increased by 397.39% and the least change was seen for the cultivated land that decreased by only 2.85%. The result showed that dominant types could change to many classes, but the minor cover types could only transform to a few classes. For example, cultivated land transformed into 10 classes within this ten years. The transformation rates were stable between some cover types, but were not stable between the other cover types. However, the rate and classes for a particular class to transform to were determined by the policy and landscape pattern. The transformation matrix shows that the urban and rural area increased by converting from the cultivated land, while the vegetable land increased with the increasing population. The transformation rate and dynamic shows that the more dominant the classes were in the landscape, the less they would transform into other classes. The indices of Kappa coefficient family decreased within this period. Kappa coefficient for quantity decreased to 96.14%, which means only 96.14% information on quantity remained and 3.86% information lost in this period. Kappa coefficient for location decreased to 94.05%, which means the landscape retained only 94.05% information on location and lost 5.95% location information in these changes. If the purpose of the study is to examine the information loss on quantity, Kappa coefficient for quantity should be used, whereas if the purpose of study is to examine the information loss on location, Kappa coefficient for location should be used. Standard Kappa coefficient and Kappa for no ability to maintain quantity and no ability to specify location decreased to 90.48% and 93.11%, respectively. In comparison, the Kappa for no ability to maintain quantity and no ability to specify location is better than the standard Kappa coefficient, because it assumes that the landscape has no ability to keep quantity and location. The overall proportion agreement is 93% between two maps and this result means there was no significant change in the landscape. The agreement due to quantity was 25% and the transformation rate due to quantity was 3%. And also the agreement due to location was 60% and transformation rate due to location was 4% in this period. The agreement due to chance was 8%, because the whole class number was 12. Therefore, the overall proportion agreement is determined by the agreement due to location.