Geoscience Reference
In-Depth Information
We can use the GDAL utility gdal_calc.py (Sect. 11.4 ) to do the re-labeling.
The forest map and NDVI mask are represented by image A and B respectively. If
the NDVI mask is above 0, we want to retain the original value of the forest map
(A
(
B
>
0
)
). If the mask equals 0, we will re-label the pixels in the forest map as
non-forest (2
). Finally, we set the out of scene pixels to a no-data value
of 0, by using the Landsat composite image, represented as image C (C
(
B
==
0
)
>
0). The
result is shown in Fig. 17.5 .
gdal_calc.py -A LC82070232013_fmap_unsupervised_sieved.tif -B
LC82070232013_ndvi_masked.tif -C
LC82070232013_composite.tif
--outfile=LC82070232013_fmap_unsupervised_masked.tif
--calc="(A*(B>0)+2*(B==0))*(C>0)" --NoDataValue=0
gdal_calc.py -A LC82070232013_fmap_supervised_sieved.tif -B
LC82070232013_ndvi_masked.tif -C
LC82070232013_composite.tif
--outfile=LC82070232013_fmap_supervised_masked.tif
--calc="(A*(B>0)+2*(B==0))*(C>0)" --NoDataValue=0
-A LC82070232013_fmap_unsupervised_sieved.tif
First input raster dataset.
-B LC82070232013_ndvi_masked.tif
Second input raster dataset.
-C LC82070232013_composite.tif
Third input raster dataset.
--outfile=LC82070232013_fmap_unsupervised_masked.tif
Name of the calculated output raster dataset.
--calc="(A*(B>0)+2*(B==0))*(C>0)"
Output will be equal to A if B is (strict) positive. Set output to noforest (class=2)
if B equals 0. C should always be (strict) positive or output will be 0.
--NoDataValue=0
Set nodata value in output raster dataset to 0.
17.3.4 Accuracy Assessment
A land cover map as a result of a classifier is only a representation of reality and
inherently contains error. It is therefore important to assess its accuracy before using
it. Depending on the application for which you want to use the land cover map, your
expectations on the accuracy will vary. A thorough discussion on accuracy and how
to design and implement an effective accuracy assessment is beyond the scope of
this topic. Here, we will just perform an accuracy assessment of the created forest
map, based on a systematic sampling design (Stehman and Czaplewski 1998). Other
strategies are valid as well, for instance a simple random sampling design. In the
 
 
Search WWH ::




Custom Search