Real-Time Monitoring and Detection of Flash Floods in Smart City

Introduction: 

Storms and floods cause 70% of the world's natural disasters. These natural disasters affect on average up to 200 million people in a year, with economic losses between US$50 billion to US$100 billion annually. National Weather Services (NWS) reported 28,826 flash floods events in the United States from October 2007 to October 2015, which resulted in 278 live loss. Monitoring these flash flood prone areas proactively in populated areas and providing just-in-time notification to the city officials can help in effectively prioritizing, controlling, and mitigating such disastrous events.  Solving flash floods problems helps improve the flow of traffic, reduce traffic congestion, environmental hazards, and formation of stagnant water, and prevent damages of any surrounding properties, premises, and loss of lives. This project is building a portable wireless sensor network-based hardware system to monitor the rising level of water, designing distributed predictive data analytics algorithms with human in the loop, and providing smartphone based real time notifications to city officials. The project proposes to integrate human observations from micro-blogging feeds to forecast the manifestation of the flood events by considering real-time contextual information of the target event. The project will be deployed in several flash flood prone areas in Ellicott City and Baltimore City in partnership with the Howard County and Baltimore County local government officials.

Project Goal: 

The  NSF CNS EAGER Project # 1640625 seeks to build  distributed Data Analytics for Real-Time Monitoring and Detection of Flash Floods in Smart City.  In this work, we built many protocols and continued to iterate on various Flash Flood Detection using Physical Sensor as well as Social Sensor(s) as our data points . We then fused these two data sources and tried building and end to end system on one . 

Team Members:

Principal Investigator:

Co-Principal Investigator:

Graduated Students:

Current Graduate Students:

Acknowledgement: 

This research is supported by the National Science Foundation under Award Numbers: 1640625. 

Community Partners:

Research Problem: Monitoring and Detecting the Flash Flood in Ellicott City Area

Solution: The research will pursue the junction of wireless sensor networks-based data analytics and real-time social network data such as tweets. 

UMBC - Howard County Government Collaboration

UMBC is collaborating with city partners to measure and compare the key performance improvements over existing municipal systems or protocols in practice and the viability of the proposed system for practical city usage and demonstration as part of the Global City Team Challenge program. 

Flash Flood Cyber Physical Design Development and Progressions

Deployment History

Over the past three years we have built and experimented with several IoTs and Flash Flood detection mechanism. Some of our prototypes and our setups along the way are depicted on these images.

Physical & Social Media Data Fusion : 

We are currently monitoring Flood Conditions from our 8 Sensors deployed in Ellicott City Main Street Area. We collect ambient site conditions such as Weather, Rainfall amounts and Water Level in the area. We then transmit the data to Twitter and scrape social media engagement from users based on Tweets.

Research Summary and Progress (Major Milestones)

Our research is focused on detecting and monitoring flood related information with the help of mainly two kind of sensors (I) The physical sensor and (II) The social sensor and underling architecture to support them.

Sensor Locations (Ellicott City,MD)

Main Component

(CPU and Solar)

 Water Measuring Unit

UMBC FloodBot: Our Community Watch Dog & Public Relation End Point

Our research is focused on detecting and monitoring flood related information with the help of mainly two kind of sensors (I) The physical sensor and (II) The social sensor and underling architecture to support them.

Natural Language Understanding from Social Media Sensing

Tweets Word Cloud Cluster 

Application of Transfer Learning in Textual Domain

Transfer Learning for Flood Using ULMFiT

Datasets

Physical & Social Media Manifested by Water Level Peak to Tweet Intensity

Software

Multiple Deep Learning Frameworks 

RealTime Weather API

Pre and Post Analysis  Twitter Scraping Code and Frameworks


Presentations

Press Release

Publications

Last Updated: 09/18/2020