The presented work studies the influence of the sampling strategy on the quality of locally calibrated UV-VIS probe measurements in combined sewer overflows (CSO) and the receiving river. Results indicate that UV-VIS spectrometers are not able to provide reliable measurements of water quality in urban stormwater without being calibrated to local conditions with laboratory analyses of water samples. The use of the global calibration (supplied by the manufacturer) led to errors of at least 30% and 45% for CSO load and river concentration of chemical oxygen demand (COD), respectively. Even with reliable local calibration, COD loads contained significant uncertainties close to 20%. Uncertainties in COD load and concentration decrease below 30% if more than 15-20 samples (i.e. 3-4 stormwater events) are sampled for local calibration. The effort and associated sampling costs to gain more than 15-20 samples are much less effective, since load and concentration uncertainties remain relatively stable with an increasing number of samples used for the calibration. The presented analysis aims at supporting practitioners in the planning, operation and calibration of UV-VIS spectrometer probes.