A comparison of three methods for estimating leaf area index of paddy rice from optimal hyperspectral bands

This paper follows previous research that identified 15 hyperspectral wavebands that were suitable to estimate paddy rice leaf area index (LAI). The objectives of the study were to: (1) test the efficiency of the wavebands selected in the previous study, (2) to evaluate the potential of least squares support vector machines (LS-SVM) to estimate paddy rice LAI from canopy hyperspectral reflectance and (3) to compare multiple linear regression-MLR, partial least squares-PLS regression and LS-SVM to determine paddy rice LAI using the selected wavebands. In the study, measurements of hyperspectral reflectance (350–2500 nm) and corresponding LAI were made for a paddy rice canopy throughout the growing seasons. On the basis of the wavebands selected previously, models based on MLR, PLS and LS-SVM to estimate rice LAI were compared using the data from 123 observations, which were split randomly for model calibration (2/3) and validation (1/3). Root mean square errors (RMSEs) and the correlation coefficients (r) between measured and predicted LAI values from model calibration and validation were calculated to evaluate the quality of the models. The results showed that the LS-SVM model using the 15 selected wavebands produced more accurate estimates of paddy rice LAI than the PLS and MLR models. We concluded that the LS-SVM approach may provide a useful exploratory and predictive tool for estimating paddy rice LAI when applied to reflectance data using the 15 selected wavebands.