A land use regression model to explain the spatial variation of nitrate concentration - A regional scale case study...

Regional-scale groundwater contamination estimation is crucial for sustainable water management. The primary obstacles in evaluating groundwater include limited data availability, small sample sizes, and difficulties in linking concentration levels to land use patterns. Linear regression identifies the relationship between measured concentrations and both natural and human-influenced factors. However, the primary difficulty with this method lies in choosing a group of regressors that meet all necessary criteria for the model when multiple potential regressors exist. This study introduces a buffer-based land-use linear regression method to develop a catchment-scale model for predicting nitrate concentrations in groundwater. The model successfully captures 85 % of the spatial variability in nitrate across the study area, as indicated by the validation results from 32 training sites. The model's prediction capability and ability to capture the spatial variability of nitrate concentration were found to be good in the model development (R² = 0.89) and validation (R² = 0.80) steps. The model performed well in the accuracy assessment and error estimation processes (RMSE = 0.025 and MAE = 0.020). In future, this LUR model can be reparameterised with the latest available time series datasets to capture climate change scenarios. While this study focused on a small sub-catchment of the Bonet River, the methodology has the potential to be applied in a border study area. Future studies with a more robust methodology and more accurate predictor variables to explain the influence of the contamination sources, transport and attenuation processes can improve the buffer-based LUR technique for better model adaptation and applicability to other study areas.