Bhattacharya, S ; Jansen, M ; Lef`evre, L ; Clette, F

Poster presented at Applications of Statistical Methods and Machine Learning in the Space Sciences hosted by Space Science Institute, Boulder, Colorado (virtual) on 2021-05-21

Abstract: The SN series is one of the longest and most detailed available series in astrophysics. The series was first constructed in 1849 by Prof. Rudolf Wolf and a time series is built in real time since then, involving a lot of observers who differ from each other in terms of their way of counting sunspots, different telescopes and eye sights, which demands proper calibration. We present a novel time-dependent error determination on the sunspot number (SN) based on nonparametric statistical techniques in smoothing. In particular we propose a generalized linear regression model with overdispersed count data as response variables in the estimation of a time varying calibration of different sunspot time series with overlapping periods. The nonparametric regression takes place through a local polynomial smoothing procedure. This data assimilation model does not restrict to SN only but is applicable to most sunspots’ parameters such as area covered on the Sun, position etc... Many catalogs such as the one from Catania Observatory, the Royal Greenwich Observatory, the US Air Force solar observations and others, include the above mentioned parameters whose overall quality assessment is lacking as of yet. The time dependence criteria of our model allows us to access the quality of daily observations with respect to other catalogs thus adding an error bar. We focus this study on one of the stable stations (Mathieu et al,2019) of the World Data Center SILSO network, the Uccle Solar Equatorial table station in Brussels (USET). The study we present aims at a further refinement of earlier work in Mandal et al. (2020), adding a realistic statistical model and a time dependent calibration factor. Applying regression before computing the calibration avoids the presence of outliers and biased estimation.

Links: link
Funding: 3FULLCOST/3FULLCOST/3FULLCOST