RESUMO
N biogeochemical flows and associated N losses exceed currently planetary boundaries and represent a major threat for sustainability. Measuring N losses is a resource-intensive endeavour, and not suitable for ex-ante assessments, thus modelling is a common approach for estimating N losses associated with agricultural scenarios (systems, practices, situations). The aim of this study is to review some of the N models commonly used for estimating direct field emissions of agricultural systems, and to assess their suitability to systems featuring contrasted agricultural and pedoclimatic conditions. Simple N models were chosen based on their frequent use in LCA, including ecoinvent v3, Indigo-N v1/v2, AGRIBALYSE v1.2/v1.3, and the Mineral fertiliser equivalents (MFE) calculator. Model sets were contrasted, among them and with the dynamic crop model STICS, regarding their consideration of the biophysical processes determining N losses to the environment from agriculture, namely plant uptake, nitrification, denitrification, NH3 volatilisation, NO3 leaching, erosion and run-off, and N2O emission to air; using four reference agricultural datasets. Models' consideration of management drivers such as crop rotations and the allocation of fertilisers and emissions among crops in a crop rotation, over-fertilisation and fertilisation technique, were also contrasted, as well as their management of the mineralisation of soil organic matter and organic fertilisers, and of drainage regimes. For the four agricultural datasets, the ecoinvent model predicted significantly lower values for NH3 than AGRIBALYSE and STICS. For N2O, no significant differences were found among models. For NO3, ecoinvent and AGRIBALYSE predicted significantly higher emissions than STICS, regardless of the fertilisation regime. For both emissions, values of Indigo-N were close to those of STICS. By analysing the reasons for such differences, and the underlying factors considered by models, a list of recommendations was produced regarding more accurate ways to model N losses (e.g. by including the main drivers regulating emissions).