*3.1. Accuracy of the Dataset*

The criteria applied using the decision tree (Figure 3) led to the final selection of 67 fields (26, 22, and 19 fields were excluded at the 1st, 2nd, and 3rd nodes of the tree, respectively). The selection was severe because a given field had to avoid problems all three years (2012, 2013, and 2014). The 67 fields selected remained representative of the diversity of soil characteristics, cropping systems (Table 1 and Figure 4) and geography of the initial network (Figure 1). To ensure that screening the data had not introduced bias, we verified that the distribution and mean of Mn had remained similar. Mn varied little among subplots over the three years (mean standard deviation of 15 kg N ha−1, with a mean CV of 10% for the 67 fields), which illustrates the high precision of the N mass balance in the selected dataset.
