Topography-based modeling to estimate percent vegetation cover in semi-arid Mu Us sandy land, China

Desertification is an important ecological and environmental problem that concerns the whole globe. Percent vegetation cover is closely related to occurrence and degree of desertification, and its estimation is essential to monitoring and assessing land desertification. Normalized Different Vegetation Index (NDVI), an indicator widely employed to estimate percent vegetation cover, did not work well because of heterogeneous topographical features and vegetation cover in Mu Us sandy land. We selected variables according to correlations between them and vegetation cover fraction and estimated percent vegetation cover with the general regression neural network (GRNN) model and partial least squares (PLSs) regression linear model. Results indicated that correlations between remote sensing indices and vegetation cover fraction varied with topographical feature types. We also found that if field samples were classified with topographical features before estimation, prediction precision of the model was improved for individual plots. This research provides new alternatives for more precise vegetation cover estimation in the semi-arid area of China.