2000 
  
pectral 
ct crop 
of rice 
(1989- 
)sest to 
d later 
ressive 
case of 
n, late 
austard 
1isition 
timates 
al and 
' crop 
M and 
curacy 
SSR (6 
S and 
ison to 
287-88 
acy at 
-II in 
Higher 
y with 
et al. 
[ (72.5 
n) and 
t-gram 
lead to 
lution 
nstant 
i and 
| and 
spatial 
higher 
curacy 
ids to 
led is 
rs but 
X way 
nation 
: MIR 
n and 
results 
have been obtained for groundnut separation from other crops 
(Sharma et al., 1990) and rice-other vegetation separation in 
Orissa (Panigrahy and Parihar, 1992). Even in MIR region, 
better crop discriminability using TMS in comparison to TM7 
has been observed, which could be related to the higher within 
crop variability in TM7 (Dadhwal et al., 1996). 
2.2 Yield Forecasting in CAPE 
The CAPE project has explored a number of weather and RS- 
based approaches for making district-level crop yield forecasts. 
Since the forecasts are at district-level, the average district yield 
series from State Department of Agriculture is used as input. 
The models have been reviewd by type and crop by Dadhwal 
and Ray (2000). 
It was realized early in the CAPE project that (a) use of single 
best approach for each crop and study area would require a 
large effort for inter-model comparison as well as preparation 
of data bases needed for each model and (b) the relative 
forecast performance of a model changes year-to-year. Thus a 
post-forecast integration of the models based on optimal 
combination was suggested (Pandey et al, 1992). This 
approach has been successfully used for a number of crops and 
study states (Sridhar et al., 1994). Over the past decade, a 
number of crop yield forecasting models using RS inputs have 
been developed and used in making crop forecasts as 
summarized below. 
Single date, RS-based models developed for wheat (Bihar, 
Haryana, MP, Punjab, UP); Kharif rice (Bihar, Orissa, West 
Bengal); Mustard (Assam, Gujarat); Sorghum (Karnataka, 
Maharashtra) and Cotton (Gujarat, Maharashtra) explain yield 
variability ranging from 6 per cent (for Sorghum in Karnataka) 
to 83 per cent (for Wheat in MP). A suite of normalization steps 
that account for differences in sensor, atmosphere and date of 
acquisition have to be used before a spectral index-yield 
relation can be developed (Sharma et al., 1993). 
Spectral profile related growth parameters derived from multi- 
date AVHRR data have been used for crop yield modelling for 
Wheat (Punjab), Rice (Orissa) and Sorghum (Maharashtra) 
(Dubey et al., 1991, 1994; Potdar 1993). For Cotton in Gujarat, 
multi-date LISS-I data have been used to generate spectral 
profiles. 
Combination of different parameters like trend, RS and 
meteorological parameters either by including all in a multiple 
linear regression equation (Wheat in Punjab, UP) or by optimal 
combination of different estimates (Wheat in MP, Mustard in 
Gujarat) improved the predictability of models. 
2.3 CAPE Forecasts 
During the CAPE project a rich experience on performance of 
various models for accuracy of production forecasts has been 
gained. The salient conclusions are: (a) It is possible to meet 
90/90 or even 95/95 accuracy criterion at state or group of 
districts level, however for each district such accuracies were 
not 
Figure 2. Summary of wheat acreage and production forecasts 
using RS data in Haryana and their comparison with 
State Department of Agriculture estimates. 
observed. (b) Using optical data, the kharif-season forecasts 
cannot be always guaranteed due to cloud cover. The accuracy 
varies with crop dominance, lower accuracy is observed if crop 
IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India, 2002 
315 
as proportion of geographic area is low. As an illustration, the 
relative deviation between RS forecasts and official estimates 
for wheat crop in Haryana are given in Figure 2. Early results 
from CAPE are given in Dadhwal et al., (1991), Panigrahi et 
al., (1991), Mahey et al., (1993) and Sridhar et al., (1994). 
3.0 PROPOSED FASAL PROJECT 
As the forecasts from CAPE project were evaluated by DAC, 
Ministry of Agriculture, a need for national scale, multiple, in- 
season forecast was felt. This consideration led to examining 
the relative strength and weakness of each system of 
information gathering on crops, which led to formulation of an 
integrated approach called ‘Forecasting Agricultural output 
using Space, Agrometeorology and Land-based observations’ 
(FASAL). As remote sensing, weather and field observations 
provide complementary and supplementary information for 
making crop forecasts, FASAL proposes an approach which 
integrates inputs from the three types of observations to make 
forecasts of desired coverage, accuracy and timeliness. The 
concept of FASAL thus strengthens the current capabilities of 
early season crop estimation from econometric and weather- 
based techniques followed by use of remote sensing data at 
appropriate stage. FASAL will enable multiple forecast of 
acreage as well as yield besides tracking the crop growth as the 
season progresses. 
Acreage forecast could begin with use of econometric models 
or weather models and then use low-resolution remote sensing 
data once the crops have emerged. To overcome the limitation 
of spatial resolution, use of higher temporal (multi date) 
resolution has been proposed. Small area i.e. district or smaller 
administrative units will compulsorily require use of sampling 
approach and high resolution RS data. The multiple yield 
forecasts envisaged in FASAL could sequentially come from 
time-series forecast, truncated weather-crop models, and RS- 
weather combined models. Multiple weather and RS inputs 
would also run crop simulation models in a Crop Growth and 
Monitoring System (CGMS). A continuous crop assessment 
with RS, weather and other inputs is thus proposed. 
In addition to proposed integrated use of multiple source of data 
for crop forecasting, FASAL also aims at institutionalising the 
operational use of RS data for diverse applications in 
agriculture while using other modern tools such as GIS, large 
databases, modelling and networking to provide accurate crop 
forecasts. 
4.0 NATIONAL WHEAT PRODUCTION FORECASTS 
The experiment to issue multiple pre-harvest national wheat 
production forecasts with state level dis-aggregation was 
initiated in 1996 (Oza et al., 1996). It uses multi-date data from 
Wide field Sensor (WiFS) onboard Indian Remote Sensing 
Satellites (IRS) 1C, 1D and P3 for wheat area estimation and 
regression-based wheat yield models at meteorological 
subdivisions between fortnightly temperatures and deviations of 
wheat yield from technology trend (Vyas et al. 1999). The six 
study states (UP, Punjab, Madhya Pradesh, Haryana, Rajasthan 
and Bihar) contribute 91.78 percent to national wheat 
production (average for years 1995-96 to 1999-2000) and for 
the remaining states historical trends are used. 
The sampling approach for wheat inventory, implemented in 
Arc/Info, consists of stratified random sampling with two-stage