Flood Detection in Changjiang 1998 from Landsat-TM Data

Shinkichi Kishi and Xiangfang Song*, Jiren Li**

*National Research Institute for Earth Science and Disaster Prevention, Tsukuba, Japan
**Ministry of Water Resources, Beijing, China

Abstract

NIED/STA, Japan and RSTAC/MWR, China have conducted a joint research study on the utilization of remote sensing data during floods in China. This paper outlines the research conducted on flood detection as regards the middle reaches basin of Changjiang in the summer of 1998. NIED provided the spatially mosaic images which made it easier to delineate the flooded areas.
Prior to the detailed analyses of TM data, the mapping of the near-daily extension of flood water was conducted by interpreting multi-temporal pictures from several kinds of satellite.
TM data of the overlapping area of the adjacent landsat paths in nine days during the flood was effectively used to examine the alteration of river water based upon the spectral characteristics of turbidity. Additionally, TM data taken before the flood was also timely used to specify the land cover condition in the area. The results of the study were supported by the examination of the MMR based on the daily discharge data represented by the hydrographs.
As a result, TM data was again practical to grasp the land cover condition in the flooded area and transition of turbidity of flood water. On the other hand, the synthesized observation frequency by several kinds of satellites was not enough in practical use to monitor the daily transition of flooding areas.

Keywords: Flood detection; Remote sensing


1. Introduction

Landsat TM data of the adjacent paths covering the middle reaches basin of Changjiang in the periods before and during the flood were obtained. To grasp the outline of the flood, two spatially mosaic images were produced and overlaid. In order to analyze the characteristics of the TM data, the overlapping areas of the adjacent paths were designated as the study area. In the areas, several kinds of multitemporal satellite pictures were effectively used to delineate the flooding water in near-daily extension.
Using TM data in the area, the land cover condition in the flooded areas was examined by overlaying the images of the nine days during the flood and by conducting the supervised classification for each image accompanied with the examination of spectral characteristics in the training areas.

 

2. General view of the middle reaches basin of Changjiang from landsat TM data

In order to broadly grasp the flooded area widely from TM data, two scenes of adjacent paths covering the middle reaches basin of Changjiang from the exit of Three Gorge to Wuhan were mosaic for the data of the summer season in 1995 three years before the flood, and of August 1998, during the flood, respectively. The flooding areas were clearly demonstrated by overlaying the images of the near-infrared band as is
in Fig.1. In the figure, the flood areas are shown in red color, according to the assigned colors of red to the data before the flood, and cyan to the data during the flood.
The authors designated the downstream basin of Jinanli, which is located in the northern part of Dangting Lake, as the area of study. The site is located in the central bottom portion of the mosaic image, and overlaps the adjacent paths observed by TM during the 9 days of the study.

Date Satelite Sensor Medium Source
1998        
8.01 ERS-2 SAR Printed www/ESRIN/ESA
8.05 RADARSAT SAR Printed JSPRS
8.06 NOAA AVHRR Printed CMA
8.11 NOAA AVHRR Printed CMA
8.12 RADARSAT SAR Printed www/CCRS/CSA
8.14 LANDSAT TM Degital Beijing/CAS
8.23 LANDSAT TM Degital Beijing/CAS
9.12 JERS-1 SAR Degital Kumamoto/NASDA

JSPRS: Japan Society of Photogrametry and Remote Sensing
CMA: China Meteorologist Administration
CSA: Chinese Academy of Science



3. Mapping of the extension of the flooded area from satellite images in time series

Prior to the digital analyses of TM data observed by change after the spread of flooding, aimed at grasping the daily progress of flood water at the beginning of August, printed images taken by several kinds of sensors of observation satellite such as NOAA/AVHRR, ERS-2/SAR and RADARSAT/SAR were been collected. The satellite data during the flood which was used in the study are listed in Table 1.
The data of observation by satellites are also entered in Fig.2, which shows the hydrograph compiled by the MWR showing the daily discharge during the flood at two major observation stations located in the upper stream (Sarshi) and lower stream (Chenlingji) with the study area between. The discharge observed at Chenlingji is combined with the data from Dongting Lake.
The extended areas of inundation interpreted from each satellite image were plotted on the rough map of the study area of about 40 km x 70 km, made from TM images of 1995 as shown in Fig.3. For an interesting example the orange colored area was not recognized as the inundation in the picture taken by the AVHRR sensor at 15:00 on Aug. 11, and appeared in the meaning of comprehensive utilization of satellite information in time series. In the study, however, any SPOT data were not available due to weather condition.

 

4. Land cover conditions in the flooded area

In order to grasp the land cover condition in the flooded area, the inundated areas extracted from TM data of mid and latter August and JERS-1 SAR data of mid-September were respectively overlaid onto the TM data in 1995, according to the standard method used internationally to understand the situation through color representation. The results are assembled chronologically from the top to the bottom in Fig.4.
In the overlaid images on the right column, blue is assigned to the binary coded data of the inundated area extracted by level slicing as shown in the central column, and green and red are respectively assigned to the data of TM Band-4 and Band-3 in 1995 in the left-central frame. In this color composition, vegetation represented by rice fields in the region appears in cyan, and inorganic substances such as soil appear in magenta respectively covered by the inundated water. In the same way, the permanent water areas of unmixed water represented by the lake and old river channel appear in pure blue, and turbid water represented by main stream of Changjiang appear in bright magenta.
Comparing the picture of the inundated area in the central column over time, in mid-September the water level is estimated to become lower due to the appearance of parts of land surface such as road and banks. In fact, this is supported by the hydrograph in Fig.2 as the reflux period of the flood. However, a problem in the simple comparison of the data without calibration between TM and SAR still remains.

 

5. Consideration of the transition of turbidity of flooded water through TM data

As an attempt to estimate the movement of the flooding water from the transition of the turbidity using TM data on two occasions during the interval of nine days at the peak of the flood, supervised classifisation by the most likelihood method for each occasion was conducted, and the overlaid image of these visible band was made as shown in Fig.5, accompanied with the training areas and their spectral signature. The turbidity here is considered as a relative quantity of sediment included in the flood water, and characterized by the intensity of Tm band-3 in the red region. In the case of the main stream of Changjiang, it is characteristic that the river water be so muddy that it has almost near intensity to the city area covered by typical inorganic substances even in Band-4 in the near-infrared region.
In the overlaid images in the bottom column, in Band-3 the red area indicates relatively high turbidity on Aug.14, and the blue area indicates high turbidity on Aug. 23. In Band-4, as mentioned above, the main stream represents some brightness at both times in the darkness of the surrounding water area.
From those images it was found that the area of low turbidity in the latter part of August through one of peaks of the flood as seen in the hydrograph, due to effects such as overflow, which sign was detectable along the natural levee on the left bank in the central part of the images of August.14, and on the contrary, the turbidity of surrounding areas of the old river of the settlement of movement of the flooded water.

 

6. Conclusion

Landsat TM data with its geometric and radiometric characteristics was again practical for us to grasp the land cover condition in the flooded area and the transition of turbidity of flood waters.
In term of the practical sensitivity of satellite data to monitor the extension of flood area, the observation frequency of at most a few days was found to be inadequate, despite synthesizing various types of observation by several kinds of satellites after considering the dependency of some satellites on weathers.
In the application of remote satellite sensing to disaster monitoring, the comprehensive recurrent period of several observation satellites would be required to be at the most one day, assuming the use of an all-weather type sensor to complement the optical sensors. Combined use of all-weather type SAR with optical VNIR images in time of peace is most practical to delineate the damaged areas, and combined use will make it possible to observe the same spot on Earth constantly.
To realize the operational use in disaster monitoring of orbital satellite remote sensing, it is necessary to establish an international cooperation system for timely observation and quick data distribution on natural disasters.

Acknowledgement

The authors acknowledge Mr.M.Hara of Vision Tech Inc, Japan, for his quick retrieval and procurement of LANSAT TM data from the Beijing Ground Station and Dr. A. Kondoh of Chiba University for offering JERS-1 SAR data in cooperative research with NASDA.

References

S. Kishi, T. Morohoshi, X. Song, J.Li & M. Hara: Flood Monitoring in China in International Cooperation, IAF-97-C.3.05, 48th International Congress in Turin.
S. Kishi: Potentiality of Space borne SAR Data for Monitoring of Natural Disaster, IAF-96-C.1.04, 47th International Congress in Beijing.
Y. Suga, S. Takeuchi & S. Kishi : Flood Monitoring in China using JERS-1/SAR and LANDSAT/TM Data IAF-99-C.2.06, 50th International Congress in Amsterdam.
S. Kishi: Development of the Database of the Examples of Application of Satellite Data to Disaster Monitoring, IAF-93-B.5.101, 44th International Congress in Graz.

Abbreviations

NIED: National Research Institute for Earth Science and Disaster Prevention
STA: Science and Technology Agency, Japan
RSTAC: Remote Sensing Technology Application Center
MWR: Ministry of Water Resources, China