Flood Early Warning System

From DRLWiki

(Redirected from floodews)
Jump to: navigation, search

Contents for This Page

Problem Description

Radio Station In Honduras

In developing countries, warning communities of an incoming flood is an expensive proposal given their limited resources. The equipment necessary for early warning flood systems is expensive and centralized to support flood detection schemes that are computationally complex. This project explores new techniques for distributing the computation of flood detection within a wireless sensor network, grounding the research in reality through the design and installation of an early warning system for flooding in a developing country.

Sensor networks for flooding do not exist in a real-time automated capacity within developing countries. We are developing such a sensor network and examining how to distribute and adapt the flood prediction calculation. Current flood prediction relies on physically-based models requiring large amounts of data in order to calibrate the model, and significant computation power to compute the model. Neither of these being feasible on a sensor network as data storage and computing power are limited, we are working on statistical algorithms capable of online, distributed computation.

Given a suite of robust adaptive distributed algorithms for flood warning, we are planning to implement these algorithms on a system that can be deployed within a developing country. This poses a number of challenges, ranging from survival in harsh flood environments, lack of trained personnel to maintain the system, and an illiterate populace to warn. Designing and implementing a system that addresses these issues poses interesting engineering problems. Key research questions here include: (1) How does each sensor node measure its environment? (2) How does each sensor communicate with the rest of the network, and convey its ongoing status given the worst weather conditions possible on a river? (3) How and where should the information from the sensors be processed and fused?

Understanding the sensor system and computational design issues requires implementing the system within a developing country. We are working with a local non-governmental organization, the Fundacion San Alonso Rodriguez in Tocoa, Honduras to develop a test system in the Rio Aguan basin. Our two organizations have worked together since January 2004, exploring issues related to adapting technologies to solve developing world issues. This area regularly suffers from severe flooding so, with the support of FSAR and the Honduran communities, this location epitomizes an ideal testbed for the installation of a prototype system.

Flood Prediction Algorithm

Performing the prediction requires an algorithm simple enough to run on a sensor network in a distributed fashion. The distributed requirement derives from the systems problems faced by covering a large geographic area such as a river basin. To do so requires long communication links, which will be both expensive (from a power standpoint) and slow to send data along. This will restrict the data we can send along the links, limiting the data available for the model to use for calibration and prediction as well as limiting the flexibility of the model. We would like to modify measurement rates, change transmission rates, recalibrate the model, and other data related activities based on the model prediction and measured data. If the model and data are separated by slow, expensive communication links, we cannot achieve all of these goals. Therefore, we need a distributed model that can self-calibrate.

We developed a multiple linear regression model that predicts river flooding: Overview of Multiple Linear Regression Model

As shown in the overview, the model consists of three stages: (1) data collection, (2) calibration, and (3) prediction. We tested the model in simulation using 7 years of data from the Blue River in Oklahoma, discovering that it works quite well for flood prediction. Currently the model distributes the data and prediction stages. In addition to the simulation, both of these stages were tested running on the system on the Charles River at Dover, Massachusetts. We are working to distribute the calibration stage of the model and are running tests in simulation.

Details of Multiple Linear Regression Algorithm

River Monitoring Sensor Network System

For the sensor network, we deal with several constraints:

  • Measure and communicate over a large geographic area
  • Communicate data in real-time
  • Survive long-term exposure
  • Recover from faults
  • Minimize costs
  • Predict the event of interest
  • Distribute the computation needed for the prediction

We designed a system architecture to meet these goals. An graphical overview of the system is:

We custom-built the hardware and software necessary to implement this system. The base system consists of an ARM7 processor, Mini-SD card storage, an Aerocomm 900 MHz radio, and expansion connectors to attach a wide range of daughter boards. We expand the functionality through the daughter boards, creating two versions: one to support the sensor measurements and the other to support the 144 MHz modem plus associated power electronics.

During several trips to Honduras, we tested the communication ranges, prototyped several water level sensors, and constructed needed infrastructure. System tests of the sensing nodes occurred the past two falls in the Charles River at Dover, Massachusetts with the latest test including the distributed prediction computation (using calibration coefficients derived from the fall 2007 data). Currently, we plan a final field experiment in Honduras in January to test the distributed computation running on the full system.

Publications and Additional Information

Elizabeth Basha, Sai Ravela, Daniela Rus - Model-Based Monitoring for Early Warning Flood Detection
Proceedings of the 6th ACM Conference on Embedded Networked Sensor Systems (SenSys) , Raleigh, NC, November 5-7 2008
Pdf Bibtex
Author : Elizabeth Basha, Sai Ravela, Daniela Rus
Title : Model-Based Monitoring for Early Warning Flood Detection
In : Proceedings of the 6th ACM Conference on Embedded Networked Sensor Systems (SenSys) -
Address : Raleigh, NC
Date : November 5-7 2008
Elizabeth Basha, Daniela Rus - Design of Early Warning Flood Detection Systems for Developing Countries
Proceedings of the Conference on Information and Communication Technologies and Development , Bangalore, India, December 2007
Pdf Bibtex
Author : Elizabeth Basha, Daniela Rus
Title : Design of Early Warning Flood Detection Systems for Developing Countries
In : Proceedings of the Conference on Information and Communication Technologies and Development -
Address : Bangalore, India
Date : December 2007
Case Study by Microsoft Research - Wireless Sensor Network Provides Early Flood Detection for Underserved Countries
Pdf Bibtex
Author : Case Study by Microsoft Research
Title : Wireless Sensor Network Provides Early Flood Detection for Underserved Countries
In : -
Address :
Date :

People

Primary Researchers

Elizabeth Basha

Daniela Rus

Sai Ravela

Piotr Indyk

Implementation Contributors

Current:

Alexander Bahr

William Boos

Carrick Detweiler

Marek Doniec

Brian Julian

Andrea Llenos

Iuliu Vasilescu


Past:

Patrick Barragan

Alejandro Flores

Erika Granger

Anthony Parolari

Sarah Rich

Tadd Truscott

Collaborators

Fundacíon San Alonso Rodríguez, Tocoa, Honduras: Gines Suarez, Director

Funding

  • Xerox
  • Microsoft Research
  • National Science Foundation
  • Carrol Wilson Award
  • Thrivent
  • IDEAS Competition
  • MIT Public Service Center
  • MIT D-Lab

Links

Personal tools