Predicting the Unpredictable: Innovative Methods in Flood Risk Management

Flood events can cause significant damage, both economically and to the built and natural environments – and as the climate changes, these events are likely to become more frequent and more severe. It is critical to stay ahead of these risks by using climate models, real-time data and machine learning.

Flooding can be viewed very simply, at first glance: too much water. But beneath that surface understanding, flooding can be incredibly complex, resulting in direct and indirect impacts on the built and natural environments, as well as people and their communities. It does not take much effort for the mind to conjure up images of flooding events in our own backyards, such as the infamous August 2018 flooding which turned Toronto streets into rivers, or the 2017 flooding in Montréal and Laval.

With population growth, rapid urbanization and changes in precipitation patterns related to climate change, mitigating flood risk is more complicated than ever. These changes present unique challenges for both for long-term and short-term planning horizons and as a result, we require an advancement in the standard method used within the field of flood risk management.

What is flood risk?

Flood risk can be broadly defined as the combination of two factors: (1) the probability of a flood event occurring, and (2) the consequences associated with that event, whether they be socio-economic and/or environmental. Simply put, when estimating flood risk, we ask “what’s the likelihood of a flood happening?” and “what are the impacts to the surrounding people and environment?” While the later question is typically the focus of economic planning, engineers in flood risk management dedicate efforts towards estimating the likelihood of the flood event occurring. Ultimately, the quality of prediction lies in the ability to accurately estimate precipitation, runoff and how flows are conveyed through a drainage system. A primary occupation of the field of flood risk management is in refining our methodology for quantifying the above with the greatest accuracy and efficiency.

Where we are today

The design storm approach has been widely used for drainage system analysis since its development in the early twentieth century. Design storms (specifically intensity-duration-frequency curves) use mathematical functions to relate rainfall intensity with its duration and frequency of occurrence. They are popular in practice because they can be easily constructed, they provide conservative peak flow estimates, and can be standardized for regulatory purposes. However, there are significant drawbacks to this approach that limit their applicability in assessing flood risk.

With the advancement of computing power and modelling capabilities, there has been a shift to continuous simulation approaches which employ long-term rainfall records as input to generate long-term statistics of system response. Such approaches address some of the glaring limitations of the design storm approach. Namely, continuous simulation allows us to account for antecedent conditions (e.g., existing system water levels and soil moisture conditions from previous storm events), and also permits the evaluation of drainage system response to a wide range of rainfall distributions rather than an average. This presents the added benefit of understanding system level of service over the entire spectrum of conditions under which the system will operate in its lifetime. But if climate change is altering precipitation patterns, where do we go from here?

Flood risk, climate change, and technology

The continuous simulation approach has proven to be relatively effective under assumptions of stationary climate, meaning that historical data is representative of the future. However, with climate change altering the spatial and temporal distribution of rainfall, more advanced methods are required in order to develop a better understanding of how system performance, and the occurrence of flooding, will change over time. To do this, we turn to climate models.

Climate models are our most advanced tools for simulating the global climate system response to increasing greenhouse gas emissions. They are also important tools for understanding past patterns in precipitation and projecting future changes. By leveraging advancements in climate modeling, we can build capacity for modifying historical precipitation data to reflect projected changes; these capabilities are key elements in advancing our understanding of flood risk in a dynamic future. Now, there are inherent limitations with the direct application of coarse resolution (i.e. large grid spacing and large time-steps) climate model output data to local flood impact assessments. However, when it comes to the future planning and management of drainage system infrastructure, they provide the best means for understanding how system performance may change over time.

A secondary approach for improving flood predictions relies on a network of remote weather stations and continuous flow monitoring data. With recent advances in machine learning methods – a branch of artificial intelligence that utilizes concepts like pattern recognition – algorithms can process big data sets to make powerful predictions by mapping the relationships between system inflows and system response. In developing a predictive relationship, decision-makers will be able to predict flood events in real time to allow for adaptation measures in response to a predicted flood event. In the future, the acquired real-time data has the added benefit of complimenting the future planning analysis by providing valuable insight into the spatial distribution of precipitation at the catchment scale and by providing an indication of how changes in climate and land use are impacting flood events.

Although applications of machine learning for flood prediction are still in the initial stages of development, the idea is exciting to consider. It’s very possible that this is just the tip of the iceberg (although they are shrinking due to climate change!) of integrating climate modeling, big data, and machine learning into the future of flood risk management.