each 
ectral 
NIR) 
ation 
(Lmin) 
-date 
lized 
d. 
tions, 
1g an 
DVI 
ialize 
:udo- 
areas 
rence 
lates. 
ation 
) was 
s and 
v/ice, 
d hill 
| and 
ation, 
p. In 
were 
poral 
rveys 
( and 
1ging 
-wise 
were 
ig of 
yana, 
ence, 
tricts 
rison 
1993; 
aking 
sums 
berg- 
nodel 
* Jan 
root 
ice of 
ıdard 
IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India,2002 
error (t-test) and significance of correlation coefficient was 
tested by F-statistic. 
4.7 Analysis of spectral parameters 
Time corresponding to peak profile NDVI values (T,,,) and 
width of profile i.e. time taken from first point of inflexion to 
second point of inflexion (6), were estimated (Equations 3 and 
4). 
Linear regression based relationships between district-wise 
wheat yields and spectral profile parameters were established 
for individual states as well as for the combined data set. The 
yield prediction models were developed from a subset of the 
combined dataset after excluding eight randomly selected 
districts that were used for evaluation of prediction. The 
statistical significance of the regression and the values of 
regression coefficients were tested. For the validation of yield 
prediction models, wheat yields for the 8 excluded districts 
were predicted and compared with observed yields. The pre- 
anthesis period of wheat crop was related with profile-derived 
parameter D (difference of T, and T, in Julian days). 
S. RESULTS AND DISCUSSION 
The estimated values of spectral profile parameters Ty, & and B 
for the study districts for 2000-2001 season are shown in Table 
I. Profile parameters T, corresponds to the crop emergence 
date; while a and B correspond to crop growth rate and crop 
senescence rate respectively (Badhwar, 1980; Dubey et al, 
1991). The values of T, varied between 322 and 360 Julian days 
i.e. Nov 18, 2000 to Dec 26, 2000, which are realistic estimates. 
The southwestern districts of Haryana had relatively higher T, 
(7345) indicating late emergence (hence late sowing) of crop in 
these districts; while. Ty values were lower («330) for the 
districts of Punjab indicating early sowing of crop in this 
region. Lower values of ao (36 to 60) and B (0.000106 to 
0.000166) were observed for the districts of Punjab; while the 
values were high for most of the districts of Haryana (except 
Panchkula, Ambala, Kurukshetra, Karnal and Mahendragarh). 
All the profile fits were statistically significant as tested by Chi- 
square test. The significance of values of spectral profiles 
parameters T,, ot and f was tested by t-test and the values were 
found to be significant at 95% confidence level. The values of 
Tmax Varied from 404 to 420 Julian days (that correspond to Feb 
8 and Feb 24, 2001) and o varied from 66 to 97 days. The 
values of o for southern districts of Haryana were 
comparatively lower (66 to 80 days) than those for northern 
districts of Punjab (80 to 97 days). 
The district-wise yields for 2000-2001 season were linearly 
related with observed maximum NDVIS and various parameters 
(To, o, B, T4, Gy and ©) of spectral profiles for individual 
states and for the combined dataset. The observed maximum 
NDVIs were found to be poorly correlated with district-wise 
wheat yields, but yields were highly correlated with profile- 
derived peak NDVI i.e. Gp, The district-wise yields varied 
from 2.59 to 5.16 t/ha and G,,,, varied from 0.46 to 0.64. The 
coefficients of determination (r?) were found to be 0.72, 0.87 
and 0.83, respectively, for Haryana, Punjab and The combined 
dataset. 
The regression coefficients, tested by t-test, were found to be 
statistically significant at 95%. The scatter plot of wheat yield 
With Gmax for the combined dataset is shown in Fig 1. 
Table 1: Wheat yields (t/ha) and spectral profile parameters for 
the study districts 
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
  
District T, a |B*10°| r^ | rmse | Yield 
Ambala 335.4 | 47.006 | 14.0 |0.937|0.0028 | 3.483 
Amritsar 330.1 | 40.184 | 11.7 |0.921]|0.0634 | 4.682 
Bathinda 341.3 |55.577 | 16.3 |0.946|0.0562 | 4.172 
Bhiwani 353.7 | 69.508 | 20.1 |0.969|0.0015 | 3.293 
Faridabad 356.2 | 79.410 | 23.0 |0.975|0.0014 | 4.022 
Faridkot 343.0 | 57.172 | 16.6 |0.939|0.0608 | 4.721 
Fatehabad 350.6 | 67.085 | 19.4 |0.936|0.0040| 4.351 
Fatehgarh S. |325.8 [43.198 | 13.1 |0.958[0.0497| 5.041 
Ferozepur 335.1 | 48.733 | 14.4 |0.948|0.0551 | 4.509 
Gurdaspur 327.1 | 37.932 | 11.2 |0.933|0.0559 | 4.257 
Gurgaon 350.3 | 71.790 | 21.2 |0.983|0.0009 | 3.507 
Hisar 359.8 | 79.826 | 22.7 |0.928|0.0041 | 4.283 
Hoshiarpur — |326.0|36.142| 10.6 [0.946|0.0471 [3.443 
Jalandhar 324.1 [38.026 | 11.4 [0.951]0.0483 | 4.626 
Jhajjar 354.2 | 75.659 | 22.1 |0.974|0.0014 | 3.826 
Jind 355.5 | 75.680 | 21.7 |0.952|0.0032 | 4.324 
Kaithal 345.9 | 63.537 | 18.6 |0.960|0.0028]| 4.498 
Kapurthala 327.6 |39.169 | 11.6 |0.948]|0.0504 | 4.439 
Karnal 339.0 | 56.335 | 16.8 |0.969|0.0021 | 4.635 
Kurukshetra | 333.3 | 48.789 | 14.7 |0.956|0.0024| 4.641 
Ludhiana 322.3 | 40.697 | 12.4 |0.950|0.0546] 5.159 
Mahendragarh | 339.7 | 53.605 | 16.0 |0.979|0.0010|3.712 
Mansa 337.9 | 53.756 | 15.9 |0.937|0.0619] 4.591 
Moga 327.9 | 45.155 | 13.6 |0.950|0.0566] 4.755 
Muktsar 333.5 | 49.984 | 15.0 |0.955|0.0529| 4.596 
Panchkula 331.4 | 38.287 | 11.4 [0.906|0.0029| 2.589 
Panipat 346.1 | 63.979 | 18.8 |0.962|0.0024 | 4.770 
Patiala 327.7 | 47.456 | 14.5 |0.967|0.0442 | 4.564 
Rewari 343.0 | 60.219 | 18.0 |0.984|0.0008 | 4.153 
Rohtak 355.1 | 73.171 | 21.1 |0.948|0.0028 [ 3.822 
Rupnagar 325.8 | 36.381 | 10.8 |0.922|0.0544| 3.631 
Sangrur 326.1 | 46.638 | 14.3 |0.954[0.0544 | 4.889 
Sirsa 357.3 | 77.910 | 222. |0.956|0.0027 | 4.161 
Sonepat 351.6 |69.561| 20.2 |0.953|0.0026] 3.986 
  
  
  
  
  
  
  
  
  
To predict the wheat yields, yield models were developed using 
a subset of the combined dataset by excluding randomly 
selected eight districts (more than 20% of the total). The details 
of the regression models for individual states as well as for the 
combined dataset are given in Table 2. 
The prediction models were validated by comparing predicted 
yields against the estimated yields by government agencies for 
the districts that were randomly selected and excluded from the 
development of prediction models. The rmse of prediction was 
0.181 t/ha at mean observed yield of 4.546 t/ha (3.9896 of mean 
yield), for the prediction model developed using subset of the 
combined dataset. The scatter plot showing observed and 
predicted yields, is given in Fig 2. The duration (D) between 
time of peak NDVI (T,,,) and time of spectral emergence (To) 
365