Modelling floods
How can we predict the unpredictable?
Flooding can have a huge impact on communities, with severe events leading to property damage, injury and even loss of life. Large swathes of Australia’s population live on floodplains, so it’s important that we understand our flood risk and plan for it as well as possible. To try to quantify flood risk, we need to understand how much flow will be produced for a given rainfall event in a given location. We can estimate this using hydrologic models.
There are many types of hydrologic models used for different purposes. The simplest and most widely used in industry are event-based models. They model a single flood event using information about the rainfall storm and the catchment. Usually we’re interested in modelling a design storm, which is a hypothetical storm with a particular probability of occurring (e.g. a flood that has a 1% chance of occurring in any given year). We create a synthetic rainfall storm based on information from rain gauges in the local area, then apply it in an event-based model of the catchment.
Different event-based models require different information, but usually this includes catchment area/shape, the proportion that is paved and the slope. The model also needs some information about the conditions in the catchment before the storm (we call these ‘antecedent conditions’). If the catchment is dry, a lot of rain will soak into the soil, so it might not produce much flow. On the other hand, if the catchment is very wet, nearly all of the rainfall is likely to run off. For this reason, very large floods are sometimes produced by rainfall that is not actually extreme (and extreme rainfall events sometimes don’t produce floods).
For an event-based model, we have to make an informed decision around the wetness of the catchment prior to the design storm we want to model. We can do this by testing our model on past events where we’ve recorded the streamflow. If we put the recorded rainfall in our model and the results match the recorded streamflow, then it’s likely that we’ve got the catchment conditions about right. However, because the wetness of a catchment changes all the time, assuming that information from recorded storms will characterise the conditions prior to design storms (which aim to represent potential future events) is a key source of uncertainty in flood risk management.
If we want a more accurate representation of the conditions in a catchment at different times, we can use a continuous model. These models simulate flow over a long period of time and keep tabs on the moisture stored in the catchment as they go. This means that, when it comes to simulating a large rainfall event, the model automatically accounts for the catchment conditions preceding the event. This is a key benefit of continuous simulation. However, even the simplest continuous models require more data than event-based models (rather than just creating one hypothetical rainfall event, you need a whole series of rainfall).
Continuous models vary a lot in their complexity. Some are lumped, which means that they represent the whole catchment as one homogeneous area without accounting for spatial variability in soil or terrain. These models need to be quite heavily parameterised to simulate flow accurately, since they don’t contain much physical information about the catchment or detailed representations of catchment processes.
Distributed continuous models are another alternative; they divide the catchment into grid cells and require information specific to the location of each grid cell. This means that they can capture spatial variability in the catchment. Some distributed models only simulate hydrologic processes (like overland runoff, streamflow and groundwater flow), but others are very complex, modelling physiological processes like vegetation growth and nutrient transfer. Complex models like these are rarely used by engineers working in flood risk management, but they are used in research to help us understand different catchment processes. This information can inform potential improvements to the practices used by engineers in industry.
Catchments are complex beasts and none of the models we have can characterise them perfectly. Data can also be difficult to obtain (one researcher wryly pointed out that, if we took enough samples to characterise the spatial variability in soil structure, we’d be modelling a catchment full of holes!). However, if we choose the right modelling tool for the right application, we can usually improve on studies that consider observed data alone.
The whole process becomes more complicated in a changing climate. What if the data we’ve recorded, which forms the basis for all of our modelling decisions and our model parameterisation, is not representative of what will happen in the future? What dynamic ecological processes could be key drivers of change in hydrologic systems? This is the focus of my research and I’ll be going into more detail in upcoming posts.
