Land-Use Regression Models for Atmospheric Ammonia in a Livestock-Dense, Mixed Land-Use Area.

Exposure to atmospheric ammonia (NH3) is associated with adverse effects on human health and ecosystems. Livestock farming is the primary source of NH3 emissions. To assess potential health risks and design effective mitigation strategies, it is important to characterise atmospheric NH3 concentrations. Land-use regression (LUR) models have been widely used to estimate air contaminant levels; however, no successful LUR models for NH3 have been reported, partly due to the absence of high-density measurement networks. In this study, we developed LUR models for NH3 using livestock- and non-livestock-related predictors in a livestock-dense, mixed land-use region in the Netherlands. We assessed three different model algorithms (standard linear regression [SLR], least absolute shrinkage and selection operator [LASSO], and Random Forest) across three predictor sets. These sets differed by their inclusion of livestock-farm counts and distances (basic variant), farm-specific emission rates (emission variant), or a combination of 1 x 1 km dispersion-modelled NH3 and the basic variant (hybrid variant). All models performed well with cross-validated (CV) coefficient of determination (R2) ranging from 0.63 to 0.75. Random Forest models did not outperform SLR and LASSO models (mean CV R2 of 0.65 vs. 0.72 and 0.66, respectively). Variables related to fattening pigs, veal calves, laying hens, and agricultural grasslands were the strongest predictors of increased NH3 concentrations, followed by population density variables, all of which were positively associated with measured concentrations. While SLR concentration maps showed similar spatial patterns across all three variants, the emission variant likely best captured local variability. Overall, LUR modelling offers a robust tool for characterising NH3 concentrations in agricultural regions, thereby offering valuable inputs to planetary health assessments and emission mitigation strategies.