Rational Kriging

V. Roshan Joseph

DOI: https://doi.org/10.1080/01621459.2024.2356296

IF: 4.369

2024-06-23

Journal of the American Statistical Association

Abstract:This article proposes a new kriging that has a rational form. It is shown that the generalized least squares estimator of the mean from rational kriging is much more well behaved than that of ordinary kriging. Parameter estimation and uncertainty quantification for rational kriging are proposed using a Gaussian process framework. A generalized version of rational kriging is also proposed, which includes ordinary and rational kriging as special cases. Extensive simulations carried out over a wide class of functions show that the generalized rational kriging performs on par or better than both ordinary and rational kriging in terms of prediction and uncertainty quantification. The only extra step needed for generalized rational kriging over ordinary kriging is the computation of Perron eigenvector of an augmented correlation matrix which can be computed in near linear time and therefore, its overall computational complexity is no more than that of ordinary kriging. The potential applications of the new kriging methods in the emulation of computationally expensive models and model calibration problems are illustrated with real and simulated examples. Supplementary materials for this article are available online, including a standardized description of the materials available for reproducing the work.

statistics & probability

What problem does this paper attempt to address?

### Problems the paper attempts to solve

This paper aims to solve several key problems in the Kriging interpolation method:

1. **Mean estimation problem**: The Ordinary Kriging (OK) interpolation method may produce unreasonable mean estimates in some cases. For example, when predicting the deflection curve of a beam, ordinary Kriging interpolation may estimate a mean that is outside the range of the observed data, which is physically unacceptable. The paper proposes a new Rational Kriging (RK) interpolation method, which can avoid this mean estimation problem while maintaining good prediction performance.

2. **Prediction uncertainty quantification**: The prediction uncertainty of the ordinary Kriging interpolation method in sparse data areas may become very large, resulting in overly wide confidence intervals. By introducing the Generalized Rational Kriging (GRK) interpolation method and adding a positive constant to the denominator term, the paper avoids this problem and makes the confidence intervals more reasonable.

3. **Model parameter estimation**: The paper proposes a parameter estimation method based on the Gaussian process framework for estimating unknown parameters in rational Kriging interpolation, such as the variance parameter \(\nu\) and the correlation function parameter \(\theta\). The estimation of these parameters is crucial for the accuracy and reliability of the model.

4. **Model calibration**: In computer experiments, model calibration is an important application area. The paper demonstrates the application of the rational Kriging interpolation method in model calibration, especially when dealing with computationally expensive models, it can provide more accurate predictions and uncertainty quantification.

### Main contributions

1. **Rational Kriging interpolation method**: A new rational Kriging interpolation method is proposed. Its predictor has a rational form and can avoid the mean estimation problem while maintaining good prediction performance.

2. **Generalized rational Kriging interpolation method**: Further proposed the generalized rational Kriging interpolation method. By adding a positive constant to the denominator term, it solves the problem of excessive prediction uncertainty in sparse data areas.

3. **Parameter estimation and uncertainty quantification**: Proposed a parameter estimation method based on the Gaussian process framework for estimating unknown parameters in rational Kriging interpolation and provided a method for quantifying prediction uncertainty.

4. **Practical applications**: Through multiple simulation experiments and actual cases, the application of the rational Kriging interpolation method in computer experiments is demonstrated, especially the advantages in model simulation and calibration.

### Formula summary

- **Ordinary Kriging interpolation predictor**:
\[
\hat{y}_{OK}(x)=\hat{\mu}_{OK}+r(x)'R^{-1}(y - \hat{\mu}_{OK}1)
\]
where
\[
\hat{\mu}_{OK}=\frac{1'R^{-1}y}{1'R^{-1}1}
\]

- **Rational Kriging interpolation predictor**:
\[
\hat{y}(x)=\frac{r(x)'R^{-1}(y / \tau)}{r(x)'R^{-1}(1 / \tau)}
\]
where
\[
\tau(x)=\frac{1}{r(x)'R^{-1}(1 / \tau)}
\]

- **Generalized rational Kriging interpolation predictor**:
\[
\tilde{y}(x)=\tilde{\mu}+r(x)'\frac{c_0 + r(x)'c}{\tilde{R}'R^{-1}\tilde{R}\tilde{c}}(y - \tilde{\mu}1)
\]
where
\[
\tilde{\mu}=\frac{\tilde{c}'\tilde{R}'R^{-1}\text{diag}(\tilde{R}\tilde{c})y}{\tilde{c}'\tilde{R}'R^{-1}\text{diag}(\tilde{R}\tilde{c})1}
\]