Prediction of Fundamental
Assemblages of Mid-Atlantic
Highland Stream Fishes

Michael Cyterski, Rajbir Parmar, Craig Barber , Brenda Rashleigh, John Johnston, Kurt Wolfe

EPA/ORD/NERL/ERD, Athens, GA

EPA Science Forum

A software tool, the Stream Fish Assemblage Predictor (SFAP),
developed using the USEPA's EMAP stream sampling data in
the mid-Atlantic Highlands, can predict stream fish communities
using stream and watershed characteristics. Step one in the tool
development was a cluster analysis that formed groups (clusters)
of streams with similar fish species. Each cluster has a
multidimensional mean, or centroid, defined by the biomass of
each species in the group. Using an iterative process, streams
were added, one by one, to the cluster with the nearest centroid.
I specified that each cluster had to have a membership of at least
1 % of the total sample size of 665. Smaller clusters were deleted
at the end of each iteration step. In addition, observations could
not join a cluster if they were more than a specified Euclidean
distance from the cluster centroid. This methodology produced
21 total clusters (see table at top right).

Step two was a discriminant analysis, which produced a system
of equations to predict a stream's cluster based on
characteristics of that stream and its watershed (e.g., stream
depth, width, and flow; percent forested area in the watershed;
amount of in-stream fine sediments).

Using the EMAP dataset, I tested the predictive accuracy of the
discriminant equations. Streams were correctly classified
approximately 42% of the time (i.e., the actual cluster was the
most likely cluster). The actual cluster was one of the three most
likely clusters approximately 70% of the time. Randomly, given
three choices, one would only have a 3 in 21 chance of picking
the correct assemblage (14%).

I envision use of this software by a wide diversity of stakeholders,
from private landowners and public interest groups to municipal
planners and developers to environmental management
professionals. One goal would be to predict fish communities in
streams for which basic watershed and stream characteristics
are known, when actual sampling of the stream is cost
prohibitive. Users could also investigate potential impacts of
environmental restoration/degradation by altering stream and
watershed characteristics, then noting subsequent changes in
the predicted fish community. For researchers, this tool's basic
fish community information can be passed to more complex,
mechanistic fish community models that examine the effects of
stressors on stream fish communities.

IDENTIFIED FISH ASSEMBLAGES



Clusters are ordered by their
number ofmember sites, from



These numbers represent the average
percentage of this cluster's total fish biomass
attributable to this particular fMr species.

Cluster 6 (si? Cluster 10 5

Cluster 3 5

Cluster 7 46 Cluster 9 40

White Sucker
Blacknose Dace
Longnose Dace
Creek Chub
Rock Bass
Fallflsh

Northern Hog Sucker
Cutlips Minnow

5.3 White Sucker 71.5 Northern Hog Sucker 26
.3 Creek Chub 4.6 Rock Bass 16
Blacknose Dace 4.5 White Sucker 6
Rock Bass 1.9 Sma 11m outh Bass 5
Northern Hog Sucker 1.5 Stonerolter 5
Slimy S culpin 1.4 Cre ek Chub 4
.7 Brown Trout 1.3 Bluntnose Minnow 3
Brook Trout 1.2 Longnose Dace 2

6 Rock Bass X. 62.1 Creek Chub 51.6
2 S m a Dm outh B ass C7.&3 Bla cknos e Da ce 23.9
2 White Sucker 5.8 White Sucker 3.9
[ Northern Hog Sucker 5.0 Stoneroller 3.1

0	Bluntnose Minnow 1.6 BlueheadChub 2.4

1	Stoneroller 1.6 Northern Hog Sucker 1.8
Blacknose Dace 1.3 Longnose Dace 1.4

9 Creek Chub 1.2 Black Sculpin 1.2

Sum

Cluster 2

Sum 87.9 Sum 69.8 Sum 86.2 Sum 89.4
Cluster 5 31 Cluster 11 30 Cluster 20 28 Cluster 13 26

Creek Chub
Blacknose Dace
Stoneroller
Black Sculp in
White Sucker
Pumpkhseed
Silverjaw Minnow
Bluntnose Minnow

Blacknose Dace 93
Creek Chub 1
FantailDarter 1
Sm allmouth Bass 0
BlueheadChub 0
Brook Trout 0
Black Sculpin 0
Stoneroller 0

9 Sma 11m outh Bass 34.9 Creek Chub 35.3 BlueheadChub 44.8
Rock Bass 12.9 White Sucker 32.9 Creek Chub 10.2
White Sucker 5.6 Blacknose Dace 12.5 Northern Hog Sucker 6.5
Northern Hog Sucker 5.6 Longnose Dace 1.8 Blacknose Dace 3.4
Stoneroller 3.6 Pumpkinseed 1.6 Mountain Re dbelly Dace 3.0
Walleye 2.6 Mountain Re dbelly Dace .1.3 Green Sunfish 3.0
Large mouth Bass 2.5 Fallflsh 1.2 White Sucker 2.9
Redbreast Sunfish 2,1 Rock Bass 1.0 Cre ek Chubsucker 2.9

Sum 98.6 Sum 98.5 Sum 69.9 Sum 87.7 Sum 76.9
Cluster 16 26 Cluster 12 22 Cluster 14 21 Cluster 19 20 Cluster 4 16

Blacknose Dace 53.8 Northern Hog Sucker 62.1 Torrent Sucker 60.0 Blacknose Dace
Creek Chub 16.4 White Sucker 5.6 Blacknose Dace 9.2 Longnose Dace 2
White Sucker 6.7 Rock Bass 5.1 Blue he ad Chub 6.0 Brook Trout
Black Sculpin 2.4 River Chub 3.3 FaBfish 4.9 Black Sculp in
BlueheadChub 2.1 Sm allmouth Bass 3,2 Brook Trout 2.3 Slimy Sculpin
Stoneroller 1.9 Creek Chub 2.5 White Sucker 2.2 White Sucker
Mountain Re dbelly Dace 1.6 Blacknose Dace 2.0 Stoneroller 1,3 Stoneroller
FantailDarter 1.6 RainbowTrout 1.6 Yellow Bullhead 1.0 Bluntnose Minnow

1 Gizzard Shad 22.3
8 Longnose Gar 15.2

4	Flathead Catfish 12.3
i Large mouth Bass 6.1

5	Silver Redhorse 4.6
Channel Catfish 4.4
Bluegill 4.0
Sma 11mouth Bass 2.8

Sum 86.5 Sum 85.4 Sum 86.8 Sum 88.6 Sum 71.6
Cluster 18 16 Cluster 8 15 Cluster 21 15 Cluster 1 13 Cluster 15 9

Creek Chubsucker 29.3 Brook Trout 87
Chain Pickerel 18.1 Blacknose Dace 5
Falifish 11.7 Black Sculpin 3
White Sucker 7.4 Mottled Sculpin 1
Pumpkhseed 6.2 Longnose Dace 1
Large mouth Bass 2.9 Slimy Sculpin 0
Stoneroller 2.5 FantailDarter 0
Torrent Sucker 1.9

3 Northern Hog Sucker 40
Sma 11m outh Bass 28
River Chub 6
Rock Bass 6
White Sucker 1
Muske llunge 1
Bluntnose Minnow 1
Creek Chub 1

9 Creek Chubsucker 6
5 BlueheadChub 7
White Sucker 3
I Blacknose Dace
5 Large mouth Bass

Blue gill 1
Pumpkinseed 1
Gre e n Sunfish 1

.9 BlueheadChub 28.6
0 Roanoke Hogsucker 27.9
3 White Sucker 16.4
3 Black Jumprock 5.7
2 Torrent Sucker 5.5
7 Mountain Re dbelly Dace 3.0
7 Stoneroller 1.7
5 Cre ek Chub 1.6

Sum 80.0 SumQ00.3) Sum 87.8 Sum 89.5 Sum 90.5

/ Cluster 17 9
r Gotten Redhorse 34.9

H

sum of 100% and a low number of

omogeneity in a larger cluster is more
table (e.g., Clusters 2 and 5).

Sma 11m outh Bass 10
Northern Hog Sucker 6
Silver Redhorse 4
Torrent Sucker 3
Rock Bass 3
Re dbre ast S unfish 3

A low sum indicates that this cluster is very heterogeneous. It
takes 33 species to incre ase the summe d cluster biom ass to > 95%.
Beyond saying that the presence of Golden Redhorse is likely, it is
difficult to predict with accuracy what a fish community in

i/',"	

SPATIAL DISTRIBUTIONS OF IDENTIFIED ASSEMBLAGES

This tool can be aGGessed from the Canaan Valley Institute's
website at www.Ganaanvi.org. A desktop version of the software
is also being developed at ERD Athens. When completed, it will
be distributed from EPA's Center for Exposure Assessment
Modeling (CEAM) website: www.epa.gov/ceampubl/

•	Ouster
q

•	Ouster

9

•	Ouster

3

Ouster

4

Ouster

5

•	Ouster

96

° Ouster

97

*	Ouster

98

Ouster

19
Ouster

20
Ouster

21

•	Cluster 6
0 Cluster 7
® Cluster 8
° Cluster 9

*	Cluster
10

•	Ouster

91

® Ouster

92

•	Ouster

13
Ouster

14
Ouster

15

Science and Innovation to Protect Health and the Environment


-------