Geoscience Reference
In-Depth Information
case where an independent reference test set is already available, the choice of the
sampling design has already been made.
on the utility
gdal2xyz.py
.
We adapt the command from Sect.
10.3
in a number of ways. First, we exclude
pixels with a value of 0 from the accuracy assessment . These pixels are either outside
of the acquisition area of the sensor or have been identified as cloudy. There is no
option to ignore no-data values in the utility
gdal2xyz.py
, so we need to do it
in a different way. One option is to open the output file in a text editor or in spread
sheet program and do it manually. Here, we use an automatic method based on the
awk
command in Bash.
gdal2xyz.py -skip 200 LC82070232013_fmap_masked.tif |
awk
-v
ₒ
OFS="," '{
if
($3>0) print $1,$2,"noforest","valid"}' >
ₒ
sample.csv
-skip 200
Downsampling factor: report every 200th row and column.
LC82070232013_fmap_masked.tif
Name of the input raster dataset.
| awk -v OFS=","
Pipe output of previous command to the Bash
awk
command.
-v OFS=","
Report in output in comma separated value (CSV) format.
'if($3>0) print $1,$2,“noforest”,“valid”'
Print first two columns if third column is (strict) positive. Add extra third and
fourth columns with the fixed text “noforest” and “valid”.
> sample.csv
Print output to a file named
sample.csv
(redirect output).
The output is formatted in CSV format in order to be recognized by OGR. The
extra field in column three will be used to describe the label of our sample as part of
the
response design
(Stehman and Czaplewski 1998). We expect most of the points
in a non-forested area. We therefore set the initial description as
"noforest"
.
This will save us typing when revisiting the points as we only have to re-label points
within a forest area. The extra field in column four will be used for comments in the
labeling process.
Based on the CSV file, we create a virtual OGR vector,
sample.vrt
,usingthe
Search WWH ::
Custom Search