Preliminary study on the detection of unnoticed changes in stage discharge relationships

Reliable discharge data is a key requirement of hydrological studies, yet previous research has primarily focused on detecting sensor errors and outliers. Undetected changes in stage discharge relationships and the resulting discrepancy between the actual and measured discharge have received significantly less attention. The present study aims to contribute to closing this research gap by developing a detection routine for unnoticed changes in stage discharge relationships. In a first step, classical statistical methods are tested. In a subsequent step, a machine learning approach is evaluated and contrasted with the statistical methods.  The study is conducted on the two Swiss rivers, Aare and Reuss, which comprise 41 gauged subcatchments.