Research on vegetation cover information extraction technologies under different terrain conditions

Abstract

Aims We compared a variety of vegetation cover extraction methods by hyperspectral remote sensing that are currently popular. Our aims were to systematically study vegetation information extraction technology and provide a reference for further study of vegetation cover. Methods The methods of vegetation index, dimidiate pixel model, principal component regression (PCR), partial least squares regression first order differential (PLSR) and linear spectral mixture decomposition model were used to extract vegetation information. Important findings The vegetation cover estimation ability of dimidiate pixel models established by different normalized differential vegetation index (NDVI) was higher than that of the regression models established by NDVI directly. The optimization model for shady slopes was PLSR model based on the first order differential (FD) with modeling R2 = 0.810, root mean square error (RMSE) = 6.29 and validation R2 = 0.773, RMSE = 8.85. The optimization model for sunny slopes was PLSR model based on the second order differential (SD), with modeling R2 = 0.823, RMSE = 6.04 and validation R2 = 0.801, RMSE = 7.35. The optimization model for plains was full confine linear spectral mixture model (FCLS), with validation R2 = 0.852, RMSE = 5.86.