Compare accuracy of land class rasters

[GISCoffee]

For a satellite image, I have a set of hand-drawn land class labels drawn by citizen scientists and a few sample areas (a few 20 * 20 m grid cells) hand-labelled by an expert. I want to compare how similar the citizen science land class labels are to a satellite image labelled by an expert. 

I've already looked at the Create Accuracy Assessment Points tool and this seems like it would require me to label the newly generated points with their "ground truth value". I wonder if it's possible to use the my existing expert label shapefile/raster as the ground truth data set to avoid having to manually label hundreds of points. 

Data: 

A fully hand-labelled satellite image by Citizen scientists (available as shapefile & raster)
Sample fishgrid cells which have been labelled by an expert (available as shapefile & raster)

[ThomasHamill]

@GISCoffee, sounds like you want to perform a crosstabulation with the two rasters.  I recommend the Tabulate Area tool.

This post might also be helpful.

[ThomasHamill]

@GISCoffee, did you have a chance to try the Tabulate Area tool?

If you need some additional background on the utility and value of this tool for your accuracy assessment scenario, I can recommend R. Pontius' research on using cross-tabulation matrices to compare raster-based land classifications (of the same area and extent) for two (or more) different time periods or classification models.

A generalized cross-tabulation matrix to compare soft-classified maps at multiple resolutions

Death to Kappa: birth of quantity disagreement and allocation disagreement for accuracy assessment