This report presents practical field aspects gained during two years of monitoring with state-of-the-art spectrometers and ion-selective sensors, combining (i) continuous measurements of the quality and flow rates of combined sewer overflows (CSO) with (ii) continuous measurements of water quality parameters within the urban stretch of the River Spree. It describes the set-up and the implementation of the monitoring and evaluates the outcomes and experiences towards “lessons learnt”. The challenge of CSO monitoring is their event-based and highly dynamic nature during rain events. Applied online sensors allow dynamic measurements of CSO and water quality impacts for a wide range of parameters. However, the success of online monitoring campaigns depends highly on three main considerations. Firstly, the representativity of the measurement station. The location of the probe must be representative of the concentration over the entire cross section of the sewer or the river. Further criteria have to be considered for the selection of the monitoring sites (e.g. easy access to the probes for maintenance) (chapter 2). Secondly, the quality of the raw measurements. External conditions can influence the quality of measurements and lead to wrong values or outliers. – To avoid drifts, probes need to be cleaned and checked regularly. We found that monitoring stations must be visited at least once a week for functional check-ups. During the two years of monitoring, the maintenance methodology have been continously improved to ensure the best measurement conditions (chapter 3). – But even under state-of-the-art operation of the probes, some values can be affected by errors and lead to misinterpretation. Thus, a validation step is required to detect wrong values and separate them from valid values. Given the large amount of data, an Access-based tool has been developed to support semi-automatic validation of monitoring data (chapter 4). Lastly, the calibration of raw measuments and the determination of uncertainties is critical. Online probes were not able to provide accurate measurements without being calibrated to local conditions with parallel laboratory measurements (online probe refers in this document to spectrometer and ISE-Probe). A Monte-Carlo method was adapted to perform regressions between raw measurement and lab values, which allows considering both uncertainties of sensor and lab chain. For instance, total uncertainty of the UV/VIS probe was between 15 and 30% for chemical oxygen demand (COD), accounting for errors from sensor, laboratory and field (representativity of site). The uncertainties in concentration and flow measurements lead to an uncertainty in CSO COD load between 20 and 70%, depending on the average concentration and flow of the event (chapter 5). In order to gain grab samples and provide high quality calibration, an automatic sampler has been installed at the sewer monitoring. However, for operational purposes, a sewer operator will expect to gain quality online data without the effort and costs of sampling each CSO. In order to estimate the optimal sampling effort, we investigated how many events (or how many lab measurements) are necessary for calibration depending on aimed at uncertainty. From a set of 12 sampled CSO events, we simulate all possible random combinations of events and calculated each time the resulting measurement uncertainty (chapter 5.5). Results shown in Figure A indicate that at least 7 random events need to be sampled to calibrate the probe reducing uncertainties of COD measurement under 30%. It has to be noted that the concentration range of the grab samples has a high influence on the quality of the calibration. A similar analysis considering only events with high lab variations (range > 500 mg/l) showed that then only 4 events must be sampled to reduce uncertainty under 30%. Considering these results, we recommend parallel short sampling campaigns with autosamplers (grab sampling) for application of spectrometers for CSO monitoring. If the lab measurements cover the entire range of water quality variations, a minimum of 3-4 rain events should be sampled to build an accurate calibration function with acceptable uncertainty. If sampled concentration range is exceeded by later measurements, new sampling campaigns should be planned. Since both sensor and autosampling results were available, CSO COD loads have been calculated using both spectrometer and lab values (chapter 6). Results indicate that load calculated with lab samples are within the error range of the loads calculated with spectrometer values. However, the frequency of grab sampling should be less than 10 minutes, to match concentration peaks and quick quality variations in our case. For the purpose of CSO load calculation, autosampler-based monitoring remains a cost-effective alternative to online probes. For a dynamic description of CSO (pollutant sources, mass/flow balance, etc.), autosampler-based data are limited by the minimal sample frequency and the sampling capacity. Investment and effort of online monitoring can overcome these limitations. For river monitoring, online probes enable measuring water quality variations with an acceptable uncertainty, if the probes are properly calibrated. Here, autosamplers are clearly limited by their sampling capacity as the impacts are spread on several days in the case of the River Spree. Since no autosampler was available during the two monitoring years no clear correlation could be established for the spectrometer parameters (TSS, COD, BOD). As the manual approach often fails to catch CSO impacts, an autosampler has been purchased for the last monitoring year in 2012. For NH4 + measurement, the ISE probe has been successfully calibrated performing monthly NH4 measurements in a bucket of river water spiked with ammonium standard solution to reach values in the range expected during CSO (1-2 mg/l).