International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Vol XXXV, Part B7. Istanbul 2004
In order to determine LANDSAT 7 vector data suitability for
updating of LTDBK 50 000, there was defined the criterion for
data selection calculating area differences and determining an
average value. The area differences are directly proportional to
theirs percent expression of inter-equivalence. The mean
percent expression is obtained 76,9%. This value is accepted as
suitability criteria for data selection. All data below this limit
were rejected as data which quality is not efficient for updating.
Value of criterion depends on quality of reference data as well.
There was defined that the updating of database using
LANDSAT 7 vector data is possible just for four object classes
excluding gardens and open pits object classes (Figure 3).
120,00
100 00 a Matching of interpretable
ERE objects of LANCGAT imagery
80.00 with LTCBK 50000 updated
2 (3)
# 60,00 as
— Minirrel limit of. data malc hing
40,00 quality for LTOBK 50000
iud updating (9o)
20.00
0,00
Figure 3. Suitability of LANDSAT 7 vector data for updating of
database
3.2 Accuracy Investigation and Statistical Evaluation
After determination the suitable topographic objects for
updating the vector database there was investigated the
planimetric accuracy of land cover features (Vainauskas, V.,
1997). The topographic features that correspond to the defined
criteria are: Water (Hydrographic objects), Build-up areas,
Forest and Agricultural land. The investigation of planimetric
accuracy is performed for such identified objects. There was
used ARC/Info software package for measurement coordinates
within created model of 115 points for each topographic object
group. The picked out points were digitized and coded,
depending on which topographic object group and data source
they are attributed (e.g., WaLa22 Wa — Water object; WaLt22
Or — point collected from LTDBK database, etc.). On a basis of
measured coordinate values of topographic objects, there was
calculated coordinate discrepancies between vector data
obtained from LANDSAT 7 imagery and LTDBK and RMSE
was calculated (Table 1).
Topographic objects group / RMSE (m)
Water Urban Forest Agriculture
m, 3,13 17,92 42,45 33,96
m, 7,32 30,76 27,50 24,57
Table 1. Evaluation of land covers accuracy from
LADSAT 7 imagery
In order to determine, does the accuracy correspond to
requirement, the analysis has been accomplished. The results of
distinctive point identification in LANDSAT 7 imagery are
influenced by: georectification error; inaccurate image
classification and point interpretation in raster data. According
to standards, RMSE should not be higher then 10,0 m.
The largest coordinate discrepancies were obtained in a point
positions where shape of object are significantly changed due to
the natural (hydrograph and forest) or human (urban and
agriculture areas) intervention.
The mathematical statistical evaluation consists of confidence
interval calculation and determination for each of topographic
object group. The interval of confidence indicates that with a
certain probability all new picked up data set will be in the
range of defined confidence interval. (Tamutis, Z., Zalnierukas,
A., Kazakevicius, S., Petroskevicius, P., 1996). Determination
of this interval, allows forecast the distribution of all model
points. The probability of confidence interval was assumed
99%. Such probability was selected to find out maximal interval
for maximum quantity of new data. The interval of confidence
is calculated on the base of linear regressive analysis. In order to
determine the suitability of analyzed data for linear regressive
analysis. The test of Smirnov-Kolmogorov was applied. This
test determines whether collected data has abnormal or normal
distribution. Properties of collected data are: coefficient
defined by theory of Smirnov-Kolmogorov test and £,,,. which
indicates the largest differences between the same point in
theoretical and practical curves (distance between theoretical
and practical curves). The decision that the distribution is
normal or abnormal is based on this distance and. the conditions
are: hypothesis HO — set of data is of normal distribution; 77 —
set of data has abnormal distribution. With a probability of 99%,
the correct hypothesis will be accepted or wrong hypothesis will
be rejected. The conditions for correct results are: if fyux 7 W,
then this set of data has abnormal distribution; if £,,, « W, then
this set of data is of normal distribution, it means that data has
the properties of normal distribution. It was defined, that all
collected data is normal distributed (e.g., see Figure 4).
—— ply
-5 -4 -3
Figure 4. Distribution of abscissa discrepancies of Water object
regarding theoretical curve
In figure 5 there are presented the upper and lower limit
tangents of coordinate discrepancies (dx, dy), also known as
regressive tangents these lines shows, that coordinate
discrepancies can not exceed this limit, with particular
probability. This is the tangent, optimally diverged from every
collected point from data set. If all collected points would
coincide with regressive tangent, discrepancies between
coordinates of the same points would be equal to zero.
Increasing coordinate discrepancies, regressive tangent will
bend; intervals extend respectively upwards and downwards.
The central part of interval presents most reliable data, but
receding to the edges, reliability of the data decreases.
458
Intern
The !
than :
classi
distril
paid t
of sur
detect
such
follov
- usaj
classe
influe
- usag
- usag
The «
maps
topog
image
then
image
suitab
By t
detern
object
mathe
calcul
featur
collec
sugge
compi
object
vectol
Accor
for lat
(depei
(Hydr
