LANDSAT

From GRASS-Wiki
Revision as of 18:38, 15 May 2013 by ⚠️NikosA (talk | contribs) (→‎Automated data import: import script updated as per Glynn's comments in the ML on using "or")
Jump to navigation Jump to search

Data Availability

Landsat imagery can be obtained from USGS' LANDSAT archive:

See also the Global datasets wiki page

Pre-Processing Overview

Typically, pre-processing Landsat imagery comprises the following steps:

  1. import in the database → r.in.gdal
  2. geometrically & orthometrically correct imagery
  3. optionally, automatically cut-off border fringesi.landsat.trim
    • of course one can use the official WRS2 Path/Row vector tiles to manually trim border fringes
  4. optionally, denoise for obvious/intensive salt & pepper effects, stripes, etc.
    • for example by applying Principal Components Analysis as a denoising technique → i.pca
  5. convert the Digital Numbers (DNs) to Top-of-Atmosphere Radiances/Reflectances (ToARs) → i.landsat.toar
  6. optionally, correct for atmospheric effectsi.atcorr
    • that is, accounting for distorting atmospheric effects and estimating actual reflectances as they would have been measured on the ground
    • also described as conversion to Top-of-Canopy Reflectances (ToCRs)
  7. assessing cloud coveri.landsat.acca
    • optionally, detect and remove clouds shadows as well
  8. topographically normalise imagery → i.topo.corr
    • also known as topographic correction, that is, accounting for illumination differences due to the acquisition's geometry
  9. radiometrically normalise → one approach via i.histo.match (in grass 7)
    • also known as relative radiometric normalisation -- one approach is the histogram matching technique of two or more raster maps
    • ToDo/More techniques?

Modules overview

Generic modules applicable to Landsat

  • d.rgb → display 3-band data
  • r.composite → flatten 3-bands of data into a single image (lossy, maybe used in combination with i.landsat.rgb)
  • i.atcorr → correct top of atmosphere to surface reflectance - see also the Atmospheric correction wiki page
  • i.topo.corr → used to topographically correct reflectance from imagery files, e.g. obtained with i.landsat.toar, using a sun illumination terrain model
  • i.oif → calculate the 3 bands showing the greatest difference (for use as R,G,B bands)
  • i.histo.match → histogram matching of two or more raster maps (in grass 7) Note, the module works with integer values and does not accept the "." character as part of the raster map's name!

Landsat specific modules

Landsat specific GRASS AddOns

  • i.landsat.trim → trims border fringes for each band separately or with the MASK where coverage exists for all bands
  • i.landsat.dehaze (addon) → haze removal

Pre-Processing

Importing data

  1. Open GRASS GIS, select Location Wizard in order to create a new location from georeferenced file
  2. Use r.in.gdal or the Import file tool to import the GeoTIFF files into GRASS GIS. See also Automated data import below.
    1. Most Landsat scenes are delivered in north-is-up orientation, hence import is straightforward
    2. If you get ERROR: Input map is rotated - cannot import., the image must be first externally rotated, applying the rotation info stored in the metadata field of the raster image file. For example, the gdalwarp software can be used to transform the map to North-up (note, there are several gdalwarp parameters to select the resampling algorithm):
        gdalwarp rotated.tif northup.tif
      


Automated data import

The following Python script imports Landsat imagery into GRASS' data base. Specifically, the script

  • creates an independent Mapset for each Landsat scene
  • imports and renames bands of a scene as B<id>, e.g. B10, B20, ..., B80.
  • additionaly sets up the timestamp based on MTL metadata file


Note, the (newest) official naming pattern for Landsat scenes -- explained in USGS' Landsat Data Dictionary as the Landsat Scene Identifier -- and all individual bands that compose a scene -- bands have a suffix which is like _B10, _B20, _B30, etc. -- differ from what some Landsat specific modules expect. For example, the modules i.landsat.toar and i.landsat.acca expect the bands to follow a naming pattern such as "scenename.1, .2, .3", etc.


To use the script save it as import_landsat.py file and make sure it is granted the execution permission.

#!/usr/bin/python
 
import os
import sys
import glob
import grass.script as grass
 
def get_timestamp(mapset):
    try:
        metafile = glob.glob(mapset + '/*MTL.txt')[0]
    except IndexError:
        return
 
    result = dict()
    try:
        fd = open(metafile)
        for line in fd.readlines():
            line = line.rstrip('\n')
            if len(line) == 0:
                continue
            if any(x in line for x in ('DATE_ACQUIRED', 'ACQUISITION_DATE')):
                result['date'] = line.strip().split('=')[1].strip()
    finally:
        fd.close()
 
    return result
 
def import_tifs(mapset):
    meta = get_timestamp(mapset)
    for file in os.listdir(mapset):
        if os.path.splitext(file)[-1] != '.TIF':
            continue
        ffile = os.path.join(mapset, file)
        if ('VCID') in ffile:
            name = "".join((os.path.splitext(file)[0].split('_'))[1::2])
        else:
            name = os.path.splitext(file)[0].split('_')[-1]
        band = int(name[-1])
        grass.message('Importing %s -> %s@%s...' % (file, name, mapset))
        grass.run_command('g.mapset',
                          flags = 'c',
                          mapset = mapset,
                          quiet = True,
                          stderr = open(os.devnull, 'w'))
        grass.run_command('r.in.gdal',
                          input = ffile,
                          output = name,
                          quiet = True,
                          overwrite = True,
                          title = 'band %d' % band)
        if meta:
            year, month, day = meta['date'].split('-')
            if month == '01':
                month = 'jan'
            elif month == '02':
                month = 'feb'
            elif month == '03':
                month = 'mar'
            elif month == '04':
                month = 'apr'
            elif month == '05':
                month = 'may'
            elif month == '06':
                month = 'jun'
            elif month == '07':
                month = 'jul'
            elif month == '08':
                month = 'aug'
            elif month == '09':
                month = 'sep'
            elif month == '10':
                month = 'oct'
            elif month == '11':
                month = 'nov'
            elif month == '12':
                month = 'dec'
 
            grass.run_command('r.timestamp',
                              map = name,
                              date = ' '.join((day, month, year)))
 
def main():
    if len(sys.argv) == 1:
        for directory in filter(os.path.isdir, os.listdir(os.getcwd())):
            import_tifs(directory)
    else:
        import_tifs(sys.argv[1])
 
if __name__ == "__main__":
    main()


Example of usage

After having collected the Landsat scenes of interest in one directory, the script can be used as follows:

  • ./import_landsat.py → the script will walk through a pool directory that contains unique Landsat scene directories (e.g. three directories named after the official naming pattern: LT51800342011158MOR00 LT51810352009079MTI00 LT51820352009326MTI00) and import all bands of each individual scene in their own Mapset
  • ./import_landsat.py LM41890261983200FFF03 → the scrip will import only bands of the specified Landsat scene directory

Hint: Minimal disk space copies

Here's a little trick using r.reclass to rename maps (for example, from L71074092_09220040924_B10 to L71074092_09220040924_B1) without touching the data or wasting disk space:

BASE=L71074092_09220040924

for BAND in 10 20 30 40 50 61 70 80; do
  BAND1st=`echo $BAND | sed -e 's/0$//'`
  r.reclass in="${BASE}_B$BAND" out=$BASE.$BAND1st << EOF
    * = *
EOF
done

Natural color composites

Solution A) use i.landsat.rgb

Solution B) equalize colors on each R,G,B band with

r.colors -e map=band1 color=grey

Composite: then run r.composite

Reset color tables

BASE=L71074092_09220040924

for map in `g.mlist pat="$BASE.[0-8]*"` ; do
  r.colors $map color=grey255
done

Create a MASK to only show data where coverage exists for all bands

BASE=L71074092_09220040924

g.region rast=$BASE.1
r.series in=`g.mlist pat="$BASE.[0-8]*" sep=,` -n out=$BASE.thresh method=threshold thresh=1
r.mapcalc "$BASE.mask = if(isnull($BASE.thresh))"
g.remove $BASE.thresh

Calculate Top-of-Atmosphere Reflectance and band-6 Temperature

Calculate Top-of-Atmosphere reflectance and band-6 temperature using the i.landsat.toar module. For details, refer to USGS' Landsat Filename and Metadata Changes dedicated page.

 i.landsat.toar input_prefix=$BASE output_prefix=${BASE}_toar metfile=${BASE}_MTL.txt

Note, the resulting temperature map is in Kelvin:

 # convert to degree Celsius
 r.mapcalc "$BASE.temp_celsius = ${BASE}_toar.6 - 273.15"
 r.info -r $BASE.temp_celsius

Haze removal

Simple haze removal can be done with i.landsat.dehaze. i.landsat.dehaze applies a bandwise haze correction using tasscap4 (haze) and linear regression.


Atmospheric correction

See Atmospheric correction

Cloud identification

Identify clouds in the image with i.landsat.acca:

i.landsat.acca -f input_prefix=226_62_toar. output=226_62.acca_cloudmask

Mask out the clouds:

r.mapcalc "MASK = if(isnull($BASE.acca_cloudmask))"

Download sample data

Preprocessed Landsat-7 data for North Carolina

The North Carolina 2008 sample dataset comes with 3 different Landsat scenes:

(Wake County -- path: 16 row: 35 for various dates)

The above import efforts are not needed since the data are already in a GRASS location.

Landsat-5: Oct 14, 1987

Glovis download
LT50160351987287XXX08
http://edcsns17.cr.usgs.gov/cgi-bin/EarthExplorer/run-phtml/results/download.phtml?node=GV&ordered=LT50160351987287XXX08&dataset_name=LANDSAT_TM


Provided metadata:

IMAGE_ID=P016R35_5T871014
PATH=16
ROW=35
DATE=10/14/87
PLATFORM=LANDSAT5

Landsat-7: Mar 31, 2000

Glovis download
LE70160352000091EDC00
http://edcsns17.cr.usgs.gov/cgi-bin/EarthExplorer/run-phtml/results/download.phtml?node=GV&ordered=LE70160352000091EDC00&dataset_name=LANDSAT_ETM

Values in the metadata below indicate that the version provided with the NC 2008 dataset's production date was after July 1, 2000, and that the channel gains were HHHLHLHHL.

Convert DNs to radiance/temperatures:

 GRASS> i.landsat.toar -v band=lsat7_2000 sensor=7 date=2000-03-31 \
    product_date=2000-07-02 solar_elevation=51.5246529 gain=HHHLHLHHL

Provided metadata:

SPACECRAFT_ID=Landsat7
SENSOR_ID=ETM+
ACQUISITION_DATE=2000-03-31
WRS_PATH=16
CPF_FILE_NAME=L7CPF20000101_20000331_12
SUN_AZIMUTH=139.6033279
SUN_ELEVATION=51.5246529
LMAX_BAND1=191.600
LMIN_BAND1=-6.200
LMAX_BAND2=196.500
LMIN_BAND2=-6.400
LMAX_BAND3=152.900
LMIN_BAND3=-5.000
LMAX_BAND4=241.100
LMIN_BAND4=-5.100
LMAX_BAND5=31.060
LMIN_BAND5=-1.000
LMAX_BAND61=17.040
LMIN_BAND61=0.000
LMAX_BAND62=12.650
LMIN_BAND62=3.200
LMAX_BAND7=10.800
LMIN_BAND7=-0.350
LMAX_BAND8=243.100
LMIN_BAND8=-4.700
QCALMAX_BAND1=255.0
QCALMIN_BAND1=1.0
QCALMAX_BAND2=255.0
QCALMIN_BAND2=1.0
QCALMAX_BAND3=255.0
QCALMIN_BAND3=1.0
QCALMAX_BAND4=255.0
QCALMIN_BAND4=1.0
QCALMAX_BAND5=255.0
QCALMIN_BAND5=1.0
QCALMAX_BAND61=255.0
QCALMIN_BAND61=1.0
QCALMAX_BAND62=255.0
QCALMIN_BAND62=1.0
QCALMAX_BAND7=255.0
QCALMIN_BAND7=1.0
QCALMAX_BAND8=255.0
QCALMIN_BAND8=1.0

Landsat-7: May 24, 2002

Glovis download
LE70160352002144EDC00
http://edcsns17.cr.usgs.gov/cgi-bin/EarthExplorer/run-phtml/results/download.phtml?node=GV&ordered=LE70160352002144EDC00&dataset_name=LANDSAT_ETM


Provided metadata: (`p016r035_7x20020524.met`)

SPACECRAFT_ID=Landsat7
SENSOR_ID=ETM+
ACQUISITION_DATE=2002-05-24
WRS_PATH=016
WRS_ROW=035
SUN_AZIMUTH=120.8810347
SUN_ELEVATION=64.7730999
QA_PERCENT_MISSING_DATA=0
CLOUD_COVER=0
CPF_FILE_NAME=L7CPF20020401_20020630_03
LMAX_BAND1=191.600
LMIN_BAND1=-6.200
LMAX_BAND2=196.500
LMIN_BAND2=-6.400
LMAX_BAND3=152.900
LMIN_BAND3=-5.000
LMAX_BAND4=241.100
LMIN_BAND4=-5.100
LMAX_BAND5=31.060
LMIN_BAND5=-1.000
LMAX_BAND61=17.040
LMIN_BAND61=0.000
LMAX_BAND62=12.650
LMIN_BAND62=3.200
LMAX_BAND7=10.800
LMIN_BAND7=-0.350
LMAX_BAND8=243.100
LMIN_BAND8=-4.700
QCALMAX_BAND1=255.0
QCALMIN_BAND1=1.0
QCALMAX_BAND2=255.0
QCALMIN_BAND2=1.0
QCALMAX_BAND3=255.0
QCALMIN_BAND3=1.0
QCALMAX_BAND4=255.0
QCALMIN_BAND4=1.0
QCALMAX_BAND5=255.0
QCALMIN_BAND5=1.0
QCALMAX_BAND61=255.0
QCALMIN_BAND61=1.0
QCALMAX_BAND62=255.0
QCALMIN_BAND62=1.0
QCALMAX_BAND7=255.0
QCALMIN_BAND7=1.0
QCALMAX_BAND8=255.0
QCALMIN_BAND8=1.0

LANDSAT Image classification

See Image classification