Geoscience Reference
In-Depth Information
-co INTERLEAVE=BAND
Creation option for band sequential encoding (see Sect. 3.5).
-co COMPRESS=LZW
Creation option for LZW compressed output (see Sect. 5.5.4).
LC82070232013_ndvi_masked.tif
Name of the first input raster dataset to be stacked.
LC82070232013_composite.tif
Name of the second input raster dataset to be stacked.
17.2 Create Training Data
We will create a training dataset of land cover points to train the supervised classifica-
tion algorithm. There is a more elegant way to do this, making use of virtual datasets
the stacked bands. We can just create an XML file that represents the virtual raster
dataset with a link to the real data. This XML file contains an entry for each raster
band (VRTRasterBand) to be included. For the purposes of the case study, we will
use data fromOSM (see Appendix A.1.1) to label the training data. These data can be
obtained for free in vector format
4
(see also Appendix A.1.1). They contain different
feature sets: points, lines, multipolygons, etc. Each feature has attribute information
that we can use to label the training data as forest and non-forest (see Sect.
17.2.2
).
In particular for the
pksvm
classification utility, we also need to prepare the training
vector such that it contains the raster band information for each training sample unit
(see Sect.
17.2.3
).
17.2.1 Get Training Data
We start the collection of training data by downloading the OSMvector for our region
of interest. This correspond to the bounding box of the Landsat 8 scene, which we
gdalinfo LC82070232013160LGN00.tif
...
Upper Left
(
438585.000, 5997015.000) (
9d56'22.52"W, 54d 7'
ₒ
3.39"N)
Lower Left
(
438585.000, 5769885.000) (
9d53'46.08"W, 52d
ₒ
4'34.08"N)
Upper Right (
672615.000, 5997015.000) (
6d21'37.85"W, 54d
ₒ
5'32.36"N)
Search WWH ::
Custom Search