3.4.11 Soil S prediction
The results pertaining to the S showed that the calibration prediction
accuracy was the highest with SVR
(R2c=0.96; MBEc=-0.41;
RMSEc=5.79; RPDc=3.21; rank=1.25) and it
was followed by MARS (R2c=0.93;
MBEc=-0.08; RMSEc=6.58;
RPDc=2.82; rank=1.75). Among different models validated,
PLSR had the highest but non-reliable predictions
(R2p=0.76; MBEp=0.39;
RMSEp=11.33; RPDp=1.27; rank=1.00). The
best overall rank was obtained by using PLSR to predict the S (Figure
4k). Cozzolino et al. (2013) predicted soil total sulfur (TS)
using the spectral reflectance data (350-1800 nm) with accuracy of
R2=0.81 using PLSR analysis. Wijewardane et al.(2018) reported satisfactory prediction of TS with
R2>0.95 and RPD>5.5 using
PLSR and ANN models in the MIR region.