Fig. 5.

Scatter plots of the predicted probabilities of good stream condition from random forest RF) models with full versus reduced variable sets. Since there are a very large number of prediction sites within each ecoregion, points are binned in the scatter plots. The black line in each panel is the 1-1 line. The prediction sites for the RF models are all 1.1 million catchments within the sampling frame for the 2008/2009 National Rivers and Streams Assessment. Ecoregion codes: Coastal Plains CPL), Northern Appalachians (NAP), Northern Plains (NPL), Southern Appalachians (SAP), Southern Plains (SPL), Temperate Plains (TPL), Upper Midwest (UMW), Western Mountains WMT), and Xeric (XER)