# How can I smooth a discrete raster in ArcGIS?

I started with a continuous spline raster of hotspots, classified by standard deviations from the mean. This displays as 7 classes, but since it's a continuous raster in ArcGIS 10.2, it appears nice and smooth. However, my goal is to take the splined continuous raster that looks nice and smooth, and make it appear 3D using cartographic representations. That means I needed to convert this spline to vector.

Now, this next part is where I'm not sure about my approach, is there are better way to do this? What I did was I ran the raster calculator on my splined continuous raster for each class. I did calculations like where spline_raster >=2.58, for example. I want to do this for hotspots and coldspots, so I have 6 classes I need to extract through raster calculator. I did this, and raster calculator converted the output into discrete rasters. So where my spline was >= 2.58, it was now 1, and everything else = 0. The issue wasn't so much this, but that the output looked much more "stairstepped" or pixelated compared to my continuous spline. I understand the difference between discrete and continuous, but I'm wondering if there's a best practice for at least making the discrete raster's pixels so much smaller that at the same scale as the continuous spline, they appear almost as smooth? I tried going into the environment settings and decreasing the cell size from the original 1414 to 5, and it hardly changed how the raster calculator output looked. I don't understand this part very well. I researched this a little, and figured I would try resampling my rasters. I tried resampling using nearest neighbor, bilinear, and cubic...they all still look the same. My conclusion is that there is some part of this process I don't quite understand.

Since I would then be vectorizing the raster calculator output (using Raster to Polygon tool), the vectorized polygons look all jagged because of the stairstepping, and I have been using the Smooth Polygon tool to make them look smoother. I feel this approach is very roundabout and not the best way to do it, but I don't know a better way. Because of my field of research and the types of data I've used, I don't typically do a lot of raster analysis.

• The reason that you aren't seeing any improvement in your edges when you resample is because your class breaks (and the cells on which they're based) are still derived from your 1414 cell size (now there are just more cells comprising the boundary between classes). You may have to compromise between data integrity and display purposes here - why is your cell size so large to begin with? Is it a processing issue, data spacing issue? If you absolutely have to use vector data for your end product, I think your options are either smoothing the polys or splining at a higher resolution – JWallace Dec 16 '15 at 19:22
• To expand on the data integrity point - either using smoothed polygons or artificially increasing your cell resolution if your known data points doesn't support it, would be some sacrifice of data integrity for display purposes (which is fine in some cases!). Have you though about turning you spline into a hillshade and then draping your continuous raster over it with a partial transparency to achieve a 3D/extrusion effect? – JWallace Dec 16 '15 at 19:25
• Thank you! Your responses are helpful. First off, I think I'll give reducing the spline cell size a shot, it's something I never considered since the default output looks smooth. But definitely a good call. As for the hillshade idea for 3D, I had not considered that at all! I work in public health, and we hardly do things on the raster side of GIS. Any pointers with hillshade that I can't get out of just reading the ArcGIS resources info? – rachel.passer Dec 16 '15 at 19:41