Abstract. Deposition of atmospheric ammonia (NH 3 ) to semi-natural ecosystems leads to serious adverse effects, such as acidification and eutrophication. A step in quantifying such effects is the measurement of NH 3 fluxes over seminatural and agricultural land. However, measurement of NH 3 fluxes over vegetation in the vicinity of strong NH 3 sources is challenging, since NH 3 emissions are highly heterogeneous. Indeed, under such conditions, local advection errors may alter the measured fluxes. In this study, local advection errors ( F z,adv ) were estimated over a 14 ha grassland field, which was successively cut and fertilised, as part of the GRAM-INAE integrated Braunschweig experiment. The magnitude of F z,adv was determined up to 810 m downwind from farm buildings emitting between 6.2 and 9.9 kg NH 3 day −1 . The GRAMINAE experiment provided a unique opportunity to compare two methods of estimating F z,adv : one inference method based on measurements of horizontal concentration gradients, and one based on inverse dispersion modelling with a two-dimensional model.Two sources of local advection were clearly identified: the farm NH 3 emissions leading to positive F z,adv ("bias towards emissions") and field NH 3 emissions, which led to a negative F z,adv ("bias towards deposition"). The local advection flux from the farm was in the range 0 to 27 ng NH 3 m −2 s −1 at 610 m from the farm, whereas F z,adv due to field emission was proportional to the local flux, and ranged between −209 and 13 ng NH 3 m −2 s −1 . The local advection flux F z,adv was either positive or negative dependCorrespondence to: B. Loubet (loubet@grignon.inra.fr) ing on the magnitude of these two contributions. The modelled and inferred advection errors agreed well. The inferred advection errors, relative to the vertical flux at 1 m height, were 52% on average, before the field was cut, and less than 2.1% when the field was fertilised. The variability of the advection errors in response to changes in micrometeorological conditions is also studied. The limits of the 2-D modelling approach are discussed.