cal yield
> medium
data was
T Season.
-divisions
forecasts
>ason.
forecasts
ifferences
ilable by
ailable by
be made
vely.
ng major
iso to be
) the user
ing states
sthan and
o national
99-2000).
ated from
on of area,
ble 1.
with two-
ch state,
and three
cating the
portions.
IAPRS & SIS, Vol.34, Part 7, “Resource and Environmental Monitoring”, Hyderabad, India,2002
Table 1: Major wheat producing states & their contribution to
national wheat acreage and production (Avg. 1995-
96 to 1999-00)
Wh
eat | Wheat Area Cumm Prod Cumm
STATE Area Prod Area (%) Prod (%)
( M ha)| (M €) | (9o India) "| (9 India) s
UTTAR
8. ; : ; : ;
PRADISH 77 |23286] 3331 33.31 33.25 | 3325
PUNJAB 33 [1334] 1251 45.82 20.13 53.38
HARYANA | 2.11 | 8.18 8.02 53.84 11.9 65.28
MADHYA
PRADESH | 405 | 76r] 1651 70.35 11.07 | 76.35
RAJASTHAN | 2.55 6.52 9.68 80.03 9.48 85.83
BIHAR 204 14051. 7.0 87.76 $96 | 91.78
Remaining |. | 565 | 1224 100 8.22 100
States
INDIA 2634 | 68.74 | 100 100
Cumm:- Cummulative
A, B, C indicate segments having crop proportions more than
70 percent, 30 — 70 percent and 10 - 30 percent respectively.
The sampling fraction was 10 percent and size of the sampling
unit was 15 x 15 km. The sampling scheme was implemented in
ARC / INFO GIS software. District boundary coverage at
1:250,000 scale in polyconic projection, a ‘Fishnet’ coverage
with grid size of 15 x 15 km was overlaid. Attribute information
added to this coverage was identification of each segment, name
of the district, meteorological subdivision, state, rabi crop
proportion, segment type A, B, C, latitude and longitude of the
top left corner of the segment. List frames were generated for all
population segments within a state. Sample segments were
selected randomly within each state using stratified random
sampling procedure. The sampling plan is summarized in Table
2. A total of 323 segments were selected from a total possible
sample frame of 2807 segments.
Table 2: Details of Sampling Plan for Study States (Population
and sample Allocation)
State |Met. Zone Population Allocated
A | B | C |TotallA | B|C [Tot.
Bihar 9 156| 94 | 64 | 314 |18]|11| 7 | 36
Haryana 13 132] 44 | 17] 193 ]15| 512 | 22
M.P. 19 81 |218|195| 494 | 9 |26|23, 58
M.P. 20 0 [45]|114| 159 |0| 5 |13| 18
Punjab 14 156| 51 [13 | 220 |18| 6 | | | 25
Rajasthan 17 12:4 35 | 19] 66 | 1 | 4/21 7
Rajasthan 18 71 | 96 |165| 332 | 8 |11/19j 38
UP. 10 366|159| 40 | 565 |43|19| 4 | 66
UP. 11 270|110| 31 | 411 |32/[13| 3 | 48
UP. 12 131 [1{ |] 29] 53 [11 | 3 | 5
3.2 Digital Image Analysis
The digital image analysis carried out for wheat acreage
estimation has been shown, as a flow diagram, in Fig 1. It can
be summarized in following steps :
3.2.1 Creation of geo-referenced master data-base by
developing a transformation model between images and
corresponding map coordinates:
In 1995-96, a master spatial data base was created by
georeferencing one-date WiFS data with Survey of India
topographic maps at 1:250,000 scale.
3.2.2 Registration of multi-date WiFS data with geo-
referenced master data-base and radiometric normalization:
Multi-date WiFS data for thirteen zones were registered with
master data and scene-to-scene radiometric normalization was
achieved by matching the digital counts of Pseudo Invariant
Features (PIFs) like manmade in-scene elements, dense built-up
area, deep and clear water-body, dry sand. This technique
corrects for atmospheric degradations, illumination effects, and
sensor response differences in multi-temporal multi-spectral
imagery (Schott et al 1988). PIF approach was used for
radiometric normalization.
3.2.3 Identification, classification, refinement and
evaluation of wheat and other crop classes based on ground
truth;
Crop phenology and vigour variations make multi-date WiFS
data more heterogeneous. Collection of representative sites for
each possible spectral class needed for maximum likelihood
classifier becomes very difficult. In cases when both the
number of features and the number of classes are large, the tree-
classifier approach or hierarchical decision rule based
classification procedure gives better performance (Wang, 1986;
Swain and Hauska, 1977). Hence, a hierarchical rule-based
classifier was attempted to classify the multi-date WiFS data.
To make classification simpler, firstly zero fills, cloud/haze,
snow/ice, cloud shadow were masked out. Then based on multi-
date NDVI [=(NIR-IR)/(NIR+IR)] non-vegetation, non-
agricultural and non-wheat classes were successively identified
and removed. Having ground truth over representative sites
wheat could be identified and discriminated form other
competing crops like gram and mustard due to its distinct
phonology.
3.3 Wheat Area Estimation
Wheat mapping involves digital image processing after creation
of a geo-referenced multi-date image database of 6-8 dates of
WiFS data during crop season. The scientific rationale for
wheat separability using multi-date data is illustrated in fig 2.
Crop identification was based on a hierarchical decision rules
using knowledge of the temporal spectral profiles of various
land covers including wheat and other crops. Wheat area was
obtained by estimating segment-wise wheat proportion for sam-
ple segments and applying appropriate sampling method of ag-
gregation.
3.4 Yield Forecasts
Weather plays a dominant role in crop growth and development
and hence in crop production. Weather variables, therefore, can
be conveniently used as indicators of change in crop yield. The
relationships between weather variables and crop yield can be
modelled by regression analysis. Such crop weather models for
providing real time crop estimates have been reported by Khan
and Chatterjee (1987) for rice, Appa Rao (1983) for wheat and
Dubey et al. (1995) for cotton.
In the present study yield forecasts were made using
temperature data at meteorological sub-division level.
Regression based models were developed using weekly weather
