Reports

Australian SAR Data

We collected data from 2000 to 2005 on land SAR incidents in Australia.

• Final Report: (Jun 2006). [ PDF ]
• View the definitions.
• Public Data:
  • Fields (a.k.a. Variables): [ Zipped PDF (1 page) ]
  • Data: [CSV (404k, 689 lines)]
  9 fields removed: officer contact info and potentially identifying information including coordinates. (These fields are not used in the report.) CSV is a standard text format. Excel and other spreadsheets read it happily.
• Python Programs and supporting files:
  • README
  • [Zip 1.1M ]
  For the analysis junkie, not the casual user. Text-mode programs. See the README.

ISRID Lost Person Behavior Reports

We are modifying the

sar-report

• DRAFT Washington State 2004-2005 data: (May 2006) Presented by Bob Koester at the Washington State SAR conference. [ PDF ]
• DRAFT New Zealand 2004-2005 data: (Nov 2006) Presented by Bob Koester at the NZ LSAR conference [ PDF ]

Other Reports

Rights and Licensing

The materials and tools are copyright © the various authors and organizations involved. Documents released as "draft" are circulated for feedback and commentary and are not to be made available elsewhere (except via links to this site).

The SARBayes Optimal Resource Allocation Library (SORAL) is copyright Monash University and various authors, and released under the Gnu General Public License (GPL) which allows anyone to use it, but requires any redistribution to also be GPL. If you intend to incorporate SORAL in a proprietary (closed-source) package, you will need a separate license. Please contact us for reasonable terms (we intend to have this software used.)

The Probability Mapper GUI makes use of the proprietary Netica API by Norsys. However, Norsys have exempted us from license fees so long as we do not charge for our application. This program has become part of the AGM SAR project.

AGM (originally Andre, Gareth, and Michael) has allowed SARBayes to redistribute their 3rd-year project, which implemented optimal resource allocation using the Charnes-Cooper algorithm in a robust, well-documented, thoroughly-tested text-based user interface. At this point they would like specific requests for the source code.

Forms

Data-entry Form

• Download the current form: [ Word97 format | PDF format ]
• Use the online data-entry instead.

Software

AGM SAR program

You want at least the executable and text files. Most people won't care about the extra documentation. The pre-packaged version does not use SORAL and so can only handle one kind of resource. (A SORAL version is available upon request.)

SORAL: SARBayes Optimal Resource Allocation Library

Public beta released: Beta_2!

See the Projects page for a description and links to the documentation.

SORAL is released under the Gnu Public License (GPL). It is free software and has no warranty. Read the license.

• Source code for Beta_2 release [ .tar.bz2 | .zip ]
• MATLAB source code for Washburn's JOTA algorithm. [ mat_wash.zip ]

For the latest code, or to report bugs, use the web-based SVN archive, wiki and bug-tracking system.

Probability Mapper

See the Projects page for a description and links to the documentation.

A proof-of-concept demonstration version of PM using the AGM SAR engine is available upon email request. It is held together with string and baling wire, and does not run on all systems. (It uses MS Access 97 for the database.)

Bayesian Networks for use in the Probability Mapper

These networks are in Netica (.dne) format, for use with Probabilty Mapper. The program will automatically construct probability tables and maps from the chosen network, using the network's lost-person categories etc.

Please do not use these in the field yet! While some models perform better than others, all indications so far say that none of these models is much better than saying "half our subjects will be within 1 mile".

koester-simple.dne:

Our implementation of data in Bob Koester's reports (as of early 2002) in a Bayesian network. This version assumes all statistics are determined by the person's category alone. (Naive Bayes model)

syrotuck.dne:

Our 2002 implementation of the data in William Syrotuck's book as a Bayesian network.

rik-minimal.dne:

A research-only minimal network we made early in our knowledge-elicitation sessions. It has a mere 3 nodes and very rough probabilities.

honors2.dne:

Representative of the networks machine-learned from Bob Koester's raw data, in Adam Golding's honors project (2001).

CVS Download (for developers)

Note: CVS and CVSTrac is unavailable because of the Monash server crash. We will probably move these projects to SourceForge, or similar. We apologize for the inconvenience.