An Inverse Gaussian Process Monte Carlo algorithm for estimation and uncertainty assessment of hydrologic model parameters

TL;DR

This paper introduces the Inverse Gaussian Process Monte Carlo (IGPMC) algorithm, a surrogate-based method for efficient parameter estimation and uncertainty quantification in hydrologic models, especially addressing ill-posed problems.

Contribution

The paper presents a novel inverse Gaussian Process surrogate approach that directly approximates the inverse model function for faster parameter estimation and uncertainty analysis.

Findings

IGPMC provides reliable parameter estimates with lower computational cost.

The method effectively handles ill-posed inverse problems with multiple solutions.

Validation against MCMC shows comparable accuracy and efficiency.

Abstract

Solving hydrologic inverse problems usually requires repetitive forward simulations. One approach to mitigate the computational cost is to build a surrogate model, i.e., an approximate mapping from model parameters (input) to observable quantities (output), so the forward simulations can be done quickly. Alternatively, if the surrogate is constructed to approximate the inverse mapping from model outputs to parameters, the parameter estimates can be obtained directly by treating measurements as inputs to this inverse surrogate. Moreover, the uncertainties of parameters can be quantified by propagating the measurement uncertainties in a straightforward Monte Carlo manner. Based on this idea, we proposed a novel surrogate-based approach for parameter estimation and uncertainty assessment, i.e., the Inverse Gaussian Process Monte Carlo (IGPMC) algorithm. The Gaussian Process (GP) regression is used to directly approximate the inverse function of the model output-input relationship. For ill-posed problems, i.e., when there exist non-unique sets of input parameters that all produce an identical system output, multiple inverse GP systems are constructed and multiple parameter estimates can be obtained accordingly. The accuracy and efficiency of this IGPMC algorithm were demonstrated through four numerical case studies. Results obtained from the Markov Chain Monte Carlo (MCMC) are used as references to assess our new proposed method. It was shown that, the IGPMC algorithm can generally obtain reliable parameter estimates with an affordable computational cost.