The spline is 'tied' to the data at points called knots. The more knots chosen, then the more closely the spline fits the good data points.
The number of knots is determined by the user by stating the number of bins required between knots.
The "bad" points are then reconstituted using the spline coefficients. The quality of these points is set to good while the errors are set to the average variance of the "Good" points.
The spline can be fitted in any particular direction as required. The above procedure is performed for each 1d slice of the data_array along the dimension chosen.