Case Study on Multispectral Land Cover Classification - Open Source Geospatial Tools: Applications in Earth Observation - page 291

-i LC82070232013160LGN00.tif -i ...
Sequence of input raster datasets.
-o LC82070232013_composite.tif
Name of the output composite based on the two input datasets.
myinline-bndnodata 8
Index of the cloud band defining nodata values in the input raster datasets (9th
band as the index starts from 0).
-srcnodata 1
Value indicating nodata, i.e. cloudy values in this case.
-dstnodata 0
Value written in output raster dataset in case of nodata (i.e. if both scenes are
cloudy).
-cr maxndvi
In case more than one input is cloud free, we select the pixel with the maximum
NDVI value.
-cb 3 -cb 4
The composite bands: the respective indices for the red and near infrared spectral
bands in the multi-band input raster to calculate the NDVI on the fly (see
Table 17.1).

17.1.2 Create NDVI Mask

If prior knowledge is available, it is better to use it. Here we are interested in a
forest/non-forest map. Therefore, we can mask non-vegetated areas in the input
image from the classification process. We will later map it to non-forest during a
post-processing step (see Sect. 17.3.3 ). For the same reason we have concentrated
on those OSM multipolygons with a vegetation attribute for the training sample in
Sect. 17.2 . Masking land cover classes using prior knowledge facilitates the task of
the classifier.
We will create a vegetation mask based on a threshold that is applied on a sim-
ple vegetation index, the normalized difference vegetation index (NDVI, see also
Sect. 11.4 ) . Bare soil, water, residential and other non-vegetated areas have a low
NDVI value. The NDVI image can be produced using gdal_calc.py as explained
in Sect. 11.4 . Notice that in this utility the band index starts counting from 1.

gdal_calc.py -A LC82070232013_composite.tif --A_band=4 -B
ₒ

LC82070232013_composite.tif --B_band=5

ₒ

--outfile=LC82070232013_ndvi.tif --type=UInt16

ₒ

--calc="(B>A)*(0.5+10000.0*(B-A)/(B+A))+(B<=A)*0"