Geoscience Reference
In-Depth Information
Fig. 4.3 Graph mosaic indicating the plotted values of percent impervious surfaces and percent
tree cover land cover metrics versus the HBI and nEPT biotic indices. These values are further
stratified by landscape weighted or non-weighted scenarios
for landscape configuration. The amount of crop and grassland showed little to
moderate relationships with the stream biotic metrics, which may be a result of
relatively low amounts of these land cover types in the study area.
4.4.2 Statistical Analyses
The GAM and MLR models identified the key predictor variables for all three
stream biotic indices and allowed us to compare the effectiveness of landscape
weighting, as well as the differences between different definitions of upstream
area (catchment vs. aggregated watersheds) (see Tables
4.1
and
4.2
). With few
exceptions, the GAM and MLR models selected percent impervious area to be the
key predictor variable for number and abundance of EPT, as well as the HBI.
Percent tree cover was also a key predictor variable, being selected second for most
of the MLR models, while the size of the watershed (area) was a significant
predictor only in the GAM model results. Crop and grasslands were consistently
poor predictor variables for all scenarios.
The GAM models produced the highest R
2
values, with 76 % reported for both
EPT abundance and richness within the catchments watershed scenarios, which
had slightly but systematically higher explained variance. There was no clear
indication that catchment predictions were systematically better than the
aggregated watershed predictions, as both scenarios produced R
2
values within
a few percentage points of one another. The poorest R
2
values were generated
from the models that predict HBI, and this was consistent for both the GAM and
MLR approaches. To explore the effect of differing data collection methods used
between states, we divided the two HBI data sets for Massachusetts and Rhode
Island. The results presented in Tables
4.1
and
4.2
show the fit (R
2
) values for
