Persistence of soil organic matter as a function of its molecular Composition as revealed by Py‐GC/MS

Póster presentado en la XVI Reunión Científica de la Sociedad Española de Cromatografía y Técnicas Afines (SECyTA2016) P‐ENV‐19Eds: González-Pérez, José Antonio.-- Almendros Martín, Gonzalo.-- González-Vila, Francisco Javier.-- Rosa Arranz, José M. de laThere is a large controversy on whether the resilience of soil organic matter (SOM) depends on physical protection by soil minerals or on its intrinsic molecular composition [1]. In the case study of 30 volcanic soils from Tenerife (Spain), the dependent variables (DV) were total amount of SOM and TMC (total mineralization coefficient, CO2 released under incubation). This research focuses on estimating DVs using SOM molecular descriptors obtained by pyrolysis‐gas chromatography‐mass spectrometry (Py‐GC/MS) using a Pyrojector® device (SGE instruments) on a GC/MS Finnigan Trace GC Ultra, a Trace DSQ MS and an HP‐1 column. Up to 47 of the 102 major Py compounds were selected as descriptors (independent variables, IVs) attending to its normal distribution and few missing values. The relationships between DVs and normalized GC peaks (IVs) were examined by three approaches: (a) Partial least squares regression (PLSR) checking different pretreatments (standardization of IVs as total or relative abundances, mean centering, variance scale, and logarithmic or square root transformations). Spurious models were discarded after repeating PLSR models with the randomized DV; (b) Pearson’s correlation coefficients between each DV and the 47 IVs; comparison with other indices computed during PLSR: the variable Importance in the projection (VIP), b coefficients, W weights and factor loadings [2]; (c) comparing soil subsets representing opposed levels of the DVs: The 30 soils were ordered in decreasing DVs values and two subsets were considered, e.g., above vs below the median of the DV (or in the uppermost Q1 quartile vs Q4), i.e., soils ‘behaving as Csinks’ vs ‘soils with ‘poor C sequestration potential’. Differences between subsets were checked by ANOVA and plotted as subtraction values. The results from PLSR demonstrated that both SOM concentration and TMC can be predicted (P 1 were not always correlated (P< 0.05) with the corresponding DV or, in the case of Pearson’s coefficients, careful control of outliers is required.[1] M.W.I. Schmidt et al., Nature 478 (2011) 49–56. [2] R.A Viscarra‐Rossel, Chemometrics and Intelligent Laboratory Systems 90 (2008) 72–83. [3] D.W. Van Krevelen, Fuel 29 (1950) 269–284N