have imposed cross-regional constraints
on SPARROW parameters to limit their variation between regionally
calibrated SPARROW models. By doing so, their regionalised
models were consistent with each other but do not ensure consistency
with process-based models. In this study, we calibrated
Nutting with an iterative algorithm that optimised parameter values
to fit the data. As initial parameter values might influence
the calibration when the algorithm converges to a local minimum,
we chose an initial parameter value in the R factor to
match the in-stream denitrification rate reported by Kronvang et al.
(2004). Finally, the version of Nutting used in this study is more
parsimonious than the original (Dupas et al., 2013) to avoid overparameterisation.
One to three calibrated parameters were enough
to fit the data in a leave-one-out cross-validation with an R2 ranging
from 0.40 to 0.66 for specific N and P loads and 0.70–0.85 for
global N and P loads.