Rapid Detection of Rare Geospatial Events: Earthquake Warning Applications A Review by Zahid Mian WPI CS525D September 10, 2012
A RE I NEXPENSIVE S ENSORS C APABLE O F D ETECTING S EISMIC E VENTS ? Cheap Sensors (16-bit MEMS accelerometer) Cell Phones (Android) Distributed Event Based (DEB) Community-Wide Participation Cloud Computing Problem/Goal
Most Cells Phones Today Have Accelerometers (used to obtain seismic measurements) Cell Service Increasing Around the World, Even in Impoverished Regions like Haiti Cost of Such Devices Decreasing The Denser the Network of “Sensors”, the More Accurate the Results Why Cell Phones?
Sending Messages (“picks”); Limitations 16-bit MEMS accelerometer Relatively Inexpensive More Accurate (than phones) Stationary Cell Phones (Android) “No” Expense (practically) Less Accurate (someone can drop a phone) Truly Mobile Sensors
Increased Number of Sensors Makes it Easier to Visual the Propagation Path of an Earthquake Why Use Many Sensors (“Dense Network”)
Google App Engine Many Benefits of PaaS; Scalable, Distributed, etc. Handles Registration of Sensors Handles Messages Checks Heartbeat Datastore Computational Analysis Detection Cloud Based Infrastructure
Overview of CSN Architecture
Synchronization: Use Entity Groups & Task Queue Jobs Timeframe Limitation: no single points of failure & tolerant of data loss Query Limitation: Numeric Geocells Limitations of App Engine
How to Store Latitude and Longitude Pairs AND Create Flexible Query? Solution: Create a Binary String to encode Information What does “011011” mean? Defines Location and Resolution Encoding Geospatial Data
Encoding a Geocell Encoding Rule: If point lies east of mean longitude, set longitude bit to 1 If point lies north of mean latitude, set latitude bit to 1 Example: N, W
STA/LTA (Short Term Average over Long Term Average) “Pick” when ration reaches above a threshold Anomaly Detection using Density Estimation (ADDE) Use sensory data to determine anomalies Several factors need to be considered Ultimately used to determine false positives and the number of messages to send Sensor-side Picking Algorithms
STA/LTA vs. ADDE
Detection Algorithm must be Insensitive to the reording of messages Insensitive to the loss of small number of messages Solution: Use 2-Second Buckets When Geocell Activated, a Task Queue job performs further Analysis Server-side Pick Aggregation
Geocell Detection Performance
Simulation: Geocell-based Regional Event Association
Simulation: Naive Event Association Acceptable for Determining Lower Bound
Quake Catcher Network (QCN) USB sensors attached to laptops No Cloud Computing NetQuakes expensive stand-alone seismographs Related Work (Seismic Networks)
Traffic Monitoring Environmental Monitoring Use Sensors to Capture Data, But Not Predict Rare Events Community and Participatory Sensing
Cell Phone Service May Not Work When Needed Too much “Noise” May Generate False Alerts Earthquake Detection is “Difficult” Many theories over the years and many false claims Collecting Data Via Sensors is Useful Can Lead to Better Modeling for Future Can Validate Certain Claims Infrastructure Can be Used for Something Else Conclusion