Predicting Flash Floods

Climate change has led to unprecedented torrential rainfall in our region, causing devastating floods. This challenge presents an opportunity to apply quantum machine learning (QML) to enhance flash flood prediction. We seek innovative QML solutions that address the complexities of imbalanced datasets and demonstrate practical implementation. The goal is to improve prediction accuracy and efficiency, ultimately contributing to disaster mitigation efforts.

Flash floods can pose a significant threat to communities, infrastructure, and human life. According to the United Nations Disaster Risk Reduction Office, the devastating impact of recent floods, such as those caused by Cyclone Freddy and Libya's floods, underscores the urgent need for effective disaster management. These events resulted in thousands of fatalities and billions of dollars in damages, highlighting the critical importance of accurate and timely flood prediction.

This challenge aims to develop a robust and accurate flash flood prediction model using quantum machine learning techniques. The model should provide early warnings for severe flash floods to enable effective disaster management, which depends on the quality of prediction and estimation of disaster losses. Prediction is associated with complex and multi-source datasets, covering weather forecasts, geological surveys, and water and land geospatial datasets, which can be accessible in different formats for different user needs. Loss estimation is often biased towards insured assets due to the availability of comprehensive data. Uninsured losses are less certain due to fragmented data. Countries with higher insurance coverage tend to have more reliable data. Participants are tasked with using a Quantum machine learning module to process quantum kernels on the provided dataset to effectively address challenges associated with flash flood prediction.

Traditional machine learning methods have been used for flash flood prediction. However, these approaches often face challenges due to heterogeneity and imbalanced datasets.

Participants are invited to develop and optimise their quantum circuits using our QISKIT transpile function, as illustrated in the second step of QISKIT Patterns- https://docs.quantum.ibm.com/guides/intro-to-patterns. Subsequently, we highly encourage participants to run their circuits on a fake backend using the QISKIT Runtime local testing mode- https://docs.quantum.ibm.com/guides/local-testing-mode to achieve comparable results to the ones we demonstrated in the references below to ensure consistency.

We used a dataset of various meteorological and geographical factors, including precipitation levels, land characteristics, and elevation data. Participants are encouraged to use this dataset for this use case.

Submissions will be evaluated based on the agreed-upon criteria set by the summit organiser: