Luca Vezzaro

Comparison of different uncertainty techniques in urban stormwater quantity and quality modelling

Urban drainage models are important tools used by both practitioners and scientists in the field of stormwater management. These models are often conceptual and usually require calibration using local datasets. The quantification of the uncertainty associated with the models is a must, although it is rarely practiced. The International Working Group on Data and Models, which works under the IWA/IAHR Joint Committee on Urban Drainage, has been working on the development of a framework for defining and assessing uncertainties in the field of urban drainage modelling. A part of that work is the assessment and comparison of different techniques generally used in the uncertainty assessment of the parameters of water models. This paper compares a number of these techniques: the Generalized Likelihood Uncertainty Estimation (GLUE), the Shuffled Complex Evolution Metropolis algorithm (SCEM-UA), an approach based on a multi-objective auto-calibration (a multialgorithm, genetically adaptive multi-objective method, AMALGAM) and a Bayesian approach based on a simplified Markov Chain Monte Carlo method (implemented in the software MICA). To allow a meaningful comparison among the different uncertainty techniques, common criteria have been set for the likelihood formulation, defining the number of simulations, and the measure of uncertainty bounds. Moreover, all the uncertainty techniques were implemented for the same case study, in which the same stormwater quantity and quality model was used alongside the same dataset. The comparison results for a well-posed rainfall/runoff model showed that the four methods provide similar probability distributions of model parameters, and model prediction intervals. For ill-posed water quality model the differences between the results were much wider; and the paper provides the specific advantages and disadvantages of each method. In relation to computational efficiency (i.e. number of iterations required to generate the probability distribution of parameters), it was found that SCEM-UA and AMALGAM produce results quicker than GLUE in terms of required number of simulations. However, GLUE requires the lowest modelling skills and is easy to implement. All non-Bayesian methods have problems with the way they accept behavioural parameter sets, e.g. GLUE, SCEM-UA and AMALGAM have subjective acceptance thresholds, while MICA has usually problem with its hypothesis on normality of residuals. It is concluded that modellers should select the method which is most suitable for the system they are modelling (e.g. complexity of the models structure including the number of parameters), their skill/knowledge level, the available information, and the purpose of their study.

Application of global sensitivity analysis and uncertainty quantification in dynamic modelling of micropollutants in stormwater runoff

The need for estimating micropollutants fluxes in stormwater systems increases the role of stormwater quality models as support for urban water managers, although the application of such models is affected by high uncertainty. This study presents a procedure for identifying the major sources of uncertainty in a conceptual lumped dynamic stormwater runoff quality model that is used in a study catchment to estimate (i) copper loads, (ii) compliance with dissolved Cu concentration limits on stormwater discharge and (iii) the fraction of Cu loads potentially intercepted by a planned treatment facility. The analysis is based on the combination of variance-decomposition Global Sensitivity Analysis (GSA) with the Generalized Likelihood Uncertainty Estimation (GLUE) technique. The GSA-GLUE approach highlights the correlation between the model factors defining the mass of pollutant in the system and the importance of considering hydrological parameters as source of uncertainty when calculating Cu loads and concentrations due to their influence. The influence of hydrological parameters on simulated concentrations changes during rain events. Four informal likelihood measures are used to quantify model prediction bounds. The width of the uncertainty bounds depends on the likelihood measure, with the inverse variance based likelihood more suitable for covering measured pollutographs. Uncertainty for simulated concentration is higher than for Cu loads, which again shows lower uncertainty compared to studies neglecting the hydrological submodel as source of uncertainty. A combined likelihood measure ensuring both good predictions in flow and concentration is used to identify the parameter sets used for long time simulations. These results provide a basis for reliable application of models as support in the development of strategies aiming to reduce discharge of stormwater micropollutants to the aquatic environment.

Quantification of uncertainty in modelled partitioning and removal of heavy metals (Cu, Zn) in a stormwater retention pond and a biofilter

Strategies for reduction of micropollutant (MP) discharges from stormwater drainage systems require accurate estimation of the potential MP removal in stormwater treatment systems. However, the high uncertainty commonly affecting stormwater runoff quality modelling also influences stormwater treatment models. This study identified the major sources of uncertainty when estimating the removal of copper and zinc in a retention pond and a biofilter by using a conceptual dynamic model which estimates MP partitioning between the dissolved and particulate phases as well as environmental fate based on substance-inherent properties. The two systems differ in their main removal processes (settling and filtration/sorption, respectively) and in the time resolution of the available measurements (composite samples and pollutographs). The most sensitive model factors, identified by using Global Sensitivity Analysis (GSA), were related to the physical characteristics of the simulated systems (flow and water losses) and to the fate processes related to Total Suspended Solids (TSS). The model prediction bounds were estimated by using the Generalized Likelihood Uncertainty Estimation (GLUE) technique. Composite samples and pollutographs produced similar prediction bounds for the pond and the biofilter, suggesting a limited influence of the temporal resolution of samples on the model prediction bounds. GLUE highlighted model structural uncertainty when modelling the biofilter, due to disregard of plant-driven evapotranspiration, underestimation of sorption and neglect of oversaturation with respect to minerals/salts. The results of this study however illustrate the potential for the application of conceptual dynamic fate models base on substance-inherent properties, in combination with available datasets and statistical methods, to estimate the MP removal in different stormwater treatment systems and compare with environmental quality standards targeting the dissolved MP fraction.

Modelling the fate of organic micropollutants in stormwater ponds.

Urban water managers need to estimate the potential removal of organic micropollutants (MP) in stormwater treatment systems to support MP pollution control strategies. This study documents how the potential removal of organic MP in stormwater treatment systems can be quantified by using multimedia models. The fate of four different MP in a stormwater retention pond was simulated by applying two steady-state multimedia fate models (EPI Suite and SimpleBox) commonly applied in chemical risk assessment and a dynamic multimedia fate model (Stormwater Treatment Unit Model for Micro Pollutants--STUMP). The four simulated organic stormwater MP (iodopropynyl butylcarbamate--IPBC, benzene, glyphosate and pyrene) were selected according to their different urban sources and environmental fate. This ensures that the results can be extended to other relevant stormwater pollutants. All three models use substance inherent properties to calculate MP fate but differ in their ability to represent the small physical scale and high temporal variability of stormwater treatment systems. Therefore the three models generate different results. A Global Sensitivity Analysis (GSA) highlighted that settling/resuspension of particulate matter was the most sensitive process for the dynamic model. The uncertainty of the estimated MP fluxes can be reduced by calibrating the dynamic model against total suspended solids data. This reduction in uncertainty was more significant for the substances with strong tendency to sorb, i.e. glyphosate and pyrene and less significant for substances with a smaller tendency to sorb, i.e. IPBC and benzene. The results provide support to the elaboration of MP pollution control strategies by limiting the need for extensive and complex monitoring campaigns targeting the wide range of specific organic MP found in stormwater runoff.

Dynamic stormwater treatment unit model for micropollutants (STUMP) based on substance inherent properties.

Modelling the removal of micropollutants (MPs) in stormwater treatment systems is essential in a context that is characterized by a general lack of measurements. This paper presents an innovative dynamic model for the prediction of the removal of MPs in stormwater treatment systems (Stormwater Treatment Unit model for Micro Pollutants--STUMP). The model, based on a conceptual model of two-compartment (water and sediment) serial Continuous Stirred-Tank Reactors (CSTRs), can predict the fate of MPs based on their inherent properties, which are often the only information available regarding this kind of substances. The flexible structure of the model can be applied to a wide range of treatment units and substances. Based on the most relevant removal processes (settling, volatilization, sorption, biodegradation, and abiotic degradation), the model allows the dynamic simulation of the MP behaviour in the different compartments of stormwater treatment systems. The model was tested for heavy metals (copper and zinc) and organic substances (benzene and di(2-ethylhexyl)phthalate). The results show that volatilization plays a big role for removal of benzene while the removal of substances with high sorption capacity is mainly driven by settling. The model was proven to be able to predict the importance of the various fate processes for selected substances with different inherent properties. A thorough assessment of the influence of the various fate process parameters will allow a reliable assessment of the treatment performances for a wide range of MPs.

A model library for dynamic transport and fate of micropollutants in integrated urban wastewater and stormwater systems

The increasing efforts in reducing the emission of micropollutants (MP) into the natural aquatic environment require the development of modelling tools to support the decision making process. This article presents a library of dynamic modelling tools for estimating MP fluxes within Integrated Urban Wastewater and Stormwater system (IUWS - including drainage network, stormwater treatment units, wastewater treatment plants, sludge treatment, and the receiving water body). The models are developed by considering the high temporal variability of the processes taking place in the IUWS, providing a basis for the elaboration of pollution control strategies (including both source control and treatment options) at the small spatial scale of urban areas. Existing and well-established water quality models for the different parts of the IUWS (e.g. ASM models) are extended by adding MP fate processes. These are modelled by using substance inherent properties, following an approach commonly used in large-scale MP multimedia fate and transport models. The chosen level of complexity ensures a low data requirement and minimizes the need for field measurements. Next to a synthesis of model applications, a didactic example is presented to illustrate the potential of the use of the developed model library for developing, evaluating and comparing strategies for reduction of MP emissions from urban areas. Display Omitted We created a model library for dynamic modelling of micropollutant fluxes in cities.The IUWS_MP model library includes sewer, treatment options and receiving waters.IUWS_MP combines existing models for evaluating pollution control strategies.Dynamic modelling allows estimation of both MP fluxes and concentrations.A didactic example showing the potential of the developed tool is presented.

Urban drainage models – simplifying uncertainty analysis for practitioners

There is increasing awareness about uncertainties in the modelling of urban drainage systems and, as such, many new methods for uncertainty analyses have been developed. Despite this, all available methods have limitations which restrict their widespread application among practitioners. Here, a modified Monte-Carlo based method is presented that reduces the subjectivity inherent in typical uncertainty approaches (e.g. cut-off thresholds), while using tangible concepts and providing practical outcomes for practitioners. The method compares the models uncertainty bands to the uncertainty inherent in each measured/observed datapoint; an issue that is commonly overlooked in the uncertainty analysis of urban drainage models. This comparison allows the user to intuitively estimate the optimum number of simulations required to conduct uncertainty analyses. The output of the method includes parameter probability distributions (often used for sensitivity analyses) and prediction intervals. To demonstrate the new method, it is applied to a conceptual rainfall-runoff model (MOPUS) using a dataset collected from Melbourne, Australia.

Evaluation of stormwater micropollutant source control and end-of-pipe control strategies using an uncertainty-calibrated integrated dynamic simulation model.

The estimation of micropollutant (MP) fluxes in stormwater systems is a fundamental prerequisite when preparing strategies to reduce stormwater MP discharges to natural waters. Dynamic integrated models can be important tools in this step, as they can be used to integrate the limited data provided by monitoring campaigns and to evaluate the performance of different strategies based on model simulation results. This study presents an example where six different control strategies, including both source-control and end-of-pipe treatment, were compared. The comparison focused on fluxes of heavy metals (copper, zinc) and organic compounds (fluoranthene). MP fluxes were estimated by using an integrated dynamic model, in combination with stormwater quality measurements. MP sources were identified by using GIS land usage data, runoff quality was simulated by using a conceptual accumulation/washoff model, and a stormwater retention pond was simulated by using a dynamic treatment model based on MP inherent properties. Uncertainty in the results was estimated with a pseudo-Bayesian method. Despite the great uncertainty in the MP fluxes estimated by the runoff quality model, it was possible to compare the six scenarios in terms of discharged MP fluxes, compliance with water quality criteria, and sediment accumulation. Source-control strategies obtained better results in terms of reduction of MP emissions, but all the simulated strategies failed in fulfilling the criteria based on emission limit values. The results presented in this study shows how the efficiency of MP pollution control strategies can be quantified by combining advanced modeling tools (integrated stormwater quality model, uncertainty calibration).

Velocity dependent passive sampling for monitoring of micropollutants in dynamic stormwater discharges.

Micropollutant monitoring in stormwater discharges is challenging because of the diversity of sources and thus large number of pollutants found in stormwater. This is further complicated by the dynamics in runoff flows and the large number of discharge points. Most passive samplers are nonideal for sampling such systems because they sample in a time-integrative manner. This paper reports test of a flow-through passive sampler, deployed in stormwater runoff at the outlet of a residential-industrial catchment. Momentum from the water velocity during runoff events created flow through the sampler resulting in velocity dependent sampling. This approach enables the integrative sampling of stormwater runoff during periods of weeks to months while weighting actual runoff events higher than no flow periods. Results were comparable to results from volume-proportional samples and results obtained from using a dynamic stormwater quality model (DSQM). The paper illustrates how velocity-dependent flow-through passive sampling may revolutionize the way stormwater discharges are monitored. It also opens the possibility to monitor a larger range of discharge sites over longer time periods instead of focusing on single sites and single events, and it shows how this may be combined with DSQMs to interpret results and estimate loads over extended time periods.

Model-based monitoring of stormwater runoff quality

Monitoring of micropollutants (MP) in stormwater is essential to evaluate the impacts of stormwater on the receiving aquatic environment. The aim of this study was to investigate how different strategies for monitoring of stormwater quality (combining a model with field sampling) affect the information obtained about MP discharged from the monitored system. A dynamic stormwater quality model was calibrated using MP data collected by automatic volume-proportional sampling and passive sampling in a storm drainage system on the outskirts of Copenhagen (Denmark) and a 10-year rain series was used to find annual average (AA) and maximum event mean concentrations. Use of this model reduced the uncertainty of predicted AA concentrations compared to a simple stochastic method based solely on data. The predicted AA concentration, obtained by using passive sampler measurements (1 month installation) for calibration of the model, resulted in the same predicted level but with narrower model prediction bounds than by using volume-proportional samples for calibration. This shows that passive sampling allows for a better exploitation of the resources allocated for stormwater quality monitoring.