home

Update on New York Aquatic GAP Pilot Project
MARCIA S. MEIXLER and MARK B. BAIN
New York Cooperative Fish and Wildlife Research Unit, Cornell Univer-
sity, Ithaca, New York

This is an update report on the pilot GIS project for aquatic systems
that began in 1995 to define a methodology and determine the fea-
sibility of predicting biodiversity distribution. Similar to gap analysis
in terrestrial environments, gap analysis for aquatic systems uses
remotely sensed data for habitat mapping, infers aquatic biodiversity
distribution from habitat data, and provides large-scale information
for targeting conservation measures. The pilot project has been a
low-level effort (e.g., a one-person project) for four years. The
aquatic GAP pilot project was developed in the Allegheny River,
Fall Creek (1:100,000 scales), and French Creek (1:24,000 scale)
watersheds of New York to study habitat classification in multiple
locations throughout the state and at two spatial scales.

The basic aquatic GAP model predicts relative levels of fish and
macroinvertebrate diversity and identifies stream reaches with high
biodiversity without management or protection. This was accom-
plished by classifying stream segments into habitat types using five
attributes: stream size, habitat quality, water quality, stream gradi-
ent, and riparian forest cover. Stream segments were then classi-
fied into one of eighteen habitat types for fish diversity prediction
and one of eight habitat types for macroinvertebrate diversity pre-
diction using the five attributes. The first round of habitat charac-
terization (used 1995-1997) involved static, manually intensive clas-
sifications from topographic and Mylar land use overlay maps. In
an effort to deviate from such limiting classification, we developed
computerized macros to automate classification from digital eleva-
tion models, land use, road and railroad coverages (termed the au-
tomated method). This provided equal or better accuracy, increased
flexibility, and enabled us to calibrate the model using previously
collected data. Through extensive literature searches, fish species
were associated with habitat types using information on preferences
and tolerances for stream size, degree of habitat specialization, and
tolerance to water pollution. Macroinvertebrate families were as-
sociated with habitat types using information on feeding guild, life
habit, and tolerance to water pollution. Predictions of habitat types
and associated fish species and macroinvertebrate family diversity
levels were performed and gaps in protection located.

An additional method of habitat characterization (termed the cali-
brated method) was developed for fish diversity prediction using
discriminant analysis. Five landscape attributes were added to those

of the automated method: point source pollution, surficial geology,
bedrock geology, depth to bedrock, and priority water status. Pre-
viously collected fish collections were used with all ten landscape
attributes to statistically optimize the prediction of high fish diver-
sity habitat on a stream segment basis. Species-specific optimiza-
tions for fish were also performed using the same methodology.

Standardized collections of habitat, fish, and macroinvertebrate data
were performed at 39 stream sites in the Allegheny River water-
shed in the summer of 1998 for use in testing the predictions. The
automated method succeeded at predicting stream size and stream
gradient with accuracy, and our adaptation of the nonpoint-source
pollution model was able to predict relative pollutant levels for to-
tal nitrogen but was less successful for total phosphorus, suspended
sediment, and biological oxygen demand. Tests of predicted habi-
tat quality and riparian forest cover indicated more than chance
agreement with observed data. Significant correlation existed be-
tween predicted and observed fish diversity using both automated
and calibrated methods in the Allegheny River watershed.
Macroinvertebrate diversity predictions, only performed using au-
tomated methods, were also well correlated with observed diver-
sity in the Allegheny River watershed. Predictions at the 1:100,000
scale in the Allegheny River watershed were uniformly higher in
accuracy than those from the 1:24,000 scale in the French Creek
watershed for both fish and macroinvertebrate diversity using both
automated and calibrated methods. Species-specific optimizations
for fish using calibrated methods revealed weak correlations with
observed species occurrences, demonstrating a need for more re-
fined methodologies for species-level predictions. It is clear that
the community-level modeling procedures presented here have po-
tential as coarse but feasible methods in identifying high-diversity
habitats that should receive priority conservation attention at the
watershed scale.