I am with the Coastal Remote Sensing Program at the NOAA Coastal Services Center in Charleston, South Carolina, where I lead the Center’s land cover mapping activities, managing the Coastal Change Analysis Program (C-CAP).
Submitted by Nate Herold on July 30, 2012
We’ve gotten a number of questions in the past related to map accuracy, and specifically the two types of accuracy associated with the assessment of individual categories. So, I’ll attempt to explain (as accurately as I can):
Overall map accuracy is the number of classes correct/total number of accuracy assessment (AA) points. In the following example, 16 of the 21 total AA points are correctly classified (AA point number match the class number where they occur). This is reported as a 76% overall accuracy. This gives you a good overall idea of the map’s accuracy, but it may also be important to note that not all classes within the map are necessarily mapped as accurately as one another.
Caption: This is a “map” consisting of three classes (noted by number and the colored polygons). Example accuracy assessment point locations for each category are shown differentiated by shape and number.
Commission error refers to things that were incorrectly included in a category, as compared to the AA points (i.e., the map classified them as a class that they are not). The points in Box A above illustrate commission errors for class 2. This would be reported as having a user’s accuracy of 80% for class 2 (8 out of 10 points within that polygon are correct), meaning that the user should know that what was called class 2 in the map was correct 80% of the time.
Omission error refers to things that were missed, or left out of the classification, as compared to the AA points (i.e., the map should have classified them as one class, but called them another instead). The points in Box B above illustrate omission errors for class 2. This is reported as a producer’s accuracy of 73% for class 2 (8 out of 11 class 2 AA points are correct), meaning that the producer correctly classified 73% of the known accuracy assessment locations correctly.
Now, it should be noted that an omission error in one category will also be a commission in another. In this example, the three points of class 2 that were omitted (missed) from that map class area are also reflected as commission errors (incorrectly included) in class 3. Class 3’s user’s and producer’s accuracies would be 50% (3 out of 6 correct) and 75% (3 out of 4 correct), respectively.
Hope that helps. Let me know if you see an error.