Presentation on theme: "CASA: A NEW PARADIGM FOR END USER DRIVEN DATA COLLECTION"— Presentation transcript:
1CASA: A NEW PARADIGM FOR END USER DRIVEN DATA COLLECTION Brenda PhilipsDirector, Industry, Gov’t, and End User PartnershipsERC for Collaborative Adaptive Sensing of the AtmosphereAMS Corporate Forum 2007March 22, 2007NSF Multidisciplinary research center. We’re creating a new paradigm for weather observation systems one that’s networked based vs. current focus on one radar and that is user needs are considered in the design and operation of the network.
2Outline Motivation and Background User Driven Adaptive Scanning End User Policy in Test bedNext Steps
3Engineering Research Centers Research and develop technologies to generate a well-defined class of engineered systems with social and economic impacts.Faculty, students and industry/practitioners work in a multi-disciplinary environment reflecting real-world technology.CASA’s Focus: New weather observation system paradigm based on low-power, low-cost networks of radars.Year 4 of a 10-year research projectOver the first five years CASA will have a total of $40million dollars, anchored by a grant from NSF and supplemented by univeristy and industry funding. We’ve been operating for just over a year now.
410/27 F1 tornado illustrates limitations of current observation technology Watch Issued. No Warning.“…did not have the correct rotational characteristic we expect from a tornadic storm. It’s tough to see small tornados…It’s a problem at large distances.”In October 27 a tornado, that started out as a water spout came ashore Apalachicola floridaSource: NWS Tallahassee Forecast Office
5Mismatch: technology vs Mismatch: technology vs. temporal and spatial scales for decision makingNEXRAD250 km spacingHorizon problem causes coverage gap~ 2 km resolution5 min. updatesFunction Autonomously“Sit and spin” surveillance with “data push”10,000 fttornadowindearth surfacesnow3.05 km4080120160200240RANGE (km)Horz. Scale: 1” = 50 kmVert. Scale: 1” -=- 2 km5.4 km1 km2 km4 kmgapWhat we have here is a mismath of the technology and the decisions. Here’s the path of the tornado that lasted touched down for 5 min. It would have been useful know of a street leve where it was headings, It would have been good to observe it.
6Mismatch: technology vs Mismatch: technology vs. temporal and spatial scales for decision makingNEXRAD250 km spacingHorizon Problem: Middle to upper troposphere coverage~ 2 km resolution5 min. updatesFunction Autonomously“Sit and spin” surveillance with “data push”10,000 fttornadowindearth surfacesnow3.05 km4080120160200240RANGE (km)Horz. Scale: 1” = 50 kmVert. Scale: 1” -=- 2 km5.4 km1 km2 km4 kmgapWhat we have here is a mismath of the technology and the decisions. Here’s the path of the tornado that lasted touched down for 5 min. It would have been useful know on a street level where it was headings, It would have been good to observe it.
7CASA addresses “mismatch” with DCAS (distributed collaborative adaptive sensing) Short range (~ 30 km) radarsLower troposphere coverage100’s meter resolutionAvg. 30 second updatesAdaptive Scanning based on user needs, “data pull”10,000 fttornadowindearth surfacesnow3.05 km4080120160200240RANGE (km)New observing capability has great potential to improve our ability to detect, track, forecast, warn, and ultimately respond to low level hazards.“Sense the Atmosphere where and when user needs are greatest”
8End User Integration links technology with warning and response streamingstoragequeryinterfacedataResource planning,optimizationpolicyresource allocationSNRMeteorologicalDetectionAlgorithms123456789AG3BCDEFGHR1R2C2I2,2,H2JH1,F1T2,R1K2,H1Feature RepositoryMC&C: Meteorologicalcommand and controlDCAS User Driven Adaptive ScanningUser Interaction with new technology for decision makingImpact of DCAS Technology on Communications, Public Response, and VulnerabilitySocial and Economic Value of CASA Data
9We interact with practitioners in the public and private sectors Value Added Data UsersIntermediariesPublicState GovernmentEmergency ResponsestreamingstoragequeryinterfacedataResource planning,optimizationpolicyresource allocationSNRMeteorologicalDetectionAlgorithms123456789AG3BCDEFGHR1R2C2I2,2,H2JH1,F1T2,R1K2,H1Feature RepositoryMC&C: Meteorologicalcommand and controlDataFederal GovernmentPublicPrivate Sector: Value-Added ProductsProducts and Services,Decision Support SystemsInstitutions/IndustryResearchAcademic Institutions/Researchers
10We have a multidisciplinary team Brenda Philips, MBA, UMass Ben Aguirre, Sociology, UDelEllen Bass, Human Factors, UVa Walter Diaz, Political Science, UPRMKevin Kloesel, Meteorology, OU Dave Pepyne, ECE, UMassHavidan Rodriguez, Sociology, UDelRoman Krzysztofowicz, Decision Sciences, UVa,
11Oklahoma Test Bed: Severe Storms 4-node mechanically scanned radar network36 km range25 – 100 m resolutionAdaptive: Multi-elevation sector scans, pinpointingPilot User Group: WFO Norman, EMs, Researchers, (Media)I’d like to shift to the oklahoma test bed7000 sq. km (90 km long)2 tornados4 tornado warnings50 severe storm warnings
12Data Pull: system optimally collects data based on user needs, evolving weather and radar capabilities.Radar network scans atmosphere (sector, pinpointing, 360 degree) and sends data to MC&C2. Detection algorithms identify weather features in radar data.3. Weather features are “posted” in Feature Repository, a 3-d grid of radar coverage area.user policyResearchersMediaOptimal radar scan configuration developed based on:How Important is a task to users?What is the quality of the scan?4. Tasks are generated based on clustering similar weather features.
13User Policy: How Important is the data to users? User Needs User Needs translated into rules for systemWhat detected weather feature?How many radars?What horizontal and vertical coverage?How often should it be scanned?“ Needs” evolving from subjective to more objective measuresStated preferencesSpace/Time variablity of weatherSocio-economic impactsUSER RULE FOR EMERGENCY MANAGERS: If rotation is detected, then scan the lowest two elevations radars every 30 seconds.Goal: Geographically specific information for public notification, responder deployment
14User Policy: How Important is the data to users? User Weights Determines the relative importance of different user groups in case of resource conflict.Mechanism established, Policy for setting weights not establishedStill understanding the level of resource contention in the systemWeighting could be set for socio-economic benefit, profits, for specific weather events, etc.
15Another view of optimization End User Policy–Howimportant is tasktto theusers?()å=ggroupskUw,UserWeightsUser Rules
16Challenges for developing user rules (preferences) Eliciting preferences from users for a new sensing paradigmIterative processSubject Matter ExpertsUse qualitative approaches initiallyGetting system designers (computer scientists, engineers) to understand user decision process and translate that into code.
17Initial approach to user rules focuses on weather features This is V0 of the rules. As you can see they are based on specific feature, such as a tornado detection and it’s importance to users with reference to geography. Th
18Feedback from NWS: “That’s not how we make decisions…” Goal: Issue warnings, communicate expertise save lives, propertyRadar data increases or reduces forecaster confidence“mental movie”focus on areas of uncertainty; not always determined by radar dataTrainingExpectationsStaffing IssuesCoordinationConditionsobservedoutsideLocation,expectedimpactOngoingSatelliteModel1,2,3,etcguidanceRadar1,2,3DataConceptualmodelsEquipment statusGut feelingWarnDon’tCommunicationsMesoanalysis1,2,3,Ground truthAfter developing those rules we continued discussions with users. Our researchers were concerned that they wouldn’t get the data to feed their model; and that if a tornado developed they wouldn’t necessarily get the multi-radar information they wanted. This slide shows how we learned about version 0 from the NWS. We used lo-res emulator out put to engage them in a discussion about scanning strategies and rules. Their deicions making depends on visual examination of the data and create a “mental movie” of an evolving event. So they wanted regular scans at 360, they’re also interested in dynamic data imput.“NWS Warning Process”, Liz Quoetone, NWS Warning Decision Training Branch, presentation at 2005 CASA workshop
19Changes to rules Incorporation of interval-based scanning Expansion of the definition of a storm cellIntroduction of contiguous scansDynamic Data Requests
20Current rules focus on “time since last scanned” triggerSectorSelectionElevations#RadarsContiguousSamplingintervalNWSN1time360Lowest two1Yes1 / minN2stormtask sizefull volume1 / 2.5 minResearcherR1rotation2+1 / 30 secR2reflectivityFull volume2velocitylowest two1/ minR3No1/ 5 minEMsE1lowestE2reflectivity over AOIE3velocity over AOI1/ 2.5 minOSO11 / 5 minTable 1. User Rules, End User Policy, Version 1
23August 15 frontal boundary with isolated storm cells 3 radars operationalSingle radar attenuation correctionClutter, velocity, and networked attenuation alg. not yet installedDCAS running based on reflectivity rules and detections.
253.004.006.008.0011.0014.00Data collected from KSAO on , between 23:19:45 and 23:20:15 UTC.
26Next Steps Study user behavior in test bed with actual data Study impact of weights on system functionDevelop revisions to end user policy with “ Needs” evolving from subjective to more objective measuresStated preferences, observed preferencesSpace/Time variability of weatherSocio-economic impactsExpand user base to include public, private companiesLaunch Decision Sciences Projectintegration socio-economic factors into adaptive scanningCreating an end-to-end decision model of network
31Rain, mountainous terrain (Puerto Rico – student led) Test beds instantiate end-to-end system conceptsRain, Urban Flooding(Houston)Rain mapping, distributed hydro. modeling, flood predicting & response in an urban zone.Wind, storm prediction(Oklahoma)Wind mapping (100’s m resolution, 10’s second update) for detecting, pinpointing, forecasting wind events; 30 km node spacing.Rain, mountainous terrain(Puerto Rico – student led)Off-the-Grid Radar Network for quantitative precipitation estimation (QPE) over complex terrain, student-led project
32End User Policy Control User Rules and WeightsMechanism in place for adjusting weights (Wg)Default: all weights = 1.0Rules based on scan frequency and/or featureUtility based on time since the rule was last scanned (Ug)How important is task t to the users?End User Policy ControlTasks (t) (blue)Scans (C) (green)End User Policy Rules, Version 1Rule is a combination of feature and timing
34We interact with practitioners in the public and private sectors Value Added Data UsersIntermediariesPublicState GovernmentEmergency ResponsestreamingstoragequeryinterfacedataResource planning,optimizationpolicyresource allocationSNRMeteorologicalDetectionAlgorithms123456789AG3BCDEFGHR1R2C2I2,2,H2JH1,F1T2,R1K2,H1Feature RepositoryMC&C: Meteorologicalcommand and controlDataFederal GovernmentPublicPrivate Sector: Value-Added ProductsProducts and Services,Decision Support SystemsInstitutions/IndustryResearchAcademic Institutions/Researchers
35August 15 StormEast-to-west warm frontal boundary with isolated storm cells; second area of stratiform rain with embedded convectionNWS issued one thunderstorm warning within the network for Grady County at 2130 UTCSeveral severe wind reports were recorded just south of IP1 at approximately 0000 UTC.CASA Test Bed3 radars operationalSingle radar attenuation correctionClutter, velocity, and networked attenuation alg. not yet installedDCAS running based on reflectivity rules and detections.CASA data Frederick Radar
37Key Influencers Existing practices Existing sources of information and their perceived uncertainty: existing weather data and models, media info, ground truthOrganizational Issues: procedures, culture, training, evaluation metrics.Societal Issues: access to information and training; differences based on education, gender, race, income.Risk Communication: tone and content of message, Multi-directional communications, social networks, etc.
383 yr project for end-to-end decision model To develop an end-to-end integrated decision model for DCAS systems (Integrated System Model) from targeted observation, detection, forecast, warning, risk perception and response to socioeconomic impact.Use socioeconomic measures to drive CASA’s resource allocation and optimization.Implement decision model in an expanded DCAS system emulator that simulates warning, response, and impact.What we have here is a model of the end to end system based on the core decisions of the system, and the information and uncertainties. The goal would be to create mathematical models linking theseBy creating this influence diagram and developing the mathematical model of these relationship, we be able to model how a change in a decision flows impact the other parts of the system through to impact, and the value of that system.