Abstract:Scaling of spatial data usually leads to varying degrees of information loss,so it needs to analyze the scale effects.Considering the characteristics of land cover,which including nonlinear,random and spatial heterogeneity,a scaling method of land cover based on discriminant space models is proposed.Discriminant space-based models,which can accommodate generalized linearity of response with explanatory discriminant variables and present the mechanism of mapping from discriminant variables to geographical categories,were introduced to provide a framework to study the statistics and mechanism model of land cover information.Scaling of discriminant variables was firstly carried out using block Kriging method from geostatistics,and then the scaled discriminant variables were input to a classification model to obtain the land cover information of corresponding scales.Overall accuracy,average mutual information and relative error were selected to evaluate the scale effects of land cover.A SPOT5image acquired in August 2008with four spectral bands and 10mpixel size in Zhangye City was used for experiment.A subset image with the area of 225km2 was also generated from the original image as study area.Maximum likelihood classifier was used as the classification model.Comparing with the conventional scaling method on the basis of majority theory,the overall accuracy and the average mutual information of the proposed method increased with the increase of scale,which showed that the proposed method kept a high consistency of land cover information.For the single land cover type,the proposed method is effective to reduce the information loss that may be given rise to in the scaling process.

Scale Effects Analysis of Land Cover Based on Discriminant Space Models