nces,
n cover,
often be
tograph,
h et al,
affected
between
irement,
ution of
es at 13
n situ at
ampling
ve could
nsity
"ibution
a
liminary
s linear
termined
ind these
be used
Sampling strata
The sampling strata were distinguished based on the brightness
(intensity) of the digital image scanned from a panchromatic
negative of the aerial photograph.
Fig 2. The sampling strata
First we examined the intensity distribution of the image. On
the histogram we identified the modals according to 2 and 4
strata (Fig 2). We determined the area of the individual strata,
and the average intensity inside the respective strata, which
values were used for the weighted sampling plans. The area
was about 50 ha large, and since Hungarian standards for the
reclamation of salt-affected soils prescribe one sample each 5
ha, our minimal sample size was 10. It was increased by 2, 4, 8
times in order to increase the spatial representation.
Table 1. The tested sampling plans
In the case of random sampling, the sampling points were
allocated independently from the information content of the
image.
In one series of sampling plans we intended to express the
work/investment requirement of gypsuming, so the number of
sampling points was weighted by the average intensity values
of the sampling strata. Compared to the non-weighted sampling
plan, now we had relatively more sampling points in the high-
intensity strata in order to increase the precision of predicting
gypsum requirement in these strata. The number of sampling
points determined is shown in Table 1.
stratified (2 strata)
weighted stratified
(4 strata)
Fig 3. Comparison of original and interpolated images based
on sample size of 80
The sampling plans were tested by Monte-Carlo simulation.
We generated completely random spatial distribution of 10, 20,
40, 80 points, which are the realizations of any homogeneous
spatial Poisson point process.
Sampling plan 4-strata 2-strata
Sampling size in stratum
Sample size 1. 2. 3. 4. 1-2 3-4.
Total sample 134 9 30 50 45 39 95
Sampling plans:
Weighted by intensity
80 6 20 32 22 26 54
40 3 10 16 11 13 27
20 2 5 7 13
10 1 2 4 3 3 7
Non weighted
80 3 14 33 30 17 63
40 2 7 16 15 9 31
20 1 3 8 8 4 16
10 0 2 4 4 2 8
We wanted to test the precision of the individual sampling
plans. Therefore we determined the intensity of the image in
the simulated sampling points, actually the average of the
pixels found in its 50 cm radius neighborhood. Then this value
was related to the given geometrical point and the intensity
values were interpolated for the image with inverse-distance-
Intemational Archives of Photogrammetry and Remote Sensing. Vol. XXXII, Part 7, Budapest, 1998 179