Research Article - (2018) Volume 21, Issue 1

Detection of Shifts in the Southeastern Coastline of the Caspian Sea with Landsat MSS, TM and ETM+ Images in 1977, 1987 and 2001

Seyyed Javad Hoseini* and Qadir Ashournejad
Department of Physical Geography, University of Tehran, Azin Alley, Ghods Street, Tehran, Iran
*Corresponding Author: Seyyed Javad Hoseini, Ph.D Candidate of Geomorphology, Department of Physical Geography, Faculty of Geography, University of Tehran, Azin Alley, Ghods Street, Tehran, Iran, Tel: +25-99179176 Email:

Abstract

Shorelines are the most important linear phenomena on earth’s surface that have a dynamic nature. Thus, favorite coastal management and environmental protection towards sustainable development requires the extraction of the coastline and its changes. For this purpose, coastal zone monitoring in an appropriate time context is of great importance. As one of reliable and rather accurate sources, remote sensing data and satellite imagery in different periods are used for investigation and interpretation of shoreline changes and quantitative measurements. In this study, we tried to apply a new method to determine the shoreline changes using remote sensing and geographic information system. This method possesses simplicity and has acceptable results as well and is able to control and evaluate the results of the research process and its reliability is approved. For this aim, Mean Shift Clustering algorithm was developed for segmentation of MSS, TM and ETM+ images of 1977 (June 6th), 1987 (14 June) and 2001 to investigate the changes of south-eastern coastline of the Caspian Sea over a length of 16 km. After determination of shorelines over the intended years, we extracted the coastline of 1977 as the baseline, and then determined the changes between 1987 and 2001. The average measurements show a 14.03 meters recession of sea water for 1987 compared to that of 1977. However, the results for 2001 show an advancing sea water around 69.8 meters in 1977. Since the results of coastline changes are considered as the fundamental basis for analysis and other related applications, the data integrity and no other possible errors are essential. To control the accuracy of the extracted data on cross sections perpendicular to the coastline, statistical tests of sample, median absolute deviation, Z-Score and box plot was used. The results, confirms that the extracted data have no errors.

Keywords: Caspian Sea; Coastline changes; Remote sensing; andsat; Segmentation

Introduction

Because of the influence of dynamic morphologic phenomenon and the adverse effects incurred from a variety of usages, coasts are always exposed to destruction. As a result, the preparation and implementation of comprehensive protection programs have involved the experts and authorities for a long time to balance the usages, reduce the destruction of resources and achieve a sustainable development [1]. For optimized coastal management, the knowledge of past, present and future position of coastline and its changes is essential. Shorelines are the most important features on earth and are constantly changing in short-term and long-term. These changes may be caused by natural or human factors. About 60 percentages of the populations live in coastal areas. Knowledge of the behaviour of coastline helps experts in beach management, designing and construction of shore facilities, and determining the safe coastal margins [2-5]. After determining and comparing the coastline in different periods, as well as investigating the reasons and factors of the changes, it is possible to forecast and make decisions about the future locations and conditions of the beach. The knowledge of the coastline behaviour also provides way to deal with the negative impacts and to benefit from the positive effects [6].

Previous studies related to monitoring the coastline changes can be explained as below; from 1807 to 1927, all coastline mappings were carried out by field surveying. In 1927, the true potential of aerial photography to complete maps of the coastline was found. During 1927 and 1980, aerial photographs were considered to be a special source for coastline mapping. However, the limitation of these photos were the large number of aerial photos even in regional scale, high cost time-consuming of collection, correction, analysis and transmission of data to the map, as well as the black and white nature of the photos. Moreover, their non-digital format, small time coverage and low geometric accuracy of aerial photos can be mentioned as well. Following the evolution of the researches related to monitoring the coastline changes with the arrival of satellite images, remote sensing and geographic information systems, the analysis of coastline changes and the related researches were and are being carried out with a very high accuracy, high spatial and temporal coverage and accurate and understandable outputs. Nowadays, application of satellite images is regarded as the newest and most economic method [7]. In addition to the low cost and high precision, the remote sensing methods have made it possible to continuously observe and manage the beaches. These changes may be caused by natural or human factors. About 60 percentages of the populations live in coastal areas. Knowledge of the behaviour of coastlines helps beaches management in designing and construction of shore facilities and determining safe coastal margins. After determining coastline in different periods and comparisons of them and investigating reasons and factors of changes, forecast and decision making about the future locations and conditions of the beach is possible. Having the knowledge of the behaviour of coastline also provides a way to deal with the negative impacts and to exploit the benefits of the positive effects [2].

In this study, we will investigate the changes of 16 km of south eastern coastline of the Caspian Sea. Fluctuations in this area are very high in recent years and studying the changes is necessary for management of coastal areas in the region [8-10]. In recent years, the performed studies have focused on advancement and retreatment of the coastal line of the Caspian Sea. Moreover, the coastal vulnerability requires extensive research. One of these methods uses aerial photos and GIS to deal with changes of natural landscape and land use in southern coast of the Caspian Sea in Noor city coasts [11-15]. As investigated the coastal changes of Kiashahr-Dastak area (North East Rasht) in a 26-year period (1967 to 1993). In this study, comparison of coastlines indicates regressive coasts (sedimentation areas) at the arrival of the rivers to the sea and progressive coasts (eroded or sunk beaches) in the eastern parts of Sefidrood delta, especially in the northern parts of the Dastak. They studied shoreline changes in Amirabad area over the years 1966, 2004 and 2012. Their study showed that the coastal area has lost the state balance and stability due to the port facilities and on the banks of the upstream (west side) and downstream (east side), sedimentation and erosion can be seen respectively. Because of the importance of the changes and forecasting the coastline mount changes of the Caspian Sea. Recent studies calculated the coastline changes in the deltaic range of Gorgan River. He predicted the future changes of the line based on field evidences and assessing the efficiency of the methods [16-19]. They studied the coastline changes of the Caspian Sea off the coast of Iran for a period of 26 years (1975 to 2001) using MSS, TM and ETM+ Landsat images. In their research, they partitioned the Caspian Sea in Iran into 22 coastal areas and calculated the coastline changes by creating cross sections perpendicular to the coastline for each area. They used high pass filters to identify the coastline on each of the images and the results showed the rapid advance of the coastline during 1977 to 2001 [20].

In this study we tried to apply a new method to determine the shoreline changes using remote sensing and geographic information system. This method possesses simplicity and has acceptable results as well, and is able to control and evaluate the results of the research process and its reliability is approved [21].

Materials and Methods

In this study MSS, TM and ETM + images of 1977 (June 6th), 1987 (14 June) and 2001 (15 August) were used to investigate the changes of 16 km of south-eastern coastline of the Caspian Sea. In selection of these images, the existence and availability of data were taken into account; therein the best images were of the Landsat satellite images archive [22]. The Mean Shift Clustering algorithm was developed for segmentation of images existing on ArcGIS software to investigate the changes as well. Having determined the shorelines over the intended years, we extracted the coastline of 1977 as the baseline and then determined the changes through developing cross sections perpendicular to coastlines [23-25].

It is critically important to identify the potential errors. To tackle this issue, the statistical tests were used including sample, median absolute deviation, Z-Score and box plot tests.

Pre-processing

Before using the remote sensing data, it is essential to conduct a series of pre-processing operations (radiometric and geometric corrections). Since the obtained images from USGS have a good georeferenced feature, the accuracy of images was controlled by visual controlling the images that have natural features such as rivers and other linear phenomena and Sharp [26-29]. This examination was conducted on ArcGIS and the results confirmed them as to be geo-referenced and also proved that the three images are properly overlapping.

Choosing the best band

Since the reflection of water on infrared bands is almost zero and the reflection of absolute majority land- covers is greater than water, coastlines can be extracted from one band of the image. According to Tables 1 and 2, the best band for detecting the coastline on MSS, TM and ETM+ images is band 4 [27-30].

Band Wavelength Useful for mapping
Band 1-Blue 0.45-0.52 Bathymetric mapping, distinguishing soil from vegetation and deciduous from coniferous vegetation
Band 2-Green 0.52-0.60 Emphasizes peak vegetation, which is useful for assessing plant vigor
Band 3-Red 0.63-0.69 Discriminates vegetation slopes
Band 4-Near Infrared 0.77-0.90 Emphasizes biomass content and shorelines
Band 5-Short-Wave Infrared 1.55-1.75 Discriminates moisture content of soil and vegetation; penetrates thin clouds
Band 6-Thermal Infrared 10.40-12.50 Thermal mapping and estimated soil moisture
Band 7-Short-wave Infrared 2.09-2.35 Hydrothermally altered rocks associated with mineral deposits
Band 8-Panchromatic (Landsat 7 only) 0.52-0.90 15 meter resolution, sharper image definition

Table 1: Wavelengths and applications of TM and ETM+ bands.

Landsat MSS 1,2,3 Spectral Bands Landsat MSS 4,5 Spectral Bands Wavelength Useful for mapping
Band 4-Green Band 1-Green 0.5-0.6 Sediment-laden water, delineates areas of shallow water
Band 5-Red Band 2-Red 0.6-0.7 Cultural features
Band 6-Near Infrared Band 3-Near Infrared 0.7-0.8 Vegetation boundary between land and water, and landforms
Band 7-Near Infrared Band 4-Near Infrared 0.8-1.1 Penetrates atmospheric haze best, emphasizes vegetation, boundary between land and water, and landforms

Table 2: Wavelengths and applications of MSS bands.

Image segmentation and extraction of coastline

In this step, the Mean Shift Segment tool on ArcGIS was used for coastline detection operations. This algorithm is the developed form of Clustering Mean Shift that has been added to ArcGIS 10.3 for segmentation [31]. The algorithm is considered as a non-parametric algorithm for clustering and is widely used in image segmentation. Compared to the k-means method, its advantage is that even number of categories not considered as input [32-34]. After acquiring the segments from images of different years (Figures 1-3), the coastline was digitized by hand on ArcGIS so that in the following steps their changes can be calculated by putting the coastline on different years (Figures 4 and 5).

coastal-zone-management-segmented-map-coastline-4-mss

Figure 1: Segmented map and coastline of the band 4 MSS-1977.

coastal-zone-management-segmented-map-coastline-4-tm

Figure 2: Segmented map and coastline of the band 4 TM-1987.

coastal-zone-management-segmented-map-coastline-4-etm

Figure 3: Segmented map and coastline of the band 4 ETM+-2001.

coastal-zone-management-a-part-caspian-sea-coastline

Figure 4: A part of Caspian sea coastline change in 1977, 1987 and 2001.

coastal-zone-management-a-part-caspian-sea-coastline

Figure 5: A part of Caspian sea coastline change in 1977, 1987 and 2001.

Results and Discussion

We used the coastline of 1977 as the baseline, and then determined the changes by developing the cross sections perpendicular to the coastlines. Therefore, the common length of the coastline that was determined for all three years was divided at regular intervals (530 m) through developing cross sections perpendicular to the coastline. Then, we determined the changes from 1987 and 2001 and compared to 1977 (Figures 6 and 7). The average measurements show a 14.03 meters recession of sea water for 1987 compared to that of 1977. However, the results for 2001 show advancing sea water around 69.8 meters in 1977 (Table 3 and Figure 8).

coastal-zone-management-measurement-part-Caspian-sea

Figure 6: Measurement part of Caspian sea coastline change in 1977, 1987 and 2001.

coastal-zone-management-measurement-part-Caspian-sea

Figure 7: Measurement part of Caspian sea coastline change in 1977, 1987 and 2001.

Year 1987 2001
First 530 meters -12.78 99.18
Second530 meters 2.66 103.97
Third 530 meters 9.11 86.76
Fourth 530 meters 15.13 68.68
Fifth 530 meters -29.04 86.52
Sixth 530 meters -30.34 118.01
Seventh 530 meters -44 76/82
Eighth 530 meters -31.41 94.92
Ninth 530 meters -22.29 89.14
Tenth 530 meters -25.73 71.45
Eleventh 530 meters -19.86 62/81
Twelfth 530 meters 1.8 54.9
Thirteenth 530 meters -12.16 55.19
Fourteenth 530 meters -0.6 96.67
Fifteenth 530 meters -5.98 93.02
Sixteenth 530 meters -24.43 79.06
Seventeenth 530 meters -15.47 94.06
Eighteenth 530 meters -21.87 86.25
Nineteenth 530 meters -23.31 88.59
Twentieth 530 meters -36.34 74.89
Twenty-first 530 meters -20.38 55.68
Twenty-second 530 meters -2.36 94.59
Twenty-third 530 meters -27.38 119.13
Twenty-fourth 530 meters -41.32 110.98
Twenty-fifth 530 meters 1.33 103.21
Twenty-sixth 530 meters -24.39 90.89
Twenty-seventh 530 meters 8.69 88.39
Twenty-eighth 530 meters 8.22 87.1
Twenty-ninth 530 meters -56/82 60.15
Thirtieth 530 meters -37.83 65.13
Average changes -14.03 m 69.08 m

Table 3: The average measurements of Caspian sea coastline change by cross sections perpendicular to coastlines (equal intervals of 530 meters) for 1987 and 2001 based on 1977 coastline.

coastal-zone-management-diagram-measurements-Caspian

Figure 8: Diagram of measurements of Caspian sea coastline change with cross sections perpendicular to coastline (equal intervals of 530 meters) in 1987 and 2001 based on 1977 coastline.

Assessing and controlling the results (identifying errors)

Since the results of coastline changes are considered as the fundamental basis for further analysis and other related applications, the data integrity and errorless being of them are essential. Various factors may cause errors in the results. Some of these factors include:

• Various errors in different levels of coastline extraction on spatial data (such as photos, maps, satellite images etc.).

• Errors related to data: aerial photos for example are most common source of coastline determination. The photos have various sources of error. Factors such as radial or axial distortion and tilt causing error.

• Errors related to measurement methods.

• Errors related to ground control points.

• Measuring errors.

• Human errors.

Typically the data that are inconsistent with the other members can be reprehensive of error data. Errors normally occur farther than the rest of the data set so that it is supposed that they were produced by a different process. Given the critical importance of identifying the potential errors, adding the ability to identify errors is essential. Researchers have offered a variety of methods to detect errors. Statistical methods include wide range of these methods. In the following, we have discussed identifying errors using resume statistical tests of Hample, median absolute deviation, Z-Score and box plot on cross sections perpendicular to the determined coastline.

Hample test: Hample Test is one of statistical methods used in identifying errors in univariate datasets. The test is theoretically a sustainable way, that is not sensitive to errors (the number and amount of errors does not affect its capability). The following steps are necessary to conduct the test:

1. Median calculation (Med)

2. Calculate the deviation from the median of the data set (ri) Equation

Equation is a member of data-set.

3. Calculate median of deviations from the median of the data set Equation

4. Check the proviso Equation , if the proviso confirmed for Equation data, it can be said that Equation is an error.

Hample Test was calculated on cross sections perpendicular to the coastline (Tables 4-6). According to Table 7, none of cross section values Equation exceed 57.275 for 1987, 50.31 for 2001. Therefore, it confirms that the extracted data have no errors.

Year Equation Med Equation
1987 12.715 -21.125 57.2175
2001 11.18 87.745 50.31

Table 4: Haple test parameters on cross sections perpendicular to coastlines 1987 and 2011 based on 1977.

Year Median MAD MADe Equation Equation
        + - + -
1987 21.125 12.715 18.851 16.57754 -58.8275 35.42881 -77.6788
2001 87.745 11.18 16.57548 120.896 54.59404 137.4714 38.01857

Table 5: Median absolute deviation test parameters on cross sections perpendicular to coastlines 1987 and 2011 based on 1977 coastline.

Year Equation Median Equation Equation
1987 -29.365 -21.125 -0.1175 29.2475 43.87125
2001 70.7575 87.745 95.3575 24.6 36.9

Table 6: Box plot parameters on cross sections perpendicular to coastlines 1987 and 2011 based on 1977 coastline.

Year 1987 2001 Z-Score 1987 Z-Score 2001 Equation1987 2001
First 530 meters -12.78 99/18 0.06956 1.690154 8.345 11.435
Second530 meters 2.66 103/97 0.928022 1.959162 23.785 16.225
Third 530 meters 9.11 86/76 1.286641 0.992642 30.235 -0.985
Fourth 530 meters 15.13 68/68 1.621352 -0.02274 36.255 -19.065
Fifth 530 meters -29.04 86/52 -0.83449 0.979163 -7.915 -1.225
Sixth 530 meters -30.34 118/01 -0.90677 2.747653 -9.215 30.265
Seventh 530 meters -44 76/82 -1.66627 0.434408 -22.875 -10.925
Eighth 530 meters -31.41 94/92 -0.96627 1.45091 -10.285 7.175
Ninth 530 meters -22.29 89/14 -0.45919 1.126304 -1.165 1.395
Tenth 530 meters -25.73 71/45 -0.65046 0.132827 -4.605 -16.295
Eleventh 530 meters -19.86 62/81 -0.32409 -0.3524 1.265 -24.935
Twelfth 530 meters 1.8 54/9 0.880206 -0.79663 22.925 -32.845
Thirteenth 530 meters -12.16 55/19 0.104032 -0.78034 8.965 -32.555
Fourteenth 530 meters -0.6 96/67 0.746766 1.549191 20.525 8.925
Fifteenth 530 meters -5.98 93/02 0.447639 1.344206 15.145 5.275
Sixteenth 530 meters -24.43 79/06 -0.57818 0.560207 -3.305 -8.685
Seventeenth 530 meters -15.47 94/06 -0.08 1.402613 5.655 6.315
Eighteenth 530 meters -21.87 86/25 -0.43584 0.964 -0.745 -1.495
Nineteenth 530 meters -23.31 88/59 -0.51591 1.095415 -2.185 0.845
Twentieth 530 meters -36.34 74/89 -1.24037 0.326018 -15.215 -12.855
Twenty-first 530 meters -20.38 55/68 -0.353 -0.75282 0.745 -32.065
Twenty-second 530 meters -2.36 94/59 0.64891 1.432378 18.765 6.845
Twenty-third 530 meters -27.38 119/13 -0.7422 2.810553 -6.255 31.385
Twenty-fourth 530 meters -41.32 110/98 -1.51726 2.352846 -20.195 23.235
Twenty-fifth 530 meters 1.33 103/21 0.854074 1.91648 22.455 15.465
Twenty-sixth 530 meters -24.39 90/89 -0.57595 1.224584 -3.265 3.145
Twenty-seventh 530 meters 8.69 88/39 1.263289 1.084183 29.815 0.645
Twenty-eighth 530 meters 8.22 87/1 1.237157 1.011736 29.345 -0.645
Twenty-ninth 530 meters -56.82 60/15 -2.37906 -0.50179 -35.695 -27.595
Thirtieth 530 meters -37.83 65/13 -1.32322 -0.22211 -16.705 -22.615
Average changes -14/03 69/08 - - - -
Standard deviations 17.98566 17.80615 - - - -

Table 7: Coastline change values, Z-Score test and Xi of Hample test on cross sections perpendicular to coastlines 1987 and 2011 based on 1977 coastline.

Median absolute deviation test: Median absolute deviation test (BEN-GAL 2005) is a basic strong method for identifying errors in univariate datasets. The test is a sustainable way and is not affected from errors in dataset.

Equation

For normal data the following relationship is established:

Equation

That:

Equation

The median absolute deviation test of extracted data (Table 5) shows that no data fall outside this range and indicates the presence of any errors.

Z-Score test: Z-Score Test is a method that can be used for identifying errors in univariate datasets. Z-Score test used mean and standard deviation for identifying errors. Therefore, it is affected from dataset members that are effective on its function as well. The Z-score Test is expressed as follows:

Equation

Where, Equation is the member of data-set Equation is Mean and SD is the Standard deviation of data. The Equation values that their absolute value does not exceed 3 are considered as error. According to Table 7, absolute values of none of cross sections perpendicular to coastline exceed 3 which indicate that the extracted data have no errors.

Box plot: Box Plot is a graphic univariate method which calculate the data distribution using five main characteristics: smallest outlier data, the lower quartile Equation median, upper quartile and the greatest outlier data. The Equation value determines the middle quartile range (Equation). Using this parameter, normal or outlier (error) being one of data can be determined. The data that are Equation times smaller thanEquation or Equation greater thanEquation can be considered as error.

The values needed for identifying errors on cross sections calculated using box plot (Table 6). According to Table 7, none of 1987 cross section values are smaller than -73.3675 and greater than 43.859. For 2011, as well, none of cross section values are smaller than 33.8575 and greater than 132.257 which indicate that the extracted data have no errors.

Conclusion

As the results show, the average measurements revealed a 14.03 meters recession of sea water for 1987 compared to that of 1977. However, the results for 2001 show advancing sea water level around 69.8 meters in 1977. The results show an oscillating nature and high sensitivity for the Caspian Sea compared to the High Seas against unstable coastal factors. Comparison of these results with those of other researchers studying the changes in the Caspian coast confirms the results. This issue confirms the segmentation method to be as one of accurate and rapid coastline determination methods that possesses simplicity, and produces similar results compared to other methods. However achieving accurate results is not only depended on extraction methods, but is also possible by considering the following:

• Identifying and using convenient and reliable data.

• Appropriate pre-processing on the used data will have important effects on results.

• Using the efficient algorithms of extraction and separation of water from land that pay attention to the required inputs will also play an important role in the output results.

• Applying appropriate way to coastline change determination and developing cross sections perpendicular to coastline are crucial.

• Controlling and assessment of extracted data by cross sections perpendicular to coastline that are usually accepted without any control are necessary in most studies.

References

Citation: Hoseini SJ, Ashournejad Q (2018) Detection of Shifts in the Southeastern Coastline of the Caspian Sea with Landsat MSS, TM and ETM+ Images in 1977, 1987 and 2001. J Coast Zone Manag 21: 457.

Copyright: © 2018 Hoseini SJ, et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.