Fully Automated Simplification of Urban Drainage Models on City Scale

Markus Pichler, Albert König, Stefan Reinstaller and Dirk Muschalla, Graz University of Technology, Graz, Austria

ABSTRACT

As the available amount of data for environmental models continues to grow, the trend towards automatic model creation is becoming increasingly prevalent. But with this automation, models are getting more complex and subsequent analysis like Monte Carlo simulations for model calibration or sensitivity analysis can be resource intensive in terms of time and computational power. In response to these challenges, we present an automated model simplification algorithm designed to retain the fundamental system behavior. Until now, there has been no attempt to simplify or aggregate highresolution, semi-distributed models at a city scale.

The methodology consists of the development and testing of a fully automated workflow for model aggregation. This process begins by eliminating non-rainwater carrying wastewater sewers at the network's endings, allocating dry weather flow (DWF) and direct inflow to the downstream node. Subcatchments (SCs) are merged with downstream SCs to form an aggregated subcatchement of no more than 5 hectares, and conduits under 400 meters are combined. This approach ensures uniform SC sizes, avoiding overestimation of infiltration rates in large SCs. Network simplification is not applied to special hydraulic structures like storage nodes, weirs, or orifices. Aggregated SCs and conduits are autocalibrated to the high-resolution model simulation results. Infiltration parameters are set according to soil type. The recalibration of aggregated conduits ensures maintenance of the original sewer storage volume and flow capacity.

The case study applied this workflow to the sewer model of Graz, Austria. The high-resolution model was significantly simplified in the aggregation process, maintaining fundamental system behavior and reducing complexity. The hydrological performance of the aggregated model closely matched the high-resolution model, with minimal errors in water balance components and flood event prediction. Additionally, the aggregated model showed a 22 times computational speedup compared to the high-resolution model.


Permanent link: