Lead Database Specialist
This study investigates the potential of continuously monitored water quality data and machine learning to identify land use signatures from agricultural fields.
High frequency (15-minute) data from the North Wyke Farm Platform were used to characterise the hydrological and water quality responses to rainfall events. Each water quality sensor on the platform is located so that it captures runoff from fields under a single land use only, enabling each event in the dataset to be attributed to either livestock or arable farming. Following feature selection and parameter tuning, a neural network classifier was trained to distinguish between livestock and arable land use and evaluated on independent validation catchments.
The model achieved a balanced accuracy of 76% and a ROC AUC of 0.84, demonstrating good discrimination between land use types. These findings show that continuous water quality data contain detectable agricultural runoff signatures and could support future estimation of river pollution sources.
