Using linear and nonlinear predictors to improve the computational efficiency of the FD-TD algorithm