Waldo in the Wilderness

Using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

How to find people faster and more often

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

Introduction

Discussion of Matt Jacobs' paper

Terrain Based Probability Models for SAR (2015)

How did we get here? A brief history of current land search strategy

A review of "probabilistic thinking"

The paper, its findings, and the "spider web" strategy

Summary: new strategies using Matt's data and probabilistic considerations

Extra credit:

Recommendations for operational periods
Applications for tracking

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

A Scenario

OP 4

A: LKP

B: Popular trail

C: Trail/creek junction

D: Heard "help"?

E: K9 scent alert

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

Introduction

About me, briefly

SAR volunteer for a California team 2016 to 2021
- SAR Type 1, Hasty team, IC roles (esp. Plans section)
CalOES Direction & Control, 2017
FEMA NIMS ICS-300, "Managing the Expanding Incident", via CalFire, 2020
NASAR MLPI, 2021
Tracker, tracking instructor, tracking team leader
Software developer, Linux sysadmin, etc.
Math, science, data, and GIS geek

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

Introduction

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

A review of probabilistic reasoning

Where are my keys?

Common behavior:

Check pockets, table by door, dresser, night stand, desk (common places)
Check sofa, recliner, bathroom, laundry (less likely, but plausible places)
Check refrigerator, cupboards, and microwave (why?)
Check pockets again, table by door again, dresser again, night stand again, desk again
Check sofa again, recliner again, bathroom again, laundry again

They haven't been found, they must be in one of these places, I just need to check them again, more carefully this time.

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

A review of probabilistic reasoning

Where are my keys?

They're in the car.

Outside your search area.

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

A review of probabilistic reasoning

Let's call this Proximity Bias: intense search activity where we expect to find what we're looking for.

Maybe a special case of the Base Rate Bias ?

Bias: "...any deviation from correct reasoning about probabilities or Bayesian updating..."

Handbook of Behavioral Economics, Daniel J. Benjamin

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

A review of probabilistic reasoning

Gather
evidence

Evaluate

Guess

Test

Scientific method!

Probabilistic reasoning uses data from past efforts to make more accurate predictions

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

A review of probabilistic reasoning

People intuitively understand these ideas

but, there are lots of possible biases and errors in reasoning ("bugs" in our "human software")
and doing this properly requires a lot of work
- collecting data
- analyzing the data
- making predictions based on the data

So it is hard to follow in practice.

Be suspicious of hunches, past experience, "must be here", and misleading evidence.

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

A review of probabilistic reasoning

Probabilistic reasoning is used a lot in search and rescue:

Probability-of-Detection (POD)
Probability-of-Area (POA)
Mattson consensus method

But

There's a lot of pseudo-science and sparse data here.

Matt Jacobs' paper suggests some new methods:

a little more data
and a better analysis of that data

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

How did we get here?

A quick history of current land search strategy

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

History of land search strategy

Probably influenced by maritime search strategies in the 1960s or 1970s, Managing the Search Function
Establishing Search Priorities by Robert Mattson 1976 - 1979
Analysis of Lost Person Behavior, Syrotuck et al, 1976
Lost Person Behavior, Robert J. Koester, 2008
ISRID
Local variations
...?

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

History of land search strategy

This is a smooth or continuous uniform

probability distribution.

Each segment in this area of the ocean has an equal probability of detection and any point in this area is as likely as any other point.

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

History of land search strategy

Marine search has changed a lot in the last 50 years:

Advanced modeling software
Lots of data on currents and flows
Experimentation
Funding
And new tactics

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

History of land search strategy

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

History of land search strategy

Advancements in land SAR search strategy:

Lost Person Behavior
- Range rings still have a smooth probability distribution within each range ring (equal POA at same distance from PLS/LKP in any direction mostly regardless of terrain)
- Lost Person Behavior includes notes emphasizing terrain features for some subject categories
- ...but area / grid searches are still the common approach
- 95% range ring is often treated as ROW because it's very big

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

History of land search strategy

Scenario:

Missing trail jogger, mid-30s, avid hiker
Got bad directions to a trailhead and is on the wrong trail
Last seen 0700, reported missing 12 hours later, vehicle located LS+26h, OP 1 begins LS+28h
LPB 95% distance from IPP is 3.6 crow miles
That means 40 square miles of rugged, difficult, mixed terrain
Assuming 60-acre assignments, that's ~425 area assignments
...or a minimum of 7,650 human hours of search time (not including overhead, transpo, IWI...)
Any volunteers?

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

History of land search strategy

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

History of land search strategy

Oh, by the way: N = 30 for "Runner" category in LPB

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

History of land search strategy

Is there another way?

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

Terrain-based probability models

Examined 1,455 find locations:

Independent of LKP/PLS/IPP (just the subject's location when found)
Focused on hunters, hikers, and gatherers (larger sample sizes)
75% were in Oregon, remainder were in New York and Arizona
Compared locations of finds to the distance to the nearest linear feature
Efforts were made to apply statistical methods to reduce factors of random chance
622 incidents of subjects on foot were examined in closer detail
Caution: Some categories had a low number of samples
- Further research with more data is required
- But this is a useful analytical update

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

Terrain-based probability models

Findings:

Uninjured subjects are significantly more likely to be found on or very near a road than injured ones
Injured and uninjured persons are approximately equally likely to be found on or near a trail
- They are both 12 times more likely to be found within a few meters of a trail
- Find probability decreases as you move farther away from the trail
Uninjured subjects are 2.5 times more likely to be found near a stream
- Injured subjects are 5 times more likely
Locations within 80 meters of the intersection of a trail and a stream are much more probable
Lower elevations are statistically more likely for injured subjects especially
High points and low points are more likely for all subjects
LPB track offsets may not be useful due to data collection methods
60% to 80% of subjects were found on or near selected linear features
- but those features comprise 45% of the search area

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

Terrain-based probability models

Terrain is not a smooth (uniform) probability distribution

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

Summary and Recommendations

Jacobs' "Spider Web"

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Core

Area searches for an unresponsive subject

Inner Web

Area searches for a responsive subject
Linear features and high terrain for an unresponsive subject

Outer Web

Natural linear features and high terrain for a responsive subject
Low terrain for an unresponsive subject

Stringers

Roads, trails, and low terrain for a responsive subject
Stream / trail intersections for an unresponsive subject

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

Summary and Recommendations

Jacobs' "Spider Web"

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

Summary and Recommendations

Other Considerations:

Time missing
Operational Period (and volume of search activity)
- Assume previously-searched areas are lower probability
Frequency of visitor activity to nearby features
- Probability Inversion: over time, less likely solutions become more likely

Waldo in the Wilderness: using terrain-based probabilistic search strategies

Introduction
A review of probabilistic reasoning
History of current land search strategy
Matt Jacobs' paper
Summary and recommendations

Summary and Recommendations

For trackers:

OP 1
1. Focus on PLS/LKP
2. nearby track traps
3. cut roads and trails within day's search area
OP 2
1. Re-assess PLS/LKP
2. cut roads, trails, and water features outside previous day's search area
OP 3+
1. Focus efforts on examining as many track traps as possible outside of areas of previous search efforts
  - especially any intersections of trails and water features