3.4.13 Soil Mn prediction
Excellent predictions (R2c=0.99; MBEc=-0.72; RMSEc=2.16; RPDc=3.34) were obtained using the SVR model for Mn (Figure 4m). The MARS models also recorded an excellent prediction (R2c=0.93; MBEc=-0.25; RMSEc=3.19; RPDc=2.25). Though the predictions for calibration using SVR and MARS were excellent, the validation predictions using PCR (R2p=0.73; MBEp=-0.84; RMSEp=6.20; RPDp=1.24) and PLSR (R2p=0.52; MBEp=-0.89; RMSEp=6.04; RPDp=1.28) were better than other models tested. These results of prediction were considered as non-reliable. The results pertaining to Mn are in agreement with those reported by Janik et al. , (1998), who recorded R2 of 0.57 and 0.66 to predict DTPA extractable-Mn and exchangeable-Mn using the PLSR analysis of the MIR spectral data (2500-25000 nm). Mehlich-III extractable-Mn could be monitored using PCR of the VIS–NIR (400–2498 nm) with an accuracy of R2=0.70 (Chang et al. , 2001). Similar to Fe, Mn also showed poor prediction accuracy and this is in accordance with Vendrame et al. (2012) who reported poor prediction with R2=0.35 and RPD=1.1 using multivariate analysis of NIR spectral data to predict soil MnO2 content.