Back to the WHR Book Page ...
Back to The Plexus Home Page ...



GARTER --
A Garter Snake Identification System

as discussed in:
Wildlife-Habitat Relationships: Concepts and Applications, Second Edition

In a recent textbook Wildlife-Habitat Relationships: Concepts and Applications, Second Edition, I discussed a knowledge-base approach to examining the inadequacies of the current taxonomy of garter snakes (Thamnophis spp.) in northwestern California.

Background -- The Snake Sample

In the book I present the results of applying the meristics -- scale counts and coloration -- and habitat of a sample of garter snakes I caught in the Klamath Mountains, to current taxonomic identification keys.  I built an expert identification key to the supposed species and subspecies of the region, using an example-driven rule-base model.  I then compared the meristics of captured snakes to the identification model, to test how well the existing taxonomy and keys fit the real world.

I built the identification model by using the program 1st Class (Programs In Motion, which may now be defunct or subsumed under some other company).  As I describe in the book,

The expert system applies an induction-based optimization algorithm (the Iterative Dichotomizer Three or ID3 induction algorithm; Quinlan 1986, Shapiro 1987) that creates the most efficient rule to best distinguish among known outcomes of examples.  In this case, outcomes were garter snake species and subspecies, and examples were the published characteristics for their identification.  The ID3 algorithm, rather than a multivariate statistical approach, was chosen for this analysis because: it can explicitly handle uncertain or unknown characteristics; it can deal with "fuzzy" or overlapping categorical descriptions, such as of coloration patterns; it can be used to easily explain classification results and to explore alternative descriptions; and it produces a simple and easily-understood classification rule tree.
Running the expert system for each of my 26 field sample observations helped determine the degree to which the key can identify each field specimen.  This helped me identify which field specimens represented outliers, potential hybrids, or undescribed forms, and the specimens' characteristics contributing to these findings.  The operational expert system allows for answers of uncertainty, from which it then prompts for additional characteristics not covered in the static identification key.

Results -- Orphan Snakes

The results suggest that some individual snakes just don't fit the keys.  While this is well known for the taxonomically messy garter snake complex, the approach served to [1] demonstrate the utility of a knowledge-base approach to identifying problems of taxonomy, and [2] identify the specimens, their specific characteristics, and the taxon-specific identifications (to species or subspecies) that confounded the standard identification keys and existing taxonomy.  This may be an invaluable step in revamping the systematics of this snake complex.  It's the first and only example I know of applying this form of modeling to these types of questions.

Presenting the Expert System -- GARTER

And here is the expert system GARTER for you to download and to inspect and use on your own.  It comes in a PC DOS PKZip file (40KB) called GARTMODL.ZIP, which consists of 3 files.  Unzip them to any directory, and run the model by typing GARTER <Enter>.

Since the Analysis Was Done

Since I conducted this analysis, the garter snake complex in the Pacific Northwest U.S. has been slightly revised.  New to the taxonomic scene is now the Pacific Coast Aquatic Garter Snake, Thamnophis atratus (Kennicott).  One of its subspecies, the Oregon Garter Snake (T. a. hydrophilus), was previously identified as no less than three different species:  Northwestern Garter Snake (T. ordinoides), Western Terrestrial Garter Snake (T. elegans), and Western Aquatic Garter Snake (T. couchii), according to Brown et al. (1995). \1 \2

As in my analysis, the Klamath Garter Snake, Thamnophis elegans biscutatus, is generally recognized as a subspecies of Western Terrestrial Garter Snake, whereas others denote the Western Terrestrial Garter Snake in the Klamath Mountains to be an intergrade between the Wandering Garter Snake (T. e. vagrans) and Mountain Garter Snakes (T. e. elegans).  Indeed, the results of my simple modeling approach also suggested a major discrepency in the taxonomy of the T. elegans complex.  In many instances, my expert identification system GARTER was unable to identify T. elegans to subspecies.

Northwestern Garter Snake (T. ordinoides) still seems to be a fairly stable species, with no subspecies currently delineated (Brown et al. 1995).  I found none in my Klamath Mountains samples.

Conclusions -- Again

It is rather heartening that the taxonomic confusion over some of the T. elegans subspecies complex -- as I identified in my little modeling analysis -- may have been in part now "solved" by the introduction of T. atratus in the newer taxonomies.

We also are a few steps ahead in answering my questions about revamping taxonomic designations of this species comples.  As I wrote,

Are current designations of species and subspecies of western garter snakes appropriate?
But I also posed some difficult questions of conservation, which are yet unanswered, and which may bear further work:
Are some currently described, or unknown, subspecies rare in this geographic area?  Should they be listed?  Are there undescribed morphs not given subspecific status, that need to be recognized for conservation of genetic diversity?  Given uncertainties in taxonomy, what habitats should be provided, and where should they be provided?  Might swamping or homogenization of species gene pools caused by interspecific hybridization affect their status and, if known, should it be controlled?  In this case, only further field studies can help resolve these questions.
And to learn more about the specifics of this modeling exercise, please consult (= purchase) our textbook!

Citations

Brown, H. A., R. B. Bury, D. M. Darda, L. V. Diller, C. R. Peterson, and R. M. Storm.  1995.  Reptiles of Washington and Oregon.  Seattle Audubon Society, Seattle, Washington.  176 pp.

Quinlan, J. R.  1986.  Induction of decision trees.  Machine Learning 1(1):81-106.

Shapiro, A. D.  1987.  Structured induction in expert systems.  Addison-Wesley, Reading MA.

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

\1 And it's still not fully resolved.  Brown et al. (1995) go on to note that T. atratus is known to hybridize (interbreed) with both Western Aquatic Garter Snakes (T. couchii) and Two-striped Garter Snakes (T. hammondii) in California south of the Klamath Mountains.  (Note that Two-striped Garter Snakes do not occur in the Klamath Mountains.)

\2 Also recently published has been:
Rossman, D. A., N. B. Ford, and R. A. Seigel.  1996.  The garter snakes: evolution and ecology.  University of Oklahoma Press, Norman OK.  336 pp.
 I have not studied this volume.  [Thanks to Plexus reader Theresa Ward for the correct citation.]


Back to the WHR Book Page ...
BACK to The Plexus Home Page ...