I am doing the geodetic mass balance of a glacier and I have the elevation difference that was produced in the glacier, but now I have to correct this data but making the mean of the elevation difference in the rock outside the glacier, zero. The difficulty is to know what is rock, because it can get confused with a rocky glacier, glacier, and where there was, for example, a collapse that greatly affects the data. I need only the rock, that is my valid cell to take the mean and the substract this mean to all the dem of glacier difference.
So I extracted everything that was outside the glacier in the DEM that is the elevation difference of the intersection of two time periods, and I supose everything here is rock, so in this case for example, rock measured in the year 2009 and the rock measured in year 2015 should be the same and the difference should be zero. But in the data there is error between the two measurements or the area I am evaluating, and the values range between -28 m and 25 m. So I need to delete this points that are not valid cells to get the mean that represent the error in valid cells of rock, and that has a low standard deviation, so I need to define a threshold to do so.
I want to know a statistical method to define this threshold or outliers that give me certainty that what I am deleting is not statistically significant. I remember that I did this in university but for the waves in the ocean and we did it with tests: gumbel, rayleigh and frechet. But I wish to do it on Qgis before looking a way that make me export more than 4 million points to R.