Zonal Statistics is a great tool! But I recently discovered it's not using the most robust method for calculating percentiles, and would like to see it get that update/upgrade so the results are consistently accurate (or enable us to choose that).
I recently used the Zonal Statistics to Table GP tool to try calculating the global 80th percentile of a raster (high spatial resolution, large variety of cell values). It gave a resulting value, which I used to reclassify the raster to create two values: above and below this threshold value. Unfortunately, when visualizing it my colleague pointed out that it wasn't the 80th percentile (i.e. top 20%), but looked closer to 25%-30%. Running the analysis in R provided the correct result that we'd expected from running the Zonal Statistics tool in ArcGIS Pro.
An awesome Esri Support Analyst reran the tool on this raster with various percentiles, and it took using 85th percentile as a parameter to get close to the 80th percentile result we wanted, expected, and were able to produce in R.
He dug into the tool's documentation for calculating percentiles and the underlying Hyndman and Fan (1996) paper, and explained that Zonal Statistics uses one method, while R's default is Type 7, a more robust method.
I'd like to suggest the Zonal Statistics tools be modified, perhaps by adopting this more robust method or at least enabling the selection of it, so that we can get correct results regardless of a raster's spatial resolution and number of unique values.
