exception of the meteorological sector Earth observation has
not yet developed fully operational systems for monitoring the
environment; well-known programmes that use Earth
observation data, such as MARS or CORINE, do so on the
basis of space systems with a fixed life. Radar data can
contribute the technical case for data continuity while
arguments are developed for Earth observation system
continuity, as is the case within the European Global
Monitoring for Environment and Security (GMES) initiative.
2. FIELD IDENTIFICATION
2.1 Potato Monitoring
In the UK Landsat and SPOT data have been used to monitor
fields growing potatoes. The British Potato Council (BPC) has
worked with Logica UK Ltd to use Landsat and SPOT data to
map fields growing potatoes and then to compare this
information with the returns produced by the farmers about
their planting patterns (Parker and Harris 1998, Williams et al
2000). Commercial potato growers in the UK pay a levy to the
BPC each year for the right to grow potatoes, and optical Earth
observation data have been used operationally to check the
accuracy of the farmers’ planting returns. The limitation of
cloud cover in optical images does on occasion prevent the
acquisition of Earth observation data and so a research and
development project has been carried out to evaluate the role of
radar data in the mapping of potato fields. Robust and reliable
algorithms to process radar data will help to create a fully
operational information service for the British Potato Council
and contribute to sustainable agriculture.
2.2 Radar speckle and speckle reduction
A characteristic feature of synthetic aperture radar (SAR)
imagery is the presence of speckle (Ulaby et al 1981). The
speckle effect originates from the way in which the electro-
magnetic radiation is produced by the sensor and how it
interacts with the target. The electro-magnetic waves produced
by a SAR sensor travel in phase, interacting minimally on their
way to the target. After reaching and interacting with the target,
they are no longer in phase because of the different distances
travelled by each wave. They then combine either
constructively or destructively to produce a characteristic
granular effect which degrades the image and makes scene
analysis more difficult than with optical imagery (Smith 1996).
Consequently, speckle reduction is a useful first step in the
processing of SAR data.
In order to examine the different methods for speckle reduction
in the context of potato field monitoring, ERS-2 SAR data for
an area around Retford, Lincolnshire was analysed (Capstick
and Harris 2001). The data were for April, May, June, July and
September 1998 and processed with a three look PULSAR 6
processor at a pixel resolution of 23 x 21 m. Six speckle
reduction filters were assessed on the multi-temporal SAR data:
Lee, Lee-Sigma, Local Region, Frost, Gamma Maximum a
Posteriori (MAP) and simulated annealing (Stewart et al 1998).
The first four of the filters are statistical, while the simulated
annealing technique models the SAR signal as it would have
been without the speckle effect. Figure 1 shows the original
SAR data together with the results of the simulated annealing
processing.
IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India, 2002
Figure 1. a. ERS-2 multi-temporal SAR image of the Retford
area, 1998. b. Data processed with a simulated annealing filter.
2.3 Classification results
Each of the filtered images was then each assessed for its
contribution to classification accuracy using a maximum
likelihood classifier. The inputs to the classifications were the
five dates of ERS-2 SAR data suitably processed using one of
the six filters. The training data and the validation data for the
maximum likelihood classifier were created from the processed
SAR data at locations corresponding with in situ data collected
during a field campaign during 1998 coincident with the Earth
observation data collection.
The results of the impact of the filtering techniques on
classification accuracy are summarised in table 1. The table
shows the results for the original data (see also figure la) plus
three of the filters examined, namely Gamma MAP witha 7 x 7
pixel window for the filter application, Lee-Sigma with a 7 x 7
pixel window and simulated annealing (see also figure 1b).
Overall Kappa Potatoes Potatoes
accurac coefficien user producer
y % t accuracy accuracy
% %
Original 38.2 0.22 73.0 53.8
data
Gamma 60.6 0.46 76.9 79.8
MAP7x7
Lee-Sigma 60.3 0.45 78.5 79.6
7x7
Simulated 58.6 0.42 78.4 79.7
annealing
Table 1. The classification accuracy results from applying three
speckle reduction filters on the original SAR data. The Kappa
coefficient is taken from Lillesand and Keifer (2000).
The results show an overall accuracy of about 60 per cent for
the filtered data: this is a distinct improvement on the accuracy
of 38 per cent achieved with the raw data but may still be
inadequate for operational purposes. However, the
classification accuracies for potatoes (user and producer
accuracy) are near to 80 per cent and are at an acceptable level
for operational applications by the British Potato Council.
The
filte
24
The
dat:
for
dev
Sys!
des
the
SAF
Sim
anne
In s
regi
use
arez
3.1
Sate
in a
moi
soil
ther
3.2
The
con
the
scer
bacl
Con
(CE
moc
seri
deta
pres
mai
The
can
Can
COVé