Lab 04, Classification accuracy assessment

Background and Goal:

    The goal of this lab was to gain experience and knowledge in the evaluation of classification results through accuracy assessment. Accuracy assessment is a necessary process that needs to occur in able for the classification to be used in any application. 

Methods:

        The accuracy assessment was done by adding random points to the high resolution image of Eau Claire, and then manually classifying those points. These points will then be checked against the classified image. The tool for creating those points is shown below. 

Add Random Points Tool

    

        The points were added using the stratified random parameters. Once they were added, they resulted in the below image.

Points on the Map

    After that, the accuracy assessment is ran, producing data that is used to fill out the error matrix.

Results:

    

        To get a correct accuracy assessment I had to add 51 to each value that I entered into the point file. The reason for this was because I had attempted to recode the file, but it didn't work. Instead of being rows 1 through 5, my classes were rows 52-56. Because of this it was necessary to add 51 to each value I entered. The accuracy assessment ran just fine after that.

The Error Matrix, populated with data from the accuracy assessment

         The accuracy assessment showed that my classified image needs some work. Ideally, the accuracy would be closer to 80%, and an accuracy of 47.20% is inappropriate for use in almost any application. In addition to that, the Kappa statistic is under 0.4, so there is poor agreement/accuracy.

        My next decision would be to re-do the entire image classification. I would like to be more careful in selecting training samples, especially those of the Agriculture, Urban, and Bare soil areas. I believe I could achieve this with a more recent aerial photo to use as a reference image.

Sources:

Earth Resources Observation and Science Center. (2014).[Landsat image of Eau Claire                                         and surrounding areas]. United States Geological Survey.  Provided by Cyril Wilson.

United States Department of Agriculture (USDA). (2014).[High Resolution image of Eau Claire and                 Surrounding Areas]. National Agriculture Imagery Program. Provided by Cyril Wilson.