463
lOOOOO
lOOOOO
lOOOOO
lOOOOO
lOOOOO
lOOOOO
lOOOOO
lOOOOO
lOOOOO
lOOOOO
lOOOOO
lOOOOO
lOOOOO
lOOOOO
lOOOOO
lOOOOO
lOOOOO
lOOOOO
lOOOOO
33334
67890
or boro
imilar to
wer amount
confusion
aquatic
spectrally
be by use
ioro areas
Multidate
ered data
d December
reen) and
(Colwell,
temped by
he cluster
e cluster
Multidate
cated an
this was
ther land
caused by
led data,
lesome in
of the
areas, are
procedure
he 3 March
S 5 and 7
homestead
processed
non-linear
in this
the table
iltivariate
resolution
less than
boro, and
pixel is
non-boro,
impensating
is quite
resolution
significant boro
fields occurring
were good, and
This spatial
(0.4 ha) of Landsat MSS data. This fact was the
single most difficult problem in making an
inventory of boro using Landsat. However, it was
expected that in regions of
production the chances of boro
adjacent to or near each other
hence they might be detected,
clustering is a result of the use of mechanical
pumps which can each irrigate from 4 to 16
hectares.
However, in Bangladesh there are likely to be
many pixels in which the amount of boro present is
so small that it is insufficient to be
distinguished from bare soil and other categories.
This is particularly true for boro fields
irrigated by traditional (non-mechanical pump)
methods. Thus, it was expected that a boro map
based on the amount of green vegetation present
would produce an underestimate of total boro.
Boro acreages collected by local agricultural
officers were available for several administrative
units near Dhaka. For those political units, the
Landsat pixel and line numbers were obtained for
the vertices of the boundary of each area and the
Landsat data for each political unit processed
independently. Registration points were located
on Landsat produced gray maps and available base
maps and a regreseion relationship obtained
between the latitude and longitude coordinates and
the Landsat pixel-row locations.
Initial examinations of several political units
indicated that the Landsat recognized acreage of
boro was less than the reported amount. This was
not unexpected and may be the result of several
factors. Some of the boro fields, particularly
those irrigated by traditional methods, are too
small to be recognized by Landsat and will be
confused with homestead or other categories. This
situation was indicated in some of the initial
processing efforts and led to a decision to
produce maps which identified the large contiguous
boro fields but not necessarily the scattered,
small fields. That decison produced accurate maps
of the major boro areas, but underestimated the
total extent. In addition, because of the
variable transplanting date of the boro, not all
of it may be at a sufficient stage of development
to be identifed on the early March Landsat data.
It is also possible that the ground statistics
consistently overestimated the true boro amount,
but there was no way of verifying this. Current
aerial photography and ground work during the time
of Landsat acquisiton would be useful to produce
better field statistics.
A census and double-sampling approach was
implemented to make an estimate of boro over a
fairly large area in Bangladesh. In this
procedure, the entire area of interest is
inventoried using Landsat data to obtain an
estimate. Then, Landsat estimates and field
estimates are made for identical sampling units
within the area of interest. Finally, the Landsat
total census is corrected by the relationship
between Landsat and field data in the double-
sampled areas (Gilmer, 1980; Colwell, 1978).
For ten small political units called unions for
which field estimates of boro area were available,
the double-sample relationship between Landsat
indicated boro area and field estimates of area
were determined. Those results are in Table 1.
When a regression between the two estimates was
performed, the slope of the regression equation
was found to be statistically significant, but the
constant term was not statistically significant.
Therefore, a zero intercept for the relation was
assumed, and a correlation found between the two
estimates of 0.98. Figure 2 is a scatter plot of
the double-sample results. The average ratio of
Landsat to field estimates of boro area was
determined to be 0.52. A Landsat census of boro
for another area for which reported field acreages
Table 1. Boro rice digital processing double
sampling results.
Union Reported acreage
Digitally identified
Aminpur
Baradi and
182
141
Baidya Bazar
610
269
Jampur
570
278
Kachpur
1140
733
Mograpara
330
195
Noagaon
580
199
Pirijpur
1190
562
Sadipur
420
160
Sambhupura
1410
813
Sanmandi
855
430
Acreage
Reported
1410
1128
0 162 325 488 650 813
Landsat Digitally Identified Acreage
Figure 2. Scatter plot of boro reported and
Landsat digitally identified acreage for ten
unions.
were available was then made and adjusted by the
average correction factor. That area, a portion
of Rubganj Thana and the Dhaka-Narayangonj-Demra
project, had a reported boro area or 27,546 acres
and an adjusted Landsat estimate of 26,528 acres;
these are considered excellent results.
The validity and expected accuracy of employing
a correction factor to Landsat recognitions to
obtain an estimate of actual ground conditions is
a function of the reasons for the discrepency
between Landsat recognition and actual conditions.
If the reasons for the discrepency are fairly
uniformly applicable throughout the study area,
then such a procedure is valid. However, if the
discrepency is spatially nonuniform then such a
procedure may be invalid unless the sampling
strategy for the census and double sampling is
well stratified, or if the estimate is for a
sufficiently large area that there will be
compensating errors.
The reported areas available in this study and
their reliability were not sufficient to provide a
good test of these considerations. However, the
use of a census and double-sampling method with
Landsat data appears to be promising for boro rice
acreage estimates. In a future application of
this method, the field data should be available
for a wider spectrum of ground conditions and more
spatially separated. The ground data should also
have a high degree of accuracy. This could be
accomplished by utlizing aerial photography at the
time of Landsat overflight with field checks to
supply the actual field statistics.
