3.4.10 Soil Mg prediction
Similar to that of Na and Ca, calibration prediction accuracy of MARS for predicting the Mg was found the highest and excellent (R2c=0.89; MBEc=-0.09; RMSEc=1.37; RPDc=2.36, rank=1.25). The acceptable validation was achieved with PCR (R2p=0.61; MBEp=-0.17; RMSEp=2.27; RPDp=1.60 and rank=1.00). The best overall rank was observed using the PCR (1.75) (Figure 4j). Our results of the best performance of the PCR model are concordant with Islam et al. (2003) and Chang et al. (2001) who reported R2=0.68 and 0.63, respectively. Janik et al.(1998) reported good prediction (R2=0.79) for the same linear technique i.e. PCR using the MIR spectral data. Prediction of soil Mg using mPLSR (R2=0.90) (Cozzolino et al. , 2013), airborne hyperspectral imaging sensors (R2c=0.69) (DemattĂȘ et al. , 2016) and NIR spectroscopy (R2=0.89 and RPD=3.08) (Hively et al. , 2011) has been reported in the literature.