Multivariate and long-term time series analysis to assess the effect of nitrogen management policy on groundwater quality in Wallonia, BE

Abstract. Groundwater nitrate contamination remains a significant environmental and health concern, as the effectiveness of nitrogen management programmes has shown mixed results. This study leverages nearly 2 decades of groundwater monitoring data to assess and explain nitrate contamination trends in vulnerable zones in Wallonia, Belgium, following the implementation of the regional sustainable nitrogen management programme. Using nitrate concentration time series (2002–2020) from 36 monitoring points across four groundwater bodies, we calculated six nitrate pollution indicators that capture the long-term trend and dynamics. To explain these trends and dynamics, spatially explicit variables describing potential contamination drivers were compiled, and their relationships with the nitrate pollution indicators were assessed using correlation analyses and multivariate regressions. Although mean groundwater nitrate concentrations remained stable due to nitrogen legacy effects, decreases were observed in the Brusselian sands, where concentrations were initially higher relative to the other bodies, while increases were found in the Geer basin chalks, typically less contaminated. While these diverging trends could be explained by differences in aquifer characteristics and nitrate transfer time lags, results also suggest that agricultural land cover continues to have a negative impact on nitrate contamination, even 20 years after implementing the Programme de Gestion Durable de l'Azote en Agriculture (PGDA). Importantly, the findings are not fully conclusive due to the limited predictive power of the regression models. This highlights the multifaceted nature of groundwater nitrate contamination and the challenges in defining input variables that accurately capture the drivers. Future studies could explore integrating modelling approaches to supplement observational data with modelled data as inputs to statistical models or to combine data-driven models and process-based models. Overall, this study emphasizes the need for sustained and adaptive nitrogen management policies, especially in vulnerable aquifers and cropland-dominated regions, alongside sustained long-term monitoring efforts given the time lags and nitrogen legacy effects. It also underscores the need for spatially and temporally detailed datasets to successfully conduct non-linear machine learning approaches that can capture the complex interactions involved.