Case Study on Multispectral Land Cover Classification - Open Source Geospatial Tools: Applications in Earth Observation - page 289

some cloud cover. Clouds directly affect image quality and can severely hamper the
image analysis and classification. To obtain a scene withminimal clouds, we combine
the downloaded scenes and create a “cloud free” composite. If you are using your own
satellite imagery acquired for another location and it is cloud free than you can skip
the next processing step. The BQA dataset is in a bit-packed format 2 and contains
information on water and clouds. This means we need to decode the information
in order to obtain, for instance, a cloud mask. More information on the bit-packing
encoding scheme is explained in Sect. 14.3 , where we also implement a tool to create
a cloud mask from the BQA dataset. Alternatively, you can use the official tool, the
L-LDOPE Toolbelt, that is provided by the USGS. 3 As an example, we create a cloud
mask with the tool bqa2cloud from Sect. 14.3 :

bqa2cloud -i LC82070232013160LGN00_BQA.TIF -o
ₒ LC82070232013160LGN00_CLD.TIF
bqa2cloud -i LC82070232013192LGN00_BQA.TIF -o
ₒ

LC82070232013192LGN00_CLD.TIF

We then stack all bands into a single multi-band raster dataset in GeoTIFF format
using the utility gdal_merge.py . The panchromatic band contains little extra
spectral information and the thermal bands have been acquired with a coarser spatial
resolution. Therefore, we do not include bands 8, 10 and 11.

gdal_merge.py -o LC82070232013160LGN00_merge.tif -of GTiff -ps
ₒ 30 30 -separate -co INTERLEAVE=BAND -co COMPRESS=LZW

ₒ LC82070232013160LGN00_B1.TIF LC82070232013160LGN00_B2.TIF

ₒ LC82070232013160LGN00_B3.TIF LC82070232013160LGN00_B4.TIF

ₒ LC82070232013160LGN00_B5.TIF LC82070232013160LGN00_B6.TIF

ₒ LC82070232013160LGN00_B7.TIF

ₒ LC82070232013160LGN00_B9.TIF

ₒ LC82070232013160LGN00_CLD.TIF

-o LC82070232013160LGN00_merge.tif
Name of the output multi-band raster dataset.
-of GTiff
We explicitly set the GeoTIFF image format for the output (not required here as
this is the default).
-ps 30 30
Set a common spatial resolution for all inputs. Most bands are already at this
resolution, but the panchromatic band needs to be resampled from 15 to 30 m.
-separate
Place each input file into a separate stacked band.

2 http://landsat.usgs.gov/L8QualityAssessmentBand.php
3 http://landsat.usgs.gov/L-LDOPE_Toolbelt.php