I
1330
• 4^'
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B4. Beijing 2008
Usually in cartographic or surveying approaches (e.g. parcel
area determination) data quality (e.g. accuracy of calculated
area) is estimated by RMS. National regulations determine
acceptable discrepancies according parcel area, elongation or
calculation method. In case of visual interpretation of remotely
sensed imagery not only factors mentioned above are important.
The results are influenced by the process of photo interpretation
as well. In this in mind we decided to test another approach -
ISO 5725-2 standard usually applied to chemical measurements.
ISO 5725-2 gives “the basic method for the determination of
repeatability and reproducibility of a standard measurement
method”. It is typically designed for interlaboratory
experiments in order to estimate repeatability and
reproducibility of measurement method (of chemical contents
for example), but it can also be used for other purposes.
Hejmanowska et al. (2005) presented its application to
repeatability and reproducibility assessment of orthophoto-
based measurements done during the experiment for validation
of land parcel areas measurement methods. Results of the
research were presented in the final report of the project
supported by Joint Research Centre (JRC) in Ispra in Italy
where. In the report the huge experiment of agriculture parcel
measurements with accuracy assessment was described.
Statistical analyses involved in the project, based on ISO 5725,
were inspired by S.Kay and J.Delineé from JRC. In a typical
application of ISO 5725-2 standard for basic interlaboratory
experiment, samples from q batches of materials representing q
different levels of the content to be measured are sent to p
laboratories. Each laboratory obtains n replicated results under
repeatability conditions for each of the q levels. In the case of
photo interpretation several test areas should be measured on
different days by different operators.
In a typical interlaboratory experiment critical examination of
the data is based on a “pooling factor”: the observations are
grouped according to this factor and mean and standard
deviation within a given group is compared to the means and
standard deviations within other groups. Typically the pooling
factor is the factor “laboratory”, because in each laboratory
replicated results are obtained under repeatability conditions.
For remote sensing applications measurements made by the
same operator on several days can be considered as made under
repeatability conditions. As a consequence, the data should be
pooled by operators.
3. TEST AREA AND DATA
Measurement experiment was performed applying the remote
sensing images covering region near Cracow, on the south of
Poland. Two kinds of data were applied: IKONOS PAN (called
PAN in the paper; pixel size of lm) and colour IKONOS PAN-
SHAR (called RGB in the paper; pixel size of lm) registered
7.05.2003 and delivered as 16 bits GEOTIF. Besides,
panchromatic airborne orthophoto (0.2m - pixel size) generated
from photographs in scale of 1:13 000 was applied as a
reference.
Test area is composed of 6 rectangles of 300x300 m (10x10
Landsat pixels): 2 dense urban (3 rd and 4 th test area, ca. 60% of
impervious surface), 2 industrial (2 nd and 5 th , ca. 80% of
impervious surface and 2 suburban (1 st and 6 th , ca. 30%),
(Figure 1). Test area no. 1 with 30 m grid overlaid on the
IKONOS (RGB) image is shown as an example on the figure
(Figure 2).
sv'iRMmmü
Figure 2 Test area no. 1 with 30m grid overlaid on the image,
right also with reference area of impervious surface digitised on
air orthophoto (0.2m - pixel size).
4. METHODOLOGY
Impervious surface area was digitised on the test area 1, 3, 4
and 6, and pervious surface in the case, when the surface is
much smaller then impervious one (test area 2 and 5). Two
groups of operators were chosen: 3 specialists and 3 beginners.
Operator digitised firstly IKONOS PAN (3 times) then
IKONOS RGB (3 times). Finally, specialist digitised
ortophotomap with pixel size of 0.2 m and reference impervious
surface was collected for each of six test area (Figure 2 right).
Data analysis is composed of 2 parts: accuracy analysis of
photointerpretation of IKONOS images (PAN, RGB) and
research of its influence on its averaging in 30 m Landsat pixel
size.
4.1 Part 1 - accuracy analysis of IKONOS
photointerpretation
Results of the remote sensing images photointerpretation could
be compared in different aspects: object recognition (object
recognised or no), variation of the border shape or variation of
area of the recognised object. Photointerpretation is both, time
and cost consuming, so there is difficult to find in the literature
some information about accuracy of the process, based on
fotointerpretaion made by many operators. Usually
fotointerpreatation is made once, and it is treated as a reference.
One can assume, that the photointerpretation process is biased
by operator, and accuracy depends on the experience of the
person performing interpretation. In approach basing on the
remote sensing, accuracy is usually determined by RMS,
calculated from differences between mean value of