wth
rios
on).
NDzoD6 NoUL NN UZoUZ
NUzoH6 NoOC
NUzol6 MNosc
york
)OS
IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India,2002
SCENARIO-1
5000 ; 3
| aM 0.0617x * 3827.5 4-4 line
| r=0.163 en 1
s rmse - 404 en
S 4500 - e
? (
= ? 25 < m 9
3 .
o T
< - ?
2j 3500 -
= ;
= "
3000 +“ : ;
3000 3500 4000 4500 5000
OBSERVED YIELD (kg/ha)
Figure 6. Comparison of observed and simulated district-wise
wheat yields under constant crop management
practices for Haryana (2000-01 season).
Under all the three scenarios, CGMS generated grid-wise wheat
yield map were aggregated to district using proportion of wheat
in each grid as weight. Under scenario-1 there was poor
correlation between the predicted and observed district yields
with a root mean square error (RMSE) of 404 kg/ha which was
9.8 percent of the observed State mean yield (Figure 6). The
correlation became significantly positive with a value of 0.53
under scenario-2 but the RMSE increased marginally to 11.5
percent (Figure 7) indicating improvement in performance of
CGMS. Under scenario-3 the correlation further increased to
0.74 with an RMSE of 11.4 percent (Figure 8) indicating a
close relation been simulated and observed yields and a further
improvement in CGMS performance. These results highlight
the importance of accurate specification of crop management
practices, besides soil and weather, for predicting realistic yield
spatially and the prototype CGMS described provides an ideal
framework for such a purpose.
SCENARIO-2
50007 y = 0.347x+ 23488 ive
r=053 i di a
rmse = 472 uke :
4500 2
4000
3500 -
SIMULATED YIELD (kg/ha)
*
*
*
3000 4522.22 Li akt
3000 3500 4000 4500 5000
OBSERVED YIELD (kg/ha)
4
E
Figure 7. Comparison of observed and simulated district-wise
wheat yields under RS-derived varying dates of
sowing and constant fertilizer and irrigation inputs
for Haryana (2000-01 season).
| SCENARIO-3 . |
| 5000 sl
| y=0.9129x + 1.0579 Le |
| r=074 A iu line |
© rmse = 470 Pa
= 4500: AT
o
=
a
wd
ul
S 4000 +
a
n
<
2 3500
J
%
3000 + i : ius
3000 3500 4000 4500 5000
OBSERVED YIELD (kg/ha)
Figure 8. Comparison of observed and simulated district-wise
wheat yields under RS-derived varying dates of
sowing with variable fertilizer inputs and constant
irrigation for Haryana (2000-01 season).
In the present study, CGMS integrates RS information in two
ways. Gird-wise crop distribution map derived from RS data is
used as weight for computing district-level average yields.
Also, the RS-derived spectral-temporal profile based phenology
indicators were coupled to CGMS for estimating dates of
sowing and its spatial variability. The specification of RS-
CGMS-derived dates of sowing for improving CGMS
performance is similar to the model “re-initialization” strategy
(Moulin et al., 1998). The CGMS framework can also use other
RS-derived information such as spatial inputs on agro-
meteorological parameters (rainfall, radiation, temperature etc.)
and crop biophysical parameters (LAI, fAPAR etc.). The agro-
meteorological parameters can be directly used as model inputs
or for accurate interpolation of point-wise meteorological data.
The biophysical products, such as LAI, can be used for in-
season model calibration / course correction to simulate
accurate crop growth. With the availability of some of these
products from MODIS sensor such a possibility is very real
though these products need to be validated for their accuracy at
muliplte sites and time (Pandya et al., 2002).
While in this study, the NDVI as well as LAI were aggregated
to district level for the purpose of estimating DOS, it is also
possible to apply this technique at individual grid level for
capturing within district spatial variability in DOS. The study
has also highlighted the need for accurate crop identification as
profiles and estimated DOS were sensitive to errors resulting
from mixed wheat — sugarcane cropping pattern in two districts
of Ambala and Yamunanagar. Improved spatial resolution from
AWIFS (50m) is expected to resolve this problem.
The CGMS described in this study predicts actual yield in
contrast to potential and water-limited yield as implemented in
MARS program. This has been made possible due to (a) use of
a production level-3 model which incorporates water and N
stress, and (b) specification of district-level variable crop
management inputs.
