DIVAnd (Data interpolation variational analysis in n-dimension): automatic choice of approximation method for the error field.
Supervisor: Jean-Marie Beckers (U Liege)
DIVAnd (Data-Interpolating Variational Analysis in n dimensions) is a tool allowing to interpolate data’s on a n-dimensional gridded field. It uses the principle of inverse variational method combined with optimal interpolation. The key point is that there are some equivalences between the two methods. Once an analysis is performed, one might want to get an error field (analysis error variance) associated with it. In order to derive it, we use the equivalence relation between optimal interpolation and variational analysis. It can be calculated “exactly” but it is very costly and some approximation methods are applied in order to gain some computational time. In the mathematical description of DIVAnd, some parameters are defined (signal-noice ratio, correlation length). The goal of this work is to generate some fictional case (1D, 2D, different distribution) and see how the different approximation methods behave for different values for the parameters (signal-noise ratio, correlation length, number of data’s). Once it is done, we proceed to create a decisional tree that allows the program to choose the “best” approximation method taking into the computation time and the error field generated.