Affect of Different Preprocessing Methods on Principal Component Analysis for Soil Classification

Fast classification of soil with different texture is essential for site-specific application of different inputs into farmland. Total 203 soil samples with five textures were collected from Silsoe Experimental Farm, Cranfield University, England. Using a Vis/NIR spectrophotometer (Tech5, Germany), spectra of soil samples were recorded for the study. Amongst the pre-processing methods, smoothing with moving average(MA), multiplicative scatter correction(MSC), standard normal variation(SNV), de-trending(DT), baseline correction(BC) and derivatives( 1st and 2nd ) were mainly investigated. PCA was applied for evaluation of the efficiency of different pre-processing methods on soil spectra. The sore plot of PCs shows that 1st derivative can help separate all textures much more effective than other methods. According to the cumulative variance of first 8 PCs, the various combinations of MA, MSC, DT and BC can be regarded as good methods. The worst is 2nd derivative due to its inducing much more noise. The study suggests that 1st derivatives should be firstly concerned amongst various pre-processing methods for the classification of soil textures.