ABSTRACT
The 14,582 km2 Neuse River Basin in North Carolina was characterized based on a user-defined land-cover (LC) classification system developed specifically to support spatially explicit, non-point source nitrogen allocation modeling studies. Data processing incorporated both spectral and GIS rule-based analytical techniques using multiple-date SPOT 4 (XS), Landsat 7(ETM+), and ancillary data sources. Unique LC classification elements included the identification of urban classes based on impervious surfaces and specific row crop type identifications. Individual pixels were aggregated to produce variable minimum mapping units or landscape “patches” corresponding to both riparian buffer zones (0.1 ha), and general watershed areas (0.4 ha). An accuracy assessment was performed using reference data derived from in situ field measurements and required to support these spatially explicit modeling approaches, imagery (camera) data. Multiple data interpretations were used to develop a reference database with known data variability to support a quantitative accuracy assessment of LC classification results. Confusion matrices were constructed to incorporate the variability of the reference data directly in the accuracy assessment process. Accuracies were reported for hierarchal classification levels with overall Level 1 classification accuracy of 82% (n = 825) for general watershed areas, and 73% (n = 391) for riparian buffer zone locations. A Kappa Test Z statistic of 3.3 indicated a significant difference between the two results. Classes that performed poorly were largely associated with the confusion of herbaceous classes in both urban and agricultural areas.
