You are here:
ASSESSING THE PREDICTIVE CAPABILITY OF LANDSCAPE SAMPLING UNITS OF VARYING SCALE IN THE ANALYSIS OF ESTUARINE CONDITION
Citation:
Hollister, J W., J Copeland, P. August, AND J F. Paul. ASSESSING THE PREDICTIVE CAPABILITY OF LANDSCAPE SAMPLING UNITS OF VARYING SCALE IN THE ANALYSIS OF ESTUARINE CONDITION. Presented at International Association for Landscape Ecology Conference, Banff, Alberta, Canada, April 2-6, 2003.
Description:
Landscape structure metrics are often used to predict water and sediment quality of lakes, streams, and estuaries; however, the sampling units used to generate the landscape metrics are often at an irrelevant spatial scale. They are either too large (i.e., an entire watershed) or somewhat arbitrary (i.e., 10 km buffer around a sampling station) and can result in poor predictive capability. To investigate this, we generated watersheds, Euclidean buffers clipped to the watershed, and hydrologically defined buffers for selected sampling stations from the Environmental Protection Agency's Mid-Atlantic Integrated Assessment. The hydrologically defined buffers were generated from flow length calculations derived from the National Elevation Dataset. For each sampling unit we calculated percent urban from the National Land Cover Dataset and used this to predict levels of zinc and copper in estuarine sediments (i.e., ZINC = ?1PERCENT_URBAN + ?0). We withheld a portion of the sampling stations and calculated the Root Mean Square Error (RMSE) between predicted and actual sediment contamination values. Full watersheds had an RMSE nearly three times greater for zinc and nearly two times greater for copper than either the Euclidean or hydrologically defined buffers. This suggests that using full watersheds as a landscape sampling unit may result in poor predictions of metals concentrations in estuarine sediments. Furthermore, little difference was seen between the predictions of the models using Euclidean and hydrologically defined buffers. The extra processing steps and data needs to compute the hydrologically defined buffers may be unnecessary since they may not provide significant increases in predictive capability.