Vous êtes sur la page 1sur 16

Soil Biology & Biochemistry 111 (2017) 44e59

Contents lists available at ScienceDirect

Soil Biology & Biochemistry


journal homepage: www.elsevier.com/locate/soilbio

Quantifying in situ and modeling net nitrogen mineralization from soil


organic matter in arable cropping systems
Hugues Clivot a, *, Bruno Mary a, Matthieu Vale  b, Jean-Pierre Cohan c, Luc Champolivier d,
Franois Piraux c, Franois Laurent c, Eric Justes e, **
a
INRA, UR 1158 AgroImpact, site de Laon, F-02000 Barenton-Bugny, France
b
AUREA, F-45160 Ardon, France
c
ARVALIS - Institut du Veg
etal, F-91720 Boigneville, France
d
Terres Inovia, Centre INRA de Toulouse, CS 52627, F-31326 Castanet-Tolosan Cedex, France
e
INRA, UMR INRA/INPT 1248 AGIR, Auzeville, CS 52627, F-31326 Castanet-Tolosan Cedex, France

a r t i c l e i n f o a b s t r a c t

Article history: Improved assessment and prediction of net nitrogen mineralization (NNM) from soil organic matter in
Received 24 October 2016 eld conditions are essential for a better management of nitrogen (N) in arable cropping systems and an
Received in revised form accurate simulation of its seasonal dynamics. The rst objective of this study was to quantify accurately
6 March 2017
NNM rates by combining measurements of soil water and mineral N contents in soil prole, daily
Accepted 15 March 2017
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
Keywords:
Modeling
performance.
Nitrogen mineralization We selected a set of 65 eld experiments representative of arable cropping systems in France which
Soil organic matter were carried out in bare fallow soils without recent addition of crop residue (minimal delay of 4 months)
Field experimental dataset or organic amendment. In these experiments, soil water and mineral N contents were monitored over
LIXIM model the soil prole (0e90 to 0e150 cm depth, 3 to 5 layers) during 100e555 days (median value of 272 days).
Cropping history LIXIM was able to satisfactorily simulate water (EF 0.58) and mineral N contents (EF 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 prole). 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 normalized day at 15  C and eld 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 explana-
tory variables. Soil organic N (SON) was the most correlated variable with Vp (r 0.51, p < 0.001) and the
rst 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 ve 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 nally 72% of Vp variance (soil-history model). Organic matter fractions (particulate or
extractible organic matter) did not explain signicantly more variance of Vp due to their strong corre-
lations with SON. Microbial biomass explained only 2% additional variance in Vp compared to the soil-
history model. The predictive error of prediction (RMSEP) varied between 0.22, 0.19 and
0.17 kg N ha1 nday1 for the soil (n 65), soil-history (n 65) and soil-history-biological (n 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.

* Corresponding author.
** Corresponding author.
E-mail addresses: clivot.hugues@hotmail.fr (H. Clivot), eric.justes@inra.fr (E. Justes).

http://dx.doi.org/10.1016/j.soilbio.2017.03.010
0038-0717/ 2017 Elsevier Ltd. All rights reserved.
H. Clivot et al. / Soil Biology & Biochemistry 111 (2017) 44e59 45

The models based on soil properties and cropping history could be further implemented in multi-
disciplinary soil-crop models and decision support systems in order to enhance the prediction of N
dynamics in soils and improve fertilizer recommendations.
2017 Elsevier Ltd. All rights reserved.

1. Introduction predict more accurately N supply to plants.


In situ eld experiments appear more labor-intensive but are
Prediction of soil nutrient supply to crops is still a major eco- needed to provide more realistic estimations of soil N mineraliza-
nomic and environmental concern in agriculture management. In tion under ambient conditions and by taking into account the ef-
particular, an accurate simulation of soil Nitrogen (N) mineraliza- fects of soil architecture. In situ N mineralization can be estimated
tion is crucial to improve N management in order to match fertilizer from N balance in agricultural elds by determining N uptake by
timing and rates with crop N demand, maximize economic effec- plants and mineral N variations in soil (Constantin et al., 2011;
tiveness and minimize mineral N losses to the environment. Meisinger, 1984). However this method is restricted to the
In agroecosystems, soil organic matter (SOM) is intimately growing season and is complicated by the need to evaluate accu-
linked to many physical, chemical and biological properties rately N losses to the environment, N remaining in unharvested
participating in soil quality and productivity (Doran and Parkin, parts of crops, in particular roots but also the effects of root exu-
1994). SOM plays a central role in soil structure, plant available dates, which are rarely measured (Jarvis et al., 1996). Furthermore,
water capacity (Reeves, 1997) and in sustaining soil fertility the large uncertainty on measurements does not allow to obtain
(Kononova, 1966; Tiessen et al., 1994) as the major supplier of N and accurate estimates of N mineralization kinetics.
other limiting nutrients for plant growth (Smith et al., 2015). N Other in situ approaches have been proposed including the
mineralization from SOM is a process mainly mediated by the ac- closed buried bags (Eno, 1960; Lehrsch et al., 2016), the covered-
tivities and communities of microorganisms (Bartholomew, 1965; cylinder (Adams et al., 1989; Adams and Attiwill, 1986; Salazar
Jansson and Persson, 1982) which are inuenced by edaphic and et al., 2015) and the ion-exchange resins methods (DiStefano and
environmental factors and by complex interactions with soil micro- Gholz, 1986; Turner and Henry, 2010) and more generally in situ
and meso-fauna (Scheu et al., 2005). Soil physico-chemical pa- core methods proposed by Raison et al. (1987) and commented in
rameters, biological characteristics (Hassink et al., 1993) and cli- the review done by Khanna and Raison (2013). The in situ incu-
matic conditions, more particularly temperature and moisture bation of undisturbed soil samples in bare fallow conditions
(Cassman and Munns, 1980; Curtin et al., 2012; Guntin ~ as et al., cannot integrate the effects of a vegetation cover, but could
2012; Stanford et al., 1973; Stanford and Epstein, 1974), are represent a good compromise between the objectives of main-
known to be important factors to take into account in order to taining realistic eld conditions and reducing the uncertainty of N
predict N mineralization from SOM. Agricultural practices such as mineralization estimates. In this approach, the measured variation
tillage, mineral or organic fertilization and residue management in soil mineral N is attributed mainly to two processes: i) net N
can impact soil N supply (Shari et al., 2014; Spargo et al., 2011; St. mineralization and ii) N losses, the latter being generally domi-
Luce et al., 2011), as well as crop rotation (Carpenter-Boggs et al., nated by nitrate leaching. Preventing or assessing nitrate leaching
2000; Senwo and Tabatabai, 2005; Shari et al., 2008; St. Luce can be achieved by three methods: a) preventing leaching by
et al., 2016). covering the cores, b) using ion exchange resins at the bottom of the
Net N mineralization (NNM) resulting from the balance between core; c) calculating nitrate leaching using measurements and a
gross N mineralization and immobilization by soil microorganisms calculation model. Methods a) and b) have been discussed by
is commonly assessed by measuring the rate of inorganic N accu- Hanselman et al. (2004) who indicated that covering the cores can
mulation in soil. This mineral N pool constitutes both the main modify the water regimes whereas resins may vary in their ef-
source of N available for crops and the potential N losses through ciency for trapping the leached nitrate. Method c) has been
leaching and gaseous emissions from soil. Laboratory incubations developed by Mary et al. (1999) in another approach combining
and chemical indices have been and are still the most common measurements and model calculations. The LIXIM calculation
approaches conducted to determine potential mineralizable N or model proposed by these authors was used later in order to
NNM rates in soils (Grifn, 2008; Shari et al., 2007). However, calculate net mineralization and leaching using in situ measure-
previous works highlighted the difculties to extrapolate results ments of water and mineral N contents (Beaudoin et al., 2005; Lesur
obtained from these methods to eld conditions (Connell et al., et al., 2014; Oorts et al., 2007; Vale et al., 2007).
1995; Hart and Binkley, 1985). Indeed, differential responses have In this study, we propose to use the calculation model LIXIM to
been previously observed between eld measurements and incu- estimate NNM rates under actual eld conditions and in
bation methods that have sometimes overestimated (Adams and conventional-till cropping systems. LIXIM relies on frequent mea-
Attiwill, 1986; Honeycutt, 1999) or underestimated in situ N surements of water and mineral N contents to calculate minerali-
mineralization (Abril et al., 2001). In laboratory experiments, the zation kinetics against true time (i.e. under the climatic conditions
diverse soil pretreatments (e.g. sieving, drying, rewetting) applied encountered in situ) and normalized time (equivalent time under
prior to incubations under controlled conditions (e.g. temperature standard moisture and temperature conditions; for details, see
and water content) can strongly inuence SOM mineralization rates Mary et al., 1999). These corrections for soil temperature and
(Benbi and Richter, 2002; Jarvis et al., 1996). As a consequence, moisture are essential to analyze the role of soil properties on NNM
some models with parameters calibrated using laboratory incuba- without the confounding effects of climatic factors. They also allow
tion experiments were found to be poor predictors of both N the comparison of NNM estimates between eld experiments and
mineralization and nitrate leaching under eld conditions (Cabrera laboratory incubations, as suggested by Ros et al. (2011).
and Kissel, 1988; Johnson et al., 1999). Realistic estimations of N This study was conducted on a network of 65 dedicated eld
mineralization are required to improve or develop models able to experiments representative of most French arable pedo-climatic
46 H. Clivot et al. / Soil Biology & Biochemistry 111 (2017) 44e59

situations. The rst objective was to calculate the actual N miner- mean daily temperature measured during the duration of eld
alization rates derived from soil organic matter expressed per unit experiments ranged from 9.1 to 17.8  C and the daily precipitation
of true time and the potential N mineralization rates expressed per ranged from 1.0 to 4.3 mm (Table 1).
unit of normalized time (Vp), by combining water and mineral N The 65 experimental sites exhibited contrasting soil parameters
content measurements and the LIXIM calculation model. The sec- (Table 1). The clay content of soils ranged from 24 to 396 g kg1
ond objective was to design a novel model of Vp prediction based (median value of 191 g kg1), silt content from 55 to 814 g kg1
on simple and routinely measured soil properties and to evaluate if (median value of 539 g kg1) and sand content from 22 to
additional factors related to cropping history or other variables 895 g kg1 (median value of 184 g kg1). The CaCO3 content of soil
more complex to acquire, such as organic matter fractions and ranged between 0 and 436 g kg1 (median value of 4 g kg1). Soil
microbial biomass, could improve the predictive quality of the pH varied between 5.7 and 8.4 (median value of 7.7). Soil organic C
model. and N stocks in the topsoil (0e30 cm) also varied widely: from 27.0
to 147.2 t C ha1 (median value of 45.8 t ha1) and 2.6 to
2. Material and methods 12.6 t N ha1 (median value of 5.0 t ha1), respectively. A com-
parison of the range covered by these pedological conditions with
The approach used in this study to quantify and model the po- the French Gis Sol database (estrada.orleans.inra.fr/geosol/)
tential NNM rates (Vp) is presented in Fig. 1. showed that soils selected in this study were rather well repre-
sentative of the French agricultural conditions (Figure S2).
Since our objective was to assess N mineralization derived from
2.1. Experimental sites and datasets
stabilized SOM and avoid the temporary mineralization/immobili-
zation processes due to organic residues decomposition, the eld
The 65 eld experiments were carried out between 1990 and
experiments were performed in bare fallow soils without recent
2008. Their location and climate zone are shown in Figure S1. The
crop residue or organic amendment. They were initiated mostly
between the end of summer and the beginning of winter, either
soon after harvest when crop residues were exported or long time
after returning crop residues to soil (at least 4 months), in order to
avoid the N immobilization phase (e.g. Garnier et al., 2003). Bare
fallow conditions were obtained primarily with herbicides and
secondarily with supercial tillage. All elds had been managed
previously with tillage and regular ploughing over an average
depth of 30 5 cm. Cropping histories varied among elds.
Depending on available information, we were able to retrieve 3 to
10 years of past cropping history. Crops were classied into 6
groups: cereals (wheat, barley, triticale), spring crops (maize,
sorghum, sunower), legumes (pea, soybean, fababean), sugar-
beet, rapeseed and others for minor crops (tomato, onion,
ryegrass). In our study, these groups represented 48%, 29%, 11%, 8%,
3% and 1% of cultivated crops, respectively. Wheat, maize, barley
and pea contributed respectively to 39%, 19%, 8% and 8% of total
crops in rotations.

Table 1
Climatic conditions of eld-experiments and physicochemical, organic matter and
microbial associated characteristics for the upper soil layer (0e30 cm) of the study
sites.

Units Min Max Median Mean SD

Climatic conditions (n 65)



Mean daily temperature C 9.1 17.8 11.7 12.6 2.3
Daily precipitation mm 1.0 4.3 1.9 2.1 0.7
Soil properties (n 65)
Clay g kg1 24 396 191 201 76
Silt g kg1 55 814 539 502 198
Sand g kg1 22 895 184 218 174
CaCO3 g kg1 0 436 4 64 123
pH 5.7 8.4 7.7 7.5 0.8
SOC t ha1 27.0 147.2 45.8 51.7 24.4
SON t ha1 2.6 12.6 5.0 5.3 1.9
C/N 7.2 15.3 9.3 9.6 1.5
Organic matter fractions (n 47)
POM-C t ha1 3.0 25.8 7.0 8.6 5.1
POM-N t ha1 0.17 1.88 0.39 0.50 0.36
EOM-C t ha1 2.7 26.4 7.0 8.3 4.8
EOM-N t ha1 0.39 2.91 0.97 1.12 0.54
Microbial biomass C t ha1 0.32 1.58 0.72 0.75 0.32

Fig. 1. Overall approach used in this study for assessing and modeling potential net SOC: soil organic carbon, SON: soil organic nitrogen, POM: particulate organic
nitrogen mineralization (NNM) rates Vp. matter (>50 mm), EOM: extractable organic matter.
H. Clivot et al. / Soil Biology & Biochemistry 111 (2017) 44e59 47

Soil physico-chemical characteristics and cropping history were Corporation). C and N contents of the ne SOM fraction (<50 mm)
recorded for the 65 eld experiments constituting a set of data were determined by subtracting C and N contents of the POM
called dataset 65. For 47 of these eld experiments, results from fraction from those of the total soil.
organic matter fractionation and microbial biomass were also The fumigation-extraction method was used to estimate soil
recovered from the top-soils constituting a subset of data called microbial biomass C (Vance et al., 1987). Chloroform-fumigated and
data subset 47 which includes these additional parameters. De- non-fumigated soil subsamples (25 g oven-dry equivalent) were
tails on climatic conditions, soil parameters and cropping fre- extracted during 30 min using 25 mM K2SO4 (1:4 w/v ratio) and
quencies are provided in Clivot et al. (submitted). ltered (0.45 mm). Dissolved organic C concentrations in extracts
were determined by persulfate oxidation at 100  C with a 1010
2.2. Soil analyses Analyzer (OI Analytical). A kEC conversion factor of 0.45 was applied
to convert measured data into microbial biomass (Joergensen,
2.2.1. Physico-chemical analyses 1996).
2.2.1.1. Upper soil layer (0e30 cm). The upper soil layer was Extractable organic matter (EOM) was determined as proposed
sampled at the beginning of each experiment to measure basic by Lemaitre et al. (1995). Soil subsamples (20 g oven-dry equiva-
soil properties. Five to six soil cores were mixed together to obtain a lent) were extracted during 45 min using 25 mM K2SO4 (1:4 w/v
representative composite soil sample. Particle-size distribution was ratio) and then autoclaved at 121  C and 1 bar during 16 h. After
determined by the pipette method (NF ISO 11277). Soil carbonate centrifugation (5 min at 6000g), soluble organic C in supernatant
content was determined by a volumetric method according to NF extracts was determined by UV-persulfate oxidation (Dohrmann
ISO 10693. Soil bulk density was measured by the cylinder method DC 80) yielding EOM-C. Total soluble N (EOM-N) was assessed by
(Blake, 1965) or estimated according to the pedological soil class. high-temperature combustion with a TN 3000 Analyzer (Euroglas-
Soil pH was measured in water (1:5 soil/water ratio) (NF ISO Apollo Instruments).
10390). Soil organic C (SOC) was measured by colorimetry after
sulfochromic oxidation (NF ISO 14235). The quantitative determi- 2.3. Calculation of in situ net N mineralization rates
nation of soil organic N (SON) was performed using the modied
Kjeldahl method after sulfuric acid digestion (NF ISO 11261). The NNM rates were determined by combining measurements of
last two methods are well adapted to calcareous soils and previous water and mineral N contents, weather data and calculations by the
work showed that these methods gave similar results as total C and model LIXIM, as described by Mary et al. (1999). This daily time-
N measured by the more commonly used Dumas combustion step model simulates water drainage and evaporation and nitrate
method (Dimassi et al., 2014). SOC and SON contents as well as transfer in soil and provides estimates of nitrate leaching and NNM
stocks of organic fractions were calculated over the 0e30 cm depth. through time by inverse modeling. Water and solute transport are
This calculation is consistent with the fact that most soils were modeled using the mixing-cells concept (Van Ommen, 1985),
frequently moldboard ploughed down to this depth. each soil layer being characterized by its water properties Wwp and
Wfc. An optimization procedure embedded in LIXIM allows looking
2.2.1.2. All soil layers (0e150 cm). Soil water and mineral N con- for the best t between observed and simulated soil water and
tents were regularly measured in 3 to 5 soil layers (30 cm thick) mineral N contents in each layer at each sampling date, by opti-
down to 90, 120 or 150 cm depth, depending on subsoil material. mizing a few parameters. It allows to disentangle the effects of
The duration of the monitoring period varied from 100 to 555 days nitrate transfer and N mineralization on the variations of soil
(median value of 272 days). The number of sampling dates ranked mineral N in each layer through time. Denitrication is supposed to
from 5 to 23 sampling dates depending on the duration (median be negligible because no fertilizer-N is added and because nitrate
number of 10). Five to six soil cores were mixed to obtain a usually accumulates at soil surface only when the soil gets dry. The
representative sample for each replicate, and two or three repli- LIXIM model application requires i) the absence of plant cover, ii) a
cates (i.e. independent blocks) were performed at each sampling rather rapid nitrication as commonly encountered in the condi-
date. Water contents were determined gravimetrically (NF ISO tions prevailing in soils of our study sites and iii) multiple mea-
11465). Soil water content at permanent wilting point (Wwp) was surements of water and mineral N contents through time in at least
measured using the Richards pressure plate method at the poten- three soil layers.
tial of 1.6 MPa (NF ISO 11274). Water content at eld capacity In our study, four parameters were optimized specically for
(Wfc) was estimated as the mean of the highest values measured each experiment: i) the maximum depth affected by soil evapora-
under eld conditions in winter as described in Mary et al. (1999). tion (Ze), ii) the evaporation distribution coefcient (a), iii) the
Soil inorganic N was extracted during 30 min with 1 M KCl (1:2 w/v depth of soil layers linked to the dispersivity length (l), and iv) the
ratio). Ammonium and nitrate concentrations were determined by maximum depth contributing to mineralization (Zm) which is close
colorimetry using a continuous ow analyzer (Skalar Analytical). to the ploughing depth. A sensitivity analysis was conducted to test
Soil bulk density was measured as described for the upper soil layer the effect of varying these parameters on the simulated minerali-
in part 2.2.1.1. zation rate. It showed that the actual N mineralization rate in soils
Vm (expressed in kg N ha1 day1) was little dependent on the
2.2.2. Organic matter fractions value of these four parameters as long as the quality of t was close
Additional analyses of organic matter fractions were performed to the best one.
on the upper soil layer collected at the start of experiment in the The actual N mineralization rate in soils is known to be strongly
subset of 47 eld experiments. Measurements were made on a dependent on soil moisture and temperature. LIXIM model trans-
representative composite sample made of ve to six soil cores forms this variable Vm into a potential mineralization rate Vp
mixed together. Soil samples were fractionated by wet sieving (expressed in kg N ha1 nday1) at a specic reference temperature
under water on a 50 mm sieve. The particulate organic matter (15  C) and water content (Wfc). LIXIM simulates the daily soil
(POM) fraction was recovered on the sieve and separated manually water content in the biologically active layer (0-Zm). The normal-
from the sand particles. It was dried, weighed and crushed before ized time is calculated by taking into account both daily air tem-
analyzing C and N contents by the Dumas dry combustion method perature and simulated soil water content and the temperature and
using a NA 2000 elemental analyzer (Thermo Electron moisture functions of the LIXIM model (f(T) and g(W), respectively).
48 H. Clivot et al. / Soil Biology & Biochemistry 111 (2017) 44e59

Both rates are linked by the equation: 2.4.2. Designing a novel model for predicting potential N
mineralization rate
t Pearson correlation analysis was used rst to identify variables
Vp Vm$ (1)
tn signicantly correlated to potential NNM rates (Vp). These pre-
liminary correlations associated with graphical analyses showed
where t represents the true time (day) and tn the normalized time that either linear or non-linear relationship can perform well
(nday) dened as follows: depending on the explanatory variable investigated. We thus chose
to develop an empirical model based on the analysis of the rela-
X
t X
Zm
tionship patterns found between Vp and each explanatory variable
tn f T $g W (2)
(Vi). We assumed that a model with a multiplicative structure
n1 z1
would be better adapted to reect the nature of the interactive
Further details can be found in the paper of Mary et al. (1999). effects of explanatory variables on soil N mineralization than an
The variable Vp provides estimates of potential N mineralization additive linear model used for example in multiple linear regres-
rates which are comparable to incubation conditions. sion (MLR) model. This assumption was tested at the end of the
modeling process by comparing multiplicative and MLR models,
both modeling approaches using the same explanatory variables.
A step-by-step approach was used to include sequentially one
2.4. Data analysis and modeling approach
function per step, each function describing the relationship be-
tween Vp and the best remaining explanatory variable to nally
R Software version 3.3.0 was used for all statistical analyses (R
obtain an empirical model with a multiplicative structure as
Core Team, 2016).
follows:

2.4.1. LIXIM evaluation and calculation of N mineralization rate b p f V $f V ,,,f V


V (5)
i 1 1 2 2 i i
To evaluate the performance of LIXIM model, i.e. the simulated
At each step i, linear and non-linear relationships were assessed
water and mineral N contents compared to experimental obser-
and compared for each newly introduced function Vi. An optimi-
vations, the root mean squared error (RMSE), the modeling ef-
zation procedure (with 10,000 iterations) was used to determine
ciency (EF) (Smith et al., 1996) and the index of agreement (d1)
the parameters associated with each linear or non-linear function
(Willmott et al., 1985) were calculated as statistical evaluation
by minimizing the differences between observed and simulated
criteria for water and nitrogen. The model errors, RMSEs, were also
values. It was performed using the quasi-Newton method of
compared to the measurement errors represented by the standard
Broyden, Fletcher, Goldfarb and Shanno (BFGS) implemented in the
deviations (SDs) measured on eld replicates. The cumulative net N
general-purpose optimization function optim in R (R Core Team,
mineralization, M(t) in kg N ha1, calculated by LIXIM was then
2016). After step i, the ratio of observed to simulated N minerali-
tted to each of two models: Vp
zation rates R (6) was plotted vs each remaining variable Vj
bV pi
- a linear model (j i 1, n) to identify the variables which could improve both
explanatory and predictive qualities of the model. We selected the
Mt Vm$t (3a) variable Vj which provided the best (linear or non-linear) correla-
tion with the ratio R.
or Using this approach, a model was designed by considering
successively variable categories ordered by increasing acquisition
Mtn Vp$tn (3b) complexity (approach A). We rst developed a model based on
basic and routinely measured soil properties as explanatory vari-
ables (step A1). After reaching the best soil model according to
statistical criteria, optimized parameters were set at xed values
- an exponential model
and were further inserted into two additional models. The soil-
history model was developed by adding functions relative to
Mt No$1  expk$t (4a)
cropping history variables (step A2). Qualitative data relative to
cropping history were transformed into quantitative variables to
or
assess the inuence of preceding crops on potential NNM rates.
Two different quantitative variables were calculated and used for
Mtn No$1  expk$tn (4b)
modeling N mineralization rates: i) the crop type frequency in
The results showed that the quality of t of both models was rotation and ii) an indicator based on the presence or absence of the
much better described by equations (b) corresponding to Vp than crop type in the two preceding years, called Preceding Crop Index
by equations (a) for Vm, showing that the N mineralization rates (PCI). Three values were attributed to PCI: 1 for the presence as
were clearly affected by the variations in water and temperature preceding crop, 0.5 for ante-preceding crop and 0 for absence
regimes and that the effect of these factors was well accounted for during the two preceding years. As previously performed, param-
by LIXIM using the functions included in the algorithm. In the eters optimized for cropping history functions were set at xed
subsequent analysis, we only consider equations (b), i.e. cumulative values after obtaining the best soil-history model. We chose to
mineralization vs normalized time (Vp). Vp is calculated as the slope calibrate the soil and soil-history models using data from all ex-
of the linear regression between cumulative N mineralized and periments (dataset 65) in order to obtain a more robust estimation
normalized time (Equation (3b)) together with its 95% condence of parameters that were set at xed values and further applied to
interval. Detailed information on statistical evaluation of LIXIM the subset of 47 eld experiments (data subset 47). Finally, we
simulations and NNM rate calculations is given in Clivot et al. attempted to develop a third model including additional variables
(submitted). related to organic matter fractions and microbial biomass (step A3).
H. Clivot et al. / Soil Biology & Biochemistry 111 (2017) 44e59 49

It was evaluated against the data subset 47 which contained these difference was 0.3% and 0.1 kg N ha1, respectively. The model
measurements. errors (RMSE) were quite comparable to the measurement errors
The approach A was confronted to a second sequential multi- (standard deviations): 1.5% and 1.3% for water content, 7.1 and
plicative approach in which all explanatory variables were 7.0 kg N ha1 for mineral N content, respectively. Finally, a good
considered together without categorizing (approach B). The com- agreement was found between measured and simulated variations
parison was made on the data subset 47 which included all of mineral N stocks from the beginning to the end of experiments,
measurements. both being related by the following equation: Y 0.96X (r 0.97,
p < 0.001).
2.4.3. Predictive model evaluation: statistical criteria and The sensitivity analysis performed on the maximum depth
sensitivity analysis contributing to mineralization (Zm) indicated that the best quality
Several statistical measurements of agreement between obser- of t was obtained when Zm was equal to the ploughing depth
vations and model predictions were performed. The coefcient of which was close to 30 cm for most situations. This suggests that the
determination (r2), the mean difference (Bias), the modeling ef- ploughed layer is the main contributor to total N mineralization in
ciency (EF) and the root mean squared error (RMSE) were calculated the soil prole, justifying the calculation of SON stocks in the 0e30
to select the models according to their explanatory quality soil layer in this study. This corresponds to the depth usually
(Wallach, 2006). The Akaike Information Criterion (AIC) (Sakamoto sampled in eld plots for routine soil analysis.
et al., 1986) was also applied to compare and select the models Cumulative NNM varied from 23 to 307 kg N ha1, with a mean
according to their performance and by limiting over- of 140 kg N ha1. The actual NNM rate (Vm) varied from 0.13 to
parameterization. All data were used for model calibration allow- 1.10 kg N ha1 day1 with an average of 0.52 0.20 kg N ha1 day1.
ing a more robust estimation of parameters. The predictive quality The average normalized time calculated by LIXIM at the end of
of models was therefore assessed by calculating the root mean experiments was 229 108 ndays and was smaller than the actual
squared error of prediction (RMSEP) estimated by leave-one-out duration (282 93 days), the ratio of normalized to actual time was
cross-validation (Stone, 1974). 0.80 0.27. The cumulative amounts of mineralized N were tted to
The sensitivity analysis of outputs of the novel model predicting a linear or exponential model vs normalized time, as indicated
NNM was based on the analysis of variance (ANOVA) decomposi- before. As expected, the exponential model gave a slightly better t
tion. For the simulations, a complete factorial design was used with than the linear model with an average model error (RMSE) of 6.3 vs
5 modalities for each soil parameter. Three levels of values were 8.5 kg N ha1. The longer the experiment duration, the better was
used for the two-year PCI and 6 frequency values were used for the the exponential compared to linear model, in accordance with
observed proportion of a crop type in rotation. For data generation, previous ndings (e.g. Mary et al., 1999). However, we selected the
we dened lower and upper limits according to those observed in linear model for its simplicity and for the fact that the two pa-
our database and equidistant values were generated for each rameters of the exponential model (No and k) are strongly corre-
parameter. The generated data were used to perform simulations of lated and cannot be determined each other accurately but only as
Vp by the models. ANOVA were performed to determine the con- their product (Wang et al., 2003). The two models produced
tributions of explanatory variables and their two-factor in- strongly correlated parameters, with the following linear regres-
teractions to the variability of potential N mineralization rates (Vp) sion: k.No 1.20 Vp (r 0.90, p < 0.001).
calculated by the predictive models. ANOVA main effects and Among the whole dataset of 65 eld experiments, the potential
interaction Sensitivity Indices (SI) were calculated by dividing the mineralization rate Vp varied between 0.17 and
sums of squares by the total variability (Monod et al., 2006). 1.67 kg N ha1 nday1, with an average of
0.72 0.32 kg N ha1 nday1. The 95% condence interval calcu-
3. Results lated with the linear regression ranged between 0.01 and
0.14 kg N ha1 nday1.
3.1. Quantication of in situ net N mineralization
3.2. Potential mineralization rate vs soil properties and cropping
LIXIM calculation model was applied to each of the 65 experi- history
ments in order to assess actual (Vm) and potential (Vp) N miner-
alization rates. Fig. 2 shows one example of the evolution of Pearson correlations and graphical analyses identied that SON
measured and simulated water and nitrate contents within the soil content was the most highly correlated variable (r 0.51 and
prole in Boulancourt site (#8). LIXIM was able to reproduce r 0.62 for the dataset 65 and the data subset 47, respectively,
correctly the evolution of water (Fig. 2A) and mineral N (Fig. 2B) in p < 0.001) with Vp (Table 3). Vp was also signicantly positively
each of the three layers. It simulated a curvilinear net mineraliza- correlated with SOC content (r 0.48 and r 0.56 for the dataset
tion kinetics vs actual time (Fig. 2C) and a quasi linear kinetics vs 65 and the data subset 47, respectively, p < 0.001), as well as the
normalized time (Fig. 2D). In fact, the amount of mineral N in the other fractions of SOM: POM-C and -N, ne SOM-C and -N, EOM-C
whole prole varied little during autumn, winter and early spring and -N and microbial biomass-C (0.48 < r < 0.64, p < 0.001). All
due to the simultaneous occurrence of N mineralization and nitrate these variables were also highly correlated with SON
leaching whereas it increased markedly in late spring and summer (0.76 < r < 0.99, p < 0.001) (Tables S1 and S2).
due the dominant N mineralization process. Weaker correlations were observed between Vp and other soil
LIXIM was also able to mimic satisfactorily the evolution of parameters: a positive relationship with silt content (r 0.39,
water and mineral N contents in the 65 eld experiments, as p < 0.01 for the dataset 65) and negative relationships with clay
indicated by the statistical criteria (Table 2). The mean modeling content (r 0.27, p < 0.05 for the dataset 65), CaCO3 content
efciency (EF) and mean index of agreement (d1) for the 65 sim- (r 0.26, p < 0.05 for the dataset 65) and pH (r 0.36, p < 0.05
ulations were EF 0.58 and d1 0.69 for water content and for the data subset 47) (Table 3).
EF 0.80 and d1 0.79 for mineral N content. The higher efciency Lower but signicant correlations were also found between
found for mineral N is attributed to the greater temporal variation potential NNM rates and cropping history variables: i) positive re-
of mineral N compared to water contents. LIXIM provided unbiased lationships with the frequency of cereals (r 0.33 and 0.41 for the
predictions of soil moisture and mineral N contents since the mean dataset 65 and the data subset 47, respectively, p < 0.01) and
50 H. Clivot et al. / Soil Biology & Biochemistry 111 (2017) 44e59

Fig. 2. Example of evolution of measured and simulated water (A) and nitrate (B) contents obtained with LIXIM in each of the three soil layers of a eld experiment (Boulancourt,
site #8) and calculated cumulative amounts of net N mineralized against time (C) and against normalized time (D). Vertical bars are standard errors (n 3).
H. Clivot et al. / Soil Biology & Biochemistry 111 (2017) 44e59 51

Table 2
Statistical evaluation criteria for model simulation of soil water and mineral N contents and net N mineralization rates calculated with LIXIM.

Units Min Max Median Mean SD

Experiment duration days 100 555 272 282 93


Water and mineral N simulations
Water RMSE % 0.5 2.9 1.3 1.5 0.5
EF 0.32 0.97 0.58 0.58 0.24
d1 0.37 0.93 0.70 0.69 0.11
Mineral N RMSE kg N ha1 2.2 16.9 6.7 7.1 3.1
EF 0.04 0.99 0.85 0.80 0.16
d1 0.49 0.95 0.80 0.79 0.08
Dstock Mineral N Observed kg N ha1 89 228 82 77 69
Simulated kg N ha1 88 223 76 77 65
Net N mineralization rates
Cumulative net N mineralized kg N ha1 23 307 135 140 59
Vm kg N ha1 day1 0.13 1.10 0.50 0.52 0.20
Normalized time days 49 535 227 229 108
Vp Mean kg N ha1 nday1 0.17 1.67 0.67 0.72 0.32
Vp Condence Interval kg N ha1 nday1 0.01 0.14 0.02 0.04 0.03

RMSE: root mean squared error of the model, EF: modeling efciency, d1: index of agreement, Vm: actual net nitrogen mineralization rate, Vp: potential net nitrogen
mineralization rate.

Table 3
Pearson correlations between potential net N mineralization rates (Vp) and introduced in the soil model (Table 4, step A11). Vp was also highly
explanatory variables for the 65 experiment dataset (dataset 65) and for the data
correlated with SOC content. However, SOC content did not explain
subset of 47 experiments (data subset 47).
further variance of Vp once SON was introduced in the soil model as
Vp dataset 65 Vp data subset 47 an explanatory variable. At the second step (A12), soil clay content
Clay 0.27* 0.27 was found to be the best explanatory variable of Vp when a
Silt 0.39** 0.21 decreasing exponential function was considered. It slightly
Sand 0.14 0.02
improved the performance of the soil model, EF increasing from
CaCO3 0.26* 0.11
pH 0.20 0.36*
0.18 to 0.22. Soil pH was introduced at the third step (A13) using a
SOC 0.48*** 0.56*** gaussian function. This function increasing up to pH 8.5 was the
SON 0.51*** 0.62*** more appropriate for simulating the pH effect. It signicantly
C/N 0.21 0.23 improved the soil model: the model efciency reached 0.43 and the
bias decreased from 0.03 to 0.00 kg N ha1 nday1. This function
Cereals Freq 0.33** 0.41**
Spring crops Freq 0.26* 0.24
Legumes Freq 0.31* 0.45** allows simulating the expected decrease in N mineralization rate in
Sugarbeet Freq 0.23 0.01 very alkaline soils. The C/N ratio of SOM was introduced in the soil
Rapeseed Freq 0.17 0.36* model at step A14 using another gaussian function which peaked at
Others Freq 0.09 0.10 a C/N ratio of 11 and decreased afterwards. With these four
PCI Cereals 0.20 0.18
PCI Spring crops 0.25* 0.19
explanatory variables, the model explained 56% of Vp variance
PCI Legumes 0.16 0.25 without any bias. Soil CaCO3 content was found to be the last soil
PCI Sugarbeet 0.23 0.08 parameter that could still improve the soil model: the efciency
PCI Rapeseed 0.19 0.34* increased to 0.61 with the introduction of a decreasing hyperbolic
PCI Others 0.02 0.12
function. The parameter values that were optimized at step A15
POM-C 0.48***
POM-N 0.52*** were maintained in the soil model and in the following models,
Fine SOM-C 0.57*** except for parameter a which is a proportionality coefcient
Fine SOM-N 0.62*** allowing to avoid model bias at each modeling step. Fig. 3 displays
BiomassMic 0.54*** the pattern of the ve functions constituting the nal soil model
EOM-C 0.60***
EOM-N 0.64***
over the range of physico-chemical characteristics encountered in
the dataset 65. It shows that the amplitude of response of Vp model
Signicant correlations are indicated in bold. (levels of signicance: *p < 0.05,
to each variable is highest for SON, intermediate for clay content,
**p < 0.01 and ***p < 0.001).
Freq: frequency of crop group in rotation, PCI: Preceding Crop Index. pH and C/N ratio, and lowest for CaCO3 content.
SOC: soil organic carbon, SON: soil organic nitrogen, POM: particulate organic The introduction of potentially explanatory variables related to
matter (>50 mm), SOM: soil organic matter, EOM: extractable organic matter, cropping history was then investigated to design a soil-history
BiomassMic: microbial biomass. model. We found that the introduction of an increasing linear
rapeseed frequency (r 0.36, p < 0.05 for the data subset 47) and ii) function of rapeseed frequency during the two preceding years (PCI
negative relationships with legumes frequency (r 0.31, p < 0.05 Rapeseed) and a decreasing linear function of legume frequency in
for the dataset 65 and r 0.45, p < 0.01 for the data subset 47) and the rotation allowed to explain 11% more variance in Vp than the
spring crop frequency (r 0.26, p < 0.05 for the dataset 65). soil model, EF increasing from 0.61 to 0.72 after steps A26 and A27
(Table 4). The soil-history model was dened by these two addi-
tional functions and their parameters determined at step A27.
3.3. Modeling potential N mineralization rate
We tested whether the introduction of additional variables
related to organic matter fractionation and microbial biomass could
In the rst empirical approach (A), a predictive model was
improve the prediction of in situ N mineralization rates. This was
designed by considering successively variable categories ordered
conducted on the data subset 47 in which all these variables had
by increasing acquisition complexity. SON being the most corre-
been measured. Since the whole dataset (n 65) had been
lated variable with Vp, a linear relationship was the rst function
52 H. Clivot et al. / Soil Biology & Biochemistry 111 (2017) 44e59

Table 4
Prediction functions, model parameters and evolution of statistical evaluation criteria of predictive models during the modeling approaches.

Step Dataset Explanatory Generic model Introduced function Optimized/xed parameter r2 EF AIC Bias RMSE RMSEP
variable values

i n Vi bp V
V b p $f V fi Vi kg N ha1 nday1
i i1 i i

A. Modeling approach by considering successively each variable category


A1. Soil model
A11 65 SON Vb p f SON f1 SON a$SON a 0.26 0.18 160 0.03 0.29 0.29
1 1
0.130
A12 65 Clay bp V
V b p $f Clay f2 Clay expb$Clay1000 a b 0.30 0.22 161 0.03 0.28 0.29
2 1 2
0.158 1.009
A13 65 pH bp V
V b p $f pH f3 pH exp  c$pH  8:52 a b c 0.43 0.43 179 0.00 0.24 0.26
3 2 3
0.279 2.770 0.089
A14 65 C/N bp V
V b p $f C=N f4 C=N 0:8$exp  d$CN  112 0:2 a b c d 0.56 0.56 195 0.00 0.21 0.23
4 3 4
0.330 2.732 0.099 0.055
A1 5 65 CaCO3 bp V
V b p $f CaCO f5 Ca 1=1 e$CaCO3 1000 a b c d e 0.61 0.61 199 0.00 0.20 0.22
5 4 5 3
0.346 2.519 0.112 0.060 1.114
47 bp
V 0.55 0.55 138 0.01 0.21 0.23
5

A2. Soil-history model


A26 65 PCIRa bp V
V b p5 $f PCIRa f6 PCIRa 1 f $PCIRa a f b, c, d, e 0.70 0.70 215 0.00 0.18 0.22
6 6
0.327 0.516 xed
A2 7 65 LegF bp V
V b p $f LegF f7 LegF 1  g$LegF a f g b, c, d, e 0.72 0.72 218 0.00 0.17 0.19
7 6 7
0.341 0.497 0.494 xed
47 bp
V 0.75 0.75 162 0.01 0.15 0.18
7

A3. Soil-history-biological model


A3 8 47 Bio% bp V
V b p $f Bio% f8 Bio 0:8 h$Bio% a h b, c, d, e, f, g 0.77 0.77 164 0.00 0.15 0.17
8 7 8
0.333 14.77 xed

B. Modeling approach by considering simultaneously all variable categories


B1 47 EOM-N Vb p f EOMN f1 EOMN i$EOMN i 0.42 0.30 126 0.05 0.26 0.26
1 1
0.549
B2 47 PCIRa bp V
V b p $f PCIRa f2 PCIRa 1 j$PCIRa i j 0.61 0.55 145 0.04 0.21 0.21
2 1 2
0.495 0.886
B3 47 EOM-C bp V
V b p $f EOMC f3 EOMC 1  k$EOMC i j k 0.60 0.59 147 0.01 0.20 0.21
3 2 3
0.591 0.749 0.011

SON: soil organic nitrogen, PCIRa: Preceding Crop Index Rapeseed, LegF: Legume Frequency, Bio%: microbial biomass % (relative to soil organic carbon), EOM: extractable
organic matter. Final models obtained with the different approaches are indicated in bold.

intentionally used to calibrate the soil and soil-history models in step (B3) explained 17% less variance of Vp than the soil-history-
order to obtain a more robust estimation of parameters, both biological model obtained with the approach A. We also found
models were rst evaluated against the data subset 47. They per- that integrating microbial biomass instead of SON or EOM-N at the
formed almost as well for the subset as for the complete dataset, beginning of the modeling process did not allow to obtain a better
showing a relatively good representativeness of this subset: the predictive model (data not shown).
model efciency was 0.55 vs 0.61 for the soil model and 0.75 vs 0.72 Comparisons between novel models obtained with the
for the soil-history model. approach A and MLR modeling approach using the same explana-
Although Vp was signicantly and positively correlated with the tory variables revealed that MLR procedure explained between 10
different organic matter fractions (POM and EOM, -C and -N), these and 15% less variance of Vp than the models designed with a
variables did not explain much additional variance of Vp and did multiplicative structure. Similar (i.e. negative or positive) effects of
not further decrease the AIC criterion. This result is explained by explanatory variables on Vp were nevertheless found for both types
the strong correlation between these organic fractions and SON. of models. We conclude that models developed with approach A
The introduction of microbial biomass slightly improved the pre- perform better than those obtained with MLR or B approach.
vious models. Adding a linear function of the percentage of mi-
crobial biomass relative to SOC raised the explained variance of Vp
to 77% for the data subset 47 in the soil-history-biological model 3.4. Evaluation of the predictive models
(Table 4, step A38).
The very small variance explained by the microbial biomass at The predicted Vp values are displayed vs observed values for the
nal step might result from the order of introducing variables. This three models developed with the approach A in Fig. 4. The close-
hypothesis was tested using the approach B in which all potential ness of t to the 1:1 line conrms the absence of bias for all novel
explanatory variables were not a priori categorized. Using the data proposed models. The predictive quality of the models, evaluated
subset 47, EOM-N was found to be the most correlated variable with by the method of cross-validation over the whole dataset (n 65),
Vp and rst introduced in the B model (step B1, Table 4). Including a provided RMSEP estimates of 0.22 and 0.19 kg N ha1 nday1 for the
function of PCI Rapeseed at step B2 allowed to explain 19% addi- soil and soil-history models, respectively. The range of Vp values,
tional variance of Vp and increased model efciency from 0.30 to both observed and predicted, is as important in the data subset 47
0.55. Thirdly, a negative linear function of EOM-C was introduced, as in the whole dataset, justifying the relevance of this subset to
increasing slightly the model efciency to 0.59 and decreasing the compare the soil-history-biological model with the other two
bias from 0.04 to 0.01. The predictive model obtained at this third models. RMSEP estimate for the soil-history-biological was
0.17 kg N ha1 nday1. It is slightly smaller than the RMSEP
H. Clivot et al. / Soil Biology & Biochemistry 111 (2017) 44e59 53

Fig. 3. Representation of linear and non-linear functions introduced in the predictive soil model of Vp with their nal parameter values. Points correspond to calculated values by
each function for the basic soil parameters of the 65 eld-experiments. Function f1 vs SON was normalized between 0 and 1 in order to compare it with the other functions f2 to f5
already normalized.

calculated over the same data subset (n 47) for the soil and soil- reliable estimates of NNM under eld conditions. Our study was
history models, which were 0.23 and 0.18 kg N ha1 nday1 conducted in the framework of arable cropping systems in
respectively. The RMSEP estimates correspond to a relative model conventional-tillage. For these systems, the actual NNM rates
error comprised between 26 and 30% of the average Vp value. calculated by LIXIM, varying between 0.13 and 1.10 kg N ha1 day1,
The ANOVA-based sensitivity analyses of the three models ob- are lower but comparable to those estimated from in situ core
tained with the approach A were conducted by comparing the methods in grasslands: 0.20e2.30 kg N ha1 day1 as compiled in
contributions of explanatory factors to the variability of predicted Jarvis et al. (1996) and 0.50e1.50 kg N ha1 day1 measured by
potential NNM rates. The sensitivity indices (SI) relative to the main Hatch et al. (2000). The range of our results is very similar to that
effects and the two factor interactions are represented in Fig. 5. obtained in Sweden (0.10e0.93 kg N ha1 day1) by Delin and
Small differences were observed between models. The most Linden (2002) who used a N balance approach during a 3-year
inuent factor for the three models was SON which contributed to experiment within a heterogeneous arable eld.
0.31e0.36 as main effect SI and 0.10e0.11 as interaction SI. It was Field estimates of NNM, which are essential for N management,
followed by clay, pH and C/N ratio, which accounted for 0.11e0.18 as can be strongly affected by climatic conditions and particularly by
main effect and 0.04e0.07 as interaction SI, for each variable. The soil moisture and temperature (Cassman and Munns, 1980). Cor-
other variables had less inuence on model outputs, both for their rections for these environmental factors have been performed by
main and interactive effects. LIXIM to produce potential NNM rates. The calculation of these
potential rates is required to disentangle the role of other driving
factors of N mineralization and obtain NNM estimates comparable
4. Discussion
to results obtained under controlled conditions. This provides op-
portunities to compare our predictive models to those designed in
4.1. Reliability of N mineralization estimates and validity of
laboratory incubation studies and also to test their predicting per-
predictive models formance on previously published datasets, which mainly derived
from in vitro experiments.
In this study, we chose to use an alternative method to assess in In term of performance, the proposed soil model is compa-
situ net N mineralization in arable elds by combining a monitoring 
rable to the prediction function developed by Dessureault-Rompre
of soil water and mineral N contents and a calculation model. LIXIM et al. (2015, 2016) which include three soil parameters (soil total N,
model was able to reproduce satisfactorily the evolution of water
a labile mineralized N pool and soil pH): their function accounted
and mineral N contents in the soil prole, therefore providing
54 H. Clivot et al. / Soil Biology & Biochemistry 111 (2017) 44e59

Fig. 5. Main effect and interaction sensitivity indices (SI) based on the analysis of
variance of the Vp simulations made by the three models designed with the approach
A on randomly generated datasets of explanatory variables.

single factor. The data obtained during a 29 week incubation made


by Springob and Kirchmann (2003) were also used to assess the
ability of our model to predict NNM in sandy soils with a high C/N
ratio (up to 30). The soil model explained 30% (for 23 soils) and 48%
(when excluding one outlier) of the variance of NNM rates
measured in this incubation study. Therefore, our model produced
consistent predictions of NNM rates measured in varied soils and
laboratory conditions. Further studies would be needed to obtain
extended datasets of eld experiments to perform an independent
Fig. 4. Observed vs predicted potential net N mineralization rates (Vp) for the three validation of the proposed predictive models. However, this type of
predictive models designed with the approach A. Black and grey circles correspond to study is costly and time-consuming, explaining why so few refer-
the whole dataset; grey circles correspond to the data subset 47 alone.
ences are available under in situ conditions.
In the following of this discussion, the effects of the different
for 67% of the variance in the rate constant of the stable N pool factors that could inuence N mineralization are discussed in the
whereas our soil model could explain 61% of the Vp variance. We light of results obtained in this work.
applied the soil model to the data reported by Schomberg et al.
(2009) obtained from a 41 week incubation and found that it was 4.2. Basic soil parameters inuencing in situ N mineralization
able to explain 59% of the variance of measured NNM rates. Even
though total N alone could account for the same level of variance, Results obtained showed that the most inuencing factor of in
our model did not deteriorate the prediction compared to this situ NNM rates was the SON stock. The strong correlation observed
between both variables is consistent with the majority of soil
H. Clivot et al. / Soil Biology & Biochemistry 111 (2017) 44e59 55

incubation studies made in various ecosystems demonstrating that that pH was the main factor positively correlated with gross N
SON is a good predictor of NNM (e.g. Dessureault-Rompre  et al., mineralization rates.
2015; McDonald et al., 2014; Schomberg et al., 2009) and even The C/N ratio of SOM was the fourth inuent driving factor in
gross N mineralization (Booth et al., 2005). The sensitivity analyses the soil model. A gaussian function with a maximum value at C/
of factors inuencing model outputs highlight the need for an ac- N 11 was found to be relevant to describe its effect on N miner-
curate characterization of soil parameters, and particularly SON alization rates. We hypothesize that the predicted decrease in NNM
stocks. Therefore, a particular attention must be paid to the rate when the C/N ratio decreases from 11 to 7 could be attributed
assessment of soil bulk density and the choice of depth (Z) used to to a shift in SOM quality towards more highly processed SOM
calculate SON stocks. In this study, the sensitivity analysis per- such as that present in subsoil (Jenkinson et al., 2008; Rumpel and
formed with LIXIM on Zm revealed that the ploughing depth Kogel-Knabner, 2010) with lower mineralization rates. Conversely,
appeared to be the most appropriate value of Z. This conclusion in soils with high C/N ratio (greater than 11), N availability could
should be recommended to regularly ploughed soils. Results ob- become a limiting factor for SOM decomposers and NNM rates
tained in reduced tillage systems suggest that N mineralization could decrease due to higher microbial N immobilization. This
decreases with depth but remains signicant down to the old result is consistent with a previous work which showed a negative
ploughing depth (e.g. Oorts et al., 2007). Therefore, in these systems exponential relationship between NNM and the C/N ratio of soils
Z should be greater than the current tillage depth and optimally whose C/N ratios varied between 10 and 30 (Springob and
equal to the past maximum tillage depth. Kirchmann, 2003).
Among important soil parameters, clay concentration has been
identied as driving factor of SOM decay rates (Adu and Oades, 4.3. Contribution of additional factors to improve N mineralization
1978; Hassink et al., 1993; Kleber et al., 2007) and NNM rates prediction
(Colman and Schimel, 2013; Delin and Linde n, 2002). Simulation
models of SOM turnover such as CENTURY (Parton et al., 1987; Previous studies demonstrated the inuence of crop rotation on
Parton, 1996), DAISY (Hansen et al., 1991), ROTHC (Coleman and N mineralization (Carpenter-Boggs et al., 2000; Senwo and
Jenkinson, 1996) or STICS (Brisson et al., 2003) include equations Tabatabai, 2005; Shari et al., 2008). Integrating the inuence of
reecting a negative effect of clay content (or clay silt content) on agricultural practices into models in combination with climatic and
these rates. The clay function introduced in the soil model is similar soil predictors could help to better predict N supply for agro-
to that used in a previous study simulating the consequences of ecosystems (St. Luce et al., 2011). In our study, cropping history
straw residues export on SOC stocks, relating carbon mineralization variables allowed to explain additional Vp variance. The positive
rate and clay content (Safh-Hdadi and Mary, 2008). The optimized correlation observed between Vp and the frequency of cereals in
parameter at the end of our modeling process (b 2.52) is close to the rotation could result from the larger amounts of crop residues
that obtained in the latter study (between 2.00 and 2.68, depending (straw) returned to soils on the long term in these rotations. Indeed,
on the stable pool size). Powlson et al. (1987) showed that the incorporation of straw exerts
To a lesser extent than clay, negative effects of CaCO3 content long-term effects on SOM and microbial biomass, increasing the
have been evidenced on potential NNM rates. The decreasing quantity of mineralizable N in soils. Surprisingly, the frequency of
relationship could be explained by the protective effects of ne cereals did not improve our predictive model when basic soil pa-
CaCO3 on SOM mineralization (Delphin et al., 1986), CaCO3 acting as rameters were already introduced. This could result from correla-
a stabilizing agent (Duchaufour, 1976). tions between explanatory variables since crop types and rotations
The proposed soil model predicts that soil pH inuences N are more or less specic to pedo-climatic conditions. Integrating
mineralization rates. It simulates that NNM rates increased with pH soil parameters at rst in our modeling approach could have helped
over the range of the present study (5.7e8.4) but the gaussian to avoid at least partly these confounding effects.
function chosen allows to simulate a decrease in N mineralization The proposed soil-history model identied a positive effect of
at very high pH, shown in some studies such as Pathak and Rao rapeseed frequency on N mineralization. This short term effect,
(1998). Surprisingly, there was no signicant simple correlation which concerned the last two years before measurements, might be
between NNM and pH in our study because pH inuence is prob- related to the large amounts of leaves produced by rapeseed and
ably masked by the effects of other correlated factors. This apparent falling during its growth period (Justes et al., 2000). These leaves
contradiction may explain why controversial results have been decompose rapidly at soil surface and could enhance soil N
reported in the literature about the specic role of soil pH on NNM. mineralization after rapeseed harvest. Conversely, the model
Curtin et al. (1998) did not observe signicant relationship between pointed out a slight negative effect of legume frequency on Vp.
soil pH and N mineralization in a 24 week incubation study per- Plaza-Bonilla et al. (2016) showed that cropping grain legumes in
formed on 61 soils from Canada. However, after manipulating soil rotations can lead within a few years to signicant losses of SOC
pH by adding Ca(OH)2 in two soils, they found that raising the pH and SON. They showed that increased legume frequency reduced
from 5.7 to 7.4 linearly increased C and N mineralization measured carbon inputs to soil. These smaller inputs could result in a further
during a 100-day incubation. Bertrand et al. (2007) found apparent depletion of the soil mineralizable N pool on the long-term and a
but not true effects of pH on C and N mineralization in an acidic soil reduction of the specic N mineralization rate (i.e. per unit of SON).
which had been limed 20 years earlier whereas Andersson and Organic matter fractions and microbial biomass did not appear
Nilsson (2001) observed a stimulation of C mineralization with to greatly improve the prediction of Vp when parameters such as
liming. Raising up soil pH with liming could increase both the SON were previously introduced in the model. A similar conclusion
amount of negatively-charged groups on the humus colloids and was suggested by Ros et al. (2011) who showed that extractable
the SOM solubility (Andersson et al., 2000). Kemmitt et al. (2006) organic N, although well correlated with mineralizable N, did not
analyzed two long-term eld experiments in which pH had been explain additional variance compared to total N. The extraction
manipulated. They concluded that changes in soil pH signicantly conditions can strongly inuence the amount and characteristics of
affected soil microbial activity and the rate of soil C and N cycling. the analyzed organic nitrogen pools. In our study, the high tem-
Aciego Pietri and Brookes (2008) also found that pH can markedly perature pretreatment used for EOM analysis could have extracted
inuence soil microbial biomass and activities. Cookson et al. readily decomposable fractions but also solubilized recalcitrant
(2007) compared ve acidic soils varying in land use and found compounds (Ros et al., 2009), so that EOM fractions could reect
56 H. Clivot et al. / Soil Biology & Biochemistry 111 (2017) 44e59

the quantity of SOM rather than a specic bioavailable fraction (Ros, N contents in bare soils from 65 eld-experiments representative of
2012). French arable situations. This calculation model provided reliable
The small explanatory power of microbial biomass on Vp vari- estimates of in situ NNM rates under actual eld conditions in
ation observed in our study might seem paradoxical. However it is arable conventional-till cropping systems. The models predicting in
consistent with the conclusion of Kemmitt et al. (2008) who situ NNM rates proposed in this paper revealed that basic soil
demonstrated that the mineralization of humied organic matter characteristics can explain 61% of the variance in potential NNM
was not inuenced by the size, activity or composition of the mi- rates. The subsequent introduction of two variables relative to crop
crobial biomass. These authors suggested that SOM mineralization type frequency (i.e. rapeseed and legume) helped to improve the
was primarily governed by abiotic processes controlling its acces- explanatory and predictive qualities of the soil model, explaining
sibility to microorganisms and introduced the Regulatory Gate 72% of the variance in Vp. Integrating variables more complex to
hypothesis. Our results could support this hypothesis, even though acquire such as organic matter fractions and microbial biomass
the importance of biotic regulation of SOM mineralization is still hardly improved the model performance, when soil parameters
debated (Kuzyakov et al., 2009; Paterson, 2009). The results of Birge such as SON had been previously introduced. Furthermore, these
et al. (2015) also suggest that the amount of available SOM is the variables did not allow to obtain a better prediction when they
limiting factor of SOM mineralization rather than the size of were introduced rst in the model. The soil-history model (A2)
enzyme pool or microbial biomass. Investigating the active portion provided the best compromise between efciency and complexity.
of soil microorganisms instead of total microbial biomass could be a Each of the three proposed models (A1eA3) was developed with
best way forward to link microbial parameters and soil functioning variables that are routinely measured or easily accessible with,
(Blagodatskaya and Kuzyakov, 2013). Fujita et al. (2014) also maybe, the exception of soil microbial biomass. These operational
showed that an improved prediction of N mineralization can be models could then be implemented in multidisciplinary models in
obtained by combining microbial biomass measurements and C/N order to improve their capacity to simulate N dynamics in soils and
stoichiometry effects in a mechanistic and more complex model help N fertilization management.
than that developed here.
Acknowledgements
4.4. Relevance of the modeling approach
We wish to acknowledge the technical staff of ARVALIS as well
Microbial biomass and organic matter fractions were shown to as Francis Barois, Eric Bazerthe, Didier Chesneau, Thierry Morvan,
be well correlated with NNM rates. However, they did not add to Pierre Mortreux and Jean-Marie Nolot for their contribution to eld
the prediction of Vp when more simple parameters associated with experiments. We thank Patrick Petibon for soil analysis. This
SOM (e.g. SON), which are less costly and time-consuming to research was supported by ARVALIS, CETIOM (Terres Inovia) and
determine, were integrated rst in the predictive model. INRA. This work was performed, in partnership with the SAS PIV-
Comparing the two multiplicative approaches (A and B) empha- ERT, within the frame of the French Institute for the Energy Tran-
sized the importance of the order of introducing variables on the sition PIVERT (www.institut-pivert.com). This work was supported,
selected variables and the model performance. In model B, SON was as part of the Investments for the Future, by the French Govern-
replaced by EOM-N (both highly correlated, r 0.92) as the main ment under the reference ANR-001-01, and was funded within the
factor, but the predictive quality of model B remained smaller than frame of the Genesys WP1 P13 Solebiom project. We also thank the
that of model A. We cannot exclude that the lower number of anonymous reviewers for their constructive evaluation of the
samples in the data subset 47 could have inuenced the results, manuscript.
even though this subset was representative of the whole dataset.
The predictive models designed with a multiplicative structure
Appendix A. Supplementary data
were shown to provide better explanation of NNM rates than ad-
ditive linear models with the same explanatory variables. This
Supplementary data related to this article can be found at http://
conrms our assumption that a model with a multiplicative
dx.doi.org/10.1016/j.soilbio.2017.03.010.
structure reects the nature of the interactive effects of factors
inuencing N mineralization better than additive models. The non-
linear functions considered in the multiplicative models also References
contributed to improve NNM rates prediction, supporting the hy-
Abril, A., Caucas, V., Bucher, E.H., 2001. Reliability of the in situ incubation methods
pothesis that many soil processes are better represented by non- used to assess nitrogen mineralization: a microbiological perspective. Applied
linear models (Archontoulis and Miguez, 2015). Our attempt was Soil Ecology 17, 125e130. http://dx.doi.org/10.1016/S0929-1393(01)00128-7.
not to design a process-based model, but to identify the main Aciego Pietri, J.C., Brookes, P.C., 2008. Relationships between soil pH and microbial
properties in a UK arable soil. Soil Biology and Biochemistry 40, 1856e1861.
driving factors of N mineralization and determine their response Adams, M.A., Attiwill, P.M., 1986. Nutrient cycling and nitrogen mineralization in
functions. eucalypt forests of south-eastern Australia. Plant and Soil 92, 341e362. http://
In a functional interpretation of the soil-history-biological dx.doi.org/10.1007/BF02372483.
Adams, M.A., Polglase, P.J., Attiwill, P.M., Weston, C.J., 1989. In situ studies of ni-
model obtained here, N mineralization rate is the result of SOM trogen mineralization and uptake in forest soils; some comments on method-
decomposition process under the inuence of: i) the amount of ology. Soil Biology and Biochemistry 21, 423e429. http://dx.doi.org/10.1016/
substrate (SON) and its quality (C/N), ii) edaphic factors that affect 0038-0717(89)90154-5.
Adu, J.K., Oades, J.M., 1978. Physical factors inuencing decomposition of organic
SOM physico-chemical protection (clay, CaCO3) or the activity of materials in soil aggregates. Soil Biology and Biochemistry 10, 109e115. http://
microorganisms (pH), iii) the specic effect of some crop residues dx.doi.org/10.1016/0038-0717(78)90080-9.
returned to soil (rapeseed) and iv) soil microorganisms which are Andersson, S., Nilsson, S.I., 2001. Inuence of pH and temperature on microbial
activity, substrate availability of soil-solution bacteria and leaching of dissolved
the major actors of SOM decomposition.
organic carbon in a mor humus. Soil Biology and Biochemistry 33, 1181e1191.
http://dx.doi.org/10.1016/S0038-0717(01)00022-0.
5. Conclusions Andersson, S., Nilsson, S.I., Saetre, P., 2000. Leaching of dissolved organic carbon
(DOC) and dissolved organic nitrogen (DON) in mor humus as affected by
temperature and pH. Soil Biology and Biochemistry 32, 1e10. http://dx.doi.org/
This study showed that the calculation model LIXIM was able to 10.1016/S0038-0717(99)00103-0.
satisfactorily represent changes through time of water and mineral Archontoulis, S.V., Miguez, F.E., 2015. Nonlinear regression models and applications
H. Clivot et al. / Soil Biology & Biochemistry 111 (2017) 44e59 57

in agricultural research. Agronomy Journal 107, 786e798. http://dx.doi.org/ 63e75. http://dx.doi.org/10.4141/cjss-2014-057.


10.2134/agronj2012.0506. Dessureault-Rompre , J., Zebarth, B.J., Burton, D.L., Grant, C.A., 2016. Depth distri-
Bartholomew, W.V., 1965. Mineralization and Immobilization of Nitrogen in the bution of mineralizable nitrogen pools in contrasting soils in a semi-arid
Decomposition of Plant and Animal Residues. Soil Nitrogen. Amer. Soc. Agron. climate. Canadian Journal of Soil Science 96, 1e11. http://dx.doi.org/10.1139/
Inc., Madison, Wis, pp. 285e306. cjss-2015-0048.
Beaudoin, N., Saad, J.K., Van Laethem, C., Machet, J.M., Maucorps, J., Mary, B., 2005. Dimassi, B., Mary, B., Wylleman, R., Labreuche, J., Couture, D., Piraux, F., Cohan, J.P.,
Nitrate leaching in intensive agriculture in Northern France: effect of farming 2014. Long-term effect of contrasted tillage and crop management on soil
practices, soils and crop rotations. Agriculture, Ecosystems & Environment 111, carbon dynamics during 41 years. Agriculture, Ecosystems & Environment 188,
292e310. http://dx.doi.org/10.1016/j.agee.2005.06.006. 134e146. http://dx.doi.org/10.1016/j.agee.2014.02.014.
Benbi, D.K., Richter, J., 2002. A critical review of some approaches to modelling DiStefano, J.F., Gholz, H.L., 1986. A proposed use of ion exchange resins to measure
nitrogen mineralization. Biology and Fertility of Soils 35, 168e183. http:// nitrogen mineralization and nitrication in intact soil cores. Communications in
dx.doi.org/10.1007/s00374-002-0456-6. Soil Science and Plant Analysis 17, 989e998. http://dx.doi.org/10.1080/
Bertrand, I., Delfosse, O., Mary, B., 2007. Carbon and nitrogen mineralization in 00103628609367767.
acidic, limed and calcareous agricultural soils: apparent and actual effects. Soil Doran, J.W., Parkin, T.B., 1994. Dening and Assessing Soil Quality. Dening Soil
Biology and Biochemistry 39, 276e288. http://dx.doi.org/10.1016/ Quality for a Sustainable Environment. sssaspecialpubl, pp. 1e21. http://
j.soilbio.2006.07.016. dx.doi.org/10.2136/sssaspecpub35.c1.
Birge, H.E., Conant, R.T., Follett, R.F., Haddix, M.L., Morris, S.J., Snapp, S.S., Duchaufour, P., 1976. Dynamics of organic matter in soils of temperate regions: its
Wallenstein, M.D., Paul, E.A., 2015. Soil respiration is not limited by reductions action on pedogenesis. Geoderma 15, 31e40. http://dx.doi.org/10.1016/0016-
in microbial biomass during long-term soil incubations. Soil Biology and 7061(76)90068-9.
Biochemistry 81, 304e310. http://dx.doi.org/10.1016/j.soilbio.2014.11.028. Eno, C.F., 1960. Nitrate production in the eld by incubating the soil in polyethylene
Blagodatskaya, E., Kuzyakov, Y., 2013. Active microorganisms in soil: critical review bags. Soil Science Society of America Journal 24, 277. http://dx.doi.org/10.2136/
of estimation criteria and approaches. Soil Biology and Biochemistry 67, sssaj1960.03615995002400040019x.
192e211. http://dx.doi.org/10.1016/j.soilbio.2013.08.024. Fujita, Y., Witte, J.-P.M., van Bodegom, P.M., 2014. Incorporating microbial ecology
Blake, G.R., 1965. Bulk density. Methods of soil analysis. Part 1. Physical and concepts into global soil mineralization models to improve predictions of car-
Mineralogical Properties, Including Statistics of Measurement and Sampling bon and nitrogen uxes. Global Biogeochemical Cycles 28. http://dx.doi.org/
374e390. 10.1002/2013GB004595, 2013GB004595.
Booth, M.S., Stark, J.M., Rastetter, E., 2005. Controls on nitrogen cycling in terrestrial Garnier, P., Ne el, C., Aita, C., Recous, S., Lafolie, F., Mary, B., 2003. Modelling carbon
ecosystems: a synthetic analysis of literature data. Ecological Monographs 75, and nitrogen dynamics in a bare soil with and without straw incorporation.
139e157. http://dx.doi.org/10.1890/04-0988. European Journal of Soil Science 54, 555e568. http://dx.doi.org/10.1046/j.1365-
Brisson, N., Gary, C., Justes, E., Roche, R., Mary, B., Ripoche, D., Zimmer, D., Sierra, J., 2389.2003.00499.x.
Bertuzzi, P., Burger, P., Bussie re, F., Cabidoche, Y.M., Cellier, P., Debaeke, P., Grifn, T.S., 2008. Nitrogen availability. Nitrogen in Agricultural Systems Agrono-
Gaudille re, J.P., Henault, C., Maraux, F., Seguin, B., Sinoquet, H., 2003. An over- mymonogra 613e646. http://dx.doi.org/10.2134/agronmonogr49.c15.
view of the crop model stics. European Journal of Agronomy, Modelling Crop- Guntin ~ as, M.E., Leiros, M.C., Trasar-Cepeda, C., Gil-Sotres, F., 2012. Effects of mois-
ping Systems: Science, Software and Applications 18, 309e332. http:// ture and temperature on net soil nitrogen mineralization: a laboratory study.
dx.doi.org/10.1016/S1161-0301(02)00110-7. European Journal of Soil Biology 48, 73e80. http://dx.doi.org/10.1016/
Cabrera, M.L., Kissel, D.E., 1988. Evaluation of a method to predict nitrogen j.ejsobi.2011.07.015.
mineralized from soil organic matter under eld conditions. Soil Science Soci- Hanselman, T.A., Graetz, D.A., Obreza, T.A., 2004. A comparison of in situ methods
ety of America Journal 52, 1027. http://dx.doi.org/10.2136/ for measuring net nitrogen mineralization rates of organic soil amendments.
sssaj1988.03615995005200040024x. Journal of Environment Quality 33, 1098. http://dx.doi.org/10.2134/
Carpenter-Boggs, L., Pikul, J.L., Vigil, M.F., Riedell, W.E., 2000. Soil nitrogen miner- jeq2004.1098.
alization inuenced by crop rotation and nitrogen fertilization. Soil Science Hansen, S., Jensen, H.E., Nielsen, N.E., Svendsen, H., 1991. Simulation of nitrogen
Society of America Journal 64, 2038. http://dx.doi.org/10.2136/ dynamics and biomass production in winter wheat using the Danish simulation
sssaj2000.6462038x. model DAISY. Fertilizer Research 27, 245e259. http://dx.doi.org/10.1007/
Cassman, K.G., Munns, D.N., 1980. Nitrogen mineralization as affected by soil BF01051131.
moisture, temperature, and Depth1. Soil Science Society of America Journal 44, Hart, S.C., Binkley, D., 1985. Correlations among indices of forest soil nutrient
1233. http://dx.doi.org/10.2136/sssaj1980.03615995004400060020x. availability in fertilized and unfertilized loblolly pine plantations. Plant and Soil
Clivot, H., Mary, B., Vale , M., Cohan, J.P., Laurent, F., Champolivier, L., Mortreux, P., 85, 11e21. http://dx.doi.org/10.1007/BF02197796.
Morvan, T.,Justes, E., Submitted. Soil-N-Min_Fr-Dataset: A dataset of in situ soil Hassink, J., Bouwman, L.A., Zwart, K.B., Bloem, J., Brussaard, L., 1993. Relationships
N mineralization in arable cropping systems in France. Submitted. between soil texture, physical protection of organic matter, soil biota, and c and
Coleman, K., Jenkinson, D.S., 1996. RothC-26.3-A Model for the turnover of carbon in n mineralization in grassland soils. Geoderma, Soil Structure/Soil Biota In-
soil. In: Powlson, D.S., Smith, P., Smith, J.U. (Eds.), Evaluation of Soil Organic terrelationships 57, 105e128. http://dx.doi.org/10.1016/0016-7061(93)90150-J.
Matter Models, NATO ASI Series. Springer Berlin Heidelberg, pp. 237e246. Hatch, D.J., Bhogal, A., Lovell, R.D., Shepherd, M.A., Jarvis, S.C., 2000. Comparison of
http://dx.doi.org/10.1007/978-3-642-61094-3_17. different methodologies for eld measurement of net nitrogen mineralization
Colman, B.P., Schimel, J.P., 2013. Drivers of microbial respiration and net N miner- in pasture soils under different soil conditions. Biology and Fertility of Soils 32,
alization at the continental scale. Soil Biology and Biochemistry 60, 65e76. 287e293. http://dx.doi.org/10.1007/s003740000250.
http://dx.doi.org/10.1016/j.soilbio.2013.01.003. Honeycutt, C.W., 1999. Nitrogen mineralization from soil organic matter and crop
Connell, M.J., Raison, R.J., Khanna, P.K., 1995. Nitrogen mineralization in relation to residues: eld validation of laboratory predictions. Soil Science Society of
site history and soil properties for a range of Australian forest soils. Biology and America Journal 63, 134. http://dx.doi.org/10.2136/
Fertility of Soils 20, 213e220. http://dx.doi.org/10.1007/BF00336080. sssaj1999.03615995006300010020x.
Constantin, J., Beaudoin, N., Laurent, F., Cohan, J.P., Duyme, F., Mary, B., 2011. Cu- Jansson, S.L., Persson, J., 1982. Mineralization and immobilization of soil nitrogen.
mulative effects of catch crops on nitrogen uptake, leaching and net minerali- Nitrogen in Agricultural Soils agronomymonogra 229e252. http://dx.doi.org/
zation. Plant and Soil 341, 137e154. http://dx.doi.org/10.1007/s11104-010-0630- 10.2134/agronmonogr22.c6.
9. Jarvis, S.C., Stockdale, E.A., Shepherd, M.A., Powlson, D.S., 1996. Nitrogen minerali-
Cookson, W.R., Osman, M., Marschner, P., Abaye, D.A., Clark, I., Murphy, D.V., zation in temperate agricultural soils: processes and measurement. Advances in
Stockdale, E.A., Watson, C.A., 2007. Controls on soil nitrogen cycling and mi- Agronomy 57, 187e235.
crobial community composition across land use and incubation temperature. Jenkinson, D.S., Poulton, P.R., Bryant, C., 2008. The turnover of organic carbon in
Soil Biology and Biochemistry 39, 744e756. http://dx.doi.org/10.1016/ subsoils. Part 1. Natural and bomb radiocarbon in soil proles from the Roth-
j.soilbio.2006.09.022. amsted long-term eld experiments. European Journal of Soil Science 59,
Curtin, D., Beare, M.H., Hernandez-Ramirez, G., 2012. Temperature and moisture 391e399. http://dx.doi.org/10.1111/j.1365-2389.2008.01025.x.
effects on microbial biomass and soil organic matter mineralization. Soil Sci- Joergensen, R.G., 1996. The fumigation-extraction method to estimate soil microbial
ence Society of America Journal 76, 2055e2067. http://dx.doi.org/10.2136/ biomass: calibration of the kEC value. Soil Biology and Biochemistry 28, 25e31.
sssaj2012.0011. http://dx.doi.org/10.1016/0038-0717(95)00102-6.
Curtin, D., Campbell, C.A., Jalil, A., 1998. Effects of acidity on mineralization: pH- Johnson, A.D., Cabrera, M.L., McCracken, D.V., Radcliffe, D.E., 1999. LEACHN simu-
dependence of organic matter mineralization in weakly acidic soils. Soil lations of nitrogen dynamics and water drainage in an ultisol. Agronomy
Biology and Biochemistry 30, 57e64. http://dx.doi.org/10.1016/S0038-0717(97) Journal 91, 597. http://dx.doi.org/10.2134/agronj1999.914597x.
00094-1. Justes, E., Denoroy, P., Gabrielle, B., Gosse, G., 2000. Effect of crop nitrogen status
Delin, S., Linden, B., 2002. Relations between net nitrogen mineralization and soil and temperature on the radiation use efciency of winter oilseed rape. Euro-
characteristics within an arable eld. Acta Agriculturae Scandinavica, Section B pean Journal of Agronomy 13, 165e177. http://dx.doi.org/10.1016/S1161-
d Soil & Plant Science 52, 78e85. http://dx.doi.org/10.1080/ 0301(00)00072-1.
090647102321089819. Kemmitt, S.J., Lanyon, C.V., Waite, I.S., Wen, Q., Addiscott, T.M., Bird, N.R.A.,
Delphin, J.E., Huck, C., Tiquet, J.L., 1986. Evaluation du pouvoir mine ralisateur de sols O'Donnell, A.G., Brookes, P.C., 2008. Mineralization of native soil organic matter
agricoles en fonction de leurs caracte ristiques physico-chimiques. Agronomie 6, is not regulated by the size, activity or composition of the soil microbial bio-
453e458. http://dx.doi.org/10.1051/agro:19860505. massea new perspective. Soil Biology and Biochemistry 40, 61e73. http://
Dessureault-Rompre , J., Zebarth, B.J., Burton, D.L., Georgallas, A., 2015. Predicting dx.doi.org/10.1016/j.soilbio.2007.06.021.
soil nitrogen supply from soil properties. Canadian Journal of Soil Science 95, Kemmitt, S.J., Wright, D., Goulding, K.W.T., Jones, D.L., 2006. pH regulation of carbon
58 H. Clivot et al. / Soil Biology & Biochemistry 111 (2017) 44e59

and nitrogen dynamics in two agricultural soils. Soil Biology and Biochemistry Ros, G.H., Temminghoff, E.J.M., Hofand, E., 2011. Nitrogen mineralization: a review
38, 898e911. http://dx.doi.org/10.1016/j.soilbio.2005.08.006. and meta-analysis of the predictive value of soil tests. European Journal of Soil
Khanna, P.K., Raison, R.J., 2013. In situ core methods for estimating soil mineral-N Science 62, 162e173. http://dx.doi.org/10.1111/j.1365-2389.2010.01318.x.
uxes: Re-evaluation based on 25 years of application and experience. Soil Rumpel, C., Ko gel-Knabner, I., 2010. Deep soil organic matterda key but poorly
Biology and Biochemistry 64, 203e210. http://dx.doi.org/10.1016/ understood component of terrestrial C cycle. Plant and Soil 338, 143e158.
j.soilbio.2012.09.004. http://dx.doi.org/10.1007/s11104-010-0391-5.
Kleber, M., Sollins, P., Sutton, R., 2007. A conceptual model of organo-mineral in- Safh-Hdadi, K., Mary, B., 2008. Modeling consequences of straw residues export on
teractions in soils: self-assembly of organic molecular fragments into zonal soil organic carbon. Soil Biology and Biochemistry 40, 594e607. http://
structures on mineral surfaces. Biogeochemistry 85, 9e24. http://dx.doi.org/ dx.doi.org/10.1016/j.soilbio.2007.08.022.
10.1007/s10533-007-9103-5. Sakamoto, Y., Ishiguro, M., Kitagawa, G., 1986. Akaike Information Criterion Statis-
Kononova, M.M., 1966. Chapter 4-The Importance of organic matter in soil forma- tics. D. Reidel, Dordrecht, The Netherlands.
tion and soil fertility. In: Soil Organic Matter, Second English Edition. Pergamon, Salazar, O., Rojas, C., Avendan ~ o, F., Realini, P., Na
jera, F., Tapia, Y., 2015. Inorganic
pp. 183e228. nitrogen losses from irrigated maize elds with narrow buffer strips. Nutrient
Kuzyakov, Y., Blagodatskaya, E., Blagodatsky, S., 2009. Comments on the paper by Cycling in Agroecosystems 102, 359e370. http://dx.doi.org/10.1007/s10705-
Kemmitt et al. (2008) Mineralization of native soil organic matter is not 015-9707-4.
regulated by the size, activity or composition of the soil microbial biomass e a Scheu, S., Ruess, L., Bonkowski, M., 2005. Interactions between microorganisms and
new perspective [Soil Biology & Biochemistry 40, 61e73]: the biology of the soil micro- and mesofauna. In: Varma, P.D.A., Buscot, P.F. (Eds.), Microorganisms
Regulatory Gate. Soil Biology and Biochemistry 41, 435e439. http://dx.doi.org/ in Soils: Roles in Genesis and Functions, Soil Biology. Springer Berlin Heidel-
10.1016/j.soilbio.2008.07.023. berg, pp. 253e275. http://dx.doi.org/10.1007/3-540-26609-7_12.
Lehrsch, G.A., Brown, B., Lentz, R.D., Johnson-Maynard, J.L., Leytem, A.B., 2016. Schomberg, H.H., Wietholter, S., Grifn, T.S., Reeves, D.W., Cabrera, M.L., Fisher, D.S.,
Winter and growing season nitrogen mineralization from fall-applied com- Endale, D.M., Novak, J.M., Balkcom, K.S., Raper, R.L., Kitchen, N.R., Locke, M.A.,
posted or stockpiled solid dairy manure. Nutrient Cycling in Agroecosystems Potter, K.N., Schwartz, R.C., Truman, C.C., Tyler, D.D., 2009. Assessing indices for
104, 125e142. http://dx.doi.org/10.1007/s10705-015-9755-9. predicting potential nitrogen mineralization in soils under different manage-
Lemaitre, A., Chaussod, R., Tavant, Y., Bruckert, S., 1995. An attempt to determine a ment systems. Soil Science Society of America Journal 73, 1575. http://
pool of labile organic matter associated with the soil microbial biomass. Eu- dx.doi.org/10.2136/sssaj2008.0303.
ropean Journal of Soil Biology 31, 121e125. Senwo, Z.N., Tabatabai, M.A., 2005. Effect of management systems on nitrogen
Lesur, C., Bazot, M., Bio-Beri, F., Mary, B., Jeuffroy, M.H., Loyce, C., 2014. Assessing mineralization and nitrication in soils. Communications in Soil Science and
nitrate leaching during the three-rst years of Miscanthus  giganteus from on- Plant Analysis 36, 1313e1326. http://dx.doi.org/10.1081/CSS-200056942.
farm measurements and modeling. GCB Bioenergy 6, 439e449. http:// Shari, M., Zebarth, B.J., Burton, D.L., Grant, C.A., Cooper, J.M., 2007. Evaluation of
dx.doi.org/10.1111/gcbb.12066. some indices of potentially mineralizable nitrogen in soil. Soil Science Society of
Mary, B., Beaudoin, N., Justes, E., Machet, J.M., 1999. Calculation of nitrogen America Journal 71, 1233. http://dx.doi.org/10.2136/sssaj2006.0265.
mineralization and leaching in fallow soil using a simple dynamic model. Eu- Shari, M., Zebarth, B.J., Burton, D.L., Grant, C.A., Porter, G.A., 2008. Organic
ropean Journal of Soil Science 50, 549e566. http://dx.doi.org/10.1046/j.1365- amendment history and crop rotation effects on soil nitrogen mineralization
2389.1999.00264.x. potential and soil nitrogen supply in a potato cropping system. Agronomy
McDonald, N.T., Watson, C.J., Lalor, S.T.J., Laughlin, R.J., Wall, D.P., 2014. Evaluation of Journal 100, 1562. http://dx.doi.org/10.2134/agronj2008.0053.
soil tests for predicting nitrogen mineralization in temperate grassland soils. Shari, M., Zebarth, B.J., Miller, J.J., Burton, D.L., Grant, C.A., 2014. Soil nitrogen
Soil Science Society of America Journal 78, 1051. http://dx.doi.org/10.2136/ mineralization in a soil with long-term history of fresh and composted manure
sssaj2013.09.0411. containing straw or wood-chip bedding. Nutrient Cycling in Agroecosystems
Meisinger, J.J., 1984. Evaluating plant-available nitrogen in soil-crop systems. Ni- 99, 63e78. http://dx.doi.org/10.1007/s10705-014-9618-9.
trogen in Crop Production acsesspublicati 391e416. http://dx.doi.org/10.2134/ Smith, J., Smith, P., Addiscott, T., 1996. Quantitative methods to evaluate and
1990.nitrogenincropproduction.c26. compare soil organic matter (SOM) models. In: Powlson, D.S., Smith, P.,
Monod, H., Naud, C., Makowski, D., 2006. Uncertainty and sensitivity analysis for Smith, J.U. (Eds.), Evaluation of Soil Organic Matter Models, NATO ASI Series.
crop models. Working with Dynamic Crop Models: Evaluation, Analysis, Springer Berlin Heidelberg, pp. 181e199. http://dx.doi.org/10.1007/978-3-642-
Parameterization, and Applications 4, 55e100. 61094-3_13.
Oorts, K., Laurent, F., Mary, B., Thie beau, P., Labreuche, J., Nicolardot, B., 2007. Smith, P., Cotrufo, M.F., Rumpel, C., Paustian, K., Kuikman, P.J., Elliott, J.A.,
Experimental and simulated soil mineral N dynamics for long-term tillage McDowell, R., Grifths, R.I., Asakawa, S., Bustamante, M., House, J.I., Sobock a, J.,
systems in northern France. Soil and Tillage Research 94, 441e456. http:// Harper, R., Pan, G., West, P.C., Gerber, J.S., Clark, J.M., Adhya, T., Scholes, R.J.,
dx.doi.org/10.1016/j.still.2006.09.004. Scholes, M.C., 2015. Biogeochemical cycles and biodiversity as key drivers of
Parton, W.J., 1996. The CENTURY model. In: Powlson, D.S., Smith, P., Smith, J.U. ecosystem services provided by soils. Soil 1, 665e685. http://dx.doi.org/
(Eds.), Evaluation of Soil Organic Matter Models, NATO ASI Series. Springer 10.5194/soil-1-665-2015.
Berlin Heidelberg, pp. 283e291. http://dx.doi.org/10.1007/978-3-642-61094-3_ Spargo, J.T., Cavigelli, M.A., Mirsky, S.B., Maul, J.E., Meisinger, J.J., 2011. Mineralizable
23. soil nitrogen and labile soil organic matter in diverse long-term cropping sys-
Parton, W.J., Schimel, D.S., Cole, C.V., Ojima, D.S., 1987. Analysis of factors controlling tems. Nutrient Cycling in Agroecosystems 90, 253e266. http://dx.doi.org/
soil organic matter levels in great plains Grasslands1. Soil Science Society of 10.1007/s10705-011-9426-4.
America Journal 51, 1173. http://dx.doi.org/10.2136/ Springob, G., Kirchmann, H., 2003. Bulk soil C to N ratio as a simple measure of net
sssaj1987.03615995005100050015x. N mineralization from stabilized soil organic matter in sandy arable soils. Soil
Paterson, E., 2009. Comments on the regulatory gate hypothesis and implications Biology and Biochemistry 35, 629e632. http://dx.doi.org/10.1016/S0038-
for C-cycling in soil. Soil Biology and Biochemistry 41, 1352e1354. http:// 0717(03)00052-X.
dx.doi.org/10.1016/j.soilbio.2009.02.012. St. Luce, M., Grant, C.A., Ziadi, N., Zebarth, B.J., O'Donovan, J.T., Blackshaw, R.E.,
Pathak, H., Rao, D.L.N., 1998. Carbon and nitrogen mineralization from added Harker, K.N., Johnson, E.N., Gan, Y., Lafond, G.P., May, W.E., Malhi, S.S.,
organic matter in saline and alkali soils. Soil Biology and Biochemistry 30, Turkington, T.K., Lupwayi, N.Z., McLaren, D.L., 2016. Preceding crops and ni-
695e702. http://dx.doi.org/10.1016/S0038-0717(97)00208-3. trogen fertilization inuence soil nitrogen cycling in no-till canola and wheat
Plaza-Bonilla, D., Nolot, J.M., Passot, S., Raffaillac, D., Justes, E., 2016. Grain legume- cropping systems. Field Crops Research 191, 20e32. http://dx.doi.org/10.1016/
based rotations managed under conventional tillage need cover crops to miti- j.fcr.2016.02.014.
gate soil organic matter losses. Soil and Tillage Research 156, 33e43. http:// St. Luce, M., Whalen, J.K., Ziadi, N., Zebarth, B.J., 2011. Nitrogen dynamics and indices
dx.doi.org/10.1016/j.still.2015.09.021. to predict soil nitrogen supply in humid temperate soils. In: Advances in
Powlson, D.S., Prookes, P.C., Christensen, B.T., 1987. Measurement of soil microbial Agronomy. Elsevier, pp. 55e102.
biomass provides an early indication of changes in total soil organic matter due Stanford, G., Epstein, E., 1974. Nitrogen mineralization-water relations in Soils1. Soil
to straw incorporation. Soil Biology and Biochemistry 19, 159e164. http:// Science Society of America Journal 38, 103. http://dx.doi.org/10.2136/
dx.doi.org/10.1016/0038-0717(87)90076-9. sssaj1974.03615995003800010032x.
R Core Team, 2016. R: a Language and Environment for Statistical Computing. www. Stanford, G., Frere, M.H., Schwaninger, D.H., 1973. Temperature coefcient of soil
R-Project.org. nitrogen mineralization. Soil Science 115, 321e323.
Raison, R.J., Connell, M.J., Khanna, P.K., 1987. Methodology for studying uxes of soil Stone, M., 1974. Cross-validatory choice and assessment of statistical predictions.
mineral-N in situ. Soil Biology and Biochemistry 19, 521e530. http://dx.doi.org/ Journal of the Royal Statistical Society. Series B (Methodological) 36, 111e147.
10.1016/0038-0717(87)90094-0. Tiessen, H., Cuevas, E., Chacon, P., 1994. The role of soil organic matter in sustaining
Reeves, D.W., 1997. The role of soil organic matter in maintaining soil quality in soil fertility. Nature 371, 783e785. http://dx.doi.org/10.1038/371783a0.
continuous cropping systems. Soil and Tillage Research, XIVth ISTRO Confer- Turner, M.M., Henry, H.A.L., 2010. Net nitrogen mineralization and leaching in
ence on Agroecological and Economical Aspects of Soil Tillage 43, 131e167. response to warming and nitrogen deposition in a temperate old eld: the
http://dx.doi.org/10.1016/S0167-1987(97)00038-X. importance of winter temperature. Oecologia 162, 227e236. http://dx.doi.org/
Ros, G.H., 2012. Predicting soil N mineralization using organic matter fractions and 10.1007/s00442-009-1435-5.
soil properties: a re-analysis of literature data. Soil Biology and Biochemistry 45, Vale, M., Mary, B., Justes, E., 2007. Irrigation practices may affect denitrication
132e135. http://dx.doi.org/10.1016/j.soilbio.2011.10.015. more than nitrogen mineralization in warm climatic conditions. Biology and
Ros, G.H., Hofand, E., van Kessel, C., Temminghoff, E.J.M., 2009. Extractable and Fertility of Soils 43, 641e651. http://dx.doi.org/10.1007/s00374-006-0143-0.
dissolved soil organic nitrogen e a quantitative assessment. Soil Biology and Van Ommen, H.C., 1985. The mixing-cell concept applied to transport of non-
Biochemistry 41, 1029e1039. http://dx.doi.org/10.1016/j.soilbio.2009.01.011. reactive and reactive components in soils and groundwater. Journal of
H. Clivot et al. / Soil Biology & Biochemistry 111 (2017) 44e59 59

Hydrology 78, 201e213. http://dx.doi.org/10.1016/0022-1694(85)90101-5. determining the potentially mineralisable nitrogen of soil. Biology and Fertility
Vance, E.D., Brookes, P.C., Jenkinson, D.S., 1987. An extraction method for measuring of Soils 37, 362e374. http://dx.doi.org/10.1007/s00374-003-0604-7.
soil microbial biomass C. Soil Biology and Biochemistry 19, 703e707. http:// Willmott, C.J., Ackleson, S.G., Davis, R.E., Feddema, J.J., Klink, K.M., Legates, D.R.,
dx.doi.org/10.1016/0038-0717(87)90052-6. O'Donnell, J., Rowe, C.M., 1985. Statistics for the evaluation and comparison of
Wallach, D., 2006. Evaluating Crop Models. Working with Dynamic Crop Models. models. Journal of Geophysical Research 90, 8995. http://dx.doi.org/10.1029/
Elsevier, Amsterdam, pp. 11e53. JC090iC05p08995.
Wang, W.J., Smith, C.J., Chen, D., 2003. Towards a standardised procedure for

Vous aimerez peut-être aussi