FloodWatch

FloodWatch is a smart city project centered around the development of cyber physical system infrastructure for disaster intelligence and forecasting. The current location of interest for this project is Vietnam.

With a growing network of sensors that provide data that drives flood intelligence, it is imperative that the readings of each sensor are verified for accuracy. This work employs time-series anomaly detection using AnomalyBERT and proposes an automatic mechanism for batch inferencing.

Flooding is a key problem in South Eastern Asia. This research focuses on providing an accurate and robust method for forecasting for flooding by leveraging elevation for locality, precipitation, and rate of increase for wind, tide, and precipitation. The novelty that is provided includes ensemble learning with voting algorithms for LSTM time series models and Regression models leveraging the previously mentioned variables.

We propose and demonstrate a pipeline for validating large amounts of crowdsourced images using a convolutional neural network (CNN) and automatic real time weather data retrieval. We make use of multithreading and cross validation for performance/efficiency and accuracy respectively.

T-Mobile

Ivy

Hospitality Innovations

UVa Engineering International Programs Office