Prediction of total silicon concentrations in French soils using pedotransferfunctions from mid-infrared spectrum and pedological attributes
Résumé
The biogeochemical cycle of Si has become a subject of increasing interest in recent years
because of its interconnection with the C cycle and global change. Recent studies have suggested
that land use and particularly agriculture can modify the Si cycle. However, the consequences of
such a modification for the global Si cycle and for the growth of crops which are Si accumulators
(most of the cereals) are still not well constrained. The bioavailable Si fraction is govern by several
factors, among which the amount in total Si (Si-tot) that represent the long term weatherable pool.
This amount has thus to be quantified at the territory scale. However, its determination is time
consuming and costly. It is performed by ICP-AES after mineralization by sodium peroxide and
sodium carbonate fusion. We thus propose an alternative estimation of the Si-tot at the French
territory scale, within the project BIOSISOL (ANR-14-CE01-002), based on the samples of the
French soil monitoring network (RMQS). This network consist in 2130 sites sampled in a regular 16
× 16-km grid. We measured topsoil Si-tot at 630 sites of the RMQS network consisting in one point
out of four from this network (that is 450 points) and a set of 150 points selected randomly among
the remaining sites. Mid-infrared reflectance and a large set of soil properties were measured at all
sites of the RMQS. These soil properties are particle-size distribution, pH water, organic carbon,
Total N, soluble and total P contents, calcareous, CEC and exchangeable cations (K, Ca, Mg, Na),
Bo and 8 trace elements (Cd, Cr, Cu, Co, Ni, Pb, Tl, Zn). The dataset was then divided in 3 sets:
Calibration, validation and test sets. The two formers were defined using a repeated folding step
within the 650 sites and the latest corresponded to the 1500 remaining sites. Using a regression
rule algorithm called CUBIST, we found that Si-tot was very well predicted by the mid-infrared
spectroscopy (validation R ~ 0.98 and test R ~ 0.97) and by conventional pedotransfer functions
using soil properties (validation R ~ 0.98). This last model was used to evaluate the MIR model.
Based on the MIR predictions, we then mapped the concentration of Sitot across France using
digital soil mapping techniques.
Origine | Fichiers produits par l'(les) auteur(s) |
---|
Loading...