Table 1.
Statistics of paddy soil samples in this study.
Figure 1.
The average reflectance spectra measured in laboratory (red) and in situ (blue) and their corresponding standard deviation values (shaded regions).
Figure 2.
Wavelength specific t-tests between continuum removed laboratory-based and in situ spectra.
Note: The shaded regions show where significant differences occur between the spectra at significance level.
Figure 3.
Number of factors (NF) used in partial least square regression versus (a) cross-validated root mean square error (RMSEcv) and (b) Akaike Information Criterion (AIC).
Table 2.
Comparison of prediction accuracy for in situ PLSR, laboratory-based PLSR and in situ LS-SVM.
Table 3.
Upper triangular correlation matrix among six soil properties.
Figure 4.
Grid search on γ and σ2 using least square support vector machine (LS-SVM).
Figure 5.
Predicted versus observed values of soil (a) OM, (b) OC, (c) TN, (d) AN, (e) AP, (f) AK, and (g) pH using least square support vector machines (LS-SVM) with in situ vis-NIR spectra.
Table 4.
Comparison of prediction accuracy of soil properties with in situ vis-NIR for paddy soils and irrigated soils (dry-farming).