Paul James Doherty Park Ranger / GIS Specialist / Graduate Student.

Paul James Doherty

Park Ranger / GIS Specialist

/ Graduate Student

Spatial Analyses in Wilderness Search and Rescue

Objective 1: Describe where Search and Rescue incidents (SARs) occur and predict future events

Objective 2: Increase the “probability of area” (POA -predicting where a lost person is or is going to be)

Objective 3: Increase the “probability of detection” (POD -detecting a lost person if in the search area)

What will be researched?

Use geo-referencing techniques to compute uncertainty measurements for incident location with a large unique (n = approx 4,000) dataset (Guo et al 2007)

Test and evaluate spatial-temporal risk modeling procedures to describe where incidents have occurred and predict where they will occur (Kelly et al 2006)

Intellectual Contribution – Objective 1

Finding a missing person/object is the classic mysteryProbability of Success (POS)

Probability of Area (POA)Probability of Detection (POD)POA x POD = POS

How can integrating GISystems and GIScience increase overall POS?Human foot-travel modeling (work with Liz Sarow Fall

2009)

Dynamic vector design (how do vector inputs improve predictive model accuracy?)

More objective approach to finding missing persons

Intellectual Contribution- Objective 2/3

Yosemite Search and Rescue Background

Case Study: Yosemite National Park• Sierra Nevada, CA• Size

• 1,200 sq. miles• 95% wilderness• 800 trail miles

• Annually• +3.5 million

people• Search and

Rescues: 200

Hung et al 2007 (1990 – 1999)

Searching for Missing Persons in Yosemite National Park

Search TheorySimplified

POA x POD = POSOverall POS = [(POA1 x POD1) +(POA2 x POD2)

….] Where 1,2,3,…. are search segments

Basically, where is the person most likely to be (based off of search model) and how are we most likely to find them (based off of search technique)

The Search for A Missing Person

Objective 1Georeference SAR data

SpatialdistributionMajority of

SARs described are referenced by distance from placenames

Data is in report format, needs to be compiled

What causes clustering?

Geo-reference points from text (occasional GPS)Classify incidents by type (trauma vs. medical

etc.)Evaluate uncertainty of geo-reference

Characteristics of where incidents have occurredEcological Niche Factor Analysis?

Develop new protocols to plot incident locations

Objective 1 - Methods

Methods

Objective 2Probability of Area

What is the current Search Model for determining POA?POA = Theoretical + Statistical + Subjective +

ReasoningHow can we use Spatial Analyses to increase

accuracy of POA?

Objective 2: Increase the probability of area

Establishing Probability of Area (POA)Theoretical

Distance subject could have traveled in the amount of time elapsed (objective, geography)

StatisticalData which reflects the distance other subjects

have traveled under similar conditions (objective, human behavior)

SubjectiveEvaluation of limiting factors (objective,

geography)Reasoning

Systematic analysis of circumstances surrounding the disappearance of missing person (subjective, human behavior)

Objective 2: Increase the probability of area

Construct and evaluate speed models for missing hikersCost-distance modeling

Generate areas of high probabilityRefine data for distance from PLS (CALEMA

data?)Hazards from Objective 1 (20 year dataset)Attractions from viewshed analysesDynamic probability mapping when clues are

detected

Objective 2 - Methods

A function of:TerrainBehavioral Profile

(Koester and Stooksbury 1995)

Time

Cost-Distance modeling

Courtesy of Liz Sarow - ESRI

Objective 3Probability of Detection

What is the current Search Model for determining POD?[Effective Sweep Width]1 + [Clue Detection]1

+ [Coverage of Assignment]1 POD1a *Extremely subjective process

POD1a + POD1b + POD1c = cumPOD1 * Law of diminishing returns applies

Objective 3: Increase the probability of detection

How can we use Spatial Analyses to increase accuracy of POD?Search assignment generation (unified

cartography)Identifying “holes” in search (using GPS)Determine optimal sweep width (ground-truth)

How do we know when we have reached our point of diminishing return?When to stop searching?Search Area vs. Rest of World (ROW)

Objective 3 - Methods

Objective 3: Increase the probability of detection

Objective 1: Georeference SAR dataGeo-referencing 4,000 points will require

significant effort, difficult to do with accuracy/ high resolution

Data quality? N = ~4000 points n = ~2,000 points

Concentration of points around popular hiking areas

Factor significance may vary across the type of incident

Objective 2: Probability of AreaTerrain model will need ground-truth and will

not apply in other habitatsObjective 3: Probability of Detection

Balancing GIS Application vs. GIScience

Challenges…

Novel research topicIncrease overall preparedness for incident

occurrencePreventative Search and Rescue initiativesHelp find missing persons (Searches are

emergencies!) Allow GIS students and Search and Rescue

personnel to collaborate and elevate collective knowledge

Technique testingGeo-referencing/ spatial uncertaintyEcological Niche Factor Analysis Predictive models

Broader Impact

National Park ServiceESRICalifornia Emergency Management AgencyNational Association for Search and RescueWilderness Medical SocietyYosemite Leadership Program at UC MercedYosemite FundYosemite Search and Rescue

Potential Collaborators

ESRIFeature article

in ArcUser Magazine

YOSAR wins Special Achievement in GIS award

Any Questions?