renewal of data using the same method requires extensive
man-power.
Remote sensing technique can collect global land cover
information timely. Satellites such as LANDSAT series especialy
provide repetitive coverage with the advantages of a synoptic
overview. Thus remote sensing techniques provide the method to
resolve the above problems faced in carring out the renewal of
NNLI.
This paper describes feasibility studies on renewal of the
land use data in NNLI using remote sensing data and digital
processing techniques, which was carried out under contract with
The National Land Ajency of Japan.
As the first generation NNLI represents the landuse at about
1972, LANDSAT MSS data can be used for change detection. However,
the resolution of LANDSAT MSS is too large to use for classifica-
tion of land use for NNLI which has about 10m ground resolution.
Hence, high altitude aerial photographs were used for classifica-
tion.
II. STUDY AREA AND IMAGE DATA
The study area is Sendai and its suburb(long. 142 52'30"E-143
00'00"E, lati. 38 15'N-38 20'N) which covers approximately lOlkm'.
Recentry, industrial developments and urban growth are active in
this area.
The smallest unit of NNLI for data collection and strage is
10x10m“ UTM grid cell. Land use categories in NNLI are fifteen
and are shown in Table l. Figure 5 shows the land use map of
study area created from the NNLI which almost corresponds to the
state at 1971.
Two LANDSAT data covering the study area, dated on Novenber
26, 1972 and December 14,1979, were used for this study. For
classification of land use, high altitude infrared color
photographs (9in.x9in.) were acquired on October 20, 1979 with
12,000m altitude. The scale of this image is 1 to 80,000.
III. CHANGE DETECTION WITH LANDSAT DATA
A. PREPROCESSING
The preprocessings needed for the change detection are divided
into four stages, i.e. destriping, geometric correction,
normalization and smoothing. The destriping was firstly applied
for two LANDSAT images by means of histgram equalization
algorithm.
Secondly, each LANDSAT image is geometrically corrected using
system correction with the aid of ground control points within the
mean accuracy of one pixel using UTM projection. Resampled pixel
size is 50x50m^ using a nearest neighbour algorithm, so that
rectified image of the study area contained 800x800 pixels.
The normalization means a linear transformation which coincids
678
2 m E C SE | | i = E 3
Enr 770
mtm
DA
m eme
e