Quantitative Representation of Spatial Heterogeneity in the LAI Scaling Transfer Process

Current remote sensing scaling research highlights the ability to quantitatively represent spatial heterogeneity which has become the core issue of many studies. However, the influence of the difference in the land cover types of subpixels is seldom considered. Therefore, in the course of quantitatively studying the spatial heterogeneity induced by the leaf area index (LAI) scaling transfer process, this paper uses the contexture of the surface parameter to express the subpixel differences of each pixel and constructs a spatial heterogeneity coefficient (C-sh) by means of information entropy, class variance and class contribution degree. Numerical simulation and analysis of the spatial heterogeneity coefficient changes with different combinations of 3 land cover types (water, building/soil, and vegetation) are carried out, and the trends are summarized. On this basis, with the help of measured data in the study area, a pixel size of 30 m is defined; then, pixels of size 8 x 8 and 16 x 16 are upscaled to test the correlation between the scaling error (e) and C-sh. The validity of C-sh is verified. The results show that both parameters achieve a high linear positive correlation at different scales, and the determination coefficient reaches 0.923 and 0.984, respectively. A regression relationship is then established between them to correct the scaling error, and as a result, an excellent correction is achieved. The RMSE decreases from 0.3329 before correction to 0.1457 after correction of e and from 0.2887 before correction to 0.0766 after correction of C-sh. This research has a certain reference importance for remote sensing scaling transfer studies and quantitative representations of spatial heterogeneity.

Automated summary

This study focuses on the quantitative representation of spatial heterogeneity in remote sensing scaling research and introduces a spatial heterogeneity coefficient (C-sh) to correct scaling errors, which is validated to be effective. Results show a high linear positive correlation between parameters at different scales, achieving excellent correction of errors.