Interpolation service

From Intamap

Interpolation service

The interpolation service will consist of the following parts:

Setup interpolation service

This is partly done manually, or can be seen as a part of the pre-processing step. Obsolete? Maybe necessary for the optimisation procedure. There is function checkSetup that can be seen as part of the setup, although that one is useful to call before the pre-processing step.

I1 Identification of clusters

We will most likely split the data set into clusters to be able to separate between the radiation cloud and the surrounding area for emergency situations. The procedure will add a row to the observations SPDF with name group (pointData$group) where the different clusters are identified by different numbers.

This procedure is not yet implemented

I2 Calculation of anisotropy parameters

This procedure is responsible for finding the anisotropy parameters of the random field. The output will be given as a DataFrame with different anisotropy parameters for each cluster. The name of the DataFrame is anisPar, and the parameters are anisPar$direction and anisPar$ratio

This procedure is not yet implemented

I3-I4 Different parameter estimation and kriging procedures

Different instances of these two procedures will be implemented by the different groups. They should all have a name of the type:

• estimateParameters.method
• spatialPredict.method

where method refers to one of the interpolation methods, defined by an object's class.

The input will be the variables and parameters described above. The output will be filled included in the interpolation object created in the pre-processing step under the name "predictions". The format of predictions depends on how to describe the posterior predictive distributions, in addition to coordinates and predictions. The aim is to give distributions as mean and variance, as quantiles or as parameters of a distribution function, probably according to UncertML.

As there are several different classes of output locations (grids, pixels, points, lines, polygons) it will be necessary to implement instances of the estimation and kriging procedures for different classes. However, in the beginning, we will focus on spatialPoints.

The test "kriging.R" is partly based on Pauls automap functions for estimation of variogram parameters and uses gstat directly for interpolation. The functions are given in "R/pointKriging.R".

I3 Smoothing over boundaries of clusters

This function has not yet been implemented.

I4 Spatial aggregation

Aggregation to larger spatial units. This function has not yet been implemented.

I5 Aggregation over simulated realisations

Aggregating simulated realisations into averages, pdfs, probability over thresholds, etc. This function has not yet been implemented.

Network optimisation

This function should be a part of the interpolation service. It is not yet implemented, and it has not been decided how to call or use this function.