**5. Flood mapping of Yialias river catchment area in Cyprus using ALOS PALSAR radar images**

#### **5.1. Introduction**

**Figure 7.** Final erosion risk map constructed with AHP method

108 Remote Sensing of Environment: Integrated Approaches

**Figure 8.** Image indicating the soil erosion severity class differences between AHP and RUSLE method

ALOS (Advanced Land Observing Satellite) PALSAR data can be used to detect the water sur‐ face due to the L-band wave length. All SAR instruments share the advantages of day-night operability (as active sensors), cloud penetration, and the ability to calibrate without perform‐ ing atmospheric corrections. The longer L-band (~23.5 cm) SAR wavelength, and, to a certain extent, the C-band (~5.5 cm), have the ability to penetrate vegetation canopies to various de‐ grees depending on vegetation density and height, dielectric constant (primarily a function of water content), and SAR incidence angle. Variations in backscattering allow discrimination among non-vegetated areas (very low to low returns), herbaceous vegetation (low to moder‐ ate returns), and forest (moderate to high returns), and to some degree among different forest structures and regrowth stages. Where water is present beneath a forest canopy, enhanced re‐ turns caused by specular "double bounce" scattering between water surface and tree trunks makes it possible to distinguish between flooded and non-flooded forest.

#### **5.2. Data and methodology**

#### *5.2.1. Data and methodology*

The purpose of this study is to explore the potential of ALOS-PALSAR imagery for observing flood inundation phenomena in the Yialias catchment area in Cyprus. Two PALSAR images (polarity: HH, pixel size 50 m) covering the study area before and after an extreme precipita‐ tion incident in 2009 were used (Table 1). A LULC map was also constructed with the use of high resolution images such as GeoEye -1 covering the study area. To analyze Radar backscat‐ ter behavior for different land cover types, several regions of interest were selected based on the land cover classes. A number of land cover classes were found to be sensitive to flooding, whereas in some other classes backscatter signatures remained almost unchanged.

The GeoEye-1 images were used for the land use monitoring of the upstream and down‐ stream of the basin; the two images refer to 12 March 2011 and 11 December 2011, respec‐ tively. GeoEye-1 is a multispectral sensor with four spectral bands. Its spectral range is: 450-510 nm (blue), 510-580 nm (green), 655-690 nm (red) and 780-920 nm (near infrared),

Integrated Remote Sensing and GIS Applications for Sustainable Watershed Management: A Case Study from Cyprus

http://dx.doi.org/10.5772/39307

111

**Figure 10.** Rain gauge stations within the study area or in close vicinity with it and drainage network

**Resolution** 50m **Swath Width** 35 km **Polarization** HH **Off Nadir –Angle (deg)** 18.0-43.3 **Incidence Angle (deg)** 20.1-36.5 **Processing level** 4.2 **Data Rate (Mbps)** 120 **Bit quantization (bits)** 5

**Projection** UTM Zone 36 North

**Table 1.** Technical specifications of ALOS PALSAR images

**ScanSar (WB1)**

while its spatial resolution is approximately 1.65 m.

#### *5.2.2. Data*

For the purposes of the study, the following satellite and digital spatial data were incorporated:


The ALOS images were acquired on 30 November 2009 and 6 December 2009 (Fig. 9a). PAL‐ SAR is a fully polarimetric instrument, operating at L-Band with 1270 MHz (23.6 cm) centre frequency and 28 MHz, alternatively 14 MHz, bandwidth. The antenna consists of 80 trans‐ mit /receive (T/R) modules on four panel segments, with a total size of 3.1 by 8.9m (Table 1). The two ALOS images were acquired after thorough indexing of Cyprus Meteorological Service archives of precipitation data. Specifically, the research team searched the precipita‐ tion archives of all the meteorological and climatological gauge stations within the study area (Analiontas, Pera Chorio, Lythrodontas, Mantra tou Kampiou, Kionia, Mathiatis), as they are indicated and spatially distributed in Fig. 10. Due to the lack of ALOS imagery data acquired during recorded flood inundation events, the research team tried to acquire images before and after extreme precipitation events in order to examine the potential of the image‐ ry to detect soil moisture and flood inundation trends. Thus, the image for 30 November 2009 corresponded to a day where no precipitation had been recorded, while the image for 6 December 2009 corresponded to a day when a mean value of 25mm of precipitation had been recorded in the rain gauge stations within the study area.

**Figure 9.** (a) ALOS PALSAR image (30 November 2009) and the study area. (b) Mosaic of the two GeoEye -1 images of the study area (RGB - 321)

The GeoEye-1 images were used for the land use monitoring of the upstream and down‐ stream of the basin; the two images refer to 12 March 2011 and 11 December 2011, respec‐ tively. GeoEye-1 is a multispectral sensor with four spectral bands. Its spectral range is: 450-510 nm (blue), 510-580 nm (green), 655-690 nm (red) and 780-920 nm (near infrared), while its spatial resolution is approximately 1.65 m.

ter behavior for different land cover types, several regions of interest were selected based on the land cover classes. A number of land cover classes were found to be sensitive to flooding,

For the purposes of the study, the following satellite and digital spatial data were incorporated:

**•** A Digital Elevation Model (DEM) of 25m pixel size provided by the Department of Land and Surveys of Cyprus, created with the use of orthorectified stereopairs of aerial photos

The ALOS images were acquired on 30 November 2009 and 6 December 2009 (Fig. 9a). PAL‐ SAR is a fully polarimetric instrument, operating at L-Band with 1270 MHz (23.6 cm) centre frequency and 28 MHz, alternatively 14 MHz, bandwidth. The antenna consists of 80 trans‐ mit /receive (T/R) modules on four panel segments, with a total size of 3.1 by 8.9m (Table 1). The two ALOS images were acquired after thorough indexing of Cyprus Meteorological Service archives of precipitation data. Specifically, the research team searched the precipita‐ tion archives of all the meteorological and climatological gauge stations within the study area (Analiontas, Pera Chorio, Lythrodontas, Mantra tou Kampiou, Kionia, Mathiatis), as they are indicated and spatially distributed in Fig. 10. Due to the lack of ALOS imagery data acquired during recorded flood inundation events, the research team tried to acquire images before and after extreme precipitation events in order to examine the potential of the image‐ ry to detect soil moisture and flood inundation trends. Thus, the image for 30 November 2009 corresponded to a day where no precipitation had been recorded, while the image for 6 December 2009 corresponded to a day when a mean value of 25mm of precipitation had

(a) (b)

**Figure 9.** (a) ALOS PALSAR image (30 November 2009) and the study area. (b) Mosaic of the two GeoEye -1 images of

been recorded in the rain gauge stations within the study area.

whereas in some other classes backscatter signatures remained almost unchanged.

*5.2.2. Data*

**•** 2 ALOS PALSAR images.

110 Remote Sensing of Environment: Integrated Approaches

covering the study area.

**•** 2 GeoEye -1 images

the study area (RGB - 321)

**Figure 10.** Rain gauge stations within the study area or in close vicinity with it and drainage network


**Table 1.** Technical specifications of ALOS PALSAR images

#### **5.3. Pre-processing techniques**

Initially, geometric corrections were carried out to ALOS PALSAR images using standard techniques with ground control points and a first order polynomial fit so asthe two im‐ ages to be co-registered. For this purpose, topographical maps were used to track the po‐ sition of ground control points in conjunction with the digital shoreline of Cyprus extracted from the provided DEM. There are ascending and descending observation modes of PALSAR images and differences in backscattering values, therefore, the image calibration is an essential task. Different factors influence backscatter strength signal in‐ cluding satellite ground track, incidence angle, radar polarization, surface roughness and the surface's dielectric properties (Yingxin & Linlin, 2010). Different objects having the same digital number which may correspond to different backscatter values. Thus, the ALOS scenes were subsequently converted from amplitude data format to normalized ra‐ dar cross section (σ°) according to Equation 8:

$$\text{Cov}^{\circ} = 10 \log\_{10} \text{ DN}^{2} \quad \text{+ CF}, \tag{8}$$

**5.4. GeoEye-1 Imagery classification and ALOS PALSAR texture analysis**

the main part of the catchment area is covered by soil and olive trees.

Bare Rock 2.52 Forest 3.99 Marl 0.33 Soil 43.86 Trees (mainly olive trees) 47.99 Urban Fabric 8.21 Agricultural Areas 2.99

**Figure 11.** LULC map of the study area after the application of ML classification algorithm to GeoEye-1 mosaic

After the application of preprocessing techniques to GeoEye-1 images and the development of an image mosaic, the Maximum Likelihood (ML) algorithm was applied to create a de‐ tailed LULC map of the study area. For this reason 7 major classes were defined (Bare rock, Forest, Marl, Soil, Trees, Urban Fabric, Agricultural Areas) (Fig. 11). The statistics of the land use regime of the study area are shown in Table 2. From these statistics, it is clearly seen that

Integrated Remote Sensing and GIS Applications for Sustainable Watershed Management: A Case Study from Cyprus

http://dx.doi.org/10.5772/39307

113

**Classes Area (km2)**

*5.4.1. GeoEye-1 Imagery classification technique*

**Table 2.** Statistics of the LULC thematic map

where, DN is Digital Number and CF is a calibration factor (CF = - 83.0 dB).

In SAR image, the speckle noise is one of obstacles to overcome in data processing, so it is necessary to take effective steps to filter the image. Several filter algorithms were tried; the Lee filter was applied to reduce speckle noise. This filter is based on the minimum mean square root (MMSE) and geometric aspects. This is a statistical filter designed to eliminate noise, while still maintaining the quality of pixel points and borders of the im‐ age (Hongga et al., 2010).

Atmospheric and geometric corrections were carried out on the GeoEye-1 images. Atmos‐ pheric correction is considered to be one of the most complicated techniques since the distri‐ butions and intensities of these effects are often inadequately known. Despite the variety of techniques used to estimate the atmospheric effect, the atmospheric correction remains a dif‐ ficult task in the pre-processing of image data. As it is shown by several studies (Hadjimitsis et al. 2004b, 2010a, 2010b; Agapiou et al., 2011), the darkest pixel (DP) atmospheric correc‐ tion methodology can easily be applied either by using dark targets located in the image or by conducting *in situ* measurements.

After the application of atmospheric and geometric corrections to GeoEye-1 images, the research team proceeded in the construction of an overall image mosaic by integrating the two individual images covering the up- and down-stream of the watershed basin (Fig. 9b). For this purpose, a histogram matching technique was applied to the common covered area of the two images in order to secure the radiometric correctness of the final extracted mosaic. Finally, the research team removed the cloud cover from the mosaic im‐ age in GIS environment.

#### **5.4. GeoEye-1 Imagery classification and ALOS PALSAR texture analysis**

#### *5.4.1. GeoEye-1 Imagery classification technique*

**5.3. Pre-processing techniques**

112 Remote Sensing of Environment: Integrated Approaches

age (Hongga et al., 2010).

by conducting *in situ* measurements.

age in GIS environment.

dar cross section (σ°) according to Equation 8:

Initially, geometric corrections were carried out to ALOS PALSAR images using standard techniques with ground control points and a first order polynomial fit so asthe two im‐ ages to be co-registered. For this purpose, topographical maps were used to track the po‐ sition of ground control points in conjunction with the digital shoreline of Cyprus extracted from the provided DEM. There are ascending and descending observation modes of PALSAR images and differences in backscattering values, therefore, the image calibration is an essential task. Different factors influence backscatter strength signal in‐ cluding satellite ground track, incidence angle, radar polarization, surface roughness and the surface's dielectric properties (Yingxin & Linlin, 2010). Different objects having the same digital number which may correspond to different backscatter values. Thus, the ALOS scenes were subsequently converted from amplitude data format to normalized ra‐

2

In SAR image, the speckle noise is one of obstacles to overcome in data processing, so it is necessary to take effective steps to filter the image. Several filter algorithms were tried; the Lee filter was applied to reduce speckle noise. This filter is based on the minimum mean square root (MMSE) and geometric aspects. This is a statistical filter designed to eliminate noise, while still maintaining the quality of pixel points and borders of the im‐

Atmospheric and geometric corrections were carried out on the GeoEye-1 images. Atmos‐ pheric correction is considered to be one of the most complicated techniques since the distri‐ butions and intensities of these effects are often inadequately known. Despite the variety of techniques used to estimate the atmospheric effect, the atmospheric correction remains a dif‐ ficult task in the pre-processing of image data. As it is shown by several studies (Hadjimitsis et al. 2004b, 2010a, 2010b; Agapiou et al., 2011), the darkest pixel (DP) atmospheric correc‐ tion methodology can easily be applied either by using dark targets located in the image or

After the application of atmospheric and geometric corrections to GeoEye-1 images, the research team proceeded in the construction of an overall image mosaic by integrating the two individual images covering the up- and down-stream of the watershed basin (Fig. 9b). For this purpose, a histogram matching technique was applied to the common covered area of the two images in order to secure the radiometric correctness of the final extracted mosaic. Finally, the research team removed the cloud cover from the mosaic im‐

å å (8)

<sup>10</sup> <sup>σ</sup>° = 10 log DN + CF, å å

where, DN is Digital Number and CF is a calibration factor (CF = - 83.0 dB).

After the application of preprocessing techniques to GeoEye-1 images and the development of an image mosaic, the Maximum Likelihood (ML) algorithm was applied to create a de‐ tailed LULC map of the study area. For this reason 7 major classes were defined (Bare rock, Forest, Marl, Soil, Trees, Urban Fabric, Agricultural Areas) (Fig. 11). The statistics of the land use regime of the study area are shown in Table 2. From these statistics, it is clearly seen that the main part of the catchment area is covered by soil and olive trees.


**Figure 11.** LULC map of the study area after the application of ML classification algorithm to GeoEye-1 mosaic

#### *5.4.2. ALOS PALSAR texture analysis*

According to Zhang & Zhu et al. (2011) texture is defined as the spatial variation in gray value and is independent of color or luminance. Texture measures smoothness, coarse‐ ness and regularity of a region in an image. For the description of texture histograms, gray level co-occurrence matrix (GLCM), local statistics and characteristics of the frequen‐ cy spectrum are used. The GCLM mainly operates by calculating a matrix that is based on quantifying the difference between the grey levels of neighboring pixels in an image window. The main aim of this matrix is the quantification of the spatial pixel structure within this window. It was initially suggested as a mechanism for extracting texture measures (Haralick et al., 1973).

In the specific study, through the use of ENVI 4.7 software, 7 different statistical indica‐ tors of texture such as contrast, angular second moment, homogeneity, entropy, dissimi‐ larity, mean and variance were applied for carrying out the statistical texture analysis of all the typical ground objects. From those textural indicators, multiple RGB composites were constructed to improve the visual monitoring and interpretation of moisture affect‐ ed areas.

(a)

Integrated Remote Sensing and GIS Applications for Sustainable Watershed Management: A Case Study from Cyprus

(b)

**Figure 12.** (a) The catchment area before the precipitation event. (b) The catchment area after the precipitation event

**Before Precipitation Event**

**Table 3.** Radar Backscatter of ALOS PALSAR images for different land cover types and days

1 Bare Rock -18.83 -24.31 5.48 2 Forest -23.04 -18.13 4.91 3 Soil -25.46 -27.94 1.47 4 Trees -27.94 -22.68 5.26 5 Urban -23.16 -16.49 6.67 6 Vegetation -26.84 -26.34 0.50 7 Marl -31.51 -24.95 6.56

**Radar Backscatter (dB)**

**After Precipitation Event**

**Difference**

http://dx.doi.org/10.5772/39307

115

**Class Name**

#### **5.5. Results and discussions**

As it is clearly seen in Figure 12, in the downward of the catchment area (northeastern part) certain patches were inundated with water. Those patches are clearly observed with the low backscattering values and their corresponding dark pixels. However, in most of the cases the backscattering values were increased mainly because of volume scattering due to the moisture effect in the vegetation and plant cover. Concerning the southwestern part of the watershed where the most forested areas are established, due to the corresponding increase of the moisture after the extreme precipitation event, the backscatter values were generally increased due to the effect of double reflection by wa‐ ter (moisture) and tree trunks. Thus, generally the SAR backscattering intensity in forest areas changes to be higher in cases of inundation events. In addition, in certain areas of the southwestern part of the catchment area where there are more bare rock and soil patterns, the backscattering values were decreased due to the corresponding moisture effect.

The values of radar backscatter coefficient for the different land cover classes as they were extracted from GeoEye-1 images, are tabulated in Table 3. The results were extracted in GIS environment (ArcGIS 10 software) through the use of zonal statistics application. According to Table 3, the backscatter coefficient in most of the classes increased after the precipitation event. The reason for this phenomenon was the overall moisture increase in the area. The backscatter of forest and urban areas was significantly increased (4.57 and 6.67dB) after the precipitation event due to the double reflection phenomenon. On the other hand, in other classes such as soil and bare rock, dB values declined due to water accumulation and the corresponding surface scattering effect. In agricultural areas of low vegetation, such as alfafa or barley crops, the db were slightly increased.

Integrated Remote Sensing and GIS Applications for Sustainable Watershed Management: A Case Study from Cyprus http://dx.doi.org/10.5772/39307 115

*5.4.2. ALOS PALSAR texture analysis*

114 Remote Sensing of Environment: Integrated Approaches

measures (Haralick et al., 1973).

**5.5. Results and discussions**

or barley crops, the db were slightly increased.

ed areas.

effect.

According to Zhang & Zhu et al. (2011) texture is defined as the spatial variation in gray value and is independent of color or luminance. Texture measures smoothness, coarse‐ ness and regularity of a region in an image. For the description of texture histograms, gray level co-occurrence matrix (GLCM), local statistics and characteristics of the frequen‐ cy spectrum are used. The GCLM mainly operates by calculating a matrix that is based on quantifying the difference between the grey levels of neighboring pixels in an image window. The main aim of this matrix is the quantification of the spatial pixel structure within this window. It was initially suggested as a mechanism for extracting texture

In the specific study, through the use of ENVI 4.7 software, 7 different statistical indica‐ tors of texture such as contrast, angular second moment, homogeneity, entropy, dissimi‐ larity, mean and variance were applied for carrying out the statistical texture analysis of all the typical ground objects. From those textural indicators, multiple RGB composites were constructed to improve the visual monitoring and interpretation of moisture affect‐

As it is clearly seen in Figure 12, in the downward of the catchment area (northeastern part) certain patches were inundated with water. Those patches are clearly observed with the low backscattering values and their corresponding dark pixels. However, in most of the cases the backscattering values were increased mainly because of volume scattering due to the moisture effect in the vegetation and plant cover. Concerning the southwestern part of the watershed where the most forested areas are established, due to the corresponding increase of the moisture after the extreme precipitation event, the backscatter values were generally increased due to the effect of double reflection by wa‐ ter (moisture) and tree trunks. Thus, generally the SAR backscattering intensity in forest areas changes to be higher in cases of inundation events. In addition, in certain areas of the southwestern part of the catchment area where there are more bare rock and soil patterns, the backscattering values were decreased due to the corresponding moisture

The values of radar backscatter coefficient for the different land cover classes as they were extracted from GeoEye-1 images, are tabulated in Table 3. The results were extracted in GIS environment (ArcGIS 10 software) through the use of zonal statistics application. According to Table 3, the backscatter coefficient in most of the classes increased after the precipitation event. The reason for this phenomenon was the overall moisture increase in the area. The backscatter of forest and urban areas was significantly increased (4.57 and 6.67dB) after the precipitation event due to the double reflection phenomenon. On the other hand, in other classes such as soil and bare rock, dB values declined due to water accumulation and the corresponding surface scattering effect. In agricultural areas of low vegetation, such as alfafa

(a)

(b)

**Figure 12.** (a) The catchment area before the precipitation event. (b) The catchment area after the precipitation event


**Table 3.** Radar Backscatter of ALOS PALSAR images for different land cover types and days

In order to improve image interpretation for water affected areas, several RGB composites were constructed, including microwave and textural bands. The optimum ones improved remarkably the final RGB composites and contributed to the delineation of the moisture af‐ fected areas, as shown in Fig. 13. Specifically, in Fig. 13a, the moisture affected areas are in‐ dicated in green tones. In Fig. 13b where only texture indicators were used the moisture affected areas are in light cyan color. On the one hand, the combination of speckle reducing Lee filter band and texture indicators in Fig. 13c, resulted in whitish color for flood prone areas. On the other hand, concerning the composite Fig. 13d, the combination of Mean, Var‐ iance and Homogeneity bands resulted in a light yellowish color for the moisture affected areas.

values and to delineate flood prone areas within a catchment area. In the first approach, sig‐ nal intensity statistics (backscattering values) were extracted to correlate moisture values with certain land cover classes. For this purpose, two high spatial resolution GeoEye-1 im‐

Integrated Remote Sensing and GIS Applications for Sustainable Watershed Management: A Case Study from Cyprus

http://dx.doi.org/10.5772/39307

117

In addition, texture analysis was employed to ALOS PALSAR images for the detection of flood prone areas. This method is based on the multi-temporal evaluation of the changes that occur between two ALOS PALSAR overpasses before and after the extreme precipita‐ tion event. The specific approach aims to highlight the changes and separate this informa‐ tion from unchanged backscatter signals. Moreover, the specific approach is used in order to improve the visual interpretation of SAR images. The visual inspection of filtered ALOS im‐ ages proved that there is a considerable change in radar backscattering when moisture af‐ fects land cover classes. Relative radar backscatter levels sampled in regions of interest and a LULC cover map indicated that different land cover classes yield different backscatter re‐

The results are useful for examining the potential of ALOS PALSAR images in recording soil moisture regime of an inundated area. However, the research team will continue observa‐ tion in longer time in case of flooding with the use of radar images. Such information is needed to understand flood mechanism and to better develop water discharge and flood

**6. Monitoring urban land cover with the use of satellite remote sensing**

This study uses an integrated approach that combines record of urban sprawl, land use and landscape metrics. Specifically, a remote sensing approach is applied to Aster satellite im‐ ages to analyze and identify patterns of urban changes within the spatial limits of Yialias watershed basin in the island of Cyprus. Moreover, there is an effort to optimize the classifi‐

Αn innovative methodology was developed for improving the classification accuracy of As‐ ter images concerning multi-temporal (2000 – 2010) record of urban land cover within the spatial limits of Yialias watershed basin in Cyprus. The phenomenon of spectral similarity of the spectral signatures of urban and marl/chalk formations, identified in the study area, stimulated the calculation of texture measurements in order to improve the traditional clas‐ sification products derived from spectral bands. Thus, with the use of ENVI 4.7 software 7 indicators of texture information were extracted for the images of 2000 and 2010. These indi‐

**techniques as a means of flood risk assessment in Cyprus.**

cation products by combining spectral and texture data to the final.

ages were used to create a LULC map to be used as a reference thematic map.

turns in response to moisture/flooding.

prevention system.

**6.1. Introduction**

*6.2.1. Methodology*

**6.2. Data and methodology**

**Figure 13.** a) RGB composite of the catchment area with the ALOS images before and after the precipitation event (R: Filtered image before precipitation, G: Filtered image after precipitation, B: Filtered image before precipitation - with green colors the areas where backscattering values were increased due to moisture effect are indicated). (b) Texture indicators RGB composite (R: Homogeneity, G: Contrast, B: Dissimilarity) (c). Combination of microwave bands and textural bands (R: Filtered image before precipitation, G: Filtered image after precipitation, B: Mean). (d) Texture indicators RGB composite (R: Mean, G: Variance, B: Homogeneity)

#### **5.6. Conclusions**

In this study, ALOS PALSAR imagery data (acquired before and after a certain precipitation event) proved to be useful for evaluating their potential to detect increased land moisture values and to delineate flood prone areas within a catchment area. In the first approach, sig‐ nal intensity statistics (backscattering values) were extracted to correlate moisture values with certain land cover classes. For this purpose, two high spatial resolution GeoEye-1 im‐ ages were used to create a LULC map to be used as a reference thematic map.

In addition, texture analysis was employed to ALOS PALSAR images for the detection of flood prone areas. This method is based on the multi-temporal evaluation of the changes that occur between two ALOS PALSAR overpasses before and after the extreme precipita‐ tion event. The specific approach aims to highlight the changes and separate this informa‐ tion from unchanged backscatter signals. Moreover, the specific approach is used in order to improve the visual interpretation of SAR images. The visual inspection of filtered ALOS im‐ ages proved that there is a considerable change in radar backscattering when moisture af‐ fects land cover classes. Relative radar backscatter levels sampled in regions of interest and a LULC cover map indicated that different land cover classes yield different backscatter re‐ turns in response to moisture/flooding.

The results are useful for examining the potential of ALOS PALSAR images in recording soil moisture regime of an inundated area. However, the research team will continue observa‐ tion in longer time in case of flooding with the use of radar images. Such information is needed to understand flood mechanism and to better develop water discharge and flood prevention system.
