Partitioning the variation of microbial measurements in forest soils into heavy metal and substrate quality dependent parts by use of near infrared spectroscopy and multivariate statistics

Combinations of different multivariate statistical methods were used to separate the variation in microbial parameters into one heavy metal dependent part, and one organic matter quality dependent part (measured by Near Infrared (NIR) spectroscopy). The microbial variation caused by differences in the organic matter quality was removed from the original parameters. The new parameters were called heavy metal related. This operation increased the significance of the correlations between microbial activity and the logarithm of the (copper + zinc) content. For basal respiration rate it increased from P(2 tail = 0.077 to P < 0.001 and for SIR from P = 0.068 to P < 0.001. For lag time and specific growth rate after glucose addition, no relevant models of the organic matter quality dependent part of the respiration could be made. The advantages and disadvantages of using combinations of Marked Variable Projection (MVP), Partial Least Squares (PLS) regression, Principal Components Analysis (PCA), Multiple Linear Regression (MLR) and different cross validation techniques are discussed.