Type Ia supernova cosmology depends on the ability to fit and standardize observations of supernova magnitudes with an empirical model .
We present here a series of new models of Type Ia Supernova spectral time series that capture a greater amount of supernova diversity than possible with the models that are currently customary .
These are entitled SuperNova Empirical MOdels ( SNEMO https : //snfactory.lbl.gov/snemo ) .
The models are constructed using spectrophotometric time series from 172 individual supernovae from the Nearby Supernova Factory , comprising more than 2000 spectra .
Using the available observations , Gaussian Processes are used to predict a full spectral time series for each supernova .
A matrix is constructed from the spectral time series of all the supernovae , and Expectation Maximization Factor Analysis is used to calculate the principal components of the data .
K-fold cross-validation then determines the selection of model parameters and accounts for color variation in the data .
Based on this process , the final models are trained on supernovae that have been dereddened using the Fitzpatrick and Massa extinction relation .
Three final models are presented here : SNEMO2 , a two-component model for comparison with current Type Ia models ; SNEMO7 , a seven component model chosen for standardizing supernova magnitudes which results in a total dispersion of 0.100 mag for a validation set of supernovae , of which 0.087 mag is unexplained ( a total dispersion of 0.113 mag with unexplained dispersion of 0.097 mag is found for the total set of training and validation supernovae ) ; and SNEMO15 , a comprehensive 15 component model that maximizes the amount of spectral time series behavior captured .