|
|
i
t
|
square method in the original pixel resolution of the image.
The difference between the original pixel values and the
interpolated pixel values pixelwise was expressed by the sum
of the square of the differences. The smaller its value, the
preciser is the interpolated image compared to the original
image. This comparison was realized 20 times for each
sampling plan (Fig 3).
Based on the comparisons, we selected a sampling plan, which
was tested in the field, but due to difficulties in the precise
field localization, we selected more or less homogeneous
patches not smaller than 3 m for sampling.
Testing of field sampling plan
The most important point in the test of the sampling plans was
to provide simple and practically applicable plans for the
everyday routine of soil reclamation. Therefore the
stratification of the samples was made only on the basis of one
variable, the intensity of the image, and we examined the
estimation precision of only one chemical variable, that of
gypsum requirement. Furthermore we used the simplest
possible estimation technique, the'averaging. Therefore all the
tested gypsum requirement estimations can immediately
utilized.
As shown in Table 1 there were 134 samples, which have been
stratified into either 2 or 4 strata. First we used analysis of
variance to find out whether the sampling strata differred
regarding the GR (Table 2).
Table 2. Results of the analysis of variance of Gypsum
Requirement (GR) by sampling strata
Stratum Mean St. Dev. Sample size Significance
GR (t/ha)
1 7.26 2.42 9 <0.001
2 5.33 2.14 30
3 3.16 1.89 50
4 2.08 1.78 45
1-2 5.78 2.33 39 <0.001
34 2.63 1.90 95
Later the sample of 134 samples was considered to be the full
sample population, and we collected small samples in order to
test the sampling plans of Table 1. There were altogether 16
such sampling plans tested.
RESULTS AND DISCUSSIONS
Correlation between gypsum requirement and intensity of
the image
Based on 134 samples, the correlation fully justitied our
preliminary assumption, and proved that the intensity strata
can be used directly for the estimation of GR (Fig 4).
Results of the testing of field sampling plans
The sampling plans were evaluated regarding GR. The
sampling strata showed strongly significant difference in the
average values of elevation above sea level, local intensity and
intensity averaged for the neighborhood, as well as GR.
£ 12 —
e e
ort , ®
e
o e 9 e
TUM NU
= e 4,9 9.6
3 641 “x © oe ©
y C v^
© e M e
4 + e $9 e*
E qe e$ À e
2 | x * |r-063
2 2+ *
> | od *
Oo 0 oi AQ 9$
50 100 150 200 250
Intensity
|
Fig 4. Correlation between the intensity of aerial photograph
and the gypsum requirement
The errors of the conclusions which can be drawn from the
sampling plan is shown on Table 3. This table shows in how
many samplings we can draw erroneous conclusions from 7-7
random samples. In the left half of the table the first criterion
(in the 1-2. sampling strata GR is larger than 4 t/ha, in 3-4
strata it is less) was used, in the right half we used the first and
second one (GR diminishes from stratum 1 to stratum 4). The
relative estimation variances and the estimation errors are
shown on Fig 5.
Table 3 The erroneous conclusions which can be drawn from
the sampling plans
; 1. criterion 1. and 2.
criteria
Sample size 4-strata 2-strata 4-strata. 2-
strata
Weighted by intensity
80 - -
40 - -
20 ] -
10 4 ]
Non weighted
80 - -
40 - -
20 2 1
10 ] 1
uncut!
!
& 5 — 1
The following conclusions were drawn from it
a) the precision of estimation decreases with the reduction of
sample size from 80 to 20 generally, that is the relative
estimation variance and estimation errors increase. This
increase is not steady, but gets larger with smaller sample size.
There is no detectable difference between the 40 and 80
sample size in estimation variance.
180 International Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998
f. AS ranı uramant
RES
