Key words: Dynamic monitoring of flood, Automatic tracing, Damage evaluation
Abstract
The greatest flood in the 20th century hit Northeast China in the summer of 1998. The
movement of the flood over the Nenjiang River drainage area in the northwest of Jilin province was
dynamically monitored using Radarsat data; with the landcover map at the scale of 1:100000, interpreted
from Landsat TM images, the damages from the flood were also evaluated. In this paper, a technique,
similar to MVC (Maximum Value Composite) used to remove cloud contamination on NOAA AVHRR
images, was applied to correct the logical errors on the flood boundaries extracted at different time. In
addition the flood boundaries were vectorized via automatic tracing method, therefore the dynamic
monitoring of the flood spatial process was performed better and the information of the changing inundated
areas was offered more rapidly. The method produced in the study proved to be an effective approach to
dynamic monitoring of flood.
1. Introduction
In the summer of 1998, there was a cataclysm
over the whole drainage area of the Nenjiang River.
The long duration and expansive area of the flood
are rare in its history. It is very important to get
dynamic flood information promptly and accurately
for safety of life and property. Radarsat data of the
wide-scanning mode play a very important role in
flood monitoring for their advantages of wide
coverage, short cycle, all-weather, full-time and
capability of distinguishing water from land. The
fusion of multi-temporal radar images is useful for
dynamic monitoring of flood, however it lacks the
capability of offering flood area. Submerged area
estimation and damage evaluation mainly depends
on manual interpretation of radar images. Owing to
its low efficiency, it is not easy to meet the demand
for dynamic, quick and accurate acquirement of
flood information. A more effective and practical
method deserves probing, which can both analyze
flood movement and evaluate damage.
2. Study area and flood process
As one of the main tributaries of the Songhua
River, Nenjiang River rises in the Da Hinggan
Mountains and the Xiao Hinggan Mountains and
has many long branches, whose water system is
characteristic of plumose shape. Its northern part is
typical of mountain river. The Nenjiang River flows
into the Songnen Plain at the Nirier Town, from
where the floodplain develops well and swamps
distribute widely.
The study area lies in the downstream area of the
Nenjiang River Basin (Figure 1). Besides Nenjiang
River, some other rivers including the Taoer and the
Huolin River flow through the region as well.
When flood occurs, the water of the Taoer River
strengthens the flood of the Nenjiang River at
Yueliangpao Lake. Located to the southwest of the
Nenjiang River, the trails of Huolin River usually
disappear in swamp areas and form stagnant water
areas. When flood emerges, floodwater rushes into
the Songhua River through Lake Chagan, thereafter,
the downstream area of the Huolin River frequently
suffers the flood.
Fig.1 The location of study area
In the summer of 1998, the Nenjiang River
drainage area was hit by heavy rains, which led to
three successive floods (from late June to middle
July, late July to early August and late August to
early September). During the second flood in
midstream area and the third one in middle and
downstream area of the Nenjiang River drainage
basin, the flood exceeded the warning water level
for a long time. Furthermore, the flood from the
Taoer River converged into the one of the Nenjiang
River many times. Besides the above causes many
destruction of its dikes also related to the decrease
of the forest coverage in the Da and Xiao Hinggan
Mountains, the shortage and disrepair of the flood
controlling projects along the banks. The Lahai
Dike on the left bank of the Nenjiang River burst
on Aug.13, and then the Tailai Dike collapsed on
15. The floodwater of 1.1 billion cubic meters
gushed into Tailai County of Heilongjiang Province
and Zhenlai County of Jilin Province. The
tempestuous flood brought about huge damages.
3. Data and preprocess
TM data covering the study area were received
from May 1995 to August 1996 and transformed
into Albers Equal Area Conic Projection. After
interactive interpretation of the mosaic TM image
on computer screen, the landcover map was
produced at the scale of 1:100000 and stored in
ARC/INFO coverage format for the damage
evaluation. The water system map was derived
from the landcover map. The administration
division map from available 1:100000
topographical maps were manually digitized and
transformed to Albers Equal Area Conic projection.
The Radarsat data of wide-scanning mode with
the work band of C were received on Aug. 16, 20,
23 and 29, 1998. They have been transformed from
slang to ground range at the Ground Receiving
Station of Satellite. Except for the Radarsat image
on Aug. 9, the others covered the whole study area.
With the false color composite TM images the
geometric calibration of the Radarsat images was
carried out. Owing to the special imaging
mechanism, the radar image holds particular
characteristics such as geometric distortion,
radiation distortion and speckle phenomenon.
Because of the slant range projection of radar
image, the farther away from the nadir is, the larger
the scale is. In this case the transformation from
slant to ground range may keep each pixel with the
area of 100m*100m. To get the accurate flood area,
the software ENVI was used to transform the
images to the Albers Equal Area Conic Projection.
The ground control points were selected
respectively from the TM and radar images. Water
is easy to be identified whereas other objects are
obscure for the coarse resolution of the radar image.
Considering the flood expansion and change of
water body boundary, the turns of main roads were
chose as control points. The wrap method used was
polynomial and the images were resampled by
nearest neighbor method. The transformed radar
images with the pixel size as large as that of TM
image makes it easy to integrate them.
The antenna pattern causes different pixel values
for same or similar objects in radar image and the
radiation distortion occurs along the range, that is
to say, the central part of the radar image is
brightest and its brightness gradually decreases
from the central line to its two sides. When the
flood movement was monitored we found that the
inundated areas held different pixel values.
Antenna pattern correction was performed by
polynomial fitting so that the signal intensities of
the identical objects were similar in the monitoring
area.
Figure 2. Location of water-land threshold profiles
Because the most distinctive characteristic of
SAR is speckle, the Gamma Map filter was adopted
to oppress the speckles, which can smooth the
noises and maintain the edges well. The window of
9*9 could bring the better result.
The five radar images through noise removal
were stretched, ranging from 0 to 255, for the
extraction of flood information.
