*** Air/Ground Probability of Detection Calibrations ***

12th June 1996
From: Martin Colwell
      sarinfo@mindlink.bc.ca

To: Patrick Keane
    pkeane@slip.net
Illinois Emergency Management Agency

Dear Patrick,
Thank you for the email regarding your new POD project. There is no doubt
that good POD data for human objects is pretty scarce and your wish to
acquire new data for your area is definitely to be commended.

Regarding your suggested POD project design I would like to suggest the
following points.

The first is to be aware of the scientists 'test matrix syndrome'. Any
time you add another variable (such as additional types of search terrain)
to your experiment you are adding another complete set of work to be
undertaken. Your design includes five types of terrain in four seasons. This
makes 20 field calibrations, which is ambititious. I am not sure that POD's
over a large lake can be easily included, as the dummy/object type may have
to be different than for the land-based field calibrations. I would be
inclined to remove this 'marine' calibration from the project unless you
believe that it will not be logistically difficult to perform.

The second point is to decide on the type of 'target'. From your mail I
get the impression that you wish to obtain POD's for humans. This is fine
but it does require being a little more specific. If you decide to use live
people as opposed to dummies then you have the logistic problem of getting
large numbers of people and keeping them in the search area. I have had no
success in recruiting large numbers of people to stand in the woods all day
and so elected for dummies instead - however for FLIR live persons are
required, at least as one subset of the whole target matrix. For dummies we
have used cardboard boxes with rolled corrugated cardboard for limbs, we
have also used coveralls with a ballon for head. The coveralls are filled
with snow in the winter or straw could be used in other seasons to pad out
the coveralls.

If you decide to use dummies then you have to decide whether to posture
them randmonly, or as standing, sitting or lying. The differences can be
significiant. A lying subject seen from the air (in summer) will probably
have a higher visibility than a standing dummy, from the ground the reverse
will probably be true. I would be inclinded to test all three postures.
Each dummy (or other target) has to have an ID number painted on it which
has to be recorded with each find, otherwise multiple counts of the same
dummy will greatly exaggerate the measured POD (which could be over 100%
if this is not done!). Painting and numbering the targets is a time-
consuming but very important task.

The next question is how many dummies to place. I would suggest 20 of each
type, this will give you a maximum accuracy of +/- 5%POD for each dummy out
of twenty that is found. For three postures this will require 60 dummies
per site, for four sites 240 dummies are required, unless you move them
from site to site, which is quite labour intensive. This does not include
the 20 'live' human targets required for the FLIR. The dummies & humans
should have their clothing colours typical of a normal distribution, which
is primarily red and blue, with green, brown, black and white present in
smaller numbers. This is also important because green, brown and white
change from low visibility to normal visibility targets as the seasons
change between summer to winter.

Given the time required to manufacture the dummies and place them in the
field I would recommend that you also consider performing ground-search POD
calibrations once the dummies are in place. This could be done immediately
after the air search, or on a following weekend. This is important because
most of the work is in actually preparing the test sites with the dummies,
once that is complete actually performing the sweeps and recording the POD
finds only takes a few hours or less. If you decide to perform ground
sweeps then I encourage you to include 20 dummies with radios (am/fm type)
and complete a sound-sweep calibration at the same time. Remember that once
all the dummies are in place it takes virtually no longer to search for
say four or five target types than it does to search for one type.

For the heli/FLIR calibration this could be done immediately before or after
the fixed-wing sweep. If twenty searchers are given numbered bibs they can
be moved into the search area and counted by both FLIR and visual spotters.
I recommend that you have both types of spotters in the helicopter and
record their POD finds of these human subjects independantly, to determine
if, during the day, the FLIR is better or worse than normal visual spotters.
If you can repeat the process at night using the FLIR alone that would be
excellent.

It is essential to record the searcher spacing, elevation and search speed
of each search resource, both to provide a reference for repeating these
configurations during a search but also to permit logistic calculations to
be performed at search time. Ground searcher speeds should be calculated
from their recorded times into and out of the search site boundaries. The
width and length of the search sites within which the targets are placed
must also be recorded.

Now we have to consider the type of search pattern. If you decide to
perform a linear sweep along a single route then the route must be
carefully defined and the distance of each target from the route must be
recorded. Without this it is impossible to provide POD data with distance
away from the route. This is a common weakness of much air POD data. If the
distance from the route at which the POD drops to an unacceptably low level
is not recorded then it is impossible to specify with any kind of accuracy
how far over (i.e. spacing) the next route assignment should be placed.

If a gridsearch pattern is chosen then the targets can be placed randomnly,
as the searcher spacing will permit us to calculate the POD at different
searcher spacings. Obtaining ground POD data is much faster using a
gridsearch as many searchers may sweep the areas simultaneously, as opposed
to the sequential deployment required along a single search route. In
either case the data can be converted from POD at a spacing to POD at a
distance from the searcher.

Refering to your proposal of a 'control' aircrew. There really is no such
thing as a control set of POD data. Each search resource will obtain their
own POD's and these results should be included as part of an average POD
dataset for that resource in that terrain during that season. Each resource
should sweep the area at least twice, preferably on the same day, so that
the reproducability of the POD data can be recorded. If the results are
meaningful (i.e. similar, reproduceable cailbration curves) then their POD
data can then be averaged and employed during searches.

I hope this helps you set up your POD cailbration project. Let me know if
you need any further information.

Best wishes,

Martin Colwell.

                    -----------------------------