Exploration of Stream Habitat Spatial Modeling; Using Geographically Weighted Regression, Ordinary Least Squares Regression, and Natural Neighbor Interpolation to Model Depth, Flow, and Benthic Substrate in Streams
Date of Graduation
Davis College of Agriculture, Natural Resources and Design
Wildlife and Fisheries Resources
Stuart A. Welsh.
Assessment and modeling of stream habitat are integral to understanding streams and the biota within them. In the past several decades, assessment sophistication of ecologic systems increased due to analysis power afforded by gains in computing capability. Spatial data analysis methodology grew alongside computing power and incorporated spatial qualities of ecological data, thereby providing new insights. New methods like geographically weighted regression (GWR) and more established methods like interpolation are now being used in ecological studies to guide assessments and management decisions. However, their accuracy and utility for analysis of stream habitat data have not been fully explored. To clarify their impacts on stream habitat data, the five chapters of this dissertation examined spatial qualities (e.g. heterogeneity, scale, sample pattern) and the use of interpolation and GWR on depth, flow velocity, and benthic substrate.;Benthic substrate, depth, and flow velocity data were collected from four streams between July 2005 and August 2010. Data were collected from Aarons Creek, Monongalia County, WV, Elk River, Kanawha County, WV, Little Wapiti and Grayling creeks in Gallatin County, MT. Using GIS, the datasets were mapped, modeled, and analyzed between fall 2009 and summer 2011.;Results from our studies demonstrated GWR outperformed non-spatial ordinary least squares regression (OLS) when modeling benthic substrate. Our study showed stream data collected at a single scale may be used to generate meaningful results at scales other than that at which it was collected. This finding is important for stream habitat studies where data are often collected at varying spatial scales. As spatial heterogeneity of benthic substrate increased, accuracy levels of models decreased showing heterogeneity must be quantified in analysis of stream habitat variables. Large (>20m width) and small (<10m width) wadeable streams may be analyzed using the same type of spatial analysis though substrate deposition pattern may vary in different size streams. Benthic substrate depositional pattern was most effectively captured by non-random point selection which created more accurate maps than grid and random point sample methods.;Combined results demonstrated the need to address spatial qualities of stream habitat data in analysis, assessment, and how spatial attributes may guide data collection. Further, failure to quantify spatial attributes in stream habitat data can cause erroneous results and thus minimize effectiveness for useful ecologic conclusions and management decisions.
Sheehan, Kenneth R., "Exploration of Stream Habitat Spatial Modeling; Using Geographically Weighted Regression, Ordinary Least Squares Regression, and Natural Neighbor Interpolation to Model Depth, Flow, and Benthic Substrate in Streams" (2011). Graduate Theses, Dissertations, and Problem Reports. 3482.