Using numerical weather prediction and in-sewer sensor data for realtime monitoring and forecasting in urban drainage-wastewater systems

Urban drainage and wastewater systems are responsible for protecting the environment against pollution and the public against diseases and flooding. These systems have traditionally been engineered as static solutions but the current wave of digitalization means that they are transitioning into actively managed assets.

Real-time operations aim to accurately monitor the current state of the system, forecast its near future behavior, and based on this control actuators that allow for flexible performance. These efforts are often built on advanced algorithms that require high-quality input data to properly function. Since rainfall and ammonium concentrations in wastewater are some of the most important variables for general system performance, this thesis deals with obtaining good data on these two aspects.

The main research objectives are about how to improve in-sewer measurements of ammonium with ammonium ion-selective electrodes (A-ISE), and how to use numerical weather prediction (NWP) for forecasting rainfall and flow in sewers. Wastewater from households contain ammonium, which can have serious detrimental environmental effects if discharged into surface waters.

It is therefore important that water resource recovery facilities (WRRFs) can accurately monitor, forecast, and ultimately remove it from the wastewater they receive. A-ISE technology has the advantage of measuring directly in the wastewater stream while being cheap to purchase and operate. However, it is also generally regarded as an unreliable data source prone to several types of errors.

A one-year measurement campaign at a WRRF highlighted that the currently recommend approach to A-ISE sensor recalibration based on grab samples is inadequate. The result was a raw signal with erratic jumps and effects of drifting. A methodology to correct the errors in the signal was therefore developed based on integrating information from the A-ISE sensors and 24-h volume-proportional composite samples.

The composite samples are available at many WRRFs and the methodology can thus be used without additional operational costs. The corrected signal provided a much more reliable estimate of ammonium concentrations, and could be used to estimate software sensors with more precise predictions. While there are still improvements to be made within use of A-ISE for monitoring ammonium in wastewater and to the developed methodology, the thesis has made major progress towards a measurement setup that can deliver reliable A-ISE data to wastewater managers.

NWP predicts rainfall through large-scale simulations of atmospheric physics and is the main alternative to radar extrapolation forecasts, which are more commonly used for urban drainage applications. However, the collective experiences with NWP for urban drainage purposes are still rather few. The thesis therefore reviewed these experiences and extracted key lessons for how to use it well, and further investigated use cases for two different NWP products.

Previous research into NWP use for urban drainage issues was grouped into four main topics: (1) generic rain and flow forecasting, (2) urban pluvial flood forecasting, (3) real-time control, and (4) post-processing. Based on this, advice were given on how to make sure that the scope and resolutions of a chosen NWP product, hydrological model, and decision algorithm are fit for the purpose they are intended to fulfil.

In general, it is an issue that many published studies have been built on small samples of a few rain events, which often leads to inconclusive results. This thesis investigated NWP performance with a large forecast archive of more than 100 rain events, which quantified how forecast performance was dependent on the type of weather event.

Dynamic events with a high degree of evolution over time and events that consisted of small and scattered rain cells were difficult to predict. The NWP product could successfully be used to control a wet weather switch at a WRRF, which led to improved performance compared to a reactive control setup based on real-time rain gauge measurements.

An intuitive and easy-to-implement post-processing method based on time-lagging was used to enhance a NWP ensemble product. The method was able to use information on forecast consistency from consecutive forecasts, and was used to make sewer flow predictions. Time-lagged forecasts were able to compete with a more well-known post-processing method based on spatial neighborhoods.

NWP is becoming available as an open data source in many countries, and improvements in data resolutions and assimilation techniques are making it increasingly attractive for urban water purposes. With the review of how NWP has been used in the past and the strides made towards using these data for predictions and decision-making, the thesis is a step towards increasing the uptake of NWP for real-time operations in urban drainage and wastewater systems.