Tweak Landcover Mask to False Positive Burn Identification Along Shorelines

[needham-michael]

Figure below shows burned pixel identification often occurs for pixels intersecting the shoreline of bodies of water. Majority of these instances will be false positives, so need to update the process_landcover_mask.py script to expand the mask outward for the water land use category.

[needham-michael]

Simply performing a binary dilation operation on the NLCD Class 11 (open water) category type successfully moves the Landcover mask boundary away from large open bodies of water (compare original green and dilated blue masks below), but leads to a strange patchwork associated with much smaller water bodies.

Will investigate whether an alternative approach of applying the dilation after the landcover mask has been reprojected fixes the issue.

[needham-michael]

Updated method allows for extending the open water mask around large features without the interference of small features.

1. Resample open water mask from NLCD to final (i.e., VIIRS) resolution using nearest neighbor
2. Apply binary dilation on the inverted resampled field (0-open water, 1-other) to remove small features, then apply binary dilation 4 times on the resultant field to make a large buffer around large features
3. Apply this interim mask to the original resampled field to retain the finer-scale features of the large water bodies
4. Apply a final binary dilation to extend the feature edges around large features