After the prediction operators are constrained to be purely lateral, additional criteria for filter design are considered to eliminate two types of prediction failures seen in this work. The most apparent failure is seen in the traces near the edge of the output volume. These traces often have polarities opposite to the extrapolated data in the interior and may be stronger than the adjacent traces. The other failure is that of blank areas in output volume. This seems to be the result of events dipping in a direction where a filter cannot make predictions.
Since bad predictions are produced between the output of the filter and the edge of the data, the bad predictions along any one boundary can be eliminated by using a prediction filter with its output along that boundary. While a bad output along the opposite boundary is expected, I show later that this can be eliminated with multiple filters. This same set of multiple filters can be used to avoid the blank areas in the output volume by providing predictions in all directions.