Nodata handling in 0c_Data_Preprocessing.ipynb

[dshean]

We need to do careful review of how nodata mask is prepared and used.

It's not clear to me that the original ASP DEM nodata is preserved, as the input ASP DEM defined in cell 2
initial_dsm_file = '20150911_2050_1020010042D39D00_1020010043455300_1.0m-DEM_trans_dx+2.40m_dy+1.51m_dz-0.36m__ASP_wt_avg_extrapolation_search_rad25_num_pass_3.tif'
Has already been filled. It seems the triangulation error is assumed to provide the "valid DEM pixel" mask. We need to verify that these are one-to-one in output from point2dem.

I don't understand this:

   # change nodata value from -9999 for triangulation error
    if key == 'triangulation_error':
        dst_raster = np.where(dst_raster==tmp_profile['nodata'], -1, dst_raster)
        tmp_profile.update({
            'nodata':-1
        })

This looping to create a common nodata mask for all input layers can likely be replaced by a simple multiplication if you have binary masks for a raster stack with identical dimensions. See my approach using masked arrays here: https://github.com/dshean/pygeotools/blob/master/pygeotools/lib/malib.py#L1307

nodata_mask = np.ones((1, nodata_mask_list[0].shape[0], nodata_mask_list[0].shape[1]), dtype=np.uint8)
for x in nodata_mask_list:
    nodata_mask = np.where(x==1, 0, nodata_mask)