3.4.12 Soil Fe prediction
The unique feature of the salt-affected soils of the study is that they possess an acidic soil reaction. The average soil pH of the full dataset was 5.42. These soils found to have high amounts of micronutrients like Fe, Mn, Cu and Zn under acidic soil reaction. The prediction accuracy for calibration was the highest (R2c=0.94; MBEc=1.88; RMSEc=22.57; RPDc=3.35) with MARS model (Figure 4l). The prediction accuracy was considered excellent. Excellent, but lesser compared to MARS, prediction accuracy was observed with RF (R2c=0.92; MBEc=-0.65; RMSEc=29.60; RPDc=2.56) and SVR (R2c=0.92; MBEc=0.18; RMSEc=26.39; RPDc=2.87). The Fe prediction using the validation dataset was highest but non-reliable using PLSR (R2p=0.82; MBEc=6.16; RMSEc=39.62; RPDc=1.13 and rank=1.25). Malmir et al. (2019) reported poor prediction accuracy for soil Fe and Mn using laboratory-based hyperspectral imaging (HIS) operated in 400-1000 nm with R2cv value in the range of 0.24-28 and 0.09-0.18 for Fe and Mn, respectively. Researchers have reported the use of spectral data through multivariate models to predict soil Fe with an accuracy R2 of 0.49-0.90 (Janiket al. , 1998; Chang et al. , 2001; Islam et al. , 2003; Cozzolino & Moron, 2003).