当前位置: X-MOL 学术Earth Sci. Inform. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Land cover pattern simulation using an eigenvector spatial filtering method in Hubei Province
Earth Science Informatics ( IF 2.8 ) Pub Date : 2020-07-02 , DOI: 10.1007/s12145-020-00483-4
Jiaxin Yang , Yumin Chen , John P. Wilson , Huangyuan Tan , Jiping Cao , Zhiqiang Xu

This paper proposes an eigenvector spatial filtering-based (ESF-based) regression model for land cover pattern simulation in China’s Hubei province. The significance and influence of biophysical, climatic, and socio-economic factors have been detected and analyzed in the study region. The ESF-based multinomial logistic regression (spatial model) is constructed for discrete choices to take spatial autocorrelation into consideration. For the massive raster pixels, a segmentation processing (grid-based partition) approach is employed to resolve the large datasets to smaller ones to improve calculation efficiency. Both 32 × 32 and 64 × 64 cell sizes are used to compare the differences and influence of these approaches. For the 32 × 32 cell size, the hitting ratio increased from 0.70 to 0.89 and the deviance decreased 65.6%. For the 64 × 64 cell size, the hitting ratio increased from 0.68 to 0.77 and the deviance decreased 33.2%. The fitted results and maps show that spatial autocorrelation (SA) plays an important role in land cover patterns. Besides, the ESF-based spatial model can isolate SA in land cover pattern simulation, and therefore can improve the fitting accuracy and decrease the model uncertainty. The experiment shows that ESF-based multinomial logistic regression method provides a promising approach for discrete choice regression for massive raster datasets.

更新日期:2020-07-02
down
wechat
bug