4.Method
4.1 Extraction of flood information
There is a clear distinction between water and
surrounding objects on radar image. However, the
transmission capability of radar results in smooth of
echo signals along the flood boundary, which
makes it difficult to determine the thresholds and
decreases the accuracy of flood extraction. The
Nenjiang River course near Yueliangpao Lake and
the non-inundated areas along the left bank were
picked out and several perpendicular lines were
drawn down the river bank (see Figure 2). In the
profile produced, where x-axis and y-axis
respectively represent the vertical line and the
intensity of radar echo, there exist the considerable
differences of radar echoes between floodwater and
non-inundated areas. The pixel values of water and
its neighbor objects were averaged and the result
was regarded as the threshold of land-flood
discrimination (see Figure 3). After the five
thresholds were calculated, the flood areas were
extracted from the radar images. The useful method
was validated through the field survey.
Figure 3. Profile of water-land threshold calculation
From the results of flood extraction the flood
mask images were built, in which flooded area and
the non-inundated area were evaluated 1 and 0
respectively. The water body information in normal
years was extracted from the TM images using the
maximum likelyhood classifier and the water mask
image was also built. Owing to the limiting
accuracy of control point selection and flood
threshold calculation on those images, the errors of
the radiation corrections for the antenna pattern on
the radar images, the flood boundaries extracted
from the multi-temporal Radarsat images
contradicted each other sometime and somewhere,
namely, the inundated areas on some image might
be regarded as the non-inundated lands on the
consequent ones. Such logical errors basically
occurred on the boundary regions between
inundated and non-inundated lands and should be
corrected.
In the light of the logic errors, the idea, similar to
MVC used to remove cloud contamination on
NOAA images, was borrowed and applied. We
numbered each radar image in the order of
receiving time and evaluated the flood data by the
following expression:
Pixel value of flood image = (6 - n) * 40 * m
where m indicates the mask image.
Table 1. The number of images
| IMAGE TYPE |
TM |
RADARSAT |
| Receiving Date |
May-August/1996-1997 |
Aug.9 1998 |
Aug.16 1998 |
Aug.20 1998 |
Aug.23 1998 |
Aug.29 1998 |
| No. (n) |
0 |
1 |
2 |
3 |
4 |
5 |
The pixel values of water body are 240 in normal
year, the ones of the flooded regions are evaluated
200 in the image on Sept.9, 1998, and so on; the
non-inundated area is 0. Six evaluated images were
integrated to produce the composite image of the
flood movement by selecting pixel with the highest
value on the same spatial location.
4.2 Vectorization of flood images
Although the composite image can show the
flood movement, it is necessary to vectorize the
flood image for dynamic and accurate acquirement
of the inundated areas and the damages from the
floods. Due to the limitation such as temporal
resolution of TM image and image classification
technique, the manual interpretation was used to
get the landcover map in vector format.
In this paper the automatic tracing method was
Figure 3. Profile of water-land threshold calculation
adopted to vectorize the flood boundaries. The low
efficiency in interactively manual interpretation
limits quick, dynamic and accurate acquirement of
flood information. According to our experience, it
takes at least three days for three people to interpret
the flood boundaries on computer screens.
The tracing digitization was performed with the
software Coreldraw7.0, whose trace module is very
powerful. Control parameters can determine the
tracing unit, that is, only the boundaries of the flood
areas larger than a certain size can be traced, the
smoothed curves may not only greatly compress
data, but also bring a good visual effect.
Before the tracing digitization, the composite
image ranging within 260-220, 220-180, 180-140,
140-100, 100-60, 60-20 is respectively encoded in
binary format again. The pixel values of the image
inside each range were evaluated 1 and the ones
outside the range were altered 0. Six binary-value
images can offer the information of normal water
body and the flood on Aug.9, 16, 20, 20, 23 and 29
respectively. The flood areas smaller than 20 pixels
were neglected and the smoothing parameter were
10. The graphics traced were transferred from the
DXF format to the PC/ ARCINFO coverage and the
attribute values were added to map units. Since
number of non-inundated areas was small, first all
the units were given the same attribute value, then
the attributes of the non-inundated units were
modified in ARC/VIEW. In the window
environment the attributes of non-inundated units
can be replaced easily.
Through above steps, six coverage files were
produced, they may be utilized to extract the flood
information in the study area.
4.3 Dynamic monitoring and damage evaluation of
flood
The dynamic monitoring of flood requires the
composition of multi-temporal flood maps. Five
flood coverages in ARC/INFO were overlaid to
create a composite map. The attribute table was
joined with a new field for recording the first time
when the map units suffered the flood, thus the
expansion of flood area during two receiving dates
can be detected. The map of administration division
was overlaid with the composite map so that the
dynamic flood area for each administration district
can be calculated.
The map produced by automatic tracing of
water body from the TM image holds more detailed
information, however, the landcover map at the
scale of 1:100000 was accomplished through the
manual interpretation of the TM image, which was
overlaid with the flood data for the damage
evaluation. Therefore the former map had to be
discarded and the map of the water body was
extracted from the landcover one.
Damage evaluation of the flood is based on the
overlaying between the landcover map and the one
of flood dynamic process, which is performed only
in an administration district for the limitations of
software and hardware.
Figure 4. Flood process in the west of Jilin province in the summer of 1998
