Ontology for Geospatial Reasoning at Disasters

1
Ontology for Geospatial Reasoning at Disasters

• David Kirsh Cognitive Science UCSD
• Nicole Peterson Anthropology Columbia Univ
• Leslie Lenert VA Hospital UCSD, CALIT(2)

2
Problem

• How do incident commanders and medical response
  team leaders
• Conceptualize the physical space of a disaster
• Is the hot zone shifting? - wheres the plume?
• What s in the shooters line of sight?
• Think about their response in spatial terms
• Whos in the transport area?
• Clear that region and rendezvous back here
• Have we made a safe refuge area?
• Lets put the staging areas over there.

Line of sight
Triage Area Treatment Areas Transport
Area Staging Areas Decontamination Areas Cleared
by SWAT Regrouping spot
3
Background

• Were designing wireless devices to attach to
  personnel, victims and equipment
• Concern information overload
• data from 100s of such devices distributed
  around a site
• Solution interpret data in terms of
• spatial concepts and reference points which
  incident commanders already use to understand
  situation
• use the language of situation awareness

How safe is the initial refuge area?
Whats it like at transport?
How long is the queue for decon?
4
Goals
Errors cost x frequency

• reduce error
• number of errors
• cost of errors
• increase coordination
• Stay in control
• Aware of who is where, doing what, how
• Better quality of messages
• improve the efficiency of response
• Increase victim flow through key areas
• Reduce bottlenecks

DSR 2 gt DSR 1
5
Methodology
6
Four Sources of Data

• Pre-drill interviews and simulations videoed
  five 2-hour interviews
• Drill observation videoed disaster drill using
  5 video cameras 4 hour each camera 2 on top of
  building, 2 stationed, 1 roaming
• Event debriefing took notes of the post-event
  debriefing
• Post-event follow-up interviews follow-up with
  the experts from our pre-drill interviews
  videoed four 2-hour interviews.

Coding interviews and drill footage ensured
inter-coder reliability through joint coding
using two to four coders
7
Interview Hand simulation of drill expectations
8
Results
9
Conceptually different types of space

• Zones
• formal, taught in class
• objective basis
• hot zone based on physics of release, trajectory
  of shooter
• Areas
• formal, taught in class
• driven by functional needs
• Staging areas Fire, Hazmat, MMST
• Stations decon, triage, treatment, transport
• Refuge size depends on needs
• Ad Hoc
• not taught per se
• driven by tactical demands of activity inferred
  from language gesture
• SWAT regroup near door
• rendezvous pt, cleared so far

Zones
Refuge Area
Safe Refuge Area
Areas
10
Helps identify and reduce errors

• Thinking about attributes of zones, areas and
  regions explains certain sorts of errors
• Dimension Errors
• Zone wind changes, but the dimensions are not
  updated
• Area a space is wrongly assumed vacant when in
  fact reserved for someone else (hazmat trucks
  took the wrong place)
• Permission Errors
• A fireman marked the hot zone when he should have
  waited for Hazmat firemen lack permission
• Who has the right to move a resource into,
  within, or out of a zone,
• Fire shouldnt move crime scene items
• Marking Errors
• Hazmat assumed no need to mark all their staging
  space (4 trucks) so were distributed
• Tree used to mark hot zone edge but information
  not propagated
• Wrong tape

11
Attributes
12
Conclusions
13
Its good to think spatially

• Responders and commanders think spatially in
  terms of
• Zones
• Areas
• Ad Hoc Regions
• Information from geospatial systems needs to be
  translated into responder meaningful terms more
  abstract than geospatial location

14
Its good to think spatially

• Ethnographic analysis can expose the attributes
  of spatial concepts that is implicit in discourse
  and gesture
• An attribute ontology provides a principled
  method for discovering and classifying responder
  errors
• When data is represented more meaningfully
• Messages are cognitively more efficient better
  speed-accuracy
• Situation awareness propagates faster higher
  awareness diffusion rate
• Distributed activity is better coordinated
  better hand-offs, time to respond, victim flow
  rate
• Fewer and less costly errors are made
• Sets the stage for more comprehensive evaluations
• Victim flows through areas, coordination of data
  between stations

15
End
16
Quantitative measures
17

• Errors
• If we think about these attributes it is apparent
  how certain sorts of errors may arise. Here is a
  simple list of 25 errors.
• Dimension errors
• zone a space has reasonably precise dimensions
  but the team miscalculates them
• zone conditions alter, such as when the wind
  changes, but the dimensions are not updated
• zone team members misread their instruments and
  so propagate incorrect information about
  dimensions
• zone team members do not realize that their
  instruments tell them about boundaries (e.g.
  beepers indicate radiological perimeters) and so
  they do not think they can discover boundaries
• areas a space is assumed vacant when in fact it
  is reserved for someone else (hazmat trucks took
  the wrong place)
• Ad hoc regions the SWAT team leader did not
  take into account the shooters height and so
  incorrectly assumed a safe retreat zone was also
  safe in the vertical dimension.
• Permissions Errors
• All spaces Person A enters but lacks permission
  to enter, and should have known it. E.g
  perimeter guard approaches the ICC
• All spaces A does not have the right to move a
  resource into, within, or out of a zone. E.g
  fireman move crime scene item.
• Zones, areas Person A marks the boundaries of a
  zone or area but lacks permission to do this, and
  should have known it. Fireman marks the hot
  zone when he should have waited for Hazmat.
• All spaces A lacks permission to do a specific
  activity in that space.
• Zones, Areas A gatekeeper should prevent A from
  entering but does not realize that a permission
  is required or that A has the wrong permissions.
  E.g. Gatekeeper at site perimeter mistakenly
  keeps out authorized responders.
• With respect to evidence different perspective
  on situation, dont share information

18
Justification Perimeter defined Perimeter marked Perimeter moves Center marked Duration Shape Nested spaces?
Zones Factual Yes Yes Yes No Event 3D Yes
Areas Long-term requirements of activity Often Sometimes Sometimes Yes Event 2D Yes
Ad hoc Short-term requirements specific to one activity No Sometimes No (expires) No Phase of activity 2D/1D Other ad hoc
Nested
19
Dril
Drill May 2004
Ambulance Staging
20
(No Transcript)