Evaluating errors in a digital vegetation map with forest inventory data and accuracy assessment using fuzzy sets

In large-area mapping projects, existing reference data, often collected for a different purpose, are increasingly being used for map accuracy assessment. Multi-attribute digital vegetation maps have been developed for all National Forest lands in California (8.1 million ha). We developed decision rules that could be applied to quantitative Forest Inventory and Analysis (FIA) plot data in order to score the fuzzy membership of plot locations in all possible map classes. We compare the accuracy of the vegetation map attributes estimated using this method to accuracy estimated from fuzzy class membership scores assigned by experts (inventory crews) during field work. Accuracy of the vegetation life form attribute was estimated to be higher when expert label assignments were used as reference data (76-87%), instead of FIA plot data (62-79%). This suggests that automated decision rules applied to detailed data from FIA plots, which have smaller area than map polygons, may systematically underestimate map accuracy. However, assignment of the actual map labels to FIA plot locations by inventory crews appears to be a robust method for using the FIA data for accuracy assessment.