during
equent
ioticed.
| effect
ver has
turates
srr,
se Daynr “
xs
ures of
280
Daynr
ers
ures of
haviour
~ wheat
atter in
(down-
yards).
ie time
> three
season
; sown
bunt of
2 com-
A shift
upward
. which
;. From
je noti-
natures
for sugar beet and winter wheat are shown. It is clear
that the dynamic range is in general rather low: much
lower as was expected for the L-band in agricultural
applications. In the MAC Europe 1991 campaign the
dynamic range was about 5 to 10 dB for L-band (van
Leeuwen, 1996). These time series have a range of
not more than 5 dB. Even the soil backscatter is a bit
higher than expected. The backscatter is most likely
very close to the noise level of the JERS-1 radar sy-
stem (about -15 dB). Opposite to the airborne cam-
paigns, the JERS-1 does not show clear backscatter
signatures. The contrast in the images is nevertheless
rather high, which is important for classification.
In group |, the increase in biomass (and soil moisture)
in figure 1 of leafy crops results apparently in an incre-
ase in backscatter. This occurred also in the signatures
of ERS-1 at the same period (day 130 - day 180) until
the signal saturates. For ERS-1, the signal seems to
saturate somewhat earlier than for the L-band signatu-
re of JERS-1. The increase in backscatter for L-band
coincides with the increase of biomass, especially for
potato. This is less pronounced for sugar beet.
In group lll, the cereals exhibit rather stable signals,
around -10 dB. However, maize is rather different from
wheat and barley and shows a clear increase of back-
scatter probably due to the increase of biomass (the
voluminous leaves and the elongation of the stems
during day 130-180). This increase in backscatter
during the same period in the growing season can be
noticed for the ERS-1 signature of maize as well.
4. CLASSIFICATION AGRICULTURAL CROPS
4.1 Introduction
For the growing season of 1993 a restricted number of
JERS-1 images were available, especially in the begin-
ning and at the end of the growing season. Unfortuna-
tely images taken in the middle of the growing season
were missing. To investigate the applicability of JERS-
1 SAR-images in relation to ERS-1, ERS-1 SAR-images
were selected close to the acquisition dates of the
available JERS-1 images (table 1).
The JERS-1 image of 28-03-1993 was not used for
this experiment. Part of the image shows a clear dis-
tortion in radar backscatter making automatic classifi-
cation methods nearly impossible.
In total 24 different crop types are present in the test
Site. Several crops are combined to one class. These
Crops are diverse (horticulture, other) and consist of
many small parcels. Crop types used for classification
are: potato, sugar beet, winter wheat, grass, maize,
rapeseed, barley, fruit, horticulture (mixed), other
(mixed), lucerne and bush.
149
4.2 Classification Method
All image sets were classified with a field-based classi-
fication approach (Janssen, 1994). For the JERS-1
data, an average backscatter per field was calculated.
A maximum likelihood classification was performed
using the average values per field as input. So the
classification result contains one class for each field.
The ERS-1 data set was classified using the same
field-based classification method. Results are com-
pared with the classification results obtained from
JERS-1 to investigate the applicability of ERS-1 and
JERS-1 data.
Schotten et al. (1995) showed that minimally 25 trai-
ning fields are needed applying a field-based classifica-
tion method. As we had only a restricted number of
reference fields the whole reference data set was used
as training data for 1993. Consequently the validation
deals with the available training data. So, a validation
with an independent data set could not be performed.
4.3 Classification Results
The classification results as obtained with JERS-1 data
are shown in table 2. An overall classification accuracy
of 5896 (in terms of hectares) is obtained.
Table 2. Field-based classification results (hectares) as
obtained with three JERS-1 SAR images or three ERS-
1 SAR images during the growing season of 1993.
Crop type JERS-1 ERS-1
acc-% rel-% acc-% rel-%
potato 47 65 82 71
sugar beet 82 85 40 52
winter wheat 58 78 47 67
grassland 15 42 65 82
maize 88 22 65 39
rapeseed 93 99 98 74
barley 80 40 42 37
fruit trees 86 76 51 58
lucerne 91 58 85 77
total 58 60
acc-% = percentage classification accuracy
rel-% = percentage classification reliability
Classification results for sugar beet, maize, rapeseed,
barley, fruit and lucerne are all above 80 % accuracy
(area %). Sugar beet and rapeseed also have a reliabili-
ty above 80 %.
The classification result for grass is very poor. Visual
interpretation of the JERS-1 images already indicated
the problems with the unambiguous interpretation of
grassland.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B7. Vienna 1996
