My idea is to use the existing 'gap mask' layer provided by the USGS for each image as input raster for unsupervised classification.
This will leave me with two classes: data and no data. I save the no data class (which represents the error gap itself) as a new layer and I transform it into a polygon layer.
I use this new layer to cut the original image in order to get rid off the stripes in the image.
I chose another imagebfrom the same day or week and use it as a background for the other image. Missing gaps will be filled this way since stripes don't overlap between two images.
Since there shouldn't be significant topographic difference between two images when collected within the same week or so , blending them this way seems reasonable.
I don't know if this method has ever been used, I just thought it would be a logical way to fill the gaps as I need data between 2003 and 2012 for my dissertation. Other satellite imagery is not available for my area for this period
Any thoughts on this method?
Thanks in advance.