Kriging

A Kriging estimator has the form:

\[\hat{Z}(\x_0) = \lambda_1 Z(\x_1) + \lambda_2 Z(\x_2) + \cdots + \lambda_n Z(\x_n),\quad \x_i \in \R^m, \lambda_i \in \R\]

with $Z\colon \R^m \times \Omega \to \R$ a random field.

This package implements the following Kriging variants:

• Simple Kriging
• Ordinary Kriging
• Universal Kriging
• External Drift Kriging

All these variants follow the same interface: an estimator object is first created with a given set of parameters, it is then combined with the data to obtain predictions at new locations.

The fit function takes care of building the Kriging system and factorizing the LHS with an appropriate decomposition (e.g. Cholesky, LU):

fit(estimator, data)

and the predict function performs the estimation for a given variable and location:

predict(estimator, var, uₒ)

Alternative constructors are provided for convenience that will immediately fit the Kriging parameters to the data. In this case, the data is passed as the first argument.

For advanced users, the Kriging weights and Lagrange multipliers at a given location can be accessed with the weights method. This method returns a KrigingWeights object containing a field λ for the weights and a field ν for the Lagrange multipliers:

weights(estimator, uₒ)

These weights can be combined with a vector of observations using the combine function:

combine(estimator, weights, z)

This functionality can be useful in real-time applications when the locations of the observations are fixed and an online stream of data is available.

Finally, all estimators work with general Hilbert spaces, meaning that one can interpolate any data type that implements scalar multiplication, vector addition and inner product.

Simple Kriging

In Simple Kriging, the mean $\mu$ of the random field is assumed to be constant and known. The resulting linear system is:

\[\begin{bmatrix} cov(\x_1,\x_1) & cov(\x_1,\x_2) & \cdots & cov(\x_1,\x_n) \\ cov(\x_2,\x_1) & cov(\x_2,\x_2) & \cdots & cov(\x_2,\x_n) \\ \vdots & \vdots & \ddots & \vdots \\ cov(\x_n,\x_1) & cov(\x_n,\x_2) & \cdots & cov(\x_n,\x_n) \end{bmatrix} \begin{bmatrix} \lambda_1 \\ \lambda_2 \\ \vdots \\ \lambda_n \end{bmatrix} = \begin{bmatrix} cov(\x_1,\x_0) \\ cov(\x_2,\x_0) \\ \vdots \\ cov(\x_n,\x_0) \end{bmatrix}\]

or in matricial form $\C\l = \c$. We subtract the given mean from the observations $\boldsymbol{y} = \z - \mu \1$ and compute the mean and variance at location $\x_0$:

\[\mu(\x_0) = \mu + \boldsymbol{y}^\top \l\]

\[\sigma^2(\x_0) = cov(0) - \c^\top \l\]

SimpleKriging(γ, μ)
SimpleKriging(data, γ, μ)

Ordinary Kriging

In Ordinary Kriging the mean of the random field is assumed to be constant and unknown. The resulting linear system is:

\[\begin{bmatrix} \G & \1 \\ \1^\top & 0 \end{bmatrix} \begin{bmatrix} \l \\ \nu \end{bmatrix} = \begin{bmatrix} \g \\ 1 \end{bmatrix}\]

with $\nu$ the Lagrange multiplier associated with the constraint $\1^\top \l = 1$. The mean and variance at location $\x_0$ are given by:

\[\mu(\x_0) = \z^\top \l\]

\[\sigma^2(\x_0) = \begin{bmatrix} \g \\ 1 \end{bmatrix}^\top \begin{bmatrix} \l \\ \nu \end{bmatrix}\]

OrdinaryKriging(γ)
OrdinaryKriging(data, γ)

Universal Kriging

In Universal Kriging, the mean of the random field is assumed to be a polynomial of the spatial coordinates:

\[\mu(\x) = \sum_{k=1}^{N_d} \beta_k f_k(\x)\]

with $N_d$ monomials $f_k$ of degree up to $d$. For example, in 2D there are $6$ monomials of degree up to $2$:

\[\mu(x_1,x_2) = \beta_1 1 + \beta_2 x_1 + \beta_3 x_2 + \beta_4 x_1 x_2 + \beta_5 x_1^2 + \beta_6 x_2^2\]

The choice of the degree $d$ determines the size of the polynomial matrix

\[\F = \begin{bmatrix} f_1(\x_1) & f_2(\x_1) & \cdots & f_{N_d}(\x_1) \\ f_1(\x_2) & f_2(\x_2) & \cdots & f_{N_d}(\x_2) \\ \vdots & \vdots & \ddots & \vdots \\ f_1(\x_n) & f_2(\x_n) & \cdots & f_{N_d}(\x_n) \end{bmatrix}\]

and polynomial vector $\f = \begin{bmatrix} f_1(\x_0) & f_2(\x_0) & \cdots & f_{N_d}(\x_0) \end{bmatrix}^\top$.

The variogram determines the variogram matrix:

\[\G = \begin{bmatrix} \gamma(\x_1,\x_1) & \gamma(\x_1,\x_2) & \cdots & \gamma(\x_1,\x_n) \\ \gamma(\x_2,\x_1) & \gamma(\x_2,\x_2) & \cdots & \gamma(\x_2,\x_n) \\ \vdots & \vdots & \ddots & \vdots \\ \gamma(\x_n,\x_1) & \gamma(\x_n,\x_2) & \cdots & \gamma(\x_n,\x_n) \end{bmatrix}\]

and the variogram vector $\g = \begin{bmatrix} \gamma(\x_1,\x_0) & \gamma(\x_2,\x_0) & \cdots & \gamma(\x_n,\x_0) \end{bmatrix}^\top$.

The resulting linear system is:

\[\begin{bmatrix} \G & \F \\ \F^\top & \boldsymbol{0} \end{bmatrix} \begin{bmatrix} \l \\ \boldsymbol{\nu} \end{bmatrix} = \begin{bmatrix} \g \\ \f \end{bmatrix}\]

with $\boldsymbol{\nu}$ the Lagrange multipliers associated with the universal constraints. The mean and variance at location $\x_0$ are given by:

\[\mu(\x_0) = \z^\top \l\]

\[\sigma^2(\x_0) = \begin{bmatrix}\g \\ \f\end{bmatrix}^\top \begin{bmatrix}\l \\ \boldsymbol{\nu}\end{bmatrix}\]

UniversalKriging(γ, degree, dim)
UniversalKriging(data, γ, degree)

External Drift Kriging

In External Drift Kriging, the mean of the random field is assumed to be a combination of known smooth functions:

\[\mu(\x) = \sum_k \beta_k m_k(\x)\]

Differently than Universal Kriging, the functions $m_k$ are not necessarily polynomials of the spatial coordinates. In practice, they represent a list of variables that is strongly correlated (and co-located) with the variable being estimated.

External drifts are known to cause numerical instability. Give preference to other Kriging variants if possible.

ExternalDriftKriging(γ, drifts)
ExternalDriftKriging(data, γ, drifts)