Method & limitations

FossWatch is an exploratory decision-support prototype. Its scores are analytical indices, not regulatory standards or evidence that a field is polluting the river.

Four kinds of evidence

Observed: EA water-quality samples. Derived GIS: distances, areas and network summaries. Modelled: crop classification, vegetation evidence and runoff potential. Statistical: cross-station Spearman associations.

• Observational data cannot establish causality by itself.
• Water quality represents cumulative upstream processes; rainfall and flow affect concentrations.
• Wastewater, urban runoff and other sources may contribute.
• Monitoring locations are not randomly distributed.
• Sentinel-2 has limited ability to detect very narrow vegetation strips.
• Crop and vegetation datasets contain classification error; CROME cells are not field boundaries.
• Groundwater nitrate may have long time lags.
• Imagery dates and sample dates may differ.
• The upstream network is oriented towards the known outlet. Land is assigned to its nearest reach, not a validated watershed.
• The 30 m Copernicus surface model includes trees and buildings. D8 flow does not account for culverts, pipes, field drains or engineered channels.
• Qualified measurements are missing chart points and excluded from exact-value statistics.
• Nested station contributing areas are not independent. Correlations and p-values are exploratory and unadjusted.

Runoff Connectivity Risk

100 × (0.20 slope + 0.20 flow accumulation + 0.20 proximity + 0.15 crop susceptibility + 0.25 buffer deficit). Inputs are scaled 0–1. Higher scores indicate relatively strong physical potential to deliver surface runoff to the river.

Priority investigation areas combine risk ≥65 with local buffer index <50. Neither threshold identifies a pollution source.