Scattered Interpolation: Defining my own triangulation?

I'm interpolating data the comes from polar coordiate systems or terrain-following coordiante system used in numerical weather models. While that data isn't gridded in the since that it's not grid aligned, it is structured.

Scattered interpolation is slow. My understanding is that this stems from computing the delaunay triangulation. For my data it would be trivial to provide the triangulation connectivity graph along with the points.

Is there a way I can provide a pre computed triangulation to scatteredInterpolant or griddata in order to speed up my scattered interpolations? Is there some hidden class property I could fill in somewhere?

[KSSV]

ScatteredInteprolant do not demand to provide triangulation connectivity. You need to provide only the (x,y,z) points.

[millercommamatt]

I know that.

I want to provide the triangulation because it's trivial for me to calculate because I know how the data are structured. I want to avoid what makes the scattered interpolation slow.

[Alexey Gorbunov]

I have the same problem. I need to interpolate scattered data with predefined triangulation.

In the previous MatLab versions, it was possible using TriScatteredInterp function.

% T - triangular endpoint indexes predefined;
% x,y - data coordinates;
% V - data values;

TR = triangulation(T, [x(:), y(:)]);     % creating triangulation
C = edges(TR);     % finding edges of the triangulation
DT = DelaunayTri(x(:), y(:), C);     % Delaunay with predefined edges
F_ne = TriScatteredInterp(DT, V(:));     % Interpolant

I have not stil find any other solution for the task...

[John Kearns]

Hey Matt,

Did you manage to find a solution to this? This idea has recently become relevant in my own work, and I would be grateful if you were able to share an successes you've had.

Cheers,

John

[millercommamatt]

I don't have any new information to share on this front.

In my case my data had a grid-like connectivity pattern so I transformed my data into a virtural coordinate system that I treated as gridded and used the much faster gridded interpolation and then converted back to my native coordinate system. It's a dirty hack that would get you in a lot of trouble unless you knew exactly what the assumptions, implications, and impacts were.

Answer 1

[Noah Prisament]

Hi, I work at MathWorks and I wanted to let you all know that we have forwarded this feedback to the relevant team. We will consider this request and it may be included in a future release.