Towards a Geospatial Knowledge Discovery Framework for Disaster Management

1
Towards a Geospatial Knowledge Discovery
Framework for Disaster Management

• Budhendra Bhaduri, Eddie Bright, Veerarghavan
  Vijayaraj
• Geographic Information Science Technology

2
Overview

• Background and Motivation
• LandScan Global Population Model
• Improving Land Cover Information
• Human Settlement Mapping
• HPC for urban region extraction
• Successes and Challenges
• Spatial-temporal Data Mining
• Online Change Detection
• Population Dynamics
• Future Directions

3
What is LandScan?
Population distribution model, database, and tool
developed from census and other spatial data
using a uniform regular grid
Improving knowledge of where people are located.
4
How Is LandScan Developed?

• Dasymetric Spatial Modeling
• Distribute best available census counts to
  LandScan cells based on a likelihood coefficient
  calculated by spatial models
• Model structure is the same everywhere, but
  weights for each variable are tailored to each
  country
• Similar operations performed for each data layer
  and outputs are mathematically combined
• Population is allocated to each cell

Product of additional data types (e.g. distance
from roads, slope, etc.)
5
Improving Land Cover Information
6
Moving from Modeling to Mapping
7
Human Settlement Mapping

• Objective Accurately identify human Settlements
  from High resolution image archives using
  efficient automated processing.
• Human Settlements typically include

• Dense Urban
• Commercial ( Urban I)
• Industrial
• Residential ( Urban II)
• High-rise apartments in urban regions, old
  residential part of cities

• Sub-urban
• Dense Suburban ( Suburban I)
• Patterned housing subdivisions
• Suburban (Suburban II)
• Mostly residential houses with loose
• pattern and roads
• Villages (Rural)
• Remote village

8
Sample of Human Settlements
Samples of varied type of settlements we are
trying to identify
9
High Performance Computing

• Different portions of an image are processed
  simultaneously and the results are assembled at
  the end. This provides an improved computation
  time performance.
• A Single Program Multiple Data (SPMD) approach is
  used for the parallel computation process.

10
Parallel Processing
Pseudo Code If (Node 0) If ( compute
features) Distribute Image blocks Collect
processed blocks Check for completion If (done
full scene) terminate nodes If( do
Clustering) send centroids receive
clusters compute convergence if
(converged) terminate nodes else compute new
centroids repeat clustering else if(
compute features) compute texture
features return computed features if (do
clustering) receive Centroids compute
clusters return
Supervisor-worker model with MPI
11
Results
QuickBird PAN
12
Results
QuickBird PAN
Rich texture features clustered along with urban
regions
Crisper urban region map after applying spectral
filtering
Gabor texture
Post Processed Filtered with NDVI
13
Hurricane Preparedness and Response
14
Hurricane Katrina
15
PRE and POST Event Imagery
Sep 30, 2003
IKONOS PAN Images
Sep 1, 2005
16
Imagery Based Change Cues
Before
17
Energy Assurance

• Spatiotemporal assessment of renewable energy
  potential
• Bioresource monitoring for energy security
• Geographically scalable spatiotemporal
  optimization for energy supply chain

18
Hail Damage Visualization from AWiFS Data
Courtesy USDA RS Program
19
Knowledge Discovery from Geospatial-Temporal Data
May 1994
May 2003
Google Earth/DigitalGlobe
Terraserver
Distribution Center Opens March, 2004
Apparent ground breaking, early 2003
Courtesy Feierebend et al., 2006 (AAG)
20
Sites in Three Different Ecoregions
Indicates Opening Date
Courtesy Feierebend et al., 2006 (AAG)
21
Online Change Detection Methodology

• Reference Model
• Difference of the Time Series (Wal-Mart vs.
  Background)
• Cube Root Transform for Variance Stabilization
• Transformed, Difference Time Series IID and
  Gaussian
• Online Alarms for Change ? O(1)
• Sustained changes (even if small) of interest
• Statistical process control methods (as in
  Lambert Liu, 2006)
• Online Change-Point Detection ? O(1) to O(n)
• Heuristic and stochastic
• Backward (downhill) search
• Similar in spirit to simulated annealing
• Updating Parameters ? O(1)
• Process in Control Mean held to zero and
  variance updated
• Alarm Generation Mean converges to new state

22
Summary Results Table

• Results (Fang et al., 2006)
• CA store Groundbreaking available ? Near perfect
  validation
• ME, NC Stores Groundbreaking not available
• Approximate match
• Experiments consistent with expectations

23
LandScan Population

• Population
• Census Polygons Tract-to-track worker flow BLS
  quarterly updates.
• Roads
• VMAP, GDT Dynamap TIGER
• Railroads
• 1100K national railway network
• Land Cover/Land Use
• Geocover, MODIS, National Land Cover Data (NLCD)
  State GIS
• Slope
• DTED, National Elevation Data (NED)
• Academic Institutions
• Department of Education ESRI GDT
• Prisons
• National Jail Census
• Hospitals
• American Hospital Association (AHA)
• Business Employment
• InfoUSA
• Dunn and Bradstreet

Improving Knowledge of Population Dynamics
24
night
25
(No Transcript)
26
(No Transcript)
27
Transportation

• Dynamic tracking of fleet movement from
  multisensor data
• Travel behavior modeling for congestion and
  safety
• Spatiotemporal data mining and visualization for
  improved operations and communication

28
(No Transcript)
29
Contact

• Budhendra Bhaduri
• Geographic Information Science Technology
• Oak Ridge National Laboratory
• Oak Ridge, TN 37831-6017
• bhaduribl_at_ornl.gov