Short URL

Results & Analysis

 

Other LULC Change in Carroll County areas:

Loading, please wait...

Chapter 5 - Results & Analysis

1999 Land Use & Land Cover Accuracy Assessment  |  1984 & 1999 Land Use & Land Cover Statistics  |  Quantitative Change Detection & Analysis  |  Land Assessment

1999 Land Use & Land Cover Map Accuracy Assessment

The general acceptance of the error matrix as the standard descriptive reporting tool for accuracy assessment of remotely sensed data has significantly improved the use of such data. An error matrix is a square array of numbers organized in rows and columns which express the number of sample units (i.e. pixels and clusters of pixels) assigned to a particular category relative to the actual category as indicated by reference data (Congalton, 1996).

An initial error matrix was generated for the initial supervised 1999 land use and land cover map prior to the mode filtering by creating a maximum likelihood report. (Table 5.1) This report calculates area and percentages of each land class incorporated in a maximum likelihood classification. The ground truth data were utilized in the maximum likelihood report as the independent data set from which the classification accuracy was compared. The accuracy is essentially a measure of how many ground truth pixels were classified correctly.

Table 5.1: Initial Maximum Likelihood Report

An average accuracy of 84.9% and an overall accuracy of 81.1% was achieved with a Kappa coefficient of 0.70638. The average accuracy is the average of the accuracies for each class, and the overall accuracy is a similar average with the accuracy of each class weighted by the proportion of test samples for that class in the total training or testing sets. Thus, the overall accuracy is a more accurate estimate of accuracy.

The importance and power of the Kappa analysis is that it is possible to test if a land use and land cover map is significantly better than if the map had been generated by randomly assigning labels to areas (Congalton, 1996). It is widely used because all elements in the classification error matrix, and not just the main diagonal, contribute to its calculation and because it compensates for change agreement (Rosenfield & Fitzpatrick-Lins, 1986). The Kappa coefficient represents the proportion of agreement obtained after removing the proportion of agreement that could be expected to occur by chance (Foody, 1992).

This implies that the Kappa value of 0.70638 represents a probable 71% better accuracy than if the classification resulted from a random, unsupervised, assignment instead of the employed maximum likelihood classification.

The Kappa coefficient lies typically on a scale between 0 and 1, where the latter indicates complete agreement, and is often multiplied by 100 to give a percentage measure of classification accuracy. Kappa values are also characterized into 3 groupings: a value greater than 0.80 (80%) represents strong agreement, a value between 0.40 and 0.80 (40 to 80%) represents moderate agreement, and a value below 0.40 (40%) represents poor agreement (Congalton, 1996).

This initial accuracy assessment was somewhat lower than expected and deemed not acceptable. A widely used, acceptable accuracy is 85%, which is striven for in the land use classification adopted by the U.S. Geological Survey. However, it should be noted that the level of accuracy sought and obtained in remote sensing projects involving per-pixel classification can be an arbitrary measure dependent on the level of classification employed, the scale of the area considered in the study as well as the spatial resolution of the imagery utilized in the analysis.

After applying a mode filter on the barren land class, an additional error matrix was generated with improved results. (Table 5.2) The average accuracy reported was 88.59%, an increase of 3.69%, and the overall accuracy reported was 84.50%, an increase of 3.4%, with a Kappa coefficient of 0.75363, an increase of 4.7%.

Table 5.2: Post-Filtering Maximum Likelihood Report

Although suitable for many classification standards at overall accuracy of 84.5%, and a Kappa coefficient of 0.75363 or 75%, the potential to achieve higher accuracy was investigated further. After performing the additional unsupervised ISOCLUS classification upon the barren land use category, an additional, yet small increase in accuracy was achieved. Both a 5 by 5 mode and sieve filter were also applied to the entire reclassed image to remove any pixel clusters less than the minimum map unit of 1 acre, approximately 5 pixels. The final accuracy achieved for the 1999 land use and land cover map for Carroll County, Arkansas had an average accuracy of 87.95%, an overall accuracy of 84.75%, and a Kappa coefficient of 0.75577 or 76%. (Table 5.3)

Table 5.3: Final Maximum Likelihood Report

The final accuracy assessment showed an additional, yet slight increase in the overall accuracy of 0.25% as well as a slight increase the kappa coefficient of 0.00214 which allowed for rounding (1%) from the previous accuracy assessment. This final accuracy assessment of the 1999 land use and land cover map was deemed satisfactory.

 

Chapter 6 - Summary & Conclusions