Conceptual model structure identification and calibration for river and sewer systems.

Keywords: river; sewer; integral water management; conceptual modelling; water quantity modelling; model structure identification and calibration; software

Water management is constantly evolving. Trends, such as population growth, urbanization and climate change, pose new challenges to water management. Risks and uncertainties have to be taken into account in order to develop strategies that remain cost-effective on the long term. Water systems consist of different subsystems and interactions, which demand for an integrated approach to achieve optimal solutions. Due to the intricate complexity of developing adequate strategies, the use of mathematical models that accurately emulate reality is indispensable. Conventionally used full hydrodynamic models suffer from several fundamental shortcomings, of which a prolonged calculation time, the high model complexity and limited flexibility in terms of model adaptability and interfacing are arguably the main ones. These deficiencies compel the use of parsimonious conceptual models for numerous water management problems, which reason from simplified relationships to mimic the response of the system. These conceptual models are therefore computationally very efficient and easily adaptable to practical needs. This dissertation focuses on the development and application of such a conceptual modelling approach for river and sewer hydraulic simulations.

The first part of the dissertation presents the newly developed conceptual water quantity modelling approach, which aims to be suitable for a wide range of scenario investigations in support of water engineering and management. First, a variety of model structures was investigated, applied and improved while elaborating several case studies. Besides the in hydrology more traditionally employed model structures such as linear reservoir theory and different linear structures, special attention was paid to the performance of machine learning techniques and expert systems. Adaptive Neuro Fuzzy Inference Systems were used for lumped floodplain modelling. Artificial Neural Networks (ANNs) have proved successful in emulating complex flows in rivers, floodplains and urban drainage systems. In addition, the performance of ANNs for capturing non-univocal behaviour in rating curve modelling was compared to that of M5’ model trees and of piecewise linear relationships with state dependent parameters. Subsequently, these model structures were combined in a generic modelling framework. Two conceptual modelling approaches were developed, one for rivers and one for urban drainage systems. These approaches aim at mimicking simulation results of full hydrodynamic river and sewer models, but at a fraction of the calculation time of the detailed models. Both conceptual modelling approaches share the same fundamental modelling principles and are based on the storage cell concept, but their model topology, structures and conceptualization procedure are tailored to each system. Their modular setting, which allows the user to combine diverse incorporated model structures freely in a single model, results in powerful emulation capabilities. Complex dynamics, such as backwater effects, varying water surface profiles in rivers, and reverse and pressurized flows in sewers, can be mimicked accurately. The approaches have important mechanistic features, such as the closing of the water balance and the possible explicit incorporation of dike levels and moveable hydraulic structures of rivers. In addition, an innovative calculation scheme with an adaptive time step was developed that avoids instabilities and minimizes calculation time.

The developed approaches were integrated in a new semi-automatic software platform to configure the conceptual models, named Conceptual Model Developer (CMD). The tool guides the user via graphical user interfaces in a stepwise manner through the configuration procedure and assists the user in setting up the conceptual model topology. A close integration was foreseen with hydrodynamic software packages to import simulation results and important model characteristics. All configured model structures are assembled fully automatically in a single highly efficient calculation script written in C programming language.

The approach and software were tested extensively on several case studies in Belgium. Three of these studies are elaborated in this dissertation. The first case study quantified the impact of installing additional retention basins and varying operation controls of moveable weirs on flood probabilities along the Marke and Dender Rivers. Conceptual models were built of both rivers and consequently used for performing long term simulations, thereby accounting for antecedent conditions. The second case study used the conceptual model of the Dender River to generate flood probability maps. The approach can be incorporated in real-time flood forecasting systems. The third study analyzed the impact of combined sewer overflows (CSOs) on the receiving river water quality in the Molse Nete catchment. Conceptual quantity models were created of the urban drainage systems of the cities of Mol and Geel, and the Molse Nete River. Next, these models were integrated and combined with conceptual water quality models for performing long term simulations. These conceptual water quality models were created in the scope of another PhD research. Other researchers also used the in this dissertation developed conceptual modelling approach for applications in which flexible simulation models with a very limited calculation time play a pivotal role. Conceptual models were configured to assess the impact on urban and river floods for three different storage strategies for the city of Turnhout. Also, model-based optimization of hydraulic structures to minimize the flood damage in the Demer catchment was elaborated using conceptual models.