Photo credit: www.sciencedaily.com
Improving Flood Evacuation Protocols with Machine Learning
Recent flooding events in Spain and other regions have underscored the critical importance of providing timely warnings to communities at risk. In response to this need, a new study published in the journal Hydrology offers insights into enhancing flood evacuation strategies through a machine-learning model developed by researchers at Concordia University.
PhD candidate Mohamed Almetwally Ahmed, alongside Samuel Li, the professor and chair of the Department of Building, Civil and Environmental Engineering, has devised a method leveraging artificial intelligence to yield more precise short-term predictions of river discharge.
The researchers utilized historical hydrometric data combined with innovative weather-based predictors. Their study focused on measuring advection—the velocity of water flow—between two hydrometric stations along the Ottawa River. They designed a test case involving two stations located approximately 30 kilometers apart; one had been inactive for years while the other remained operational.
To enrich their model, the researchers integrated decades of historical data from the Government of Canada, augmented by recent information regarding rainfall, temperature, and humidity levels. This compilation allowed their machine-learning model to generate reliable forecasts of daily river discharge and to provide real-time assessments of water flow at specific points along the river.
“Sub-diurnal forecasting—predictions made within a 24-hour window—is primarily utilized for evacuation purposes. Our approach enhances the accuracy of these forecasts compared to traditional daily or multi-day predictions,” Ahmed notes. “The precision of our results increases as the time frame of the forecast narrows.”
A Transparent and Transferable Model
The research builds upon a well-established algorithm known as the group method of data handling. This algorithm constructs predictive models by categorizing and amalgamating data into groups, which are then processed repeatedly in different combinations until the optimum and most reliable set of data emerges.
“Our model incorporates nine predictors—seven weather variables alongside historical data from the two stations. It continuously ranks and reorganizes these parameters to form different combinations, ultimately making an optimal selection of predictors. Importantly, it does not necessarily utilize all predictors uniformly, focusing on those that offer the highest accuracy,” Ahmed elaborates.
The model is dynamic and varies according to the forecasting time frame. For instance, the parameters used to predict discharge twelve hours in advance differ from those for eight, nine, or ten hours ahead.
Additionally, the model’s adaptability extends to different waterways. To validate its effectiveness, Ahmed applied further analyses to data from the Boise and Missouri rivers in the United States.
“As we refine this technique, we anticipate developing it into an operational tool, enabling users to access river discharge forecasts on their mobile devices, similar to how they check weather updates,” Li explains. “Rather than simply reporting temperature or rainfall projections, we can provide actionable insights on water levels.”
For Ahmed, who is dedicated to advancing flood evacuation readiness, this model represents just one of many tools he envisions for communities facing potential flooding crises.
“I aspire for authorities to leverage this information as a vital resource for their models concerning flood-prone regions,” he states. “With this tool, we can enable them to ascertain which routes are available for evacuation, thereby equipping local transportation systems with immediate, actionable plans that can be pivotal in safeguarding lives and property.”
Source
www.sciencedaily.com