Professional Documents
Culture Documents
Interpretation Guide
for
ALOS PALSAR / ALOS-2 PALSAR-2
global 25 m mosaic data
Version 1.1
1 October, 2016
1
2
1 Briefly about Synthetic Aperture Radar
While optical remote sensing sensors function similar to the human eye – they are
passive sensors which record reflected sunlight – a radar sensor operates more like a
flash camera in a dark room. The radar emits a light pulse and records the part of the
pulse that is reflected, or scattered, back to the sensor (hence the term backscatter).
Unlike sunlight which is non-polarised and comprises a large range of different
wavelengths, the radar is a laser which operates within narrow and well-defined
wavelength bands, and at a specific polarisation.
The most common present and near-future spaceborne radar systems operate with the
following bands:
• P-band: ~23.5 cm (BIOMASS)
• L-band: ~23.5 cm (ALOS PALSAR, PALSAR-2, SAOCOM-1A/B)
• S-band: ~10 cm (NovaSAR)
• C-band: ~5.6 cm (Sentinel-1, Radarsat)
• X-band: ~3.1 cm (TerraSAR-X, TanDEM-X, COSMO-SkyMed)
The choice of wavelength band strongly affects what type (size) of objects the radar is
sensitive to. As a rule of thumb, the radar “can see” objects of about the same spatial
magnitude as the radar wavelength, and larger. Objects significantly smaller than the
radar wavelength on the other hand, become transparent to the radar. The smaller the
object, the less influence on the backscatter.
Longer wavelength radar signals (L-band) consequently penetrate through the forest
canopy (the small leaves are invisible) and interact with the larger structures such as the
trunks and larger branches of trees – and hence display a positive, although limited,
correlation with above-ground biomass. Systems operating at shorter wavelengths
(C-band, X-band) on the other hand, are more sensitive to sparse and low vegetation.
1.2 Polarisation
The radar polarisation is another parameter affecting the strength of the backscatter.
Current spaceborne radar systems operate with linear polarisation, where the radar
signals are transmitted and received at horizontal (H) and/or vertical (V) polarisation.
3
The polarisation of SAR imagery are commonly denoted by two letters, the first
indicating the transmitted polarisation and the latter the received polarisation.
1.3.1 Sigma-naught
σo = 10 * log10(DN2) + K
where DN is the image pixel digital number measured in the SAR image (or more
accurately, the average pixel value over a group of pixels). K is a calibration factor which
varies depending on the SAR sensor and processor system used. For ALOS/PALSAR
and ALOS-2/PALSAR-2 data provided by JAXA, the calibration factor is -83.0 db.
1.3.2 Gamma-naught
Even for homogeneous targets, σo varies slightly depending on the angle between the
ground and the sensor – the incidence angle – being higher (brighter) in the near-range
part of the image (closest to the satellite) and lower (darker) in the far-range of the image,
further away from the satellite. By normalising σo with respect to the incidence angle we
can remove the range-dependency to obtain γo (gamma-naught):
γo = σo/cosϕ
4
2 ALOS PALSAR and ALOS-2 PALSAR-2 mosaics
ALOS/PALSAR mosaics have been generated for four years: 2007, 2008, 2009 and 2010
while one ALOS-2/PALSAR-2 mosaic, from the year 2015, has been generated to date.
The global mosaic data can be downloaded free of charge from JAXA at:
http://www.eorc.jaxa.jp/ALOS/en/palsar_fnf/fnf_index.htm
The ALOS PALSAR and ALOS-2/PALSAR-2 data sets provided with the Kenya Data
Cube comprises six separate images/layers:
• Radar backscatter (γo) – HH polarisation
• Radar backscatter (γo) – HV polarisation
• RGB false-colour composite (Data Cube only)
• Observation date image
• Mask image
• RGB image in KML format for visual display in Google Earth (Data Cube only)
All data layers have the same (0.8 arc sec) pixel spacing, which for most practical
purposes can be considered equal to 25 m in countries along the Equator.
5
2.3 Data Layer description
Radar backscatter is provided as two separate images: one for the Horizontal-Horizontal
(HH) polarisation, and one for Horizontal-Vertical (HV) ditto. The backscatter images are
identical to those provided by JAXA, with the exception that the Kenya Data Cube
images are provided in GEOTIF format.
The backscatter images constitute the data files on which all analysis should be
undertaken.
SAR backscatter data have a large radiometric dynamic range with DN values ranging
up to several thousand and are therefore provided as 16 bits unsigned integer data
(accommodating DN values 0 ~ 63000).
6
As mentioned above, the linear backscatter values in the HH and HV images can be
converted from image digital numbers (DN) to γo, and expressed in decibel (dB), by the
formula:
γo can vary from around -35 dB (DN ~500) in very low backscatter areas, up to about -5
dB (DN ~8,000) for extremely high backscatter. γo is generally negative and very seldom
reaches 0 dB.
G: HV γo backscatter
B: HH/HV ratio
Data type: 8 bits (DN 0~255)
Format: GEOTIF
Size: 4500 pixels x 4500 lines per channel (60.8 MB)
File name: AAABBBB_YY_[MOS]
The RGB image is false-colour composite generated for visualisation purposes only. It is
not included amongst the original data lays provided by JAXA.
7
The red and green image channels contain the HH and HV backscatter images
respectively, while the third (blue) channel is synthesized from the former two. The
HH/HV backscatter ratio is commonly used for the blue channel as it results in a
composite image in which vegetated areas appear green.
Note that since the HH and HV backscatter data in RGB image have been scaled to 8 bits,
the formula to calculate γo in 2.3.1 above is not applicable.
As the mosaic tiles are composed of SAR data from multiple satellite paths, an
observation date image is required to provide information about the date of acquisition
for each pixel in the mosaic tile.
The (16 bits) pixel digital numbers in the observation date image correspond to the
number of days after the launches of ALOS and ALOS-2 satellites. ALOS was launched
on Jan. 24, 2006 (used for the 2007, 2008, 2009 and 2010 mosaics), and ALOS-2 on May 24,
2014 (used for the 2015 mosaic).
8
The Digital Numbers 409, 470 and 498 in the ALOS-2 mosaic image shown in Figure 3
above thus correspond to:
24/05/2014 + 409 = 07/07/2015
24/05/2014 + 498 = 06/09/2015
24/05/2014 + 498 = 04/10/2015
The mask image provides information about pixels affected by radar shadowing and lay
over, as well as the location of water bodies and no-data areas. It is identical to the mask
image provided by JAXA, but in GEOTIF format.
9
2.3.5 Google Earth visualisation
Figure 5. KML
An RGB false-colour composite image is provided for visual display in Google Earth. It is
not included amongst the original data lays provided by JAXA.
10
3 PALSAR (L-band SAR) quick interpretation guide
3.1 Caveats
Below follow examples of how some typical land cover types found in the Kenyan
landscape appear in L-band SAR data, or more specifically, in ALOS PALSAR/ALOS-2
PALSAR-2 mosaic data at HH and HV polarisations.
It should be noted that the backscatter images in the figures have been scaled from the
original 16 bits data. While the HV backscatter typically is 5 – 10 dB lower (darker) than
that of HH, the two channels can look similar in the examples as different scaling factors
have been used to maximise the visual impression.
The backscatter values shown below the images have however been measured in the
original (16 bits) images – averaged over >1000 pixels to reduce the influence of speckle –
and should hus indicate “true” estimates of the backscatter. The backscatter standard
deviation is given in brackets when applicable.
Also an RGB colour composite image is shown for reference, although it is important to
keep in mind that there are neither any standardised rules on how represent microwave
data visually, nor how to compose a 3-channel colour image from only two input
channels. Yet, the [HH, HV, HH/HV] combination used here for [R, G, B] is frequently
utilised by remote sensing users.
L-HH L-HV R: HH
G: HV
B: HH/HV
11
Sparse Forest
HH HV R: HH
G: HV
B: HH/HV
Agriculture
HH HV R: HH
G: HV
B: HH/HV
Agricultural areas typically have a purple appearance in the RGB composite. The
agricultural landscape can also be identified by geometrical patterns and textural
features in the image, such as fields, roads and waterways. Small patches of trees appear
green due to higher HV backscatter.
12
Irrigated rice
HH HV HH/HV
HH HV R: HH
G: HV
B: HH/HV
13
Rocky terrain
HH HV R: HH
G: HV
B: HH/HV
Water
HH HV R: HH
G: HV
B: HH/HV
14
Acknowledgements
15