Efficient Bayesian inference for large chaotic dynamical systems

Abstract

Abstract. Estimating parameters of chaotic geophysical models is challenging due to these models' inherent unpredictability. With temporally sparse long-range observations, these models cannot be calibrated using standard least squares or filtering methods. Obvious remedies, such as averaging over temporal and spatial data to characterize the mean behavior, do not capture the subtleties of the underlying dynamics. We perform Bayesian inference of parameters in high-dimensional and computationally demanding chaotic dynamical systems by combining two approaches: (i) measuring model-data mismatch by comparing chaotic attractors, and (ii) mitigating the computational cost of inference by using surrogate models. Specifically, we construct a likelihood function suited to chaotic models by evaluating a distribution over distances between points in the phase space; this distribution defines a summary statistic that depends on the attractor's geometry, rather than on pointwise matching of trajectories. This statistic is computationally expensive to simulate, compounding the usual challenges of Bayesian computation with physical models. Thus we develop an inexpensive surrogate for the log-likelihood via local approximation Markov chain Monte Carlo, which in our simulations reduces the time required for accurate inference by orders of magnitude. We investigate the behavior of the resulting algorithm on model problems, and then use a quasi-geostrophic model to demonstrate its large-scale application.

A Bayesian Approach to Modelling Biological Pattern Formation with Limited Data

“…In creating the parameter posteriors by MCMC methods, the amount of required model evaluations may be higher, around N chain ˆNset and N chain ˆNsyn , where generally N chain ě 12000. However, random-walk-based MCMC are conceivable by using suitable techniques, such as multi- level MCMC methods [43] or local approximation methods for the posterior [23,44]. It results a further improvement of the method.…”

“…We solve this problem by a statistical approach, which allows defining a stochastic cost function that makes it possible to measure a distance between experimental pattern data and the model output. This technique was developed in [22] and [23] for classical, chaotic dynamical systems and extended to reaction-diffusion systems in [17]. Once defined, the CIL cost function can be minimised by available algorithms of stochastic optimisation, resulting in point estimates for model parameters.…”

An Efficient Estimation of Time‐Varying Parameters of Dynamic Models by Combining Offline Batch Optimization and Online Data Assimilation

“…The design of Equation apparently depends on the problems. The design of the appropriate climatological index for the offline batch optimization is generally challenging (e.g., Springer et al., 2021). It is promising that the climatological index based only on the widely used hydrological indices, runoff ratio and baseflow index, appropriately works in the application of HOOPE‐PF to the conceptual hydrological model.…”