Researchers led by Penn State have created an AI-powered water model that significantly enhances the accuracy and efficiency of flood predictions, potentially revolutionizing how communities prepare for and respond to natural disasters.
In a significant breakthrough, a team of researchers led by Penn State’s College of Civil and Environmental Engineering has developed an advanced computational model that enhances flood prediction accuracy and efficiency on a national scale.
Floods rank among the most destructive natural disasters in the United States, inflicting billions of dollars in damage each year. The National Weather Service highlights the pressing need for accurate and timely forecasts to mitigate these impacts.
The team’s innovative model, published in Water Resources Research, employs artificial intelligence to simulate and predict water movement with unprecedented precision.
The new model, known as a high-resolution differentiable hydrologic and routing model, integrates big data and physical readings from river networks across the country. By leveraging cutting-edge AI techniques, the model generates highly accurate flood predictions, surpassing traditional methods.
One widely-used traditional model, the National Oceanic and Atmospheric Administration (NOAA)’s National Water Model (NWM), relies on weather data to simulate streamflow rates. However, this model necessitates time-consuming parameter calibration, which involves sifting through decades of historical streamflow data.
“To be accurate with this model, traditionally your data needs to be individually calibrated on a site-by-site basis,” co-corresponding author Chaopeng Shen, a professor of civil and environmental engineering at Penn State, said in a news release. “This process is time-consuming, expensive and tedious. Our team determined that incorporating machine learning into the calibration process across all the sites could massively improve efficiency and cost-effectiveness.”
The research team integrated neural networks into their model, enabling it to recognize complex patterns across vast and dynamic datasets.
“By incorporating neural networking, we avoid the site-specific calibration issue and improve the model’s efficiency substantially,” added co-corresponding author Yalan Song, an assistant research professor of civil and environmental engineering at Penn State. “Rather than approaching each site individually, the neural network applies general principles it interprets from past data to make predictions. This greatly increases efficiency, while still accurately predicting streamflow in areas of the country it may be unfamiliar with.”
What sets this model apart is its hybrid approach, combining the strengths of both physics-based models and machine learning. This integration allows for more accurate extreme event predictions, crucial for anticipating severe weather scenarios.
“The old approach is not only highly inefficient, but quite inconsistent,” Shen added. “With our new approach, we can create simulations using the same process, regardless of the region we are trying to simulate. As we process more data and create more predictions, our neural network will continue to improve. With a trained neural network, we can generate parameters for the entire U.S. within minutes.”
The researchers trained their model using 15 years of streamflow data from 2,800 gauge stations, supplemented by weather data and detailed basin information from the United States Geological Survey. The resulting predictions showed a 30% improvement in accuracy compared to the NWM, especially in unique geological areas.
“Once the model is trained, we can generate predictions at unprecedented speed,” Shen added. “In the past, generating 40 years of high-resolution data through the NWM could take weeks, and required many different super computers working together. Now, we can do it on one system, within hours, so this research could develop extremely rapidly and massively save costs.”
Beyond flood predictions, the model’s capabilities extend to forecasting other significant events like droughts, which could inform water resource management and have substantial implications for agriculture and sustainability research.
“Because our model is physically interpretable, it can describe river basin features like soil moisture, the baseflow rate of rivers, and groundwater recharge, which is very useful for agriculture and much harder for purely data-driven machine learning to produce,” added Shen.
Looking ahead, the team’s model is a contender to be integrated into the next generation framework of the NWM being developed by NOAA, promising to set new standards for flood forecasting across the country. As Shen emphasized, ensuring users’ comfort with the AI component through rigorous independent evaluations will be crucial for broader adoption.
The study includes contributions from other Penn State researchers and collaborators from various institutions, highlighting the collaborative effort behind this groundbreaking advance.
