> We use another disc(same size with the plate) with an array of
> temperature sensors to measure the temperature of the target plate and
> there are no sensors at the very edge of the disc. So it is very
> likely that we can not solve the problem by measuring the edge region
> of the target plate. Maybe I can try setting the NaNs to be weighted
> average of several adjacent points, but this will use more CPU time.
> Do you have any suggestion?
Because the points on the edge of the disk are beyond any
data points, estimatng their value will require some
extrapolation as opposed to pure interpolation. This can
be a bilinear weighting of nearby data values, but some of
the weights may be negative. In a simple linear case with
equally spaced points A - B - C and data at A and B, to
estimate the value at C we could use:
value(C) = 2 * value(B) - value(A)
In a 2-D case of square grid points:
A B C
D E F
G H I
we can estimate the value at I by:
v(I) = 4 * v(E) - v(A) - v(C) - v(G)
These are the simplest estimates.