An inversion method of leaf area index based on spectral absorption features

Leaf area index (LAI) is an important parameter in the research of green vegetation ecosystem. Traditional methods for inversion of leaf area index based on hyperspectral image mainly use the reflectance of a band or an index generated by the combination of several bands, whose inversion accuracy is generally limited. In this paper, the BP (back-propagation) neural network is constructed to invert the leaf area index by using the multivariate absorption features, benchmarked with the one based on the reflectance of blue (450nm), green (550nm), red (670nm) and near infrared (870nm) wavebands. The LAI estimated by the proposed methods was compared with the other two methods. In order to evaluate the inversion models, the RMSE (root mean square error) and R
²
(coefficients of determination) between the estimated and the real LAI were used. Experimental results show that the RMSE of LAI obtained by the proposed methods outperformed the benchmarked method by 0.22 vs. 0.28.