Interpolating FROM points TO points (i.e. not to a new raster)?

[AndrewFord]

Imagine I have two sets of XY points, lets call them A and B, roughly coincident in space. Points A also have an attribute, lets call it N. I’d like to estimate (i.e. interpolate) the same attribute N for points B. I could simply “copy across” the value from the nearest corresponding points in A and B, but this is no good for the use case I have in mind. Instead I’d like to at least apply some summary statistic (e.g. mean), or better still something like IDW. So far all tools I can think of in Spatial Analyst take point attributes and interpolate them to a new raster (i.e. continuous surface, which I do not want), rather than interpolating to existing points. Does anyone know of any way to accomplish the latter please?

[DanPatterson]

If the points are roughly coincident in space, on what basis is interpolation needed?

What is N a function of?  The small differences in location?

[AndrewFord]

Many thanks.

When I say coincident I mean in terms of approximately the same extent or coverage, not the same actual locations.

[SteveLynch]

What about

IDW to raster followed by ExtractValuesToPoints

[AndrewFord]

Many thanks.

Indeed I had thought of just such a work-around. However, the result will be sub-optimal, as the approximation (interpolated value) is assigned to a regular grid/array of pixels rather than discrete, randomly distributed points. Indeed creating a continuous surface (i.e. raster) is overkill in the first place, as it is doing waaaaay more computations for a much larger series of pixels compared to a  [relatively] small number of discrete points.

[SteveLynch]

In that case you could use IDW in Geostatistical Analyst to create a ga layer and then use GALayerToPoints which will use the interpolation parameters in the ga layer to predict a value at exactly the given locations. The ga layer, does no have a cell size, and is held in memory. You could also then do cross validation on your method.