| Methods and Results
The Synthetic Aperture
Radar or SAR sensors are active sensors. This means that they emit
a signal (called a wave front) in the direction of the surface being
observed and register the part of the signal that is reflected back
by the surface. This enables them to register mathematically complex
images. For each pixel one gets information about the amplitude
(signal intensity), which is linked directly to the reflectivity
and the geometry of the observed area, and about the phase, which
is linked to, among other things, the optical path taken by the
radar wave between the sensor and observed surface.
During the image acquisition period the two satellites, ERS-1 and
ERS-2, were placed on the same orbit so as to cover the same area
24 hours apart. The great stability of the two satellites
orbits and a computerised superimposition process enabled this mission,
which was dubbed the Tandem Mission, to get information
about the phase shifts between the two images wave fronts.
Phasimetry or interferometry covers the measurement techniques based
on radar image phase information.
The following products are used:
- The interferogram measures the
relative difference in the optical paths of the two SAR images
wave fronts. It enables one to determine the third dimension of
the scenes observed and to generate altimetric images of the terrain
(images of the terrains elevation). These images were not
used directly in our project
Interferogram for the area of interest.
shows the phase shift between the two images of
the Tandemcouple of 19 and 20 March 1996 (not
georeferenced image, pixels: 40 x 40 m)
- The intensity image is proportional
to the intensity of the signal sent back by the target. It is
not an interferometric product per se, for it is calculated separately
for each of the two images in the pair.
- The coherence image for its part,
measures the degree of correlation between the two complex images.
This measurement can be used to determine the targets characteristics.
Intensity image of 28 May 1996, covering
the area of interest (not georeferenced image, pixels: 40 x 40 m)
Coherence image of the area of interest
to the image couple of 28 and 29 May 1996.
The covered zone corresponds exactly to
the zone covered in the intensity image
LETS PROCEED STEP BY STEP
We break data processing into two phases,
namely, the pre-processing steps and the interpretation steps.
- Coherence and intensity image generation.
The various interferometric products will not be georeferenced.
- Digitisation of the plots boundaries
in the image projection system by means of their identification
- Extraction of image statistics for each
digitised plot. By using the boundary vectors for each plot as
image masks we can determine the mean intensity and coherence
values for the pixels included in each plot.
- Relevancy analysis of the information
extracted from the images on the different dates compared with
the parameters measured in the field, i.e., plant height, ground
coverage, and soil moisture.
- Crop identification based on the coherence
and intensity images complementarity. Creation of images
composed of various layers of information, i.e., multitemporal
images of coherence and intensity and composite images of coherence
and intensity. Analysis of crop differentiation by the various
Delineation of agriculture plots on an
(right part of the image), based on an aerial photograph
(left part of the image)
Accurate to a few centimetres!
The presence of vegetation
weakens the correlation of the echoes registered by the Tandem satellites.
The more lush the vegetation, the greater this lack of correlation,
which is expressed by lower coherence. This particularity is used
to measure the vegetation.
Indeed, the strong
correlation between the mean coherence values for each plot of wheat
and the heights of the plants measured in the field enabled us to
construct a model for estimating the wheat fields height by
a simple linear regression. This models accuracy is such that
it can be used to measure crop height with an absolute mean error
of only ±7 cm!
We also saw a strong
correlation between coherence and plant height for sugar beets,
potato plants, and maize. This finding shows that it is possible
to develop similar models for determining plant height for these
As the figure of the
potato field shows, ground coverage by the crop is also closely
correlated with the level of interferometric coherence. The cover
could thus be estimated by simply measuring coherence.
DETECTION OF THE SOWIN DATE FOR BEETS
You can even see the tractor!
is strongly influenced by changes in the targets geometry
between the Tandem satellites data acquisition times. Any
activity that changes the soil structure or vegetations structure,
for example, tilling the soil, ploughing, harvesting, etc., will
cause the coherence to drop and thus be detectable. So, based on
this principle, it was possible to observe the sowing of a few plots
of sugar beets on 5 April 1996, thanks to the corresponding coherence
Here we see low coherence
for Plots 2, 3 and 6. These fields were sown between the two Tandem
acquisitions and the resulting change in the soil structure explains
the drop in coherence. Fields 1 and 5 were sown later and show a
higher level of coherence. However, the most impressive case is
that of Plot 4, where we see two different hues. These two different
levels of coherence indicate that the field was being sown as the
second image in the pair was being acquired, for the soil structure
of half of the plot (dark part) had already been altered. We could
thus even tell where the tractor was at the time of the satellites
Interferometric coherence image of the
couple of 4 and 5 April, coded in gray values (high
coherence in white and low coherence in black).
This image allows the detection of the sowing date
for beets.The limits of 6 plots with beets are delineated.
images have been used to recognise crops with excellent results.
Two types of combination of images were analysed in our study, that
is, a combination of coherence images taken at different times and
a combination of coherence and intensity images for a pair of dates.
We must point out that we did not consider
all crops in our study. So, while all the maize fields show up magenta,
we cannot conclude that all magenta fields are maize. Similarly,
all blue plots of land are not necessarily grain fields.
MULTITEMPORAL COLOUR COMPOSITES OF COHERENCE IMAGERY
|Different crops have different
vegetative growth periods. So, wheat, oats and barley start shooting
up much earlier than potatoes, beets, or maize. The differences in
these crops vegetative growth rates influence their coherence
values directly. This figure shows a colour composite of three coherence
images taken in the course of the cropping season.
The variety of colours reveals the differences in the crops
vegetative growth at the times of imaging. Thus,
- The forests appear black because of their low coherence for the
- The urban areas contain a lot of noise, for they are composed
of the juxtaposition of light pixels with very high coherence for
the three acquisition dates (due to the presence of buildings and
roads) and dark pixels with low coherence due to the presence of
light-scattering elements that are unstable over time, such as vegetation.
- Most of the fields are yellow (sum of the additive colours red
and green), revealing high coherence (little vegetation) for the
first two pairs of dates and a drop in coherence for the third pair
of dates (growth of the vegetation). This is the case for the sugarbeet
and potato fields. The wheatfields can also be discerned on this
colour composite by the slow decrease in their coherence values
as the growing season progresses. They show up as light pink, almost
white, in this image.
Colour composite image with 3 coherence
(Red channel: 19-20 March, Green: 23-24 April, Blue: 2-3 July)
COMBINED COHERENCE AND INTENSITY INFORMATION
FOR A GIVEN DATE
This figure shows a colour composite made
from the Tandem Missions paired images of 13 and 14 June 1996.
The strong contrast
in this image may be interpreted as follows:
- The town of Gembloux (in the bottom left-hand corner) and a few
villages scattered around the image appear in yellow because of
their strong signal intensities and coherence.
- The railroad is also clearly identified by the yellow line linking
Ottignies and Gembloux, amongst other places.
- The forests are coloured green. This is the result of their low
coherence, the small changes in intensity between the two acquisitions,
and their relatively high average intensities.
- The farmland offers a great diversity of colours.
- The sugarbeet and potato fields appear in green and show low coherence,
strong intensity, and a relatively large drop in intensity between
the first and second images taken by the Tandem satellite pair.
- The wheat and winter barley are seen in blue. This corresponds
to low coherence, very low intensity, and a smaller change in intensity
than that seen for the sugarbeets.
- In mid-June the maize fields are less developed than the other
crops. They thus exhibit a higher level of coherence. Their signal
intensities are low and change little between acquisitions. They
are magenta in the image.
Colour composite image acquired on 13
and 14 June 1996
(Red: interferometric coherence between the two images,
Green: average intensity for the two images,
Blue: intensity difference between the two images)
|This project shows
that the images derived from interferometric pairs of radar images
(intensity and coherence images) are very useful for analysing crops
and cropping practices.