FireWxNet: A Deployment of Wireless Sensor Networks In the Forest Fires of the Western U.S.* Speaker: Professor Rick Han MANTIS Research Group Department of Computer Science University of Colorado at Boulder *Best Paper Award at ACM MobiSys 2006, with coauthors Carl Hartung, Saxon Holbrook, and Carl Seielstad Forest Fires - Costs • Substantial cost in human lives, property damage, firefighting expenses, etc. – – – – – Deaths and displacement resulting from wildland fires Billions in property damage (Oakland Hills 1991 cost $2.4 billion) In 2005 in the U.S., 66,000 fires burned 8.7 million acres $1 billion spent to fight fires by U.S. government 18,000 fire fighters available for 2006 • WSN technology can help save lives, reduce all of these costs – By providing finer granularity scientific data - weather Bitterroot National Forest, Montana, 2000 Weather and Forest Fires • Temperature – As Temperature goes up: fuels dry out, fuels temp rises, etc.. • Relative Humidity – 10 minute fuels (small twigs) change with RH quickly. – 10,000 hour fuels (large trees) not very affected. • Wind Speed and Direction – Feeds fire with extra oxygen. Dictates direction. Can change quickly. • Example: Temperature Inversions and Thermal Belts – Temperature is colder at lower elevations. – Fires above the inversion continue to burn actively. – Band of warmer air trapped midway up the mountain. – When they break – increased winds, rapid increase in temp. Temperature Inversion Temperature Inversion Forest Fires – Current Ways to Measure Weather Belt Weather Kit Disadvantages: poor coverage, often forgotten, slow, etc. Current Ways to Measure Weather Remote Automated Weather Station (RAWS) Disadvantages: poor coverage, expensive, etc. FireWxNet Design Goals • Deploy a WSN to monitor weather data around prescribed forest fires – Temp, RH, wind – Actually deployed around unplanned lightning-strike wildfires – In collaboration with the University of Montana • Must be rugged, ondemand/portable, inexpensive, long-lived, easy to deploy, etc. – Did not focus on survivability of nodes, but system is robust to node failures • Deployment difficulties => sparse WSN – Focus on improved coverage of the thermal inversion belt – well-known but difficult to detect – capturing vertical elevational CU’s Carl Hartung deploys WSNs temperature differences, not horizontal, which would require 100’s in the fire area via helicopter of sensor nodes FireWxNet’s Hierarchical Architecture Soekris Base Station FireWxNet’s Wireless Sensor Network • MICA2, 900 MHz, 2 AA batteries • Mantis OS (MOS) • Commercial sensors – e.g. Relative Humidity: Humirel 1520, extremely accurate below 30% RH Networking Design: Deployment, Duty Cycling, and Routing ON Deployment Hears Beacon After 14 Minutes Sleep After ~1 Minute Awake: Sense and Send 6% Duty Cycle – maintained by the beacons Ease of Deployment LOCATE: A FOUND: A New Node A • Used to set up symmetric links where possible • Crude but effective, but could be improved Sensor Network – Routing • During 1 min Awake period, – Beacons flooded every 4 seconds from Base – Establishes tree-structured routing – Data was sent 1 packet per second for 60 seconds A (1) Base C (2) B (1) D (3) • Fault tolerance: multipath above, plus – If no beacon heard within 10 seconds (2 ½ beacon cycles) with DTB <= MyDTB -1, reset DTB and look for new beacon with any DTB. – Adapts to failed, moved, shut off nodes FireWxNet Deployment: Bitterroot National Forest, Idaho A) Base Camp B) Ranger Station C) Boulder Peak Relay D) Hell’s Half Acre (SN) E) Kit Carson (SN) F) Spot Mountain • WSNs deployed within 1 km of fire • One-week deployment, then it snowed Network Topology and Performance Kit Carson Hell’s Half Acre • Kit Carson Yield at Base station • Hell’s Half Acre Yield • 36% Overall • 80% Effective • 47% Overall • 74% Effective Base Station Packet Reception Rates Hell’s Half Kit Carson • Extreme packet loss, and nodes further away suffer more loss, e.g. 43 & 44 • Variation across days is consistent • Node 22 was deliberately set up with shaky connectivity Base Station Packet Reception Rates Hell’s Half Kit Carson Network Variability – Kit Carson Node Hop Variability 5 44 4 43 DTB 3 42 2 41 1 40 0 9/6/05 6:00 PM 9/7/05 12:00 AM 9/7/05 6:00 AM 9/7/05 12:00 PM 9/7/05 6:00 PM 9/8/05 12:00 AM 9/8/05 6:00 AM Time Large variation even after placement using the LED deployment technique Sensed Data Inversion Close up Inversion Start Inversion End • This data was used in the morning briefing by the Incident Commander – we hadn’t fully tested our system yet! Summary and Lessons Learned • FireWxNet: Designed, built and deployed a hierarchical WSN to monitor weather on wildland fires – Characterized thermal belt - useful to fire fighters – Somewhat achieved our design goals: robust, on-demand/portable, inexpensive, long-lived, easy to use, etc. • Learn what are the real problems – Extensive RF variability, even after placement => Design for robustness in networking – Ease of use in the field: • Add simple tools (LED blinking) for status/debugging • Keep the design basic • Add complexity only when needed, e.g. FireWxNet II this summer (solar, commands, adaptive FEC, calibration, etc.) – Software bugs: test & refine iteratively in lab, small deployment outside, etc. • Partner with the right people! (NCLFA at UMontana) – Gain acceptance by overcoming cultural issues, obtain red flag training, etc. Contact: rhan@cs.colorado.edu or Carl.Hartung@colorado.edu C. Hartung, S. Holbrook, R. Han, C. Seielstad, “FireWxNet: A Multi-Tiered Portable Wireless System for Monitoring Weather Conditions in Wildland Fire Environments”, Fourth International Conference on Mobile Systems, Applications and Services (MobiSys) 2006, pp. 28-41. Longevity Our Deployment only ran for 1 week, then the snows came...