Towards Personalized and Active Information Management for Meteorological Investigations Beth Plale Indiana University USA
Jan 11, 2016
Towards Personalized and Active Information
Management for Meteorological Investigations
Beth Plale
Indiana University
USA
Problem Statement• Mesoscale meteorology research is highly data-
driven.– Large percentage of data streams in from
observational platforms. Available in OPeNDAP servers.
– Data that is over 10 minutes old is too old. – Researchers are currently working on increasing real-
time responsiveness to developing weather conditions.
• Mesoscale meteorology is a vast information space.– Forecasting models assimilate data from growing
number of sources
Solution Statement• Internet has proven the utility of user-oriented
view towards information space management– Browser, bookmarks to organize– Blogs, web page tools (FrontPage, Dreamweaver) to
publish
• We apply concept of user-oriented view to management of mesoscale meteorology information space.
• myLEAD: tool to help an investigator make sense of, and operate in, the vast information space that is mesoscale meteorology.
Motivation for LEAD• Each year, mesoscale weather – floods, tornadoes,
hail, strong winds, lightning, and winter storms – causes hundreds of deaths, routinely disrupts transportation and commerce, and results in annual economic losses > $13B.
Conventional Numerical Weather Prediction
OBSERVATIONS
Radar DataMobile Mesonets
Surface ObservationsUpper-Air BalloonsCommercial Aircraft
Geostationary and Polar Orbiting Satellite
Wind ProfilersGPS Satellites
OBSERVATIONS
Radar DataMobile Mesonets
Surface ObservationsUpper-Air BalloonsCommercial Aircraft
Geostationary and Polar Orbiting Satellite
Wind ProfilersGPS Satellites
Analysis/Assimilation
Quality ControlRetrieval of Unobserved
QuantitiesCreation of Gridded Fields
Conventional Numerical Weather Prediction
Analysis/Assimilation
Quality ControlRetrieval of Unobserved
QuantitiesCreation of Gridded Fields
Prediction
PCs to Teraflop Systems
Conventional Numerical Weather Prediction
OBSERVATIONS
Radar DataMobile Mesonets
Surface ObservationsUpper-Air BalloonsCommercial Aircraft
Geostationary and Polar Orbiting Satellite
Wind ProfilersGPS Satellites
Analysis/Assimilation
Quality ControlRetrieval of Unobserved
QuantitiesCreation of Gridded Fields
Prediction
PCs to Teraflop Systems
Product Generation, Display,
Dissemination
Conventional Numerical Weather Prediction
OBSERVATIONS
Radar DataMobile Mesonets
Surface ObservationsUpper-Air BalloonsCommercial Aircraft
Geostationary and Polar Orbiting Satellite
Wind ProfilersGPS Satellites
Analysis/Assimilation
Quality ControlRetrieval of Unobserved
QuantitiesCreation of Gridded Fields
Prediction
PCs to Teraflop Systems
Product Generation, Display,
Dissemination
End Users
NWSPrivate Companies
Students
Conventional Numerical Weather Prediction
OBSERVATIONS
Radar DataMobile Mesonets
Surface ObservationsUpper-Air BalloonsCommercial Aircraft
Geostationary and Polar Orbiting Satellite
Wind ProfilersGPS Satellites
Analysis/Assimilation
Quality ControlRetrieval of Unobserved
QuantitiesCreation of Gridded Fields
Prediction
PCs to Teraflop Systems
Product Generation, Display,
Dissemination
End Users
NWSPrivate Companies
Students
Conventional Numerical Weather Prediction
OBSERVATIONS
Radar DataMobile Mesonets
Surface ObservationsUpper-Air BalloonsCommercial Aircraft
Geostationary and Polar Orbiting Satellite
Wind ProfilersGPS Satellites
The process is entirely serialand pre-scheduled: no response
to weather!
The process is entirely serialand pre-scheduled: no response
to weather!
Analysis/Assimilation
Quality ControlRetrieval of Unobserved
QuantitiesCreation of Gridded Fields
Prediction
PCs to Teraflop Systems
Product Generation, Display,
Dissemination
End Users
NWSPrivate Companies
Students
The LEAD Vision: No Longer Serial or Static
OBSERVATIONS
Radar DataMobile Mesonets
Surface ObservationsUpper-Air BalloonsCommercial Aircraft
Geostationary and Polar Orbiting Satellite
Wind ProfilersGPS Satellites
Analysis/Assimilation
Quality ControlRetrieval of Unobserved
QuantitiesCreation of Gridded Fields
Prediction
PCs to Teraflop Systems
Product Generation, Display,
Dissemination
End Users
NWSPrivate Companies
Students
The LEAD Vision: No Longer Serial or Static
OBSERVATIONS
Radar DataMobile Mesonets
Surface ObservationsUpper-Air BalloonsCommercial Aircraft
Geostationary and Polar Orbiting Satellite
Wind ProfilersGPS Satellites
LEAD data: initial working data set
• ETA model gridded analysis
• METAR surface observations
• Rawinsondes – upper air balloon observations
• ACARS – commercial aircraft temperature and wind observations
• NEXRAD Level II data
• GOES visible satellite data
Returning to Solution Statement• We apply concept of user-oriented view to
management of mesoscale meteorology information space.
• myLEAD: tool to help an investigator make sense of, and operate in, the vast information space that is mesoscale meteorology.
Information space management tool
• At core is metadata catalog– Why? Observational products already being
stored elsewhere.• Public file and could be large, so do not want to
copy user’s file system. Instead maintain “bookmark”
• Scale to support thousands of distributed users, including individual investigators, pre-college classroom investigators, casual observers.
Technical Challenges
• Querying must be efficient– Over data products described by rich domain-specific metadata– Over data products whose description can be augmented over time
• Obtaining metadata is hard– Automate as much as possible
• Privacy must be fully enforced– Any data product that user designates as private must remain private
• Publishing– Publish product to larger community:
• data file, model output, full experiment– Must be under user control– Discovery of information that has been made public
• Build trust– User may work within myLEAD space for 5 years of graduate work, for
instance– User must be convinced of privacy, reliability, longevity, etc.
Rundown on Implementation Specs
• Building on top of MCS and OGSA-DAI– MCS for extensible db schema, general db schema,
and security infrastructure already in place– OGSA-DAI for grid/web service architecture
• Database used is mySQL 5.0– Supports stored procedures– Ogsa-dai to mySQL is JDBC
• Data product descriptions in and out of database conform to LEAD-specific XML schema.
• myLEAD server and myLEAD agent are written in java.
Related Work• mySpace – AstroGrid, UK
– Similar to myLEAD in reigning information space– Creates swatches in large federation of data archives for the cache and
persistent data for a “community”– Provides common query access over cache space and persistent space
• RDF (Resource Description Framework)
– Basic building block is the subject-predicate-object triple:– [S] – P -> [O] [Dickens] – hasWritten -> [Pickwick Papers] – Good for storing detailed relationship information (good for
understanding the relationship between two terms)• NEESgrid – NCSA
– Uses RDF– Little available in public literature
• myGrid Information Repository (MIR) – myGRID, Manchester – Most similar to myLEAD– Support for text search scientific papers, uses Life Sciences Identifier
(LSID)– myLEAD stronger personal orientation (gurantees, publishing, automatic
metadata generation)
myLEAD service
Server sideservices
Client sideservices
datamodeldata
model
MCSMCS
myLEAD stored procedures
OGSA-DAI
JDBC
MCS client
myLEAD agent
Portal access to myLEAD
User interface
relational DB
myLEADmyLEAD
myLEAD Architecture
FactoryFactory
myLEAD“agent” instance
myLEAD“agent” instance
WRF modelWRF modelData miningtask
Data miningtask
workflowworkflow
myLEADservice
myLEADservice
Storage Repository
Service (RLS)
Storage Repository
Service (RLS)
myLEAD portletas component of LEAD portal
/var/tmp/wrf_tmp
IU NCSA
myLEAD use scenario
Workflow confers with myLEAD “agent” to determine location of scratch space
Metadata Catalog Data Model• Users
• Investigations– Tornado April 20 Chicago Illinois
• Experiments– Ensemble: run of 100 simultaneous forecast models
parameterized slightly differently• Collections• Logical files
– Input observational files, input parameters, derived files, analysis results, images, model results, workflows, execution status messages
Abe Bing Caru
Investigation
User – DublinCore
Attributes storedin “type” tables: i.e., string, float, temporal, int. Great extensibility, but need to carefully control naming; efficient querying could be an issue as well.
Logical file
Collection
Data Model
myWorkspace: J. Kowaleski
preferences
Experiment 1: Norman, OK 21Oct04:23:11:45
Workflow template vizEta 03Aug04:13:35:40Workflow template WRF 15May04:05:25:59
Favorite spaces
Home disk space
Thor cluster scratch space
Input parameters
NEXRAD 26Oct04:13:45:40
GOES-infrared 26Oct04:12:00:00
METAR 26Oct04:09:10:05
Wrf-out1-26Oct04:13:35:40
Input observational
WRF-out
Wrf-out2-26Oct04:13:37:25
Wrf-out3-26Oct04:13:43:15
workflow instance
Collection level
Logical file level
Have associated a set of attributes that describe this data product
Browser provides usera hierarchical view of space that is essentially flat. Users like hierarchy.
Data Model
myLEAD agent
• Separate transient grid/web service– Has state about user, current investigation and experiment– Embeds myLEAD client API
• Purpose: – Controls naming– Helps use database structure in repeatable, meaningful way
• Maintains FSM of current state of execution; stores into new collection based on state
– Input model run analysis final results
– Derives metadata attributes for new data product object when created during course of workflow by means of:
• Case-based reasoning• Internal state• Consulting ontology
Resources
Geo- Data products
Workflow scripts
compute resources,storage resource
Data analytics resources (statistics table)
services
Observational data
Model generated data
Collections
Derived data
Data analytics
Model input resources
Resources: “things that need describing (i.e., metadata)”
Data mining
Data Product Metadata
Data Product Metadata Notes
Global ID “LSID” for geosciences
Temporal coverage Same as spatial
Spatial coverage GML, THREDDS, FGDC, COARDS-CF
Geophysical quantity Defined by common vocabulary
Platform Goes10, Goes8; WSR-88, CASA
Instrument type
site East-west; KXYZ
Model run info Model derived data product
Syntactic description Binary format of data product
Contact info Dublin core
Physical location of service
Protocol to access service
Dataset summary Dublin core
list of predecessors GID of input data products, workflow instance
Event mesocyclone, storm cell, tornado
Quality Complex
Completeness
Current Research Challenges
• Publishing– Publishing data product to larger community:
• data file, model output, full experiment– Discovery of information that has been made public
• Guarantees– Any data product that user designates as private must remain private– When request for product is issued, product must exist
• Flexible yet efficient schema– Inherited from MCS, supports evolved understanding of data product
over time by means of extended attributes• Immutable investigations
– Collections, views, and logical files can be reused from earlier investigations without destroying integrity of earlier investigation
• Proactive agent– Infers metadata attributes from context of active experiment using case-
base reasoning.