2. Case study

As a follower of the Landsat 8 mission, the upcoming Landsat 9 which is expected to be launched in 2021 will continue to observe the Earth and collect valuable data for researchers

Spectral band Wavelength (μm) Resolution (m)

Band 1—Coastal/Aerosol 0.433–0.453 30 Band 2—Blue 0.450–0.515 30 Band 3—Green 0.525–0.600 30 Band 4—Red 0.630–680 30 Band 5—NIR 0.845–0.885 30 Band 6—SWIR-1 1.560–1660 30 Band 7—SWIR-2 2.100–2.300 30 Band 8—Panchromatic 0.500–0.680 15 Band 9—Cirrus 1.360–1.390 30 Band 10—LWIR-1 10.30–11.30 100 Band 11—LWIR-2 11.50–12.50 100

The latest SAR satellite developed by the European Space Agency, Sentinel-1, is an imaging radar satellite at C-band (5.405 GHz) consisting of a constellation of two satellites, Sentinel-1A, launched on 3 April, 2014, and Sentinel-1B, launched on 22 April 2016. The C-SAR instruments support the operation of dual polarization: HH + HV and VV + VH. Their main cover applications are monitoring sea ice zones and the Arctic environment; surveillance of marine environment; mentoring land surface motion risks; mapping of land surfaces like forest, water, and soil and agriculture; and mapping in support of humanitarian aid on crisis situation [31, 32]. As a result of a number of environmental circumstances and system factors, satellite images can be often distorted in geometry and brightness which requires preprocessing of the images before their use [33]. SAR data are exposed to radiometric and geometric distortions that should be removed or minimized. The preprocessing of the Sentinel-1 SAR data can be easily done with the Sentinel-1 Toolbox integrated in SNAP, and it contains few steps: (i) data preparation, (ii) radiometric calibration, (iii) multilooking, (iv) spackle reduction, (v) terrain

The data preparation consists of selecting the study area, selecting the data type needed for the

Radiometric calibration corrects the SAR image so that the pixel values represent the radar backscatter of the reflected surface. SAR calibration provides imagery in which the pixel values

all over the world.

Table 1. Landsat 8 band specification.

90 Wetlands Management - Assessing Risk and Sustainable Solutions

1.3. Sentinel-1 data

correction, and (vi) dB conversion.

study, selecting the date, and downloading the SAR product.

can be directly related to the radar backscatter of the scene.

In order to investigate wetlands' condition and seasonal changes, in this chapter, we review case study of thermal, optical, and radar remote sensing data and their relation for better understanding of wetland dynamics. As a study area, Seyfe Lake, located in the central Anatolian region, has been selected (Figure 1).

### 2.1. Study area

Turkey is the richest country in wetlands among Europe and the Middle East countries and with its geographical location plays an important part in the migration path for water birds. Seyfe Lake is located in the central Anatolian region, or 200 km northeast from Turkey's capital Ankara and 30 km east from Kirsehir. Seyfe Lake is a salty internal lake formed the base of a closed catchment area in a tectonic pit covering area approximately 10,700 ha. Over the years over 186 different bird species have been observed. Seyfe Lake is a first-degree natural site area under the protection of natural structure and ecological character pledged by the Ramsar Agreement in 1994. The lake is surrounded with agricultural areas, and according to the field

Figure 1. 14 June 2018, Seyfe Lake.

Figure 2. Seyfe Lake, June 1987 (left) and June 2018 (right).

observation on 14 June 2018, most of the agricultural fields around the lake are wheat, barley, and chickpeas.

This internationally valuable wetland area has been losing its value over the years. As seen in Figure 2, the water area has drastically decreased since 1987.

#### 2.2. Data and methods

Satellite images from both Landsat 8 and Sentinel-1 satellites were used for observing the seasonal changes in Seyfe Lake. As the time schedule is different for both of the satellites, the images with the smallest time gap were chosen for further investigation. Thus, from five dates, three were approximately 2 days apart, while two images were taken on the same day (Table 2).

The field measurements were used for comparison and more accurate classification of the water objects in the study area. Using Landsat 8 satellite data, LST, NDWI, and MNDWI were extracted, and afterward a threshold analysis and classification were made. Sentinel-1 data were preprocessed and the values from the dual polarization data were obtained. The relation between the optical, thermal, and radar satellite data was also investigated. The NDWI, MNDWI, LST, and the dual polarization (VV, VH) values were extracted using approximately

Landsat-8 Sentinel-1 Difference 19 February 2017 21 February 2017 2 days 23 March 2017 23 March 2017 10 h 27 June 2017 27 June 2017 10 h 30 August 2017 01 September 2017 2 days 17 October 2017 19 October 2017 2 days

Mapping and Monitoring Wetland Dynamics Using Thermal, Optical, and SAR Remote Sensing Data

http://dx.doi.org/10.5772/intechopen.80264

93

Table 2. Satellite data acquisitions over the study area, Seyfe Lake.

Figure 3. Flowchart of the used methodology.

The methods in this study include both field measurements and remote sensing measurements and techniques (Figure 3). The field measurements were performed on 14 June 2018, while the remote sensing data used in the seasonal change analyses were from different months in 2017. Mapping and Monitoring Wetland Dynamics Using Thermal, Optical, and SAR Remote Sensing Data http://dx.doi.org/10.5772/intechopen.80264 93


Table 2. Satellite data acquisitions over the study area, Seyfe Lake.

Figure 3. Flowchart of the used methodology.

observation on 14 June 2018, most of the agricultural fields around the lake are wheat, barley,

This internationally valuable wetland area has been losing its value over the years. As seen in

Satellite images from both Landsat 8 and Sentinel-1 satellites were used for observing the seasonal changes in Seyfe Lake. As the time schedule is different for both of the satellites, the images with the smallest time gap were chosen for further investigation. Thus, from five dates, three were approximately 2 days apart, while two images were taken on the same day

The methods in this study include both field measurements and remote sensing measurements and techniques (Figure 3). The field measurements were performed on 14 June 2018, while the remote sensing data used in the seasonal change analyses were from different months in 2017.

Figure 2, the water area has drastically decreased since 1987.

Figure 2. Seyfe Lake, June 1987 (left) and June 2018 (right).

92 Wetlands Management - Assessing Risk and Sustainable Solutions

and chickpeas.

(Table 2).

2.2. Data and methods

Figure 1. 14 June 2018, Seyfe Lake.

The field measurements were used for comparison and more accurate classification of the water objects in the study area. Using Landsat 8 satellite data, LST, NDWI, and MNDWI were extracted, and afterward a threshold analysis and classification were made. Sentinel-1 data were preprocessed and the values from the dual polarization data were obtained. The relation between the optical, thermal, and radar satellite data was also investigated. The NDWI, MNDWI, LST, and the dual polarization (VV, VH) values were extracted using approximately 160 random points. The same points were used in all of the satellite images used in this study and were then used for several statistical analyses.

soil from the further investigated land cover. Further ahead, when touched with hand, the level of moist in the soil could be felt, and the level of moist becomes even higher at the second measurement station, where the difference could be both felt and seen as in Figure 4, details 2 and 3. At the fourth measurement station (Figure 4, detail 4), the soil was significantly wet,

Mapping and Monitoring Wetland Dynamics Using Thermal, Optical, and SAR Remote Sensing Data

http://dx.doi.org/10.5772/intechopen.80264

95

Using several developed spectral water indices from two or more spectral bands requires an appropriate threshold in order to extract water bodies. The NDWI values range from �1 to +1 where values higher than zero are classified as water bodies. However, with careful adjustment of the NDWI threshold, more accurate results could be achieved [25, 36]. Taking into consideration the field measurements, NDWI threshold analysis on two different water indices was made in order to separate not just the water bodies but also the wetlands in the study area. The first index uses green and near-infrared band (NDWI) [26], while the second band uses green and middle near-infrared band (MNDWI) [27]. In the first case, a threshold value of zero has been set to separate the water bodies from the other land cover features, while in the

With the development of the thermal remote sensing technology, retrieving LST has become the topic for a number of research. A recently developed tool in Erdas Imagine calculates the

where ML represents the band-specific multiplicative rescaling factor, Qcal is the value of the thermal band, AL is the band-specific additive rescaling factor, and Oi is the correction for the thermal band [37]. After the digital numbers (DNs) have been converted to reflection, the data from the thermal band are converted from spectral radiance to brightness temperature (BT)

Normal difference vegetation index (NDVI) is needed for further calculation of the proportion

NDVI <sup>¼</sup> NIR band ð Þ� <sup>5</sup> R band ð Þ<sup>4</sup>

The proportion of vegetation is calculated according to Eq. (4). NDVI values for vegetation and

soil NDVI ð Þ <sup>v</sup> ¼ 0:5 and NDVI<sup>s</sup> ¼ 0:2 are suggested to apply in global conditions [38]:

Lλ ¼ ML∗Qcal þ AL � Oi (2)

<sup>þ</sup> <sup>1</sup> � <sup>273</sup>:<sup>15</sup> (3)

NIR band ð Þþ <sup>5</sup> R band ð Þ<sup>4</sup> (4)

and the field conditions did not allow further investigation/measurements.

second case, values close to zero (�0.1 and 0.1) were also set as a threshold value.

First, the top of atmospheric (TOA) spectral radiance (Lλ) is calculated using Eq. (2):

using the thermal constants provided in the metadata file Eq. (3) file (Table 3):

BT <sup>¼</sup> <sup>K</sup><sup>2</sup> Ln <sup>K</sup><sup>1</sup> Lλ

LST from Landsat 8 satellite data in few steps (Figure 5).

2.2.2. NDWI threshold analysis

2.2.3. LST estimation

of vegetation Eq. (4):

### 2.2.1. Field measurements

Before going to the field, the availability of satellite imagery was investigated. As the Landsat 8 satellite has a temporal resolution of 16 days, the overpass over the Seyfe Lake was estimated on 14 June 2018. As expected, the overpass happened on 14 June 2018, around 8:20 Coordinated Universal Time (UTC), or around 11:20 local time. The field measurements were taken around 11:00–12:00 local time. At the time of the measurements, the weather was clear, sunny, and hot with a maximum air temperature of 33�C. According to the meteorological records, in the past 2 weeks, three light rains have occurred in the study area, where 3 days before the field measurements, it has rained with over 1 mm (1 kg/m2 ).

Several parameters of the study area were taken into consideration during the field measurements: soil moisture, soil texture, soil color, and surface roughness.

The measurements were done 2 km along the study area (Figure 4).

According to the field measurements based on the feel of wet soil, the soil at all measurement points can be described as silty clay [35]. The soil moisture observation in the study area showed that three different types of soil can be distinguished: dry soil, moist soil, and wet soil. At the first measurement station (Figure 4, detail 1), approximately 3–5 m of the soil can be classified as dry soil with a minimal to no vegetation cover. However, since the spatial resolution of used satellite images, Landsat 8 is 30 m, it is impossible to separate the mentioned dry

Figure 4. Field measurements details, satellite image (red, green, blue: 6, 3, 4).

soil from the further investigated land cover. Further ahead, when touched with hand, the level of moist in the soil could be felt, and the level of moist becomes even higher at the second measurement station, where the difference could be both felt and seen as in Figure 4, details 2 and 3. At the fourth measurement station (Figure 4, detail 4), the soil was significantly wet, and the field conditions did not allow further investigation/measurements.

#### 2.2.2. NDWI threshold analysis

160 random points. The same points were used in all of the satellite images used in this study

Before going to the field, the availability of satellite imagery was investigated. As the Landsat 8 satellite has a temporal resolution of 16 days, the overpass over the Seyfe Lake was estimated on 14 June 2018. As expected, the overpass happened on 14 June 2018, around 8:20 Coordinated Universal Time (UTC), or around 11:20 local time. The field measurements were taken around 11:00–12:00 local time. At the time of the measurements, the weather was clear, sunny, and hot with a maximum air temperature of 33�C. According to the meteorological records, in the past 2 weeks, three light rains have occurred in the study area, where 3 days before the

Several parameters of the study area were taken into consideration during the field measure-

According to the field measurements based on the feel of wet soil, the soil at all measurement points can be described as silty clay [35]. The soil moisture observation in the study area showed that three different types of soil can be distinguished: dry soil, moist soil, and wet soil. At the first measurement station (Figure 4, detail 1), approximately 3–5 m of the soil can be classified as dry soil with a minimal to no vegetation cover. However, since the spatial resolution of used satellite images, Landsat 8 is 30 m, it is impossible to separate the mentioned dry

).

and were then used for several statistical analyses.

94 Wetlands Management - Assessing Risk and Sustainable Solutions

field measurements, it has rained with over 1 mm (1 kg/m2

Figure 4. Field measurements details, satellite image (red, green, blue: 6, 3, 4).

ments: soil moisture, soil texture, soil color, and surface roughness. The measurements were done 2 km along the study area (Figure 4).

2.2.1. Field measurements

Using several developed spectral water indices from two or more spectral bands requires an appropriate threshold in order to extract water bodies. The NDWI values range from �1 to +1 where values higher than zero are classified as water bodies. However, with careful adjustment of the NDWI threshold, more accurate results could be achieved [25, 36]. Taking into consideration the field measurements, NDWI threshold analysis on two different water indices was made in order to separate not just the water bodies but also the wetlands in the study area. The first index uses green and near-infrared band (NDWI) [26], while the second band uses green and middle near-infrared band (MNDWI) [27]. In the first case, a threshold value of zero has been set to separate the water bodies from the other land cover features, while in the second case, values close to zero (�0.1 and 0.1) were also set as a threshold value.

### 2.2.3. LST estimation

With the development of the thermal remote sensing technology, retrieving LST has become the topic for a number of research. A recently developed tool in Erdas Imagine calculates the LST from Landsat 8 satellite data in few steps (Figure 5).

First, the top of atmospheric (TOA) spectral radiance (Lλ) is calculated using Eq. (2):

$$L\lambda = M\_L \ast Q\_{cal} + A\_L - O\_i \tag{2}$$

where ML represents the band-specific multiplicative rescaling factor, Qcal is the value of the thermal band, AL is the band-specific additive rescaling factor, and Oi is the correction for the thermal band [37]. After the digital numbers (DNs) have been converted to reflection, the data from the thermal band are converted from spectral radiance to brightness temperature (BT) using the thermal constants provided in the metadata file Eq. (3) file (Table 3):

$$BT = \frac{K\_2}{Ln\left[\left(\frac{K\_1}{L\lambda}\right) + 1\right]} - 273.15\tag{3}$$

Normal difference vegetation index (NDVI) is needed for further calculation of the proportion of vegetation Eq. (4):

$$NDVI = \frac{NIR(band5) - R(band4)}{NIR(band5) + R(band4)} \tag{4}$$

The proportion of vegetation is calculated according to Eq. (4). NDVI values for vegetation and soil NDVI ð Þ <sup>v</sup> ¼ 0:5 and NDVI<sup>s</sup> ¼ 0:2 are suggested to apply in global conditions [38]:

Figure 5. Flowchart of the LST calculation.

$$P\_V = \left(\frac{\text{NDVI} - \text{NDVI}\_s}{\text{NDVI}\_v - \text{NDVI}\_s}\right)^2 \tag{5}$$

with Eq. (7) where the emissivity constant values are 0.991 for water, 0.962 for built-up areas/

ε<sup>s</sup>λ, NDVI < NDVIs

ε<sup>s</sup><sup>λ</sup> þ C, NDVI > NDVIv

NDVI values lower than 0 are considered to be water, NDVI values between 0 and 0.2 are considered to be bare soil, and NDVI values between 0.2 and 0.4 are considered to be mixtures of soil and vegetation cover; when the NDVI value is greater than 0.4, it is considered to be

The LST or the emissivity-corrected land surface temperature Ts is computed with Eq. (8) [40]:

where Ts is the LST in Celsius, λ is the wavelength of emitted radiance (λ = 10.895) [41], and r

where σ is the Boltzmann constant (1.38 � 10–23 J/K), h is Planck's constant (6.626 � 10–34 J s),

).

Comparing the field measurements with remote sensing data, different threshold NDWI and MNDWI values were set for more accurate classification. Since both NIR and SWIR parts of the

<sup>r</sup> ln ελ

n o h i (8)

<sup>σ</sup> ð Þ <sup>1</sup>:438 x 10–2mK (9)

Ts <sup>¼</sup> BT <sup>1</sup> <sup>þ</sup> <sup>λ</sup>BT

<sup>r</sup> <sup>¼</sup> <sup>h</sup> <sup>c</sup>

ε<sup>v</sup>λPv þ ε<sup>s</sup>λð Þþ 1 � Pv C, NDVIs ≤ NDVI ≤ NDVIv

Mapping and Monitoring Wetland Dynamics Using Thermal, Optical, and SAR Remote Sensing Data

http://dx.doi.org/10.5772/intechopen.80264

(7)

97

bare soil, 0.966 for a mixture of soil and vegetation, and 0.973 for vegetated areas [19]:

K1 1321.08 K2 777.89

ML 0.000342 AL 0.1

Oi 0.29

ελ ¼

Table 3. Metadata of Landsat 8 satellite image.

is a constant calculated with Eq. (9):

and c is the velocity of light (2.998 � 108 ms�<sup>1</sup>

3.1. Field measurements: NDWI threshold

covered with vegetation.

Thermal Constant–Band 10

Rescaling Factor–Band 10

Correction–Band 10

3. Results

8 >><

>>:

The emissivity can be calculated following Eq. (6):

$$
\varepsilon\_{\lambda} = \varepsilon\_{v\lambda} P\_v + \varepsilon\_{s\lambda} (1 - P\_v) + \mathbb{C}\_{\lambda} \tag{6}
$$

where ε<sup>v</sup> and ε<sup>s</sup> are the vegetation and soil emissivities, respectively, and C represents the surface roughness taken as a constant value of 0.005 [39]. The condition can be represented


Table 3. Metadata of Landsat 8 satellite image.

with Eq. (7) where the emissivity constant values are 0.991 for water, 0.962 for built-up areas/ bare soil, 0.966 for a mixture of soil and vegetation, and 0.973 for vegetated areas [19]:

$$\varepsilon\_{\lambda} = \begin{cases} \varepsilon\_{s\lambda\nu} & \text{NDVI} < \text{NDVI}\_s \\\\ \varepsilon\_{v\lambda} P\_v + \varepsilon\_{s\lambda} (1 - P\_v) + \mathbb{C}, & \text{NDVI}\_s \le \text{NDVI} \le \text{NDVI}\_v \\\\ \varepsilon\_{s\lambda} + \mathbb{C}, & \text{NDVI} > \text{NDVI}\_v \end{cases} \tag{7}$$

NDVI values lower than 0 are considered to be water, NDVI values between 0 and 0.2 are considered to be bare soil, and NDVI values between 0.2 and 0.4 are considered to be mixtures of soil and vegetation cover; when the NDVI value is greater than 0.4, it is considered to be covered with vegetation.

The LST or the emissivity-corrected land surface temperature Ts is computed with Eq. (8) [40]:

$$T\_s = \frac{BT}{\left\{1 + \left[\frac{\lambda BT}{\rho} \ln \varepsilon\_\lambda\right]\right\}}\tag{8}$$

where Ts is the LST in Celsius, λ is the wavelength of emitted radiance (λ = 10.895) [41], and r is a constant calculated with Eq. (9):

$$
\rho = h \frac{c}{\sigma} \text{ (1.438 x 10-2 } m \text{ K)}\tag{9}
$$

where σ is the Boltzmann constant (1.38 � 10–23 J/K), h is Planck's constant (6.626 � 10–34 J s), and c is the velocity of light (2.998 � 108 ms�<sup>1</sup> ).

#### 3. Results

(5)

PV <sup>¼</sup> NDVI � NDVIs NDVIv � NDVIs <sup>2</sup>

where ε<sup>v</sup> and ε<sup>s</sup> are the vegetation and soil emissivities, respectively, and C represents the surface roughness taken as a constant value of 0.005 [39]. The condition can be represented

ελ ¼ ε<sup>v</sup>λPv þ ε<sup>s</sup>λð Þþ 1 � Pv C<sup>λ</sup> (6)

The emissivity can be calculated following Eq. (6):

96 Wetlands Management - Assessing Risk and Sustainable Solutions

Figure 5. Flowchart of the LST calculation.

#### 3.1. Field measurements: NDWI threshold

Comparing the field measurements with remote sensing data, different threshold NDWI and MNDWI values were set for more accurate classification. Since both NIR and SWIR parts of the electromagnetic spectrum are sensitive to water and wet areas, in this study, both NDWI and MNDWI were considered. The study area, Seyfe Lake contains water area, wetland area, and dry area. The dry area is known to be salt soil [42]. For the threshold value, salt soil areas were not taken into consideration.

As seen in Figure 6, the SWIR band is more sensitive to water bodies, as well as to soil moisture. While NIR can only detect the open water bodies, with SWIR beside open water bodies, both shallow water and moist soil can be distinguished from the other land covers.

The higher values of zero thresholds for both NDWI and MNDWI indices indicate water body areas. As seen in Figure 7, where the Landsat image is a combination of SWIR, red and green bands, the NDWI threshold results did not classify the shallow water areas as a water body while using the MNDWI successfully extracted both open water bodies and shallow water bodies. However, setting one threshold can only separate the land cover into two classes, in this case, water (NDWI > 0) and other (NDWI < 0), and the wet soils cannot be distinguished from the other land covers. Setting additional threshold values in both NDWI and MNDWI values helped us to distinguish two more classes, shallow water and moist soil, or wetland (Figures 7 and 8).

As it can be seen in Figures 7 and 8, the threshold of zero in the NDWI was not successful in extracting shallow water area, while the MNDWI successfully extracted both water object and shallow water areas, but both indices failed in extracting the wet soil within the study area. On the other hand, setting two additional thresholds close to zero (�0.1 and 0.1) showed successful results in distinguishing wetland, shallow water, and water bodies from the other land cover features. In the visual comparison with the field data, the results from the additional

Mapping and Monitoring Wetland Dynamics Using Thermal, Optical, and SAR Remote Sensing Data

http://dx.doi.org/10.5772/intechopen.80264

99

The results from the seasonal changes in the Sefye Lake showed drastic changes over the months. Using both NDWI and MNDWI, according to the field measurements, four classes have been determined, wet soil, salt soil, shallow water, and water. In Figure 9, the Sentinel-1, the NDWI classification, and the RGB images from every investigated month are given.

Using the pixel number of every class, the area of each class was calculated, and the results are presented in Figure 10. As it can be seen, after March, the water class has decreased for more than 75%, and the area of the shallow water and wet soil has not changed, while the area or the salt soil has significantly increased. The results from August have the highest area of salt soil and the lowest area of water, shallow water, and wet soil, while in October the areas of the salt soil class have decreased, and the areas of the wet soil and shallow water classes have increased. While the water area has been calculated to be more than 20 km<sup>2</sup> in March, in

LST is one of the most important parameters for wetland dynamics (Figure 11). The relation between the average LST values and the seasonal changes in Seyfe Lake using NDWI indices has also been investigated. The statistics between these two variables indicates a strong correlation between the LST changes and the changes in every class separately. Thus, the correlation between the wet soil (�0.51), shallow water (�0.58), and water (�0.75) indicates a negative relation, meaning that with the increase of the LST, the values of the mentioned classes decrease. The correlation between the LST and the salt soil has been calculated to be more than 0.8, meaning, with the increase of the LST, the increase of the salt soil (dry soil) area occurs.

.

thresholds were satisfactorily accurate.

Figure 8. NDWI (left); MNDWI (right).

October the water area has been calculated to be 0 km<sup>2</sup>

3.2. Seasonal changes

Figure 6. Seyfe Lake: RGB-NIR, red and green (left); RGB-SWIR, red and green (right).

Figure 7. NDWI > 0 (left); MNDWI > 0 (right).

Mapping and Monitoring Wetland Dynamics Using Thermal, Optical, and SAR Remote Sensing Data http://dx.doi.org/10.5772/intechopen.80264 99

Figure 8. NDWI (left); MNDWI (right).

electromagnetic spectrum are sensitive to water and wet areas, in this study, both NDWI and MNDWI were considered. The study area, Seyfe Lake contains water area, wetland area, and dry area. The dry area is known to be salt soil [42]. For the threshold value, salt soil areas were

As seen in Figure 6, the SWIR band is more sensitive to water bodies, as well as to soil moisture. While NIR can only detect the open water bodies, with SWIR beside open water bodies, both shallow water and moist soil can be distinguished from the other land covers.

The higher values of zero thresholds for both NDWI and MNDWI indices indicate water body areas. As seen in Figure 7, where the Landsat image is a combination of SWIR, red and green bands, the NDWI threshold results did not classify the shallow water areas as a water body while using the MNDWI successfully extracted both open water bodies and shallow water bodies. However, setting one threshold can only separate the land cover into two classes, in this case, water (NDWI > 0) and other (NDWI < 0), and the wet soils cannot be distinguished from the other land covers. Setting additional threshold values in both NDWI and MNDWI values helped us to distinguish two more classes, shallow water and moist soil, or wetland

Figure 6. Seyfe Lake: RGB-NIR, red and green (left); RGB-SWIR, red and green (right).

Figure 7. NDWI > 0 (left); MNDWI > 0 (right).

not taken into consideration.

98 Wetlands Management - Assessing Risk and Sustainable Solutions

(Figures 7 and 8).

As it can be seen in Figures 7 and 8, the threshold of zero in the NDWI was not successful in extracting shallow water area, while the MNDWI successfully extracted both water object and shallow water areas, but both indices failed in extracting the wet soil within the study area. On the other hand, setting two additional thresholds close to zero (�0.1 and 0.1) showed successful results in distinguishing wetland, shallow water, and water bodies from the other land cover features. In the visual comparison with the field data, the results from the additional thresholds were satisfactorily accurate.

#### 3.2. Seasonal changes

The results from the seasonal changes in the Sefye Lake showed drastic changes over the months. Using both NDWI and MNDWI, according to the field measurements, four classes have been determined, wet soil, salt soil, shallow water, and water. In Figure 9, the Sentinel-1, the NDWI classification, and the RGB images from every investigated month are given.

Using the pixel number of every class, the area of each class was calculated, and the results are presented in Figure 10. As it can be seen, after March, the water class has decreased for more than 75%, and the area of the shallow water and wet soil has not changed, while the area or the salt soil has significantly increased. The results from August have the highest area of salt soil and the lowest area of water, shallow water, and wet soil, while in October the areas of the salt soil class have decreased, and the areas of the wet soil and shallow water classes have increased. While the water area has been calculated to be more than 20 km<sup>2</sup> in March, in October the water area has been calculated to be 0 km<sup>2</sup> .

LST is one of the most important parameters for wetland dynamics (Figure 11). The relation between the average LST values and the seasonal changes in Seyfe Lake using NDWI indices has also been investigated. The statistics between these two variables indicates a strong correlation between the LST changes and the changes in every class separately. Thus, the correlation between the wet soil (�0.51), shallow water (�0.58), and water (�0.75) indicates a negative relation, meaning that with the increase of the LST, the values of the mentioned classes decrease. The correlation between the LST and the salt soil has been calculated to be more than 0.8, meaning, with the increase of the LST, the increase of the salt soil (dry soil) area occurs.

Figure 9. Seasonal wetland dynamic: (A) February, (B) March, (C) June, (D) August, and (E) October.

#### 3.3. Remote sensing data statistics

Using the random point values that were added on the Seyfe Lake area, various statistical analyses were performed in order to find a correlation between the used remote sensing data. Thus, the relation between VH, LST, and MNDWI and VV, LST, and MNDWI, between LST and MNDWI, and between VH and VV has been under detailed investigation (Table 4).

Mapping and Monitoring Wetland Dynamics Using Thermal, Optical, and SAR Remote Sensing Data

http://dx.doi.org/10.5772/intechopen.80264

101

As seen in Table 4, the relation between the investigated parameters has been shown using three different statistical calculations; the multiple R is actually the correlation coefficient between the two variables, and it tells how strong the relationship between these variables is. Values close to 1 indicated a strong relation, while values close to 0 meant no relation at all. R2

on the other hand, tells us how much of the change in the dependent variable can be explained by the independent variable. The R<sup>2</sup> value can be easily supported with significance variable. So when the significance variable is less than 0.05, the results are significant, or it means that

the results did not occur by chance.

Figure 11. LST results from 23 March 2017.

Figure 10. Seasonal Seyfe Lake dynamics.

,

Mapping and Monitoring Wetland Dynamics Using Thermal, Optical, and SAR Remote Sensing Data http://dx.doi.org/10.5772/intechopen.80264 101

Figure 10. Seasonal Seyfe Lake dynamics.

Figure 11. LST results from 23 March 2017.

3.3. Remote sensing data statistics

100 Wetlands Management - Assessing Risk and Sustainable Solutions

Using the random point values that were added on the Seyfe Lake area, various statistical analyses were performed in order to find a correlation between the used remote sensing data.

Figure 9. Seasonal wetland dynamic: (A) February, (B) March, (C) June, (D) August, and (E) October.

Thus, the relation between VH, LST, and MNDWI and VV, LST, and MNDWI, between LST and MNDWI, and between VH and VV has been under detailed investigation (Table 4).

As seen in Table 4, the relation between the investigated parameters has been shown using three different statistical calculations; the multiple R is actually the correlation coefficient between the two variables, and it tells how strong the relationship between these variables is. Values close to 1 indicated a strong relation, while values close to 0 meant no relation at all. R2 , on the other hand, tells us how much of the change in the dependent variable can be explained by the independent variable. The R<sup>2</sup> value can be easily supported with significance variable. So when the significance variable is less than 0.05, the results are significant, or it means that the results did not occur by chance.


separating water bodies, shallow water, and wet soil from the other land cover classes. It has been concluded that MNDWI performs better in both water and wet soil extraction than NDWI. According to the MNDWI seasonal changes, Seyfe Lake is a very dynamic wetland, and its water area depends on the seasonal temperature. Thus, the water area is inversely proportional to the air temperature. So, for a temporal monitoring of the Lake, in this case, it is

Mapping and Monitoring Wetland Dynamics Using Thermal, Optical, and SAR Remote Sensing Data

http://dx.doi.org/10.5772/intechopen.80264

103

Figure 12. LST-MNDWI relation from June 2017 (left); VV-VH relation from October 2017 (right).

Wetlands ensure critical habitat for wildlife such as migrating water birds. Information gathered from monitoring wetlands may help land managers about the quality of wetlands. Remote sensing data has been the most useful tool to achieve spatial and temporal information

Wetlands are some of the most important ecosystems on Earth if wetlands' function is managed properly which provides tremendous fish and wildlife habitat but also improves groundwater quality and natural floodwater control. Global climate change and anthropogenic impact degrade wetlands, which creates a serious problem in identifying and quantifying wetland areas. Consequently, it is critical to be able to assess the status and quality of our

Remote sensing is a major source of spatial information about the land cover. Although a number of studies have investigated the relation between SAR and optical sensors in different land classes, the relation between backscatter values, NDWI, and LST values in wetland classes has not been a subject of a delicate investigation. In this chapter we investigate the relation between several remote sensing parameters, from a different aspect, trying to find a significant relation that will explain the wetland dynamics for better mapping, monitoring, and managing of wetland areas. The results in this study show a strong relation between the investigated

recommended to use annual data from the same season/month.

about wetlands.

remaining wetlands.

4. Discussion and conclusion

Table 4. Statistical results.

Since every season has different characteristics, the results were reviewed separately for each month and for each relation. February can be characterized with the lowest LST and high water level. The correlation coefficient between the investigated variables was highest in February for VH-MNDWI, VH-LST, and VV-MDWI. With the assumption that the LST is the independent variable, and the MNDWI is the dependent variable, the R2 value shows that approximately 39% of the data or the rise of the temperature affects the water area. The results for March were not much different from the results from February with the only difference in the VV-MNDWI and VV-LST relation where the R values were significantly lowered. According to the statistics, the results from June have the most significant value in this study. As seen from the seasonal changes of the Seyfe Lake, in June the MNDWI values showed that the water area has slowly transformed into shallow water and wet soil. In this month the highest correlation coefficient of 0.61 was observed and also the highest R<sup>2</sup> of 0.422. Also, in June, the relation between VV and VH polarization was high with a correlation coefficient of 0.725 and R2 of 0.52. While the other relation in August and October did not show significant results, the relation between the two polarizations, VV and VH, was noticeable with correlation coefficients higher than 0.76 and R2 higher than 0.58 in both cases (Figure 12).

#### 3.4. Summary

The initial investigation shows a strong relation between the parameters retrieved from the SAR, thermal, and optical satellite data and leads to better understanding of the wetland dynamics. Following the field measurements performed on 14 June 2018, a series of threshold analyses of NDWI and MNDWI were used in order to determine the optimal values for

Mapping and Monitoring Wetland Dynamics Using Thermal, Optical, and SAR Remote Sensing Data http://dx.doi.org/10.5772/intechopen.80264 103

Figure 12. LST-MNDWI relation from June 2017 (left); VV-VH relation from October 2017 (right).

separating water bodies, shallow water, and wet soil from the other land cover classes. It has been concluded that MNDWI performs better in both water and wet soil extraction than NDWI. According to the MNDWI seasonal changes, Seyfe Lake is a very dynamic wetland, and its water area depends on the seasonal temperature. Thus, the water area is inversely proportional to the air temperature. So, for a temporal monitoring of the Lake, in this case, it is recommended to use annual data from the same season/month.

Wetlands ensure critical habitat for wildlife such as migrating water birds. Information gathered from monitoring wetlands may help land managers about the quality of wetlands. Remote sensing data has been the most useful tool to achieve spatial and temporal information about wetlands.

### 4. Discussion and conclusion

Since every season has different characteristics, the results were reviewed separately for each month and for each relation. February can be characterized with the lowest LST and high water level. The correlation coefficient between the investigated variables was highest in February for VH-MNDWI, VH-LST, and VV-MDWI. With the assumption that the LST is the independent variable, and the MNDWI is the dependent variable, the R2 value shows that approximately 39% of the data or the rise of the temperature affects the water area. The results for March were not much different from the results from February with the only difference in the VV-MNDWI and VV-LST relation where the R values were significantly lowered. According to the statistics, the results from June have the most significant value in this study. As seen from the seasonal changes of the Seyfe Lake, in June the MNDWI values showed that the water area has slowly transformed into shallow water and wet soil. In this month the highest correlation coefficient of 0.61 was observed and also the highest R<sup>2</sup> of 0.422. Also, in June, the relation between VV and VH polarization was high with a correlation coefficient of 0.725 and R2 of 0.52. While the other relation in August and October did not show significant results, the relation between the two polarizations, VV and VH, was noticeable with correla-

Relation VH/MNDWI VH/LST LST/MNDWI VV/MNDWI VV/LST VV/VH

R square 0.198 0.162 0.385 0.184 0.141 0.233 Significance F 0.000 0.000 0.000 0.000 0.000 0.000

R square 0.103 0.043 0.383 0.004 0.006 0.046 Significance F 0.000 0.009 0.000 0.418 0.322 0.007

R square 0.010 0.059 0.422 0.038 0.007 0.525 Significance F 0.213 0.002 0.000 0.014 0.293 0.000

R square 0.047 0.073 0.001 0.059 0.307 0.582 Significance F 0.006 0.001 0.000 0.002 0.000 0.000

R square 0.097 0.048 0.000 0.184 0.086 0.609 Significance F 0.000 0.006 0.907 0.000 0.000 0.000

February Multiple R 0.445 0.402 0.621 0.429 0.375 0.483

102 Wetlands Management - Assessing Risk and Sustainable Solutions

March Multiple R 0.321 0.209 0.619 0.065 0.079 0.215

June Multiple R 0.100 0.243 0.650 0.196 0.084 0.725

August Multiple R 0.218 0.270 0.035 0.242 0.554 0.763

October Multiple R 0.311 0.220 0.009 0.429 0.293 0.780

tion coefficients higher than 0.76 and R2 higher than 0.58 in both cases (Figure 12).

The initial investigation shows a strong relation between the parameters retrieved from the SAR, thermal, and optical satellite data and leads to better understanding of the wetland dynamics. Following the field measurements performed on 14 June 2018, a series of threshold analyses of NDWI and MNDWI were used in order to determine the optimal values for

3.4. Summary

Table 4. Statistical results.

Wetlands are some of the most important ecosystems on Earth if wetlands' function is managed properly which provides tremendous fish and wildlife habitat but also improves groundwater quality and natural floodwater control. Global climate change and anthropogenic impact degrade wetlands, which creates a serious problem in identifying and quantifying wetland areas. Consequently, it is critical to be able to assess the status and quality of our remaining wetlands.

Remote sensing is a major source of spatial information about the land cover. Although a number of studies have investigated the relation between SAR and optical sensors in different land classes, the relation between backscatter values, NDWI, and LST values in wetland classes has not been a subject of a delicate investigation. In this chapter we investigate the relation between several remote sensing parameters, from a different aspect, trying to find a significant relation that will explain the wetland dynamics for better mapping, monitoring, and managing of wetland areas. The results in this study show a strong relation between the investigated variables. However, while some variables have strong relation when the water area is high, others have stronger relation when the lake is completely dried. The results of the statistical analyses from the month of June were taken to be highly significant from two points of view; the correlation between LST and MNDWI was more than 0.65, while the R2 was more than 0.42. Taking in consideration the relation between the average temperature from each month and the area of each class, it can be concluded that with a correlation of �0.75, the water area depends on the temperature, or as the temperature increases, the water area decreases.

References

2017. pp. 271-277

2015;7(9):12503-12538

1A images. Remote Sensing. 2016;8:570

Sentinel-1 data. Remote Sensing. 2016;8(9):756

Spatial Information Sciences. Vol. 42, Issue 3; 2018

777

[1] Ustin S. Manual of Remote Sensing: Remote Sensing for Natural Resource Management

Mapping and Monitoring Wetland Dynamics Using Thermal, Optical, and SAR Remote Sensing Data

http://dx.doi.org/10.5772/intechopen.80264

105

[2] Czajkowski K, Torbick N, Lawrence P. Application and assessment of a GIScience model for jurisdictional wetlands identification in Northwestern Ohio. In: Wetland and Water Resource Modeling and Assessment: A Watershed Perspective. CRC Press; 2007. pp. 2-12

[3] Schmidt K, Skidmore A. Spectral discrimination of vegetation types in a coastal wetland.

[4] Kaplan G, Avdan U. Mapping and monitoring wetlands using Sentinel-2 satellite imagery. In: ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences;

[5] Kaplan G, Avdan U, Avdan ZY. Estimation of wetland biomass using Sentinel-2 Images. Icocce Capadokia, Nevsehir, Turkey: Digital Proceeding of ICOCEE; May 2017

[6] Gorham E. Northern peatlands: Role in the carbon cycle and probable responses to

[7] Guo M, Li J, Sheng C, Xu J, Wu L. A review of wetland remote sensing. Sensors. 2017;17(4):

[8] Zhang S, Na X, Kong B, Wang Z, Jiang H, Yu H, et al. Identifying wetland change in

[9] Jones JW. Efficient wetland surface water detection and monitoring via landsat: Comparison with in situ data from the everglades depth estimation network. Remote Sensing.

[10] Baker C, Lawrence RL, Montagne C, Patten D. Change detection of wetland ecosystems using Landsat imagery and change vector analysis. Wetlands. 2007;27(3):610-619

[11] Frohn RC, Reif M, Lane C, Autrey B. Satellite remote sensing of isolated wetlands using

[12] Cazals C, Rapinel S, Frison P, Bonis A, Mercier G, Mallet C, et al. Mapping and characterization of hydrological dynamics in a coastal marsh using high temporal resolution Sentinel-

[13] Muro J, Canty M, Conradsen K, Hüttich C, Nielsen AA, Skriver H, et al. Short-term change detection in wetlands using Sentinel-1 time series. Remote Sensing. 2016;8(10):795

[14] Dabrowska-Zielinska K, Budzynska M, Tomaszewska M, Malinska A, Gatkowska M, Bartold M, et al. Assessment of carbon flux and soil moisture in wetlands applying

[15] Kaplan G, Avdan U. Sentinel-1 and Sentinel-2 data fusion for wetlands mapping: Balikdami, Turkey. In: International Archives of the Photogrammetry, Remote Sensing &

object-oriented classification of Landsat-7 data. Wetlands. 2009;29(3):931-941

China's Sanjiang Plain using remote sensing. Wetlands. 2009;29(1):302-313

and Environmental Monitoring. Hoboken, NJ: Wiley; 2004

climatic warming. Ecological Applications. 1991;1(2):182-195

Remote Sensing of Environment. 2003;85(1):92-108

As stated in previous studies, with careful adjustment of the NDWI threshold, more accurate results could be achieved [25, 36]. Thus, it can be concluded that the values close to zero in the MNDWI threshold can give valuable information about the soil moisture. The positive values close to zero (0.1) indicate the shallow water areas, while the negative values close to zero (�0.1) indicate wet soil.

Even though in more of the cases the relation between the Landsat 8 and the Sentinel-1 parameters was high, in some cases their correlation was close to zero. Thus, while the changes of the VH values in March could be explained about 20% with the help of LST, the VV values could be explained about 20% with the help of MNDWI in October. The relation between the two polarization has highest values when the water level is at its lowest level because at that time the study area is mostly covered with dry soil. For a better understanding of wetland dynamics, we recommend using new techniques and different data fusion for exploring the full potential of remote sensing in wetland monitoring, supported with field measurements.
