Parameter Estimation from an Optimal Projection in a Local Environment

TL;DR

This paper introduces a local projection method called MATISSE for estimating parameters from model grids, improving accuracy in non-linear environments by optimizing the projection process and kernel choice.

Contribution

It presents a novel two-step local projection approach, connecting local linear regression with optimal estimation, and compares it to objective analysis methods.

Findings

MATISSE improves parameter estimation accuracy.

Kernel choice significantly affects estimation quality.

Method effectively handles non-linear parameter spaces.

Abstract

The parameter fit from a model grid is limited by our capability to reduce the number of models, taking into account the number of parameters and the non linear variation of the models with the parameters. The Local MultiLinear Regression (LMLR) algorithms allow one to fit linearly the data in a local environment. The MATISSE algorithm, developed in the context of the estimation of stellar parameters from the Gaia RVS spectra, is connected to this class of estimators. A two-steps procedure was introduced. A raw parameter estimation is first done in order to localize the parameter environment. The parameters are then estimated by projection on specific vectors computed for an optimal estimation. The MATISSE method is compared to the estimation using the objective analysis. In this framework, the kernel choice plays an important role. The environment needed for the parameter estimation can result from it. The determination of a first parameter set can be also avoided for this analysis. These procedures based on a local projection can be fruitfully applied to non linear parameter estimation if the number of data sets to be fitted is greater than the number of models.