A multi-scale classification method for rocky desertification mapping in the red-bed area of northwestern, Jiangxi, China

Currently, the monitoring of rocky desertification(RD) is concentrated in the karst area, whereas the study in red-bed areas is rare. In this paper, we present a multi-scale classification framework for RD monitoring based on spectral-spatial features. At the pixel scale, we explored several spectral indices based on spectral statistics and separability analysis of land cover samples. The homogeneous land covers were classified by the decision rules from the selected spectral indices (such as NDIOI, NRRI and NDGI); At the patch scale, RD classes were further distinguished by spatial decision rules based on multiple neighborhood features including proximity, linear density, and buffer distance. The method was applied on an OLI image over the red-bed area of northwestern Jiangxi, south central China, and validated using ground-based observations. The experimental results of verification and comparison are satisfactory. This work demonstrates a methodological supplement to the monitoring of red bed RD.

Automated summary

A multi-scale classification framework based on spectral-spatial features was proposed in this paper for monitoring red bed rocky desertification. Spectral indices and spatial features were used at pixel and patch scales, respectively, and validated using an OLI image in northwestern Jiangxi. The experimental results were satisfactory, providing a methodological supplement for monitoring red bed rocky desertification.