Sarris, T.; Close, M.; Burbery, L. 2018 Denitrification rate inputs to groundwater models. Lower Hutt, N.Z.: GNS Science. GNS Science report 2018/43. 51 p.; doi: 10.21420/MZJ5-6M59.
Diffuse nitrate leaching from agricultural areas is a major environmental problem in many parts of the world. Understanding where natural attenuation occurs in the form of denitrification is key to understanding the fate of nitrate in groundwater. Furthermore, uncertainty in the parameters used to mathematically represent the naturally occurring processes remains a major challenge in modelling nitrate fate in aquifers. In this work using mappable physical attributes such as geology, topography and soil characteristics, we predict the spatially variable groundwater redox status in three case study regions in New Zealand (Hauraki, Southland and Ruamāhanga). The analysis was performed using linear discriminant analysis and we found that for each regional redox model, different parameters contributed to the model predictive capability. Elevation, surface slope, geological age, rock type, NZ soil order and land use classification were found to be statistically important parameters for all three regional models. Depth to water table and hydraulic conductivity were also significant for the Southland regional model, as was well depth for the Ruamāhanga model. An extensive literature review of previous in situ field measurements of nitrate reactivity in New Zealand revealed few studies focussing on nitrate reactivity below the root zone thereby limiting understanding of nitrate attenuation in New Zealand aquifers. Denitrification uncertainty was considered here as it closely associated with groundwater redox status predictions. Uncertainty was calculated for the nitrate reduction rates of a first order reduction model using non-linear uncertainty analysis. The effects on uncertainty from scaling and spatial averaging were also analysed. It was found that denitrification potential uncertainty decreases with increasing scale, however parameter uncertainty increases with aggregation scale. The latter holds for model predictive uncertainty when the parameter scale dependency is accounted for in the analysis. (auth)