Presented by

Geospatial Data Sciences

Ranga Raju VatsavaiAuroop Ganguly

Budhendra BhaduriGeographic Information Science and Technology

Computational Sciences and Engineering

2 Bhaduri_GDS_SC07

Geospatial data sciences

HPC

GeoDatabasesDamage

Total population Senior

100% 791,361 90,773

75% 1,594,806 187,677

50% 1,505,196 184,372

25% 1,701,593 224,279

Total 5,592,956 687,101Petabytes of spatiotemporal data archives

Efficient storage and retrieval methods

Errors anduncertainty

Terabytes of spatially referenced data per day

Kno

wle

dge

Dis

cove

ry

Data mining

Spatial analysis

Machine learning

Pattern recognition

Statistics Nonlinear dynamics

Actionable insights/

knowledge

Disaster mapping

and monitoring

Geospatialintelligencediscovery

Extremes

Social networks

Nationalsecurity

HP Visualization

3 Bhaduri_GDS_SC07

Pattern detection and decision support

Data integration

Pattern detection

Process modeling Decisio

n

Decision support

Stream of new

data

Consequences

HIT

Action

Event summary

Body of actionable knowledge

OfflinePredictive analysis

OnlineDecision making

End-user Analyst

Additional data hypothesis generation

4 Bhaduri_GDS_SC07

Spatial classification and prediction Spatial autoregressive regression (SAR) Markov random fields (MRF)

Challenge Large neighborhood matrices Overlapping computationNode 1 Node 2

Ove

rlap

60 k

m

60 km2000

2000

60,000

60,0

00

3,600,000,000

3,60

0,00

0,00

0

TM 30 m

IKNOS1 m

SARW matrix

{ }yWIMWIySSE

WIconstssellf

xWyy

TTT )()(21

||lndetlogdetlog

ρρσ

ρ

εβρ

−−=

−=−++−=

++=

2

5 Bhaduri_GDS_SC07

Extremes A new measure to describe the relationship among

extremes in space and time: Simultaneous occurrence of 100-year return levels:

Perfect dependence 100 year event. No dependence 10000 year event.

Rainfall observations in South America

Simulations from CCSM3 Climate Model

ChallengeSpatiotemporal dependence estimation is computationally expensive (1000 grid cells results in scanning more than half a million pairs).

Correlation Tail Dependence Correlation Tail Dependence

6 Bhaduri_GDS_SC07

Managing uncertainty in spatiotemporal environments (MUSE)

Challenge Expensive query processing and spatial analysis

GeoDB

Knowledge Discovery

Representation

Propagation

Source

Updates

Data collectionsource of uncertainty

Distributed asynchronous update framework

Reasoning Data mining Spatial analysis Visualization

Modeling Conceptual Logical

Ontology Semantic Reasoning

7 Bhaduri_GDS_SC07

ContactsRanga Raju VatsavaiGeographic Information Science and TechnologyComputational Sciences and Engineering(865) 576-3569vatsavairr@ornl.gov Auroop GangulyGeographic Information Science and TechnologyComputational Sciences and Engineering(865) 241-1305gangulyar@ornl.gov

Budhendra BhaduriGeographic Information Science and TechnologyComputational Sciences and Engineering(865) 241-9272bhaduribl@ornl.gov

7 Bhaduri_GDS_SC07