INTEGRATION OF VARIOUS TYPES OF REMOTE SENSING DATA FOR THE MORAVA RIVER CATCHMENT EVALUATION

 

Lena Halounová

MGE DATA, s.r.o.

Vrchlického 60

Praha 5

Czech Republic

 

KEY WORDS: flood detection, remote sensing, RADARSAT, SPOT, Thematic Mapper, flood area delineation

ABSTRACT

During the flood in 1997, situation in Moravia was studied from the quickest available satellite data - from the Canadian satellite RADARSAT Standard and Wide Modes with 30 meters resolution. These images which form an archive record of an immediate real state, were processed to evaluate the flood situation. The northern part of Moravia was imaged twice. The first image showed the flood peak and the second one the end of the flood. The southern part of Moravia was imaged three times - the flood peak, the end of the flood and one more image shortly before the flood end. Black and white radar images show easily distinguished water bodies. Radar images are useful for interpretation of water, forest, urban areas, agricultural areas as different units, however also for a smooth surface versus a rough surface, or a dry surface versus surfaces with higher moisture.

To study deeply flood impacts in the inundation region, optical data were also used. Two types of scanner data were combined with radar images - one Thematic Mapper with 30 meters resolution from the second half of August 1997, and two SPOT multispectral images from spring 1997.

Combination of radar and optical data was very fruitful. Radar images presented real flood situation already within two days from data order regardless bad atmospheric conditions. Optical data which are unlike RADARSAT data archived, were implemented later. None of them were from the flood because of the weather.

1. RELATED DATA

The area affected by the flood was about 25,000 km2. Satellite images were a source of flood documentation captured in near-real time. Satellite data brings an overview of the area at one moment in time. Table 1 shows briefly the obtained radar images. Their spatial prelaunch nominal resolution is 30 meters. This radar sensor operates at a single microwave frequency, known as C - band (5.3 GHz frequency and 5.6 cm wavelength). RADARSAT transmits and receives its microwave energy in a HH polarization. The spectral resolution of processed images is 16 bits per pixel. RADARSAT digital products can be delivered as six different data types. Path Image product (used in this case) is aligned parallel to the satellite orbit path. The product was calibrated which refers primarily to the electrical stability of the radar sensor and its ability to provide repeatable measurements over time. The system was designed to achieve radiometric accuracy within one scene < 1.0 db, over three days < 2.0 db with global dynamic range 30.0 db. Absolute radiometric calibration was required so that the magnitude of the digital data processed can be related to the radar backscatter coefficients. To achieve this accuracy, detailed measurements of the radar and processing system performance were made on a regular basis (RADARSAT Illuminated, 1995).

2. IMAGE PROCESSING, IMAGE INTERPRETATION

2.1 Radar image, preprocessing

Radar images can be viewed as single channel black and white images with a characteristic "salt and pepper" appearance. The pixel values represent the strength of the returned radar signal from the earth's surface. For each surface feature, there is a statistical distribution of the probable strength of that returned signal. Each pixel representing that surface is assigned a value randomly selected from the statistical distribution. Therefore, a seemingly homogeneous surface area has an irregular distribution of light and dark pixels, producing a granular effect. This effect is termed "speckle" and is an inherent property of radar images (RADARSAT Illuminated, 1995). Original image data were compressed from 16 bits to 8 bits data. Results of data calibration control showed that the July 24 image values had a 6 - 9 per cent lower average values in selected targets. The same control performed for July 27 showed that average values in the same targets were for 10 - 15 per cent higher, both compared to the July 14. An example of forest targets is shown on Figure 1.

Figure 1. Calibration forest targets

To modify speckles, the images listed in Table 1 were smoothed by a 5 x 5 pixel spatial filter. All images were transformed into a Czech cartographic projection.

2.2 Image Classification

Pixel classification techniques, an often used method in image processing was performed for surface water. To delineate flooded forest and areas which were flooded between image pairs, a visual interpretation was used. In the lowlands, automated classification of present water surface could be reliable for more than 90 per cent of the area excluding urban and forest regions. Whenever these two features occur, an important principle of radar backscatter appears - a corner reflector. Two or three dimensional corner reflection is caused by the existence of buildings (two and three dimensional). Scattering from a forest canopy can present a complex case of volume scattering. Double-bounce scattering between tree trunks and the ground is one important effect of the volume scattering. This can give a very strong return if the ground is covered with water. (Ahern, 1995). Double-bounce scattering is a geometrically similar situation to a two dimensional corner reflection. Buildings and trees can redirect a radar beam which was backscattered from a smooth water surface back to the radar sensor. This is why flooded towns and forests can look even brighter than unflooded areas.

3. RESULTS

3.1 Areas of Heavy Storm = Areas of Earliest Flood

The northern Moravia was imaged twice with RADARSAT. The first image revealed the flood peak in the lowlands while the second image featured the post-flood situation. The mountainous area of northern Moravia, which was also the main source of water volume suffered the most destruction.

Table 1. Applied RADARSAT images

 

Date

area

image size

orbit

incidence angle

Standard 2

10 July 1997

North Moravia

100 x 100 km2

ascending

24 - 31

Standard 7

14. July 1997

South Moravia

100 x 100 km2

descending

45 - 49

Standard 5

24 July 1997

South Moravia

100 x 100 km2

descending

36 - 42

Wide 1

27 July 1997

South Moravia

150 x 150 km2

descending

20 - 31

Wide 1

27 July 1997

North Moravia

150 x 150 km2

descending

20 - 31

 

Destruction due to the flood water included roads, railways, bridges, houses, and many other types of infrastructure features. The affected communities within the narrow valleys of the high mountain ranges could not be analyzed by RADARSAT imagery for two reasons. Radar shadow was a factor and the duration of the flood in this region was short and had ended before the first RADARSAT image was obtained.

3.2 Areas with Imaged Flood - Northern Moravia

The central Moravia and southern Poland, lowland regions were then studied on the image of July 10. An example of flooded areas, and permanent water bodies, are shown as black areas (Figure 2a). Water surfaces without waves act as a smooth surface. When the radar sensor transmits a beam of radar energy towards this smooth surface the result was no backscatter return to the radar sensor but rather the scattering of the radar energy away from the sensor. Pixels for these areas have zero values and water areas are black solid phenomena on images (Leconte et al, 1991).

Pixel classification techniques, an often used method in image processing was performed for surface water. They detected not only areas with surface water excluding forest and urban regions but also shadows in high relief areas. These shadows have the same values of reflection as water bodies: their measured values are the same - zero or very low values in both cases.

It was necessary to use two images from two different time intervals in order to distinguish flooded areas from permanent surface water. A color composite (RGB) of the two images (one of them must be used twice) can distinguish permanent versus flood water immediately. Permanent surface water was black (Figure 3) whereas flooded areas were lighter (blue in color version).

Figure 2b represents the same area on the image for July 27, 1997. Brighter features within the imagery coincided with previously flooded areas. Brighter features were related to either terrain with greater surface roughness due to ploughing or the sedimentation of course materials or higher soil moisture content (RADARSAT Illuminated, 1995, Brown, Engman et al., 1995). Sedimentation of coarse materials did not occur as the result of rather low water velocities in this area. Nor was ploughing the reason for higher backscatter values. The region was divided into small long private fields which were not significantly damaged by the flood because crops continued in their growth after flood levels declined. It was therefore concluded that excessive soil moisture was the reason for the brighter backscatter values.

There was an area around Olomouc (town in central Moravia on July 10 and 27 images) which did not show the same effect on the same RADARSAT images. This area is quite flat, similar to southern Poland but probably with different soil perviousness. Hydrogeological, geological and pedological conditions for the area around Olomouc are different. The previously flooded fields could not be detected on the post-flood image.

3.3 Areas with Imaged Flood - Southern Moravia

Southern Moravia was imaged by RADARSAT at three time intervals. Pixel values on the image of July 24 are lower for 6 per cent compared to July 14 as it reveals from examples of calibrating targets on Figure 1. Pixels values on the image of July 27 are higher for about 20 ÷ 25 per cent in comparison to July 14 in forest targets.

To delineate flood areas meant to include free water surfaces, areas flooded but not in moments of imaging, and flooded forest. To determine free water areas is easy. They are black homogenous areas (e.g. Figure 2a).

Detection of previously flooded areas can be performed due to higher moisture which is represented as an area with a high reflection. High reflection can be caused also by higher surface roughness. This case must be excluded regarding additional information. Application of this attitude is shown on Figure 4a, 4b, and 4c.

This is an area which was flooded on July 14 and had higher moisture on July 24 and July 27 The same area was brighter on the images of July 24, and July 27. Comparison of these two images can be a confirmation of the fact that steep incidence angles (July 27) provide the greatest amount of information regarding soil moisture and minimize roughness effects (Ulaby, 1974). This was an example of possibility to delineate flooded area on a post-flood radar image. To determine reliably these areas required images from a given area at the moment of existing higher soil moisture. This moment differs for various soil types, hydrogeological conditions, terrain slopes, and canopy. To determine the time when soil moisture levels were due to the previous flood higher must be a subject of more detailed studies in areas of interest. More frequent post-flood images could offer this information.

Flooded forest was brighter than unflooded as is shown on Figure 5 where flooded forest is labeled F and unflooded U. Values of July 24 are in all areas lower than of July 27, three days later without rain. The image of July 24 had a shallower incidence angle (36° ÷ 42° ) and that was why the area was not as bright as the same area on the image of July 27 (with incidence angle 20° ÷ 31° ). Comparison of unflooded and flooded forest at a different incidence angle are on Figure 6a and 6b.

Higher difference between mean pixel values of flooded and unflooded forest is at the steeper incidence angle on Figure 6a and lower difference at the shallower incidence angle on Figure 6b.

 

Figure 6a Comparison of flooded and unflooded forest

 

Figure 6b. Comparison of flooded and unflooded forest

 

Figure 7 shows a higher reflection of flooded forest in case of a steeper incidence angle and a lower reflection at a shallower incidence angle.

 

Figure 7. Reflection of flooded forest at different incidence angles

A color composite (RGB) of the three images can distinguish permanent and flood water immediately (Figure 5). Permanent surface water shown as black versus flooded areas at different time intervals can be displayed in different colors. The RGB image was able to show the flood progress. RADARSAT images were the only images available for the flooded areas with such a short time lag after the flood onset. It was the only sensor able to repeat images in very short time intervals.

Radar sensors are the only ones which can penetrate clouds, fog, and smog. Moravia was covered by clouds nearly the whole time, and thus no optical data were available.

SPOT images and Thematic Mapper were used only to determine what kind of land was flooded - agricultural, meadows, and other. Spring (SPOT XLS) and late summer (TM) images enabled to distinguish permanent vegetation cover from agricultural land.

4. DISCUSSION

Any automated classification performed without an additional visual interpretation can result in erroneous information. For example radar shadow found in mountainous areas can have pixel values the same as a smooth water surface. A post-classification modification must be applied.

Visual interpretation was necessary for forest areas and urban regions on single images. Single images display only surface water bodies existing at the time of image capture. It is not possible to distinguish permanent water bodies from flooded areas. In contrast, image pairs from the moment of flood and before or after flood enable one to discriminate permanent and flooded areas. This task can be easily performed by creating colour composites.

Images from after a flood can be useful in cases when no images from the flood itself are available. Higher soil moisture as a consequence of flooding causes a higher backscatter, and thus can be interpreted as brighter regions on the post-flood image. To decide whether a brighter backscatter value is due to high moisture requires information about the locality and to be able to exclude surface roughness which can be related to field activities such as ploughing, and to the previous flood in case of coarse sediments. Incidence angle are another phenomenon which must be taken into account. Steeper incidence angle emphasize soil moisture influence of radar reflection. In contrast, shallower incidence angles are more influenced by surface roughness. Delineation of flooded areas was compared to geological maps. Very good coincidence of flood delineation and alluvial sediments is presented on Figure 5.

Detailed study of soil moisture changes of flooded areas in short time intervals after flood can be a data source for models of determination of sediment layers perviousness, or their thickness, e.g.

 

© RADARSAT Images RADARSAT International Copyright 1997

 

References:

  1. Ahern, F. J. 1995. Fundamental concepts of imaging radar: basic level (unpublished manual). Canada Centre for Remote Sensing, Ottawa, Canada.
  1. Brown, R. J. et al (1993), Potential applications of RADARSAT data to agriculture and hydrology. Canadian Journal of Remote Sensing, 4, 317 - 329.
  1. Engman, E. T. and Chauhan, N. (1995), Status of microwave soil moisture measurements with remote sensing, Remote Sensing of Environment, 51, 189 - 198.
  1. Leconte, R. and Pultz, T. J. (1991) Evaluation of the potential of RADARSAT for flood mapping using simulated satellite imagery. Canadian Journal of Remote Sensing, 3, 241 - 249.
  1. RADARSAT Illuminated. 1995. User Guide. RADARSAT International
  1. Ulaby, F. T. et al (1974), Radar measurement of soil moisture content. IEEE Transaction on Antennas and Propagation, 2, 257 - 265.

 

 

 

Figure 2a. Standard 2 image of southern Poland showing maximum flood area, surface water represented by solid black color.

Figure 2b. Post-flood Wide 1 image of southern Poland showing recently flooded areas in brighter hue which coincide with flood areas on Figure 2a.

 

Figure 3. Color composite of two time intervals with flooded areas in dark hue (blue in color version) and permanent surface waters in black color.

 

Figure 4. Standard 7 image (July 14) shows flooded area in southern Moravia near town Uhersky Ostroh (a), the same area on Standard 5 image (July 24) is in bright hue (b), and on Wide 1 (July 27) even in brighter hue (c) than on (b). Brighter color is caused by higher soil moisture as a result of previous flood. That confirms even a difference between Standard 5 and Wide 1. Soil moisture is more decisive element than roughness at the incidence angle of Wide 1 which is steeper than at Standard 5.

 

Figure 5. Color composite of three time intervals images (RGB July 14, July 24, July 27) shows permanent water in black color