Researchers have developed a new method utilizing artificial intelligence to enhance predictions of short-term river discharge by analyzing historical data from two hydrometric stations located on the Ottawa River, along with various weather-related factors. This approach builds upon an existing algorithm known as the group method of data handling, which focuses on creating predictive models through sorting and merging data into groups. The researchers continuously calculate different combinations of data until they identify the most reliable set.
Recent floods in Spain and other regions have underscored the importance of timely warning for potential flooding, which can help protect lives and property. A study published in the journal Hydrology offers insights that may enhance flood evacuation strategies through a machine-learning model crafted by researchers at Concordia University.
PhD candidate Mohamed Almetwally Ahmed, along with Samuel Li, who chairs the Department of Building, Civil and Environmental Engineering, developed an innovative method that leverages artificial intelligence for more precise predictions of short-term river discharge.
The researchers utilized historical data combined with a new set of weather-related predictors to measure advection—the speed at which water moves—between two hydrometric stations along the Ottawa River. They tested their model using data from two stations situated about 30 kilometers apart, where the downstream station had been inactive for many years, while the upstream station remained functional.
The historical data gathered over decades by the Government of Canada was enhanced with information related to rainfall, temperature, humidity, and other relevant factors. When these variables were input into the machine-learning model, they produced dependable daily discharge estimates and real-time information regarding the water flow at specific points in the river.
“Forecasting periods under 24 hours is primarily utilized for evacuation efforts. Our method provides more precise forecast probabilities compared to daily or longer-term predictions,” says Ahmed. “These forecasts are probabilistic, and the accuracy improves as the forecast period gets shorter.”
A clear and adaptable model
The researchers relied on the group method of data handling algorithm, which constructs predictive models by organizing and combining various data points into groups. The model calculates different combinations repeatedly until it identifies the most effective and trustworthy data mix.
“In our approach, we incorporate nine predictors: seven weather factors plus historical data from the two hydrometric stations. The model organizes and ranks these elements to generate numerous combinations, ultimately selecting the most accurate predictors. It is important to understand that not all predictors are necessarily used or weighted the same; only those yielding the best results are chosen,” Ahmed clarifies.
The model adapts based on the time frame for predictions. For instance, the model designed to forecast discharge 12 hours in advance will differ from one predicting for 8, 9, or 10 hours.
Additionally, this model varies between different rivers. Ahmed further applied the technique to data from the Boise and Missouri Rivers in the United States for testing purposes.
“As this method continues to develop, we expect to implement it operationally, enabling individuals to check river discharge predictions on their mobile devices, similar to how they view weather forecasts,” Li notes. “Instead of predicting future temperatures or rainfall, we can provide water level estimates.”
For Ahmed, who is focused on enhancing flood evacuation strategies, this model represents just one resource he envisions authorities could use to prepare for severe flooding events.
“I hope they can integrate this data into their models for areas at high risk of flooding,” he states. “With this tool, we can help determine which roads would be available for evacuation, supplying local transportation systems with real-time action plans that could save lives and shield property from harm.”
