Abstract:
This collection of three manuscripts serves to improve methods for collecting,
interpreting, and utilizing autocorrelated data from headwater stream
networks. Each stream network is comprised of linear segments. These segments
lie within a unique branching structure that connects the segments
via flowing water, and the connectivity provided by water varies seasonally.
These aspects separate stream networks from other landscapes, and provide
unique challenges to the statistical analysis of stream-based phenomenon.
Two chapters of this work relied on a unique and comprehensive set of
data. These data constitute a complete census of habitat unit fish counts
from 40 randomly selected headwater basins in western Oregon. The first
objective of this work was to evaluate how different sampling designs captured
spatial autocorrelation, given the samples were drawn from a population of
spatially autocorrelated observations. Spatially distributed clusters of sampling
locations were more apt to capture spatial autocorrelation than samples
without clusters or small clusters located at tributary junctions. A similar
investigation was made concerning sampling design performance in relation
to estimating autocorrelation function values. All sampling designs lead generally
to negatively biased estimates, and practical differences among the
sampling designs were not observed. The second objective was to investigate
spatial autocorrelation model range parameters as measures of patch sizes.
It is common practice to use range parameters to infer the size of patches
within spatially autocorrelated data, but this methodology lacks sufficient
justification. The census data were used to compute range parameter values,
and another proposed autocorrelative measure of patch size: the integral
scale. The same data were used to compute patch sizes under several patch
definitions, and the relationship of range parameters and integral scale values
with patch sizes was explored. Range parameter values did not equal and
were not strongly correlated with average patch sizes, though range parameter
values were more correlated with maximum patch and gap sizes. Integral
scale values matched the magnitude of, but were not strongly correlated with,
average patch sizes.
The third objective was to refine the analysis of temporally autocorrelated
hydrology data from paired watershed studies. Paired watershed studies are
used to evaluate forest harvesting effects on stream biota and hydrology (i.e.
fish, amphibians, insects, stream flow, and sediment yield). Traditionally,
treatment effects are discerned using prediction intervals. This work provided
an improved method for constructing prediction intervals for use in change
detection in paired watershed studies. The improved prediction intervals included
variation associated with estimating linear and autocorrelation model
parameters.
