index path/row date scene_id
1 19/33 04/05/2000 7019033000009650 2 19/34 03/10/2002 7019034000206950 3 20/33 03/14/2001 7020033000107350 4 20/34 02/26/2001 7020034000105750 4 20/35 02/26/2001 7020035000105750 5 21/33 04/30/2001 5021033000112010 6 21/34 04/19/2000 7021034000011050 6 21/35 04/19/2000 7021035000011050 7 22/33 04/29/2001 7022033000111950 8 22/34 04/13/2001 7022034000110350 9 22/35 04/26/2000 7022035000011750 10 23/35 03/22/2002 7023035000208150
By combining the training data with Landsat spectral data and other ancillary data, percent imperviousness prediction models were developed using a regression tree algorithm (named "Cubist"). Three ancillary data sets were used for urban/suburban delineation masking, which is essential to confine the imperviousness mapping within the developed areas. Ancillary data used for urban area masking are listed under "Source Information".
After the best combination of possible input data layers had been determined and the final percent impervious layer produced, the urban mask created was used to eliminate those pixels from the final percent impervious file that fall outside identified impervious surface areas. Thus, the impervious pixels identified within the mask were retained, while all non-impervious surface pixels were removed from the final product. This corrected for any small areas of pixels that may have been included as part of the impervious surface layer because their spectral signature was not covered in the Cubist modeling process. Finally, visual inspection of imperviousness layer was made with limited manual editing to eliminate non-urban areas based on area of interest delineated by the mapping team.