Hey everyone,
Regarding unwanted terraces and steps in heightmap surface, read here (demo scenes available in package): TerraLand 2 – High Quality Photo-Realistic Terrains from Real-World GIS Data page-5#post-2878638
What you need to consider first in order to download heightmap for the area is that if the resolution is too much while there is no data in the original server, ESRI resamples heightmap and then outputs the render for TL and also TL sometimes resample again to match with Unity’s proper terrain resolution. So the more the resolution of heightmap is in comparison to the original data, the more resampling and unwanted jaggies and terrace banding artifacts will occur.
The average resolution of elevation data on ESRI servers throughout the entire Earth surface is approx. 10 meters per pixel, so with this in mind, if you have for example a 5x5 km2 area, the best optimal resolution for the heightmap would be 512x512 pixels => 5000 / 10 = 500 (512 as PO2 value).
With this simple math before setting the heightmap resolution, you will get the most optimal resolution for the surface without those terracing artifacts and without needing the extra Smooth operations resulting less extra pixels/polygons and subsequently better performance.
The same happens for imagery when the total satellite image resolution is too high for a small specified area and you have to consider setting the optimal resolution for that based on the coverage resolution.