Improving spatial and temporal variation of ammonia emissions for the Netherlands using livestock housing information and a Sentinel-2-derived crop map

Ammonia emissions to the atmosphere have a range of negative impacts on environmental quality, human health, and biodiversity. Despite the considerable efforts in quantifying spatially explicit ammonia emissions, there are significant uncertainties in ammonia emission estimates at regional scales. We aimed to improve the modeling of atmospheric ammonia emission variability in space and time across the Netherlands by updating an agricultural ammonia emission model with a newly derived high-resolution crop map and a livestock housing location database of the Netherlands. To generate a crop map of 12 agricultural land cover classes, we applied random forest classification to the multi-temporal multispectral observations of surface reflectance and vegetation indices derived from Sentinel-2. The crop statistics were used to calculate ammonia emission distribution based on nitrogen demand (manure and mineral fertilizer needed) of different crop types using the INTEGRATOR model. Next, the crop map was utilized to spatially allocate the ammonia emissions to a high-resolution grid across the Netherlands. In addition, ammonia emissions from livestock housing systems were introduced as point sources using location data from the Geographic Information Agricultural Business system. The temporal emission variability was updated using a recently developed TIMELINES module. After the spatial and temporal distribution of ammonia emission was obtained with the crop map and housing information, it was imported into the chemistry transport model LOTOS-EUROS to model ammonia surface concentration for validation with in situ measurements.

The performed crop classification has an average accuracy score of 0.73. The derived crop map was compared with Dutch national statistics, and the results showed that the absolute median of the relative difference between Sentinel-2 derived crop areas and national statistical information is around 5%. The newly modeled ammonia monthly surface concentrations compared better with in situ measurements in terms of the magnitude and temporal variability than those derived from the original emission distribution, indicating that the temporal distribution of ammonia emissions was improved. The comparison of modeled and measured annual averaged surface concentrations illustrated that the spatial distribution of ammonia emission was also improved. All model performance indicators significantly improved, and the performance of the updated model was more stable and robust. The improvement was more evident at the stations where livestock housing is the main emission source. This study illustrates that apart from a satellite-derived crop map, information on the locations of animal housing systems also plays an essential role in better estimates of the spatial and temporal distribution of ammonia emissions. It can be worthwhile to extrapolate the method to other regions in Europe and elsewhere.