home

<- Previous | First | Next ->

Expert Review for Land Cover: Integrating Information from Specific Comments and Evaluating the Results

Geoffrey M. Henebry¹, James W. Merchant¹, Justin W. Fischer¹, and Douglas Garrison²

¹Center for Advanced Land Management Information Technologies, University of Nebraska, Lincoln

²USDA Natural Resources Conservation Service, Lincoln, Nebraska

Introduction

The Landsat TM data used to generate the base land cover map for Nebraska were acquired nearly a decade ago in the early 1990s. This situation poses a challenge for evaluation of the accuracy of the land cover product. Land cover changes subsequent to image acquisition result in temporal decorrelation that is interpreted as classification error in a formal accuracy assessment, if the ground data for the accuracy assessment are not collected concurrently with image acquisition. As we did not have concurrent ground data, we sought a procedure to not only evaluate the accuracy of our draft land cover map but also to increase its relevance to a potential user community.

To solicit expert assessment of the draft land cover map, the Nebraska Gap Analysis Project and the Nebraska State Office of the Natural Resources Conservation Service (NRCS) sent out relevant county-level maps to district conservationists at NRCS offices during late 1999 and early 2000. The district conservationists coordinated review of the hard-copy maps, utilizing staff from 81 NRCS offices statewide. Local experts reviewed the draft maps and identified misclassifications by annotating the hard-copy map with a series of general and specific comments. As of December 15, 2000, 75 of 93 county maps had been returned, a response rate of over 80%.

Methods

Of the 75 maps returned, 65 were annotated with specific comments and 10 indicated no change. General and specific comments were recorded from each map. Specific comments, defined as comments noting misclassification of particular groups of pixels, were then tabulated into a special confusion matrix reporting only misclassification errors; thus, all elements of the matrix were located off the principal diagonal (Table 1).

Table 1: Special confusion matrix based on original draft land cover map. Matrix elements corresponds to total numbers of comments made by reviewers. No elements occur on the diagonal.