115
Symposium on Remote Sensing for Resources Development and Environmental Management / Enschede / August 1986
SLAR as a research tool
G.P.de Loor & P.Hoogeboom
Physics and Electronics Laboratory TNO, The Hague, Netherlands
ABSTRACT: In the early seventies for seme time an EME real aperture X-band SLAR with imaging on film was
flcwn in the Netherlands. It shewed many new and unexpected phenomena, in particular over the sea. It soon
became clear that for a closer investigation of these phenomena an absolute and digital system is necessary.
Being simple and straightforward and still giving sufficient coverage and resolution for the research
purposes under consideration, a real aperture system was chosen. Its final construction and implementation
required the efforts of several institutes. Radiometric and creometric errors caused by unwanted aircraft
motions can new be corrected during the data processing and resampling, resulting in a presentation of the
data in a map corrected format. The radicmetric accuracy of the system and its internal and external
calibration are discussed.
1. INTRODUCTION
Already since the early sixties side-looking radars
have been flcwn over the Netherlands with as a
result the detection of many new and interesting
phenomena (de Loor 1981). They delivered images on
film of which the respective grey tones had only a
vague undefined relation with the radar backscatter
of the observed phenomena. Therefore, when the Dutch
remote sensing program proceeded, it soon became
clear that such simple imaging systems were not
sufficient. The major advantage of radar - apart from
the fact that it can be used day and night and
through clouds - is the fact that it can produce
absolute figures for the backscatter. Good examples
in this respect are the windscatterometers in
satellites as SEASAT and the ERS-1.
Ihe Dutch remote sensing program emphasizes the
various aspects of the interaction of the sensor
(the radar in this case) and the objects observed in
order to design optimum sensors and data handling
procedures which require a minimum of signal
transmission bandwidth and -time. Therefore
groundbased measurements have been carried out on
agricultural crops (de Loor et al 1982) and the sea
surface (de Loor and Hoogeboom 1982) and a large
data base is new available in the Netherlands. In
this way an insight was gained in the use of radar
for e.g. classification, monitoring, oil detection,
etc. It was also found that the good use of radar
asks for special procedures which require accurate
observation systems. Therefore the choice was made
for a SLAR system with digital data recording,
internal calibration, and accurate absolute data
handling. A real aperture system is the least
costly, relatively simple and straightforward, and
still gives sufficient coverage for research
purposes.
2. DYNAMIC VERSUS GEOMETRIC RESOLUTION
When speaking about resolution one usually means
geometric resolution. Dynamic resolution ("contrast"
in photography), hewever, is just as important.
Measurements (de Loor et al 1982; de Loor and
Hoogeboom 1982) have shewn that in the microwave
window the variation in radar backscatter between
areas of interest (e.g. due to waves at sea, due to
different crop species on land, etc.) are small: a
10 to 20 dB at maximum. To differentiate between
these different targets (e.g. between crops) the
radar must have a high dynamic resolution: better
than J. dB, This is a seyepe requirement in itself
b-’h there is another problem due to the coherence of
the illumination we use in radar. The targets
mentioned are all compound targets with dimensions
larger than the pixel size. Within a pixel they all
contain many scatterers as e.g. the stems and leaves
in a vegetation canopy. The reflection measured by a
radar will be the vector sum of the reflections at
the individual scatterers. Because of their movement
this vector sum will vary with time. The-return
signal to the radar so fluctuates with time and its
strength varies according to a Rayleigh distribution.
This means that single independent observations can
vary considerably. This is the well knewn speckle
in many radar images. To obtain an accurate value
for the backscatter coefficient averaging over a
sufficient number of independent observations is
necessary. The more the number of independent
observations, N, the better and smoother the grey-
tone. See figure 1. The eye does better in this
respect (Moore 1979) than a machine, since
unconsciously the eye averaaps over neighbouring
pixels. An image of pixels with №=5 to 10 looks nice
but is difficult to handle for a machine (variation
between pixels: +2.5 dB).
So speckle is inherent to radar images and must be
taken into account. Among others it can obscure
texture or - wrongly - be seen as texture (Churchill
and Wright 1984). Three approaches are possible to
deal with it: averaging within a pixel, averaging
per area, or a combination of both. In the Dutch
SLAR averaging within a pixel is used. This averaging
is either over 8 independent samples (geometric
resolution 7.5 x 7.52m ) or over 30 (geometric
resolution 15 x 15 in ) which gives an accuracy per
pixel for a compound target of respectively + 2.5 dB
or + 1 dB. This is insufficient for many
applications, among others in crop classification,
and therefore Hoogeboom (1986) also uses averaging
per field.
3. MEASURES TAKEN IN THE DUTCH DIGITAL SLAR
The above assumed that the independent observations
taken by the radar are absolute values (are
radiometrically correct). This requires a series of
measures in the system, which indeed were taken in
the Dutch digital SLAR. They will now be described
shortly. This can best be done with the aid of the
radar equation:
