We previously decided to skip creating the noDataMask (which is in the final product coordinates) because the logic was hard to reconcile with Landsat-7 scenes with no-data stripes due to the Scan Line Corrector failure (SLC-Off).
We incorrectly thought this only affected the search range and would just cause some additional unnecessary computational overhead. It turns out that it also affects the final output product size and the %-valid pixels calculation.
Fortunately, now that we always wallis filter up front, we have the zero_mask (no data in image coordinates after filling the stripes) available to us, which we can use to create the noDataMask. The figure below shows the zero_mask on the left and the (now computed) noDataMask on the right:
We previously decided to skip creating the
noDataMask
(which is in the final product coordinates) because the logic was hard to reconcile with Landsat-7 scenes with no-data stripes due to the Scan Line Corrector failure (SLC-Off).We incorrectly thought this only affected the search range and would just cause some additional unnecessary computational overhead. It turns out that it also affects the final output product size and the %-valid pixels calculation.
Fortunately, now that we always wallis filter up front, we have the
zero_mask
(no data in image coordinates after filling the stripes) available to us, which we can use to create thenoDataMask
. The figure below shows thezero_mask
on the left and the (now computed)noDataMask
on the right: