Expanding on the comment left earlier, the docs for Zonal Statistics say not all statistic types are calculated for floating point rasters, for some reason the statistic type median cannot be calculated for floating point rasters... but let's go a little deeper into why not every statistic is offered:
- Majority/Minority: almost all cells in floating point rasters are
different values (to the nth decimal place) so statistics of 'the
most frequently appearing' and 'the least frequently appearing' are
kind of meaningless as every value is most and least appearing.
- Variety: again, as every value is different the entire raster will
have the same value in the output indicating the count of cells.
- Median: As every cell is different there is (possibly) no statistical
median/mode value in the frequency distribution.
To get around this you can convert your raster into an int type, this removes the decimal place, there are three tools that do roughly the same job: Int, Round Down and Round Up. Note Int does exactly the same thing as Round Down: truncate (remove anything after the decimal place).
Make a decision on how many decimal places are important (or significant) then multiply by a 10 based factor:
- For 3 decimal places multiply by 1000.
- For 1 decimal place multiply by 10
- For 6 decimal places multiply by 1000000.
- Not 10 based will give unpredictable results, but works just the same
as long as you keep track of the number to unscale by later!
The maximum range for your integer is the same as a 32bit integer : -2,147,483,648 to 2,147,483,647 (see System.Int32), so long as your multiplied value Perform your integer function, either Up or Down, now all statistic types are available.
Now that you have your statistic you need to divide it by the same value as the original scale factor (10 or non-10 based)... but some statistics are meaningless when divided, so be careful (variety and range can't be divided).
Irrespective you will face the same problems with too many significant digits as if the raster were float values; you will need to make a decision on how to int the data and manage the scaling to achieve meaningful results.