I will admit that even though I’ve had a few courses on land use/cover classification and accuracy assessments during my Master and PhD, I always get confused about what user and producer accuracy are, and even more so what Kappa is.
In trying to assess the accuracy of my phenology based classification of cropland, I had to go back to re-learning how a confusion matrix is constructed. And my mind completely blacked out for a while before I got the hang of it. To help me understand it again, I found a few things that helped me:
This document from Yale – that describes the basics of an accuracy assessment and how to calculate overall accuracy, user accuracy and producer accuracy (but doesn’t mention Kappa, for some reason).
This pdf slideshow from my own department at Lund University – that gives a more visual description of the why’s and how’s of the accuracy assessment.
A confusion matrix online calculator – where I could enter my values and then see what the correct results should be, and then us them in my own excel spreadsheet.
I always go back to these documents and even though I’m still not 100% clear about how to calculate these things without double checking the documents first, the process of updating myself about it is much quicker now than before.
Maybe this short post can help other people who are also confused, if so let me know what you think!