HR: 1330h
AN: H22B-0933    [PDF]
TI: The effect of land-cover misclassification on predicted impervious surface
AU: * McMahon, G
EM: gmcmahon@usgs.gov
AF: U.S. Geological Survey, 3916 Sunset Ridge Road, Raleigh, NC 27607 United States
AB: If land-cover data are available for a study area, a coefficient-based impervious surface modeling approach is relatively straightforward and quick to implement. Impervious-surface estimates developed by using this approach, however, may be biased or imprecise for several reasons, including land-cover misclassification. Data collected as part of a U.S. Geological Survey's National Water-Quality Assessment (NAWQA) Program study in New England were used to assess whether impervious-surface model predictions were improved by explicit adjustment of National Land Cover Data (NLCD) land-cover values to account for misclassification. Adjustments were based on a user probability matrix derived from published accuracy-assessment information. NLCD and field-measured impervious-surface values at 55 New England study sites were used to assess models for predicting impervious surface. Four models were distinguished by land-cover resolution (Level I and Level II) and whether land-cover data were adjusted. The relatively poor classification accuracy of the NLCD developed-land class (74% for Level I developed land and 40-61% for Level II developed categories) generally results in adjustments that decrease the amount of developed land. These downward adjustments begin when relatively low amounts of developed land are present in a study site. Because developed-land impervious-surface coefficients have a high value, decreases in the amount of developed land will lower predicted impervious-surface values relative to those in models using unadjusted NLCD data. The results of the investigation are mixed as to whether impervious-surface models that account for land classification error rates result in meaningful improvements in impervious-surface predictions over models that do not make these modifications. The model-verification results suggest that NLCD Level I land-cover data, adjusted for land-cover misclassification, is preferable to other land-cover options for use in models predicting impervious surface. There was no significant difference between paired observations of observed and predicted impervious-surface values when the adjusted Level I data were used. Significant differences between paired observed and predicted impervious-surface values occurred in the other three models. Overall, the sensitivity of the models to land-cover classification errors and adjustments for these errors was small; the difference between predicted impervious surface from a model using unadjusted land-cover and simulated values from an adjusted model ranged from 1-2%.
DE: 1640 Remote sensing
DE: 1899 General or miscellaneous
SC: Hydrology [H]
MN: 2003 Fall Meeting