A Comparison of Feature Classification Methods for Modeling Solar Irradiance Variation

Physical understanding of total and spectral solar irradiance variation depends upon establishing a connection between the temporal variability of spatially resolved solar structures and spacecraft observations of irradiance. One difficulty in comparing models derived from different data sets is that the many ways for identifying solar features such as faculae, sunspots, quiet Sun, and various types of “network” are not necessarily consistent. To learn more about classification differences and how they affect irradiance models, feature “masks” are compared as derived from five current methods: multidimensional histogram analysis of NASA/National Solar Observatory/Kitt Peak spectromagnetograph data, statistical pattern recognition applied to SOHO/Michelson Doppler Imager photograms and magnetograms, threshold masks allowing for influence of spatial surroundings applied to NSO magnetograms, and “one-trigger” and “three-trigger” algorithms applied to California State University at Northridge Cartesian Full Disk Telescope intensity observations. In general all of the methods point to the same areas of the Sun for labeling sunspots and active-region faculae, and available time series of area measurements from the methods correlate well with each other and with solar irradiance. However, some methods include larger label sets, and there are important differences in detail, with measurements of sunspot area differing by as much as a factor of two. The methods differ substantially regarding inclusion of fine spatial scale in the feature definitions. The implications of these differences for modeling solar irradiance variation are discussed.