Beyond the Algorithm Challenge

We present a quantum-enhanced machine learning pipeline that quickly and accurately predicts flood progression. We use quantum-enhanced scenario encoding, which ingests environmental data including topography, land use, and weather conditions, and sends it through a quantum-enhanced variational autoencoder. This module classifies flooding behavior into representative scenarios, and the quantum layers cut training time by up to 4X compared to pure classical models. This better captures intricate dependencies and correlations in the data, boosting both accuracy and diversity. Then, for time series prediction, our optionally quantum-powered diffusion model uses Gaussian noise and neural networks to learn how flooding evolves over time for each scenario. Instead of solving complex differential equations, the model focuses on key parameters of a shallow water physics engine, and outputs a flood map that includes water levels and uncertainty quantification across regions. There is clear performance benefit of our quantum model over constant flow and classical-only baselines.

‍