eld X | 
old Y | 
1 Line | 
— 
  
  
neasured 
is: 1996 
to nine 
d on the 
ckscatter 
se were 
gust 12- 
ust date 
7l was 
AR and 
o adjust 
s (Xu et 
nakes a 
I. 
imental 
for two 
used in 
d SRM 
vertical 
295 and 
  
  
  
  
n, there 
d sugar 
  
IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India,2002 
  
Figure 3: Comparison of observed sugar yield with various 
predicted for sugar beet fields in 1995 and 1996 
  
  
  
  
  
  
  
14 + = 
= Y 
S B 
(Eae D X o 
= as IA us 
= a o n 4 ERS 
- D. WA, | 
© toi 
2 10 A A SRM 
? ° 4 © SPOT 
2 — 1:1 Line 
P "| 
& 
6 ! + + 
6 8 10 12 14 
Observed sugar yield (t/ha) 
  
However, no significant relation was found between observed 
and predicted sugar yield by other methods. It may be due to 
1995 Field B, where water stress was observed and the water 
cloud model predicted an LAI value vary different from the 
observed value. 
8.3.2 Extension of sugar yield prediction to other fields in 
1995 
In 1995, a questionaire, along with a SPOT and an ERS SAR 
image was sent to farmers growing beet on 25 fields, asking 
them to send information about variety, soil type, sowing and 
harvest date, irrigation, root and sugar yield. A complete 
response was received for 12 fields and this information was 
used to predict yield by Werker and Jaggard model (1997) 
(Table 3). 
Table 3: Sugar yield prediction for 12 Sugar beet fields near 
Broom’s Barn in 1995 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
Serial |IACS |Sowing |Harvest |Observed|ERS |SPOT 
No.  |no. Yield 
(t/ha) 
I|TL .|23.3.95 |01.10.95 6.31: 777 8.9 
7565 
A|TL  |20.3.95 |02.12.95 6.8| 11.8 114 
7564 
3|TL  |18.3.95 |22.10.95 6.7] 8.4 9.3 
7469 
4|TL [19.3.95 |28.10.95 6.4| 103 8.1 
7468 
SITL- 127.395 {10.1095 7.9| 89} 10.6 
9269 
6|TL [30.3.95 [10.12.95 8.5} 10.9 10.2 
9369 
7[TL  |02.4.95 |20.12.95 8.3}; 10.5 11.9 
9268 
S,1L_..124.3.95 |12.11.95 | 13.0| 14.5 14.8 
7967 
9iTL = 122.3.95 {15.12.95 11.0| 12.6 12.1 
7868 
IO TL. .|21.3.95. 125.12.95 6.0] 9.9 7.2 
7767 
11|TL. |27.3.95 {20.10.95 69] 9.7 8.8 
7365 
12|TL. [29.3.95 [28.10.95 6.6] 11.4 7.4 
  
  
  
  
  
  
  
  
  
  
[7364 | | | bdo os] 
Correlation coefficients of 0.77 and 0.88 was found between 
observed and predicted sugar yield by ERS SAR and SPOT 
respectively. No correction factor (k) was used: this accounts 
for yield losses during harvesting and storage. This may be the 
reason that predicted yield was larger than observed yield. 
8.3.3 Extension of sugar yield prediction to other fields in 
1996 
In 1996, fifteen sugar beet fields were used to predict sugar 
yield. This year only one SPOT image was available due to 
cloud cover. Nine sugar beet ground truth data were taken from 
a map of land use made during the summer. 
Correlation coefficient of 0.72 was found between observed 
sugar yield and predicted by combined data (SPOT and ERS) 
(Figure 4). This is agreement with other studies, like Dockter 
and Kuhbauch (1990) who suggested combining SPOT and 
SAR data to predict yield with weather data. Combinations of 
radar and optical data also have the advantage of filling the 
gaps between cloudy periods and can improve yield predictions 
(Bouman, 1992; Kohl et al., 1994; Wooding et al., 1997). 
  
  
  
  
  
© 11 e 
S © 
ë ; y 
UO 
$ 10 + 
5 o [© 1996 
o 9 —— 1:1 Line 
a o 99 o 
Ys > 
> o 
o 
2 o o 
o 
& 74 
6 + + i + + 
6 7 8 9 10 11 12 
Observed sugar yieid (t/ha) 
  
  
  
Figure 4: Observed vs predicted sugar yield by combined ERS 
SAR and SPOT for 15 sugar beet fields near IACR-Broom's 
Barn in 1996 
9. DISCUSSION AND CONCLUSIONS 
Comparison of the predicted values and measured data together 
with their error analysis for 1995 and 1996 experiments 
demonstrated that ERS-1 and ERS-2 SAR data can be used in 
the water cloud model to estimate LAI with an acceptable 
accuracy and thus has the potential for operational application 
to predict sugar beet yield. The modified version of Leeuwen 
and Clevers’ (1994) model (Xu et al., 1996) worked well for 
the 1995 and 1996 experiments. 
ERS SAR data can be used in crop monitoring and yield 
prediction. The predictions are less accurate than with optical 
data but cloud cover does not affect the data availability. For 
example, in 1996 only one SPOT image was available for the 
season but there could be as many radar images as there were 
overpasses. SPOT data are also expensive. 
The yield prediction model needs to be improved by including 
a correction factor (k) to accommodate storage and other losses 
after harvest. It also needs to be robust and should be tested in