I am trying to estimate the distribution of the under- and overestimation errors that I found in the outputs of a spectral mixture analysis (SMA), i.e. abundance fractional maps (AFM).
Specifically, AFMs are images composed of 10 bands (10 land use classes) whose pixels for each band show the fractional value [0,1] representing the abundance of that specific class in the specific pixel.
To estimate the error I have a "ground truth" that has identical dimensions to the AFM but that contains the real values of abundance on the ground.
The first step was to calculate the Bias Error, subtracting the AFM from the ground truth, with resulting values ranging from -1 (max underestimation) to 1 (max overestimation) for each pixel and class.
I would now like to calculate the overall distribution of these errors, for example: when a class is unerstimated by -0.2, to which other class/classes were assigned those 0.2 missing points? And so for all the under-overestimation errors of all the classes.
What is the most appropriate method to use in these cases?
P.S.: I work in Python
