3.4.4 BD prediction
Although the prediction accuracy during calibration for BD was observed
as excellent (R2c = 0.99;
MBEc = -0.01; RMSEc = 0.05;
RPDc = 5.94) using the SVR model, the results during
validation were categorized as non-reliable. The highest prediction
accuracy (R2p = 0.52;
MBEp = 0.03; RMSEp = 0.22;
RPDp = 1.13) for BD was recorded with PLSR (Figure 4d).
The BD is one of the fundamental soil physical properties and is often
used as an indicator of soil quality, site productivity and soil
compaction (HÃ¥kansson, 1990; Suuster et al. , 2011). Previously,
only a few studies were carried out mainly focusing on the mid-infrared
region (MIR) for predicting soil BD (Minasny et al. , 2008). They
observed unsuccessful prediction as BD is related to a structural
pore-space condition that cannot be captured by MIR spectra. Saekiet al. (2003) predicted BD using the NIR spectroscopy with a good
prediction accuracy using the PLSR technique (R2 =
0.96). Moreira et al. (2009) reported good and satisfactory
prediction with R2c= 0.34 obtained
using modified PLSR technique (mPLSR) and NIR spectral information.
According to Minasny et al. (2008) under certain conditions soil
BD is strongly related to the organic matter content of the soil, of
which major mass (approximately 58%) exists as carbon (C). Poor
prediction accuracy using VIS-NIR spectroscopy in our study could be
attributed indirectly with low prediction accuracy for some soil
parameters (e.g. total C and N) which shows relatively a small variation
in the soil as suggested by van Groenigen et al. (2003).