On the Use of ERS SAR Interferometry for the Retrieval of
Geo- and Bio-Physical Information
| Maurice Borgeaud
||European Space Agency
P.O. Box 299
2200 AG Noordwijk
||GAMMA Remote Sensing AG
3074 Muri BE
the present time, most of the research activity in SAR interferometry
is directed towards development and improvement of SAR interferometric
techniques, mapping of scene topography (DEM), and displacement
mapping with differential interferometry. Recently, it has been
however shown that SAR interferometry has also a large potential
for the retrieval algorithms for bio- and geophysical parameters.
Using only the coherence information derived from the complex
correlation of two co-registered ERS SAR images, it is possible
to retrieve additional information complementary to information
contained in the amplitude value of the backscattering coefficient.
Phase-unwrapping, one of the most critical steps of the interferometric
processing chain, is not required for this type of analysis.
Applications of SAR interferometry for forest mapping, forest
type discrimination, freezing, land-use classification, soil moisture
monitoring, crop classification, crop density, crop growth and
field development monitoring, monitoring of open water surfaces,
and erosion are discussed. Examples from four different test sites
(Bern, Flevoland, Middle Zeeland, and Death Valley) are shown
using ERS-1 data collected during 3-day and 35-day repeat orbits
as well as ERS-1/2 tandem data.
SAR, interferometry, ERS, coherence, forest, land-use, crop, erosion
In the last few years, SAR interferometry has become a very attractive
technique to obtain extra information from SAR images. Not only
the amplitude of the signal is considered, but its phase as well.
In order to use this technique, two SAR images of the same region,
acquired with slightly different sensor positions, are coherently
combined together. SAR interferometry can be performed either
using data collected by repeat-pass or single-pass sensors. The
former implies the same antenna is used twice while the latter
requires two distinct antennas to be flown aboard the aircraft
This paper deals with repeat-pass SAR data acquired by ERS-1 during
the 3-day (1991 and 1994) and 35-day repeat periods as well as
during the ERS-1/2 tandem mission (1995). ERS-1, a satellite carrying
a spaceborne C-band SAR, was built by ESA and launched in July
1991 while ERS-2 was launched in April 1995. Both satellites are
identical (from the SAR point of view) and acquire data over the
Earth with incidence angles varying between 19 deg and 26 deg
with slant-range pixel spacing set to 7.9 m (for the single look
complex SLC product generated by ESA).
A promising applications of SAR interferometry is to generate
digital elevation maps (DEM) owing to the fact that the height
information can be related to the phase difference between two
SAR images. However, other applications of are emerging and this
paper describes some of those for the retrieval of geo- and bio-physical
After a brief description on how the SAR data are processed, the
paper lists the test sites which used for the study and then describes
the potential of ERS SAR interferometry for several applications
including forestry, agriculture, land-use classification, soil
moisture monitoring, freezing, and erosion.
Processing of the SAR data
Interferometric processing of complex SAR data combines two single
look complex (SLC) images into an interferogram. In a first step
the two images are co-registered at sub-pixel accuracy. In the
same step common band filtering of the azimuth and range spectra
can be conducted, in order to include only those parts of the
spectra which are common to the two images, and thereby optimizing
the interferometric correlation and minimizing the effects of
the baseline geometry on the interferometric correlation. Then
the two images are cross correlated, i.e. the normalized complex
interferogram is computed. The azimuth and range phase trends
expected for a flat Earth are then removed from the interferogram.
From this "flattened " interferogram and the two registered
intensity images the multi-look interferometric correlation and
backscatter intensities are estimated. The backscatter intensity
and interferometric correlation estimators have to include a sufficiently
large number of looks in order to reduce speckle noise and biased
correlation estimation at low correlation levels. To optimize
the trade off between spatial resolution and estimation accuracy
estimators with adaptive window sizes were used. Notice that no
phase unwrapping step if necessary for the estimation of the backscatter
intensities and the interferometric correlation.
Test sites and data description
Several test sites have been used to investigate the potential
of SAR interferometry for retrieving geo- and bio-physical information.
Each of them is listed in Table 1 as
well as the ERS data used and the bio/geo-physical parameter investigated.
|Bern (CH) ||ERS-1
||Nov. 91||Forest types, freezing
|Bern (CH) ||ERS-1/2 tandem
||July 95||Forest/non-forest, forest types, land-use classification
|Middle Zeeland (NL) ||ERS-1
||Jan.-Apr. 1994||Soil moisture, freezing
|Flevoland (NL) ||ERS-1
||Sep.-Nov. 1991||Crop type, crop growth, field development
|Flevoland (NL) ||ERS-1/2 tandem
||1995||Crop type, crop growth, field development
|Death Valley USA) ||ERS-1
||Jan.-May 1993||Vegetation density, geometric change (erosion), mapping of surface types
Table 1: Test sites investigated
Applications of SAR interferometry
Some of the potential applications of SAR interferometry for the
retrieval of bio- and geo-physical information, including forestry,
agriculture, land-use classification, soil moisture monitoring,
freezing, and erosion, are examined and examples are given in
It is difficult to identify forested areas using only single-frequency
and single-polarization C-band amplitude SAR data as shown in
Figure 1 . The city of Bern can be recognized
at the bottom of the image while the lake of Bienne and Morat
are displayed on the left of the image. However, it is almost
impossible to differentiate forested from non-forested areas.
Figure 1: Part of the ERS-1 image (SLC format) over
the Bern test site acquired on 24 November 1991 (full size image)
Using the coherence information between two ERS-1 images acquired
3 days apart (in this case: 24 and 27 November 1991), one can
derive the so-called ``RGB" image to get a visual impression
of the additional information that the coherence carries (Wegmueller 1995a, Wegmueller 1995b) .
The red channel is proportional to the coherence, the green with
the mean amplitude image, and the blue with the amplitude difference
between the two images. Figure 2 shows the
corresponding RGB image for the Bern test site.
Figure 2: RGB interferometric image over the Bern
test site (full size image)
During the winter season, the coherence is quite high for agricultural
fields (mainly bare soils) while forests exhibits a low coherence
due to a three-day temporal decorrelation. Figure 3 shows
the variation of backscattering coefficient as a function of the
interferometric coherence, derived from ERS 3-day repeat-pass
interferometry for different types of land surfaces.
Figure 3: Variation of backscattering coefficient
as a function of the interferometric coherence for different types
of land surface
It can be observed that both urban areas and sparse vegetation
have a high backscattering coefficient and a high correlation.
However, forested areas, while having a similar relatively high
backscattering coefficient value, are described by a much lower
coherence. Taking these differences into account, forested areas
can easily be identified in green on Figure 2
using the RGB color scheme.
More recently, it was shown (Wegmueller 1996a) that
it might also be possible to separate deciduous from coniferous
trees using ERS SAR interferometry. From Figure 3 ,
it can be noted that the interferometric correlation is higher
for deciduous trees than for coniferous while having an approximate
constant backscattering coefficient. One should however not forget
that this result is true only for winter time when deciduous trees
have lost their leaves and hence are less sensitive to wind effects
and is based on 3-day repeat-pass interferometry. In order to
generalize this result, the different seasonal development of
the two forest types (deciduous vs coniferous) was used. Deciduous
forest sheds its leaves in fall, coniferous not. For the deciduous
forest stands this results in increased interferometric correlation
during winter time because the scatter contribution of more stable
structures such as branches, stronger twigs and the soil is increased.
The classification shown in Figure 4
was based on the interferometric correlation of the June 95, November
95 and April 96 ERS-1/2 tandem pairs over the Bern test site.
Figure 4: Forest type classification based on multi-temporal
interferometric correlation for Bern test site. June 1995, Nov.
1995 and April 1996 ERS-1/2 tandem pairs were used for the classification.
The dark green areas correspond predominantly to coniferous forest,
the bright green areas to deciduous forest. Blue corresponds to
permanently very low correlation like water and lay-over. Grey-brown
corresponds to higher correlation, i.e. urban areas and agricultural
fields. The result was only validated for a few known forest stands.
More quantitative validation is planned.
It has been recognized that ERS-SAR data can be very valuable
for agricultural applications (Wooding 1995) ,
not only because of the all-day capability of microwave vs optical
data, but also due to the fact that SAR data are very sensitive
to moisture, soil roughness, and vegetation structure.
Additional, and most of time complementary, information can be
derived from SAR interferometry. Several crops where observed
during the Fall 1991 over the Flevoland test site by ERS-1. The
temporal variation of the backscattering coefficient as well as
the interferometric coherence is shown in Figure 5.
Figure 5: Temporal signatures of interferometric correlation
and backscattering coefficient for several crops over Flevoland
(full size image)
It can be seen that abrupt changes in the coherence for potatoes
and wheat are not observed in the backscattering coefficient values.
These losses of coherence can be attributed either to harvesting
or mechanical preparations of the fields (e.g. ploughing).
In addition to the extraction of the signatures for selected fields,
multi-temporal composites were generated. Figure 6 shows
the multi-temporal image of the interferometric correlation combining
the correlation of the 19 Sep. & 4 Oct. (red channel), 4 Oct.
& 19 Oct. (green channel), and 19 Oct. & 9 Nov. (blue
channel) interferograms. The fields appear in different colors
according to its vegetation cover and farming activity occurring.
Figure 6: Multi-temporal coherence image over Flevoland
in the Fall 1991 (full size image)
Using this technique, the area can also be classified in eight
different classes depending on the value of the coherence between
two dates. If the coherence is higher than 0.5, one assumes that
no change has taken place during the time interval. On the other
hand, if it is lower than 0.5, one assumes that some kind of "change"
has occurred such as harvesting or mechanical preparations of
the fields. Using the color table definition described in Table 2 ,
the classification of the Flevoland test site in eight different
classes can be derived and is shown in Figure 7.
Forest, water areas always exhibit change of correlation, therefore
these areas are shown in black while bare soil fields (or spare
vegetation covered fields) which are not mechanically cultivated
during the whole period are displayed in white.
|19 Sep. & 4 Oct.
||4 Oct. & 19 Oct.||19 Oct. & 9 Nov.
|no change||no change
|no change||no change
Table 2: Color table definition according to a coherence
greater (no change) or smaller (change) than 0.5 between two dates
Figure 7: Map of multi-temporal change over Flevoland
in the Fall 1991 (full size image)
The temporal monitoring of the coherence may also give some information
about the status of a crop. As shown in Figure 8
for the case of rapeseed (two different fields located in the
Flevoland test site) during the growing season 1991-1992, the
coherence decreases as the soil cover fraction increases. This
can easily be explained due to the fact that as the vegetation
increase, the coherence (measured by repeat-pass interferometry)
decreases. With knowledge of the farming calendar, such information
would allow the identification of the crop. For the specific case
of the rape seed, it is possible to identify it as early as November
Figure 8: Interferometric correlation as a function
of rape seed soil cover fraction over the Flevoland test site
A promising approach would be to couple this information linked
to the biomass content with crop growth models to predict the
crop yield (Van Leeuwen, 1996) .
Using SAR interferometry and combining coherently the data acquired
on 24 November 1991 by data measured three days later (i.e. 27
November 1991), it is now possible (Wegmueller, 1996a) to
derive land-use maps, as illustrated in Figure 9 .
Land classes such as lakes, urban, forest, agriculture can be
identified with a classification accuracy of 90%. Furthermore,
lay-over areas are shown in yellow. The color table used is displayed
in Figure 10 . This classification was
made possible by combining the mean amplitude and differences
between the data as well as the coherency image acquired on both
Figure 9: Landuse map using SAR interferometry technique
with ERS-1 data acquired on 24 and 27 November 1991 over the Bern
test site. Water (blue), urban (red), forest (dark green), agriculture
(light green), and lay-over (yellow) are shown. (full size image)
Figure 10: Color table for Figure 9
Soil moisture monitoring
The scattering properties of a soil surface are dominated by its
geometry and its permittivity or dielectric constant. The permittivity
itself depends strongly on the soil moisture content because of
the very high permittivity of liquid water. The main limitation
in soil moisture retrieval is that the geometry of the soil as
well as vegetation cover influence the backscattering coefficient,
too. SAR interferometry can help to resolve the effects of the
soil moisture, the soil roughness and the vegetation cover, which
are very difficult to separate, otherwise (Wegmueller 1996b) .
As discussed above vegetation can be identified by its lower interferometric
correlation. This allows to identify bare and sparsely covered
fields. In addition, high interferometric correlation indicates
that no geometric change, that means also no surface roughness
change, occurred between the acquisition of the image pair. Even
if the surface roughness remains unknown, the fact that surface
roughness change may be excluded allows to monitor soil moisture
change more reliable. Of course, these ideas improve the situation
primarily for those fields which are bare and without geometric
change. Using a similar technique as described in Table 2 (setting
a change/no change threshold for the coherence to 0.5), a multi-temporal
image of the change of the interferometric correlation over the
Middle Zeeland test site Figure 11 shows that
such fields exist in relatively high number (identifiable by a
white color), confirming that the presented ideas are be applicable
in practise. The color table is the same as described in Table 2 ,
however different dates are used.
Figure 11: Map of multi-temporal change over Middle
Zeeland in the winter 1994 (full size image)
Under the assumption of constant surface roughness, quite reliable
soil moisture estimates may be obtained from the microwave backscattering
coefficient as was shown in the past (Borgeaud, 1995) .
The backscatter intensities of fields of high interferometric
correlation allow to monitor near-surface soil moisture. Additional
advantages of SAR interferometry with respect to soil moisture
monitoring are the possibility to use the interferometric phase
to retrieve topographic information, and to carry out geometric-
and radiometric calibration of the data.
The most drastic cases of permittivity change are freezing and
thawing events. The liquid water content changes between almost
zero for the frozen case to the usually very high values near
saturation which are typically observed during winter season.
As illustrated in Table 3 , the
interferometric correlation (cc) and the backscatter change (delta
sigma) in [dB] between the first and the second acquisition were
extracted for three fields over the Middle Zeeland test site.
The drastic decrease in the backscattering of more than 4 dB between
11 and 14 Feb at a high interferometric correlation (> 0.70)
clearly indicates the freezing event.
||Baseline [m]||Time interval [day]
||In-situ info||Field 4 [cc]
||Field 4 [delta sigma, dB]
||Field 5 [cc]||Field 5 [delta sigma, dB]
||Field 6 [cc]||Field 6 [delta sigma, dB]
|5 & 11 Feb. 1994 ||30.1
|11 & 14 Feb. 1994 ||136.8
||3||frozen on 14 Feb.
|14 & 17 Feb. 1994 ||112.9
||3||frozen both days
Table 3: Coherence [cc] and change in backscattering
coefficients [delta sigma] for three fields over the Middle Zeeland
test site during the period 5-17 February 1994.
Vegetation density and erosion
Over semi-arid to arid sites such as the Death Valley and Amargosa
Desert the interferometric correlation is high even after 35 days
or longer. Besides radar system and processing related effects,
decorrelation occurs from geometric change. In such areas mainly
two sources of geometric change occur: changing vegetation (motion,
growth, etc.) and erosion. Wind and rain change the shape of unstable
sandy surfaces. As a result such surfaces can be distinguished
from more stable surfaces with larger rocks. Figure 12 shows
the interferometric correlation (color scale from 0.1 to 1.0).
The plains of the Amargosa Desert are relatively unstable and
affected by wind and rain erosion. On the other hand larger rocks
are typically found in dry River beds and on the alluvial fans
of the Panamint Range, Amargosa Range, and Yucca Mountain. An
extreme example for erosion with very low correlation after 35
days are sand dunes as for example the Big Dune in the Amargosa
Figure 12: ERS repeat-pass ERS interferometry (35-day)
over Death Valley (29 January & 5 March 1993) (full size image)
In this paper, techniques using SAR interferometry have been presented
to retrieve several bio- and geo-physical parameters. Using only
the coherence information derived from the complex correlation
of two co-registered ERS SAR images, it is possible to retrieve
additional information complementary to information contained
in the amplitude value of the backscattering coefficient. Applications
of SAR interferometry for forest mapping, forest type discrimination,
freezing, land-use classification, soil moisture monitoring, crop
classification, crop density, crop growth and field development
monitoring, monitoring of open water surfaces, and erosion have
been clearly demonstrated.
- Borgeaud, Attema, Salgado-Gispert, Bellini, Noll, 1995:
- Analysis of Bare Soil Surface Roughness Parameters with ERS-1
SAR Data, Symposium on Retrieval of Bio- and Geophysical Parameters
from SAR Data for Land Applications , Toulouse, France, 10-13
- Van Leeuwen, Bakker, van den Broek, Rijckenberg, Bouman, Clevers,
- Vegetation retrieval by combined microwave and optical remote
sensing, ESA contract 11154/94 , Noordwijk, The Netherlands.
- Wegmueller, Werner, Nuesch, Borgeaud, 1995a:
- Forest Mapping Using ERS-1 SAR Interferometry, ESA Earth
Observation Quarterly , No. 49.
- Wegmueller, Werner, Nuesch, Borgeaud, 1995b:
- Land-Surface Analysis Using ERS-1 SAR Interferometry, ESA
Bulletin , No. 81, pp. 30-37.
- Wegmueller, Werner, 1995c:
- SAR interferometric signatures of forest, IEEE Transactions
on Geoscience and Remote Sensing , Vol 33, No 5, pp 1153-1161.
- Wegmueller, Urs, 1996a:
- ERS SAR interferometry for land applications, ESA contract
11740/96 , in progress, Noordwijk, The Netherlands.
- Wegmueller, Urs, 1996b:
- The potential of ERS SAR interferometry for hydrology, Environmental
Remote Sensing and Applications , Parlow (ed.), Balkema,
Rotterdam, (ISBN 90 54 10 5984), pp. 319-324, in progress.
Wooding, Attema, Aschbacher, Borgeaud, Cordey, de Groof, Harms,
Lichtenegger, Nieuwenhuis, Schmullius, Zumda, 1995
Satellite Radar in Agriculture: Experience with ERS-1, ESA
SP-1185 Publication , Noordwijk, The Netherlands.
Keywords: ESA European
Space Agency - Agence spatiale europeenne,
observation de la terre, earth observation,
satellite remote sensing,
teledetection, geophysique, altimetrie, radar,
chimique atmospherique, geophysics, altimetry, radar,