Deep-learning based monitoring of FOG layer dynamics in wastewater pumping stations.

Accumulation of fat, oil and grease (FOG) in the sumps of wastewater pumping stations is a common failure cause for these facilities. Floating solids are often not transported by the pump suction inlets and the individual solids can accumulate to stiff and thick FOG layers. The lack of data about the dynamics in FOG layer formation still hampers the design of effective measures towards its mitigation. In this article, we present a low-cost camera-based automated system for the observation of FOG layer dynamics in wastewater pumping stations at high-frequency (minutes) over extended time windows (months). Optical imagery is processed through a deep-learning computer vision routine that allows describing FOG layer dynamics (e.g. accumulation rate and changes in shape) and various hydraulic processes in the pump sump (e.g. the water level, surface flow velocity fields, vorticity, or circulation). Furthermore, the system can perform in-camera image processing, thus allowing the transfer of compressed-processed datasets when deployed in remote locations (Edge AI computing), which could be of great utility for the hydro-ecological monitoring community. In this study, the technology applied is illustrated with a dataset (six months, two-minute frequency) collected at a wastewater pumping station at the municipality of Rotterdam, The Netherlands. This monitoring system represents a source of information for the management of (waste)water pumping stations (e.g. detection of free-surface vortices and scheduling of sump cleaning operations) and facilitates the collection of standardized high-frequency FOG layer dynamics data for a detailed description of FOG build-up and transport processes.

Parametric emulation and inference in computationally expensive integrated urban water quality simulators.

Water quality environmental assessment often requires the joint simulation of several subsystems (e.g. wastewater treatment processes, urban drainage and receiving water bodies). The complexity of these integrated catchment models grows fast, leading to potentially over-parameterised and computationally expensive models. The receiving water body physical and biochemical parameters are often a dominant source of uncertainty when simulating dissolved oxygen depletion processes. Thus, the use of system observations to refine prior knowledge (from experts or literature) is usually required. Unfortunately, simulating real-world scale water quality processes results in a significant computational burden, for which the use of sampling intensive applications (e.g. parametric inference) is severely hampered. Data-driven emulation aims at creating an interpolation map between the parametric and output multidimensional spaces of a dynamic simulator, thus providing a fast approximation of the model response. In this study a large-scale integrated urban water quality model is used to simulate dissolved oxygen depletion processes in a sensitive river. A polynomial expansion emulator was proposed to approximate the link between four and eight river physical and biochemical river parameters and the dynamics of river flow and dissolved oxygen concentration during one year (at hourly frequency). The emulator scheme was used to perform a sensitivity analysis and a formal parametric inference using local system observations. The effect of different likelihood assumptions (e.g. heteroscedasticity, normality and autocorrelation) during the inference of dissolved oxygen processes is also discussed. This study shows how the use of data-driven emulators can facilitate the integration of formal uncertainty analysis schemes in the hydrological and water quality modelling community.

Uncertainty analysis in a large-scale water quality integrated catchment modelling study.

Receiving water quality simulation in highly urbanised areas requires the integration of several processes occurring at different space-time scales. These integrated catchment models deliver results with a significant uncertainty level associated. Still, uncertainty analysis is seldom applied in practice and the relative contribution of the individual model elements is poorly understood. Often the available methods are applied to relatively small systems or individual sub-systems, due to limitations in organisational and computational resources. Consequently this work presents an uncertainty propagation and decomposition scheme of an integrated water quality modelling study for the evaluation of dissolved oxygen dynamics in a large-scale urbanised river catchment in the Netherlands. Forward propagation of the measured and elicited uncertainty input-parametric distributions was proposed and contrasted with monitoring data series. Prior ranges for river water quality-quantity parameters lead to high uncertainty in dissolved oxygen predictions, thus the need for formal calibration to adapt to the local dynamics is highlighted. After inferring the river process parameters with system measurements of flow and dissolved oxygen, combined sewer overflow pollution loads became the dominant uncertainty source along with rainfall variability. As a result, insights gained in this paper can help in planning and directing further monitoring and modelling efforts in the system. When comparing these modelling results to existing national guidelines it is shown that the commonly used concentration-duration-frequency tables should not be the only metric used to select mitigation alternatives and may need to be adapted in order to cope with uncertainties.

A dynamic emulator for physically based flow simulators under varying rainfall and parametric conditions.

This work presents a method to emulate the flow dynamics of physically based hydrodynamic simulators under variations of time-dependent rainfall and parametric scenarios. Although surrogate modelling is often employed to deal with the computational burden of this type of simulators, common techniques used for model emulation as polynomial expansions or Gaussian processes cannot deal with large parameter space dimensionality. This restricts their applicability to a reduced number of static parameters under a fixed rainfall process. The technique presented combines the use of a modified Unit Hydrograph (UH) scheme and a polynomial chaos expansion (PCE) to emulate flow from physically based hydrodynamic models. The novel element of the proposed methodology is that the emulator compensates for the errors induced by the assumptions of proportionality and superposition of the UH theory when dealing with non-linear model structures, whereas it approximates properly the behaviour of a physically based simulator to new (spatially-uniform) rainfall time-series and parametric scenarios. The computational time is significantly reduced, which makes the practical use of the model feasible (e.g. real time control, flood warning schemes, hydraulic structures design, parametric inference etc.). The applicability of this methodology is demonstrated in three case studies, through the emulation of a simplified non-linear tank-in-series routing structure and of the 2D Shallow Water Equations (2D-SWE) solution (FLOW-R2D) in two computational domains. Results indicate that the proposed emulator can approximate with a high degree of accuracy the behaviour of the original models under a wide range of rainfall inputs and parametric values.

Detection and quantification of lateral, illicit connections and infiltration in sewers with Infra-Red camera: Conclusions after a wide experimental plan.

Separate sewer systems are sensitive to illegal or mis-connections. Several techniques (including the Distributed Temperature Sensor) are now available to identify and locate those connections. Based on thermal fingerprints, DTS allows the localization of each lateral connection along a reach. The use of Infra-Red camera has been investigated with 748 laboratory experiments (artificial connections along a flume). The tested connections vary in diameters (from 75 to 200 mm), lengths of intrusion (from 0 to 200 m), shapes (circular or linear i.e. cracks), depths, discharge rates between the lateral connection and the main flume, and temperatures. IR frame analysis (for detection) and 2D temperature mapping (at the free water surface, for quantification) demonstrate that: i) the detection limit is very low (ratio between lateral and main discharges: 0.025) and ii) the quantification of the lateral discharge is impossible. Application of an IR camera seems to be a promising technique to detect lateral connections.