TITLE:
Accuracy of Stream Habitat Interpolations Across Spatial Scales
AUTHORS:
Kenneth R. Sheehan, Stuart A. Welsh
KEYWORDS:
Natural Neighbor Interpolation; Residuals; Ordinary Least Squares; Stream Modeling; Habitat; Benthic Substrate
JOURNAL NAME:
Journal of Geographic Information System,
Vol.5 No.6,
December
27,
2013
ABSTRACT: Stream
habitat data are often collected across spatial scales because relationships
among habitat, species occurrence, and management plans are linked at multiple
spatial scales. Unfortunately, scale is often a factor limiting insight gained
from spatial analysis of stream habitat data. Considerable cost is often
expended to collect data at several spatial scales to provide accurate
evaluation of spatial relationships in streams. To address utility of single
scale set of stream habitat data used at varying scales, we examined the
influence that data scaling had on accuracy of natural neighbor predictions of
depth, flow, and benthic substrate. To achieve this goal, we measured two
streams at gridded resolution of 0.33 × 0.33 meter cell size over a combined
area of 934 m2 to create a baseline for natural neighbor
interpolated maps at 12 incremental scales ranging from a raster cell size of
0.11 m2 to 16 m2. Analysis of predictive maps showed a
logarithmic linear decay pattern in RMSE values in interpolation accuracy for
variables as resolution of data used to interpolate study areas became coarser.
Proportional accuracy of interpolated models (r2) decreased, but it was
maintained up to 78% as interpolation scale moved from 0.11 m2 to 16
m2. Results indicated that accuracy retention was suitable for assessment
and management purposes at various scales different from the data collection
scale. Our study is relevant to spatial modeling, fish habitat assessment, and
stream habitat management because it highlights the potential of using a single
dataset to fulfill analysis needs rather than investing considerable cost to
develop several scaled datasets