VEGETATION INDEXES-BIOMASS MODELS FOR TYPICAL SEMI-ARID STEPPE—A CASE STUDY FOR XILINHOT IN NORTHERN CHINA

Abstract

Aims There is a crucial need in grassland study for a vegetation index (VI)-biomass model simulating steppe biomass based on remote sensing. Methods Thematic mapper (TM) images (spatial resolution of 30 m× 30 m) for the research area in 2005 and 1991 were rectified so that geometric errors were less than one pixel, then extracted the image of the research region in the soft of ERDAS. We used five vegetation indexes: RVI (ratio vegetation index), NDVI (normalized difference vegetation index), SAVI (soil-adjusted vegetation index), MASVI (modified soil-adjusted vegetation index) and RSR (reduced simple ratio index). They were correlated to plant biomass sampled on the ground at the same time as the TM images. We developed four kinds of regression models: linear, logarithm, second-degree polynomial and cubic polynomial. Important findings The correlations between sampled biomass and the five VIs were highly significant, with four (NDVI, SAVI, MSAVI, RSR) being positive and one (RVI) negative. Multiple correlation coefficients (R2) of the 15 regression models were 0.6, indicating that a VI-biomass regression model was a simple method to monitor the biomass of steppe grassland. The R2 of the NDVI-biomass model was the highest, indicating that it was better suited to simulate the biomass of typical steppe than the other VIs. For TM image, all four kinds of models were significant at the 0.01 level, with the cubic polynomial model as the best to simulate the biomass, followed by the second-degree polynomial, linear and logarithm models. Therefore, the cubic polynomial regression model based on NDVI-biomass was the best model, and was used to simulate the biomass of the research region. Simulated biomass was higher in the east than in the west of the research region and higher in the southeast than in the northwest. Simulated biomass was consistent with sampled biomass in 2005.