2005 GBIF Ebbe Nielsen Prize:
NDM / VNDM: Computer Programs to Identify Areas of Endemism
Pablo A. Goloboff
Instituto Superior de Entomología (Facultad de Ciencias Naturales, Universidad de Tucumán), and
Consejo Nacional de Investigaciones Científicas y Técnicas, Miguel Lillo 205, 4000 S.M. de Tucumán,
Argentina
pablogolo@csnat.unt.edu.ar
A taxon is said to be endemic to an area when it is found in that area but nowhere else.
If limits in distribution of multiple, different groups of plants and animals are
determined by the same factors, it is expected that the distributions of those groups will
show similar patterns. Therefore, it is also expected that some areas or regions – socalled
areas of endemism – will consistently house taxa that represent multiple different
groups that are found only there and nowhere else. Although the notion of endemism is
an old one that dates back to the early 1800s, attempts to identify areas of endemism by
formal, quantitative means have begun only recently. The most widely used method
records presence/absence data in a grid, and subjects the resulting data matrix to
analysis by algorithms that implement criteria established for reconstructing phylogeny.
However, these criteria are problematic when used to try to determine areas of
endemism. Recent publications by this author and his collaborator Claudia Szumik have
proposed alternative criteria that are designed specifically for recognition of areas of
endemism. These criteria are implemented in two interacting computer programs, NDM
and VNDM. These programs take as input vouchered locality data taken from natural
history specimens (such as those made available by GBIF), and convert these into a grid
of presence/absence data (optionally, probable presence can be used as a third category).
The criterion used is very simple: for a given set of cells (or "area"), a given taxon can
be considered as "endemic" to that area if it is more or less evenly distributed across the
area, but is not found in distant cells. Thus, choosing sets of cells for which many taxa
are endemic provides a natural way to identify areas of endemism. However, because
the program has to examine, in principle, all possible combinations of cells, a
computational problem arises. For a grid of c columns and r rows, there are
c.r-1 (c.r)!
? ---------
i = 2 i! (c.r-i)!
possible combinations of cells. Even for a modest c = 10, r = 20, there are more than
22 x 109 possible sets of cells. Therefore, it is impossible to actually examine every
possible cell combination when given input of the larger datasets that more realistically
reflect biological reality. For this reason, a trial-and-error technique that has proven
useful in other kinds of analyses is employed by the program NDM. The procedure uses
promising sets of cells as starting points, modifies them by deleting or adding cells
according to a specified set of rules, re-evaluates the degree of endemicity, and keeps
the sets of cells with highest endemicity scores while discarding those sets with low
scores. Also, endemicity analyses present additional problems when partially
overlapping areas, or the combination of several areas into one, are attempted. An
analysis of a real data set is presented to illustrate the approaches used by NDM and
VNDM to dealing with these challenges.