The impact of network simplification on property flood risk predictions


Courtesy of Innovyze

A recent project, set up to compare the merits of simplified and desimplified models in InfoWorks CS, has proved that the solution’s fast and robust simulation capabilities are now so advanced that highly-complex models can provide far greater accuracy at competitive speeds than smaller, simpler models.

As the water industry moves towards a risk-based approach to flooding, the existing sewer models are the main tools being used to establish the levels of risk. During the initial stage of this United Utilities project, consultant MWH was tasked with determining whether these models were fit for purpose, if they contained sufficient detail to identify property flood risk, whether they accurately replicated the effects of extreme events, and the level of confidence that could be obtained in the results.

MWH proposed a study examining a number of scenarios to gauge the effect of the existing InfoWorks CS models against a more complex model by adding data in local sewers and property connections. The consultancy also planned to use digital terrain models (DTMs) as well as historic records as the basis for the models, and to determine how modeling in 1D and 2D affected the overland flow paths.

The study area chosen was a recently-modeled part of the city of Blackburn, Lancashire, for which a great deal of real-life data existed including 16 flow monitoring points and a 1m resolution LIDAR digital terrain model. Importantly, there were records of existing flooding in the catchment that provided sound data for calibrating and validating the flood risk scores. There were a number of properties in the area registered as being at risk of flooding (DG5), and the flow monitor records of external flooding gave reasonable confidence in the original model.

The flood risk assessment examined two flooding mechanisms under the various scenarios: overland flow from flooding manholes, whose route encompassed properties for which the potential impact could be assessed; and sewer surcharging causing floodwaters to back up into basements or properties themselves.

The study

MWH created a series of scenario models, based on the existing model. The consultancy also created desimplified models that incorporated local sewers, property laterals and digital LIDAR terrain data. In two models, an InfoWorks CS 2D mesh was created to determine the effect on the model results. At each stage, the modelers undertook a flood risk assessment to determine the risk of surcharge and overland flows at each property.

Model validation

MWH determined the flood risk for every property from the baseline model - the results agreed well with the properties on the DG5 “at risk” register, indicating reasonable calibration. However, the results also suggested that many more properties were potentially at risk than actually were – 916 properties out of 4500, a proportionately large number - due to the lack of model detail.
The baseline model of the 1D overland flows also indicated a small amount of external flooding at a known “at risk” location, which again suggested the calibration was reasonable.


The first stage of this process involved reinserting the local sewers from the original asset database into the model. When this extra level of detail was included, any surcharge or back-up in the catchment during an extreme event would (as in real life) utilize the storage in the local network rather than being represented as discharging from manholes. The modelers also removed the standard storage compensation from the model.

Because the initial runs had predicted such a large number of properties were at risk, the modelers wanted to find ways to improve the flood risk prediction. Just re-running the desimplified model with the local sewers added halved the number of properties flagged as at high risk.

The impact on the overland flow was not significant in the 1D desimplified model, partly because relatively few properties were affected. The model still predicted flooding at the main manholes that had previously been predicted to flood, and the same flood routes were identified.

Using the InfoWorks CS 2D module the number of properties at both high and medium risk increased, due to the use of real flow paths. The model accurately identified an area that required further, more detailed investigation.

Looking in more detail at the increased number of properties deemed at risk, it was realised that even very small amounts of floodwater would route through the 2D model and predict small depths of 1mm or 2mm relative to the property. Because the model was attempting to determine relative depths, such insignificant amounts of water were still highlighted as a potential risk. When the results were analysed, it was found that at 78% of the properties the depths were less than 50mm. Inserting a threshold to remove some of the very small depths around properties reduced the number deemed at risk to 200.

Modeling laterals

The next stage of modeling involved adding lateral connections. Not all of the property-level sewer details were available so a fairly simplistic approach was taken, while ensuring that adding the laterals did not insert too much storage into the model as this would have distorted the results.

Moving to this greater level of detail meant even more properties were shown to have a lower flood risk – the surcharge impact modeling results reduced the figure to 118, a much more realistic representation of the potential risks in the catchment. The modelers found that the amount of detail in the model had to be balanced against run times.

With overland flow modeling, inserting the laterals again reduced the number of properties found to be at risk though not as significantly as for surcharges. The model still showed flooding from manholes, a mechanism for overland flow.

Summary of results

Moving to increasing levels of model detail ultimately reduced the number of properties at high risk of flooding during an extreme event from the 900-plus predicted in the original model to around 120. As this was a much more accurate representation of the true level of risk, the consultants concluded that if models are being used for flood risk assessments there is benefit in this greater level of detail – perhaps focused on particular areas where a robust flood risk assessment is required.

MWH also assessed the level of confidence in the results for each property, using criteria such as the distance between the property and model manhole or from the actual sewer. The relative difference between the cover level and LIDAR data was also assessed to provide further confidence. This increased the confidence level from 20% in the original baseline model to a high confidence level of 64% for the most detailed version, the lateral model.

Desimplification was found to reduce the surcharge risk by around 54%, just using the existing sewer model as a flood risk predictor. An 80% reduction in overland flows was also achieved. The rebuild using the DTM did not significantly affect the predictions.
It was found that the key was to balance model run and build time against the level of confidence in the results. Although the lateral model ran significantly more slowly than the original models, it was not prohibitively slow.

The quality of the LIDAR data was also found to be critical. The ground survey (see box) was relatively inexpensive and a significant number of manholes could be checked in a day, giving greater confidence in the model and allowing the consultant to correct some anomalous results.

Next steps

MWH aims to repeat the analysis across a range of catchments and different topographies to determine whether the results from the project can be replicated. The results of this extended work could have implications for the amount of storage compensation that is built into models for extreme events, as this initial work proved that modeling local sewers in InfoWorks CS gave a significantly better prediction of surcharge, top water levels and hydraulic grades.

Customer comments

No comments were found for The impact of network simplification on property flood risk predictions. Be the first to comment!