Interpolation and synthesis of sparse samples in exoplanet atmospheric modeling

TL;DR

This paper explores geostatistical methods like kriging for interpolating and synthesizing sparse exoplanet atmospheric model data, aiding in model intercomparison and combining observational data with simulations.

Contribution

It introduces the application of ordinary and universal kriging techniques to exoplanet atmospheric modeling, enabling better analysis of sparse data and model synthesis.

Findings

Kriging methods effectively interpolate sparse atmospheric data.

Variograms serve as diagnostic tools for model data distribution.

Universal kriging synthesizes data from models of varying complexity.

Abstract

This paper highlights methods from geostatistics that are relevant to the interpretation, intercomparison, and synthesis of atmospheric model data, with a specific application to exoplanet atmospheric modeling. Climate models are increasingly used to study theoretical and observational properties of exoplanets, which include a hierarchy of models ranging from fast and idealized models to those that are slower but more comprehensive. Exploring large parameter spaces with computationally-expensive models can be accomplished with sparse sampling techniques, but analyzing such sparse samples can pose challenges for conventional interpolation functions. Ordinary kriging is a statistical method for describing the spatial distribution of a data set in terms of the variogram function, which can be used to interpolate sparse samples across any number of dimensions. Variograms themselves may also be useful diagnostic tools for describing the spatial distribution of model data in exoplanet atmospheric model intercomparison projects. Universal kriging is another method that can synthesize data calculated by models of different complexity, which can be used to combine sparse samples of data from slow models with larger samples of data from fast models. Ordinary and universal kriging can also provide a way to synthesize model predictions with sparse samples of exoplanet observations and may have other applications in exoplanet science.