*3.2. Soil N Mineralization: Mass Balance Components and N Rates*

Mean, minimum or maximum Mn from March to October were similar for the 67 fields in all three years (e.g. mean Mn was 162, 146 and 154 kg N ha−<sup>1</sup> in 2012, 2013 and 2014, respectively) (Table 2). Mean N uptake by maize was the main component of the N mass balance, particularly in 2014, when it repesented 95% of Mn vs. 88% in 2012 and 73% in 2013 (Table 2). Lower N uptake in 2013 was explained by the weather conditions, with low rainfall from June to September that induced hydric stress. Mean predicted nitrate leaching in early spring was low in 2013 and 2014 but was a significant component of the N

mass balance in 2012 (21 kg N ha<sup>−</sup>1) (Table 2); this difference can be explained by higher Ni due to less leaching during the preceding winter and high rainfall in April (138 ± 32 mm).

**Figure 4.** Boxplots of (**a**) soil organic nitrogen (SON), (**b**) extractable organic nitrogen (EON), and (**c**) soil microbial biomass (SMB) for all 137 fields and for the 67 fields selected from the network. Whiskers extend to 1.5 times the interquartile range.

**Table 2.** Mean, minimum, maximum, and standard deviation (SD) of the mass-balance components and nitrogen (N) mineralization rates for the 67 fields selected. N uptake is N taken up by maize plants, Ni and Nf are, respectively, the initial and final amount of mineral N in the soil profile (0–90 cm), N leached is N leached out of the soil profile predicted by the STICS crop model, Mn is net N mineralization, ndays is normalized time between initial and final N measurements, and Vn is normalized N mineralization rate.


Although the fields were unfertilized and soil N mineralization was the main source of N for plants, the mean N Nutrition Index (NNI) [55] was high for unfertilized crops, particularly in 2012 (0.88) and 2014 (0.92) (vs. 0.73 in 2013), reflecting high availability of N during the crop cycle. Mean Vn was similar in 2012, 2013, and 2014 (0.99, 1.06, and 0.92 kg N ha−<sup>1</sup> nday−1, respectively), in agreement with the hypotheses on which the experimental design was developed (Table 2).

## *3.3. Correlations between Vn, Soil Properties, EON, and I\_Sys*

Vnmean correlated most strongly with EON (r = 0.47), SMB (r =0.45), POM-N (r = 0.43), and, to a lesser extent, SON (r = 0.31) (Table 3). Texture, particularly clay content, can influence N mineralization strongly, and Vnmean had a significant but weak negative correlation with clay content (r = −0.19), yet a stronger positive correlation with the coarse sand content (r = 0.32). Vnmean also correlated strongly with I\_Sys (r = 0.39), which was highlighted by a significant effect of I\_Sys class on Vnmean (*p* < 0.05) (Figure 5b): mean Vnmean of the high I\_Sys class was 29% higher than that of the low I\_Sys class (1.11 vs. 0.86 kg N ha−<sup>1</sup> nday<sup>−</sup>1, respectively). This difference was due to much lower Vn measured in fields with low I\_Sys, which corresponded to fields without grassland in their rotation and without organic manure application. SON was strongly correlated with EON (r = 0.64) and POM-N (r = 0.75), while POM-N was also correlated with EON (r = 0.50).

**Table 3.** Correlation coefficients (r) between Vn for the years 2012, 2013, 2014, Vnmean, the cropping system indicator (I\_Sys), and soil properties for the 67 fields selected. (*p* < 0.05 for r > 0.24, *p* < 0.01 for r > 0.31, and *p* < 0.001 for r > 0.39). F: fine, C: Coarse.


**Figure 5.** Boxplots of Vnmean as a function of (**a**) soil organic nitrogen (SON) class (class\_1: SON < 5.7 t N ha−1; class\_2: [5.7; 7.15]; class\_3: [7.15; 8.2]; class\_4: >8.2 t N ha−1) and (**b**) I\_Sys class. Whiskers extend to 1.5 times the interquartile range.

#### *3.4. Modeling Normalized N Mineralization Rate*

A model that predicted Vnmean using only soil properties and the I\_Sys indicator (model 1) explained only 47% of the variance in mineralization (Table 4, Figure 6a). The soil properties selected were SON, clay, coarse sand and coarse silt. A quadratic relation was observed with coarse sand and clay, with a negative effect of clay contents that exceeded 22 g kg−1. A positive linear relation was observed with I\_Sys, which was the most influential variable after coarse sand (Table 4). The model's RPIQ value of 1.8 was moderate.


**Table 4.** Assessment of the models, covariates selected, and dMSEP values. Relation indicates the type of relation selected: L, linear; P2, 2nd-degree polynomial.

**Figure 6.** Comparison of observed Vnmean to that predicted by (**a**) model 1, whose covariates were the basic soil parameters and I\_Sys, and (**b**) model 2, with SMB and EON as additional variables. Solid lines are 1:1 lines, while dashed lines indicate <sup>±</sup> 0.2 kg N ha−<sup>1</sup> nday−<sup>1</sup> around each 1:1 line. (RMSE: Root Mean Square Error).

We then developed a model with EON, SMB and POM-N as additional input variables (model 2). The same soil physical properties as in model 1 were selected, as was the I\_Sys indicator, and they had similar relationships (Table 4). The additional variables selected were EON and SMB, which significantly increased the proportion of variance explained (R<sup>2</sup> = 0.67) (Figure 6b) but increased the RPIQ value only moderately (2.2). EON and SMB were selected because they were the two variables most correlated with Vnmean but not correlated with each other (r = 0.14). Because they provided complementary information, it was useful to include them both in the model. SON was not selected due to its strong correlation with EON.

#### **4. Discussion**

#### *4.1. Soil N Mineralization and N Rates*

The approaches used to estimate Mn in crop fields agree that crop N uptake is the main component of the N mass balance [5,9,37,56,57]. Nonetheless, we observed relatively high variability in the contribution of N uptake to the mineral N mass balance among the three years: 95% in 2014, but only 73% in 2013 [40]. This can be explained by differing weather conditions between years, with high hydric stress in summer 2013, which may have decreased both soil N mineralization and plant growth. In addition, this drought period was followed by strong rainfall events at the end of summer, which could have created favorable conditions for N mineralization just before Nf was measured [58,59]. This can explain why the mean difference between Nf and Ni equaled 24% of Mn in 2013, and why Nf was higher that year. Calculating the N mass balance from March to October, thus including mineralization at the beginning of autumn, explains why mean Mn was similar among the three years. Mn likely would have differed even more if Nf had been measured immediately after harvest, as illustrated by the large inter-annual variability observed by Delin and Linden [9].

Comparing the Mn measured in the field to data from the literature is difficult because measurement periods can vary from 5–8 months depending on the crop (e.g., wheat, spring barley, maize, sugar beet), which has a huge influence on mineralization [36,37]. Comparisons must, thus, be made with daily Vn, either provided directly by studies (which is rare) or estimated later from their data. Delin and Linden [9] reported mean daily rates of 0.34 ± 0.12, 0.50 ± 0.17, and 0.69 ± 0.16 kg N ha−<sup>1</sup> day−<sup>1</sup> in a field experiment with 34 cereal fields studied for three consecutive years. Vnmean estimated from data of Engels and Kuhlmann [36] equals 0.37 ± 0.17 under wheat and 0.67 ± 0.23 under sugar beet, while those estimated from data of [37] and [60] under maize equal 0.71 and 0.68 ± 0.19 kg N ha−<sup>1</sup> day−1, respectively, which lie in the same range as those we measured in our network (Table 2). These reference values show the high variability in Mn among fields.

Expressing Vn in normalized time is the best basis for comparison, since it controls for the influence of weather, but it has rarely been used in the literature. From a database of 65 soils, Clivot et al. [12] report a range of 0.17–1.67 kg N ha−<sup>1</sup> nday−<sup>1</sup> (mean = 0.72 ± 0.32 kg N ha−<sup>1</sup> nday−1). Oorts et al. [35] calculated normalized Vn of 0.57 and 0.62 kg N ha−<sup>1</sup> nday−<sup>1</sup> at two experimental sites with field crops,. The normalized Vn that we calculated from the network (mean = 0.99 kg N ha−<sup>1</sup> nday<sup>−</sup>1), thus, lie near the top of the range reported in the literature.
