DC series arc fault(DCSAF) in photovoltaic systems is a controversial issue owing to the challenges related to its modeling and protection. This article focuses on efficient models based on Schwarz model for DCSAF utilizing actual recorded data. The original Schwarz model has four parameters, containing two coefficients and two model orders, and all four parameters are constant. Despite the original Schwarz model that uses invariant parameters for arc modeling, the proposed models employ a Schwarz model in which the time-variant nature of the arc's parameters is taken into account. The first developed Schwarz model with two constant model orders and two time-variant coefficients is derived in the first part. While the first model provides more capability compared with the original Schwarz modeling, the second developed Schwarz model with full time-variant parameters is also presented in the second part. A novel optimization technique based on Powell's dog leg algorithm is proposed to estimate the arc coefficients in every simulation window. Therefore, the parameters in both modified Schwarz models are time-variant and change every simulation window. As a result, they are classified as a time series, which is examined using the autoregressive moving average process. By actual recorded data, the performance of the proposed models is assessed. Analyses show that the proposed models are more efficient to model DCSAF in comparison with the original Schwarz model.