Accelerating hydrodynamic simulations of urban drainage systems with physics-guided machine learning.

We propose and demonstrate a new approach for fast and accurate surrogate modelling of urban drainage system hydraulics based on physics-guided machine learning. The surrogates are trained against a limited set of simulation results from a hydrodynamic (HiFi) model. Our approach reduces simulation times by one to two orders of magnitude compared to a HiFi model. It is thus slower than e.g. conceptual hydrological models, but it enables simulations of water levels, flows and surcharges in all nodes and links of a drainage network and thus largely preserves the level of detail provided by HiFi models. Comparing time series simulated by the surrogate and the HiFi model, R2 values in the order of 0.9 are achieved. Surrogate training times are currently in the order of one hour. However, they can likely be reduced through the application of transfer learning and graph neural networks. Our surrogate approach will be useful for interactive workshops in initial design phases of urban drainage systems, as well as for real time applications. In addition, our model formulation is generic and future research should investigate its application for simulating other water systems.

Modeling in-sewer transformations at catchment scale - implications on drug consumption estimates in wastewater-based epidemiology.

To which extent illicit drugs are transformed during in-sewer transport, depends on a number of factors: i) substance-specific transformation rates, ii) environmental conditions, iii) point of discharge (location of drug user) and iv) sewer network properties, primarily hydraulic residence time (HRT) and the ratio of biofilm contact area to wastewater volume (A/Veq). Assessing associated uncertainties typically requires numerous simulations. Therefore, we propose a new two-step modeling framework: 1) Quantify hydrodynamic conditions. This computationally demanding step was performed once in SWMM to derive HRT and A/Veq for each potential point of discharge (node) in three catchments of different size. 2) Estimate biomarker loss. In this step, Monte Carlo simulations were performed for defined scenarios. Depending on assumptions about drug user distribution and prevalence, a number of nodes was sampled. For each node an empirical first-order transformation model was applied with flow-path-corresponding HRT and A/Veq from step 1. Biotic and abiotic transformation rates were sampled from distributions combining variability of different biofilms. In our modeling study, median losses were >30% for amphetamine, 6-monoacetylmorphine and 6-acetylcodeine in all three catchments with high uncertainty (5%-100% loss), which would imply a systematic underestimation of consumption when neglecting in-sewer processes. Median losses for 21 other investigated biomarkers were <10% with different uncertainty ranges - "no substantial transformation" was confirmed for nine substances in a real sewer segment with a 2-h residence time. Transferability of these results must be tested for other catchments. To further reduce uncertainty, mainly additional knowledge on transformation rates, particularly in biofilm, and their distribution across a sewer network is needed to update model input objectively. Our approach allows efficient testing and, furthermore, can be expanded for many other human biomarkers. Accounting for biomarker stability during in-sewer transport will avoid biased estimates and further improve wastewater-based epidemiology.