Improvements to ASTER-Derived Fractional Estimates of Bare Ground in a Savanna Rangeland

Abstract

Accurate estimates of ground cover and its inverse, bare ground (BG), derived from satellite imagery are required for monitoring rangeland-health indicators over large areas. This paper shows how accurate estimates of BG were obtained from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) image data over a savanna rangeland in north-east Australia. The normalized difference vegetation index and the lignin and cellulose absorption index were used to extract spectral-reflectance signatures of three scene endmembers from the image data: BG, nonphotosynthetic vegetation, and photo- synthetic vegetation. The endmember signatures were used with the Monte Carlo spectral mixture analysis (MCSMA) algorithm to derive image estimates of BG that were compared with field measurements. The results showed that the accuracies of the BG estimates were improved, compared to those obtained in a previous ASTER study that used only two endmembers in the unmixing procedure (root mean square error (RMSE) was improved from >0.1 to ~0.05). The results are an improvement on previous work that used Landsat and IKONOS satellite- multispectral imagery, compared favorably with estimates derived from airborne hyperspectral imagery, and can be used with existing rangeland monitoring methods. We conclude that the end- member extraction method is simple and widely applicable and can be used with MCSMA to obtain accurate estimates of BG from ASTER imagery. However, the use of this approach for estimating BG from satellite imagery depends on the future development of satellite-hyperspectral or ASTER-like sensors.