Abstract:
The ability to predict the distribution of submersed aquatic vegetation in the
Upper Mississippi River on the basis of physical or chemical variables is
useful to resource managers. Wildlife managers have a keen interest in advanced
estimates of food quantity such as American wildcelery (Vallisneria americana)
population status to give out more informed advisories to hunters before the
fall
... hunting season. Predictions for distribution of submerged aquatic
vegetation beds can potentially increase hunter observance of voluntary
avoidance zones where foraging birds are left alone to feed undisturbed. In
years when submersed aquatic vegetation is predicted to be scarce in important
wildlife habitats, managers can get the message out to hunters well before the
hunting season (Jim Nissen, Upper Mississippi River National Wildlife and Fish
Refuge, La Crosse District Manager, La Crosse, Wisconsin, personal
communication).
A statistical model was developed to predict the probability of occurrence of
submersed aquatic vegetation in Pool 8 of the Upper Mississippi River on the
basis of a few hydrological, physical, and geomorphic variables. Our model
takes into consideration flow velocity, wind fetch, bathymetry, growing-season
daily water level, and light extinction coefficient in the river and calculates
the probability of submersed aquatic vegetation existence in Pool 8 in
individual 5- x 5-m grid cells. The model was calibrated using the data
collected in 1998 (516 sites), 1999 (595 sites), and 2000 (649 sites) using a
stratified random sampling protocol (Yin and others, 2000b). To validate the
model, we chose the data from the Long Term Resource Monitoring Program (LTRMP)
transect sampling in backwater areas (Rogers and Owens 1995; Yin and others,
2000a) and ran the model for each 5- x 5-m grid cell in every growing season
from 1991 to 2001. We tallied all the cells and came up with an annual average
percent frequency of submersed aquatic vegetation occurrence and compared the
results with actual LTRMP survey data. Both a paired Student's test (P =
0.4620) and a Wilcoxon's two-sample test (P = 0.4738) did not contradict our
null hypothesis that the model prediction and the sampling data are
statistically the same. We have not found an effective statistical test to
compare model-predicted spatial pattern with aerial photography geographic
information, but we are satisfied with the model's outcome on the basis of
visual inspection.
A unique feature about this model is that a prediction can be made by the end
of June each year; therefore, providing wildlife managers an assessment of
current year vegetation growth condition 3 to 4 months ahead of the arrival of
migrating waterfowl that feed on submersed aquatic vegetation. We are working
with the LTRMP partnership to create a mechanism so that model predictions
(fig. 4) can be updated annually and the results posted on the LTRMP Web site.
Our model underestimated the prevalence of vegetation from 2001 to 2004. We
speculate that the summer water level reduction conducted in 2001 and 2002
triggered vegetation responses that are outside the model's domain. Future
enhancement of the model will incorporate the summer water level drawdown
effects as well as the effects of growth conditions in previous years.
[Summary provided by the USGS.]