Quantifying in situ and modeling net nitrogen mineralization from soil organic matter in arable cropping systems
Résumé
Improved assessment and prediction of net nitrogen mineralization (NNM) from soil organic matter in
field conditions are essential for a better management of nitrogen (N) in arable cropping systems and an
accurate simulation of its seasonal dynamics. The first objective of this study was to quantify accurately
NNM rates by combining measurements of soil water and mineral N contents in soil profile, daily
weather data and the LIXIM calculation model (Mary et al., 1999). The second objective was to develop a
predictive model of in situ NNM rates based on basic soil properties and to evaluate if additional factors
related to cropping history, organic matter fractions or microbial biomass could improve the model
performance. We selected a set of 65 field experiments representative of arable cropping systems in France which were carried out in bare fallow soils without recent addition of crop residue (minimal delay of 4 months)
or organic amendment. In these experiments, soil water and mineral N contents were monitored over
the soil profile (0e90 to 0e150 cm depth, 3 to 5 layers) during 100e555 days (median value of 272 days).
LIXIM was able to satisfactorily simulate water (EF ¼ 0.58) and mineral N contents (EF ¼ 0.80) in the
several layers, allowing to disentangle the two main processes involved in mineral N changes: NNM
(occurring in the upper soil layer) and nitrate transfer (through the whole soil profile). The actual NNM
rates varied widely between sites from 0.13 to 1.10 kg N ha1 day1. The potential NNM rates (Vp),
calculated under standard conditions of water and temperature, varied between 0.17 and
1.67 kg N ha1 nday1 (nday ¼ “normalized” day at 15°C and field capacity). A novel model with a multiplicative structure and a sequential approach was developed by including functions explaining the relationship between mineralization rate Vp (predicted variable) and explanatory variables. Soil organic N (SON) was the most correlated variable with Vp (r ¼ 0.51, p < 0.001) and the first variable introduced in the model as a linear function. Non-linear functions of soil clay content, pH, C/N ratio and CaCO3 content were successively introduced, leading to a “soil model” based on five basic soil parameters that allowed explaining 61% of the variance in Vp. Two functions relative to the frequency of rapeseed and legumes in rotations were found to further improve the predictive quality of the soil model, explaining finally 72% of Vp variance (“soil-history model”). Organic matter fractions (particulate or extractible organic matter) did not explain significantly more variance of Vp due to their strong correlations with SON. Microbial biomass explained only 2% additional variance in Vp compared to the “soilhistory model”. The predictive error of prediction (RMSEP) varied between 0.22, 0.19 and
0.17 kg N ha1 nday1 for the “soil” (n ¼ 65), “soil-history” (n ¼ 65) and “soil-history-biological” (n ¼ 47)
models. This multiplicative, non-linear approach performed better than a multiple linear regression
approach which explained 10e15% less variance of Vp when using the same explanatory variables.