Coastline Change Modelling Induced by Climate Change Using Geospatial Techniques in Togo (West Africa)

Climate change is a major concern of humanity. One of the consequences of climate change is global warming causing melting glaciers, rising sea levels and shoreline regression. In Togo, the regression of shoreline leads to coastal erosion with significant damage on socio-economic infrastructures and human habitats. This research, basing on geospatial techniques, focuses on coastal erosion monitoring from 1988 to 2018 in Togo. It is interested in the extraction of shoreline and in the analysis of change. Various satellite images indexes have been developed for shoreline extraction but the major scientific problem concerns the precision of the different classification algorithms methods used for the extraction of the shoreline from these water index. This study used NDWI index from multisource satellite images. It assesses the performance of Otsu threshold segmentation, Iso Cluster Unsupervised Classification and Support Vector Machine (SVM) Supervised Classification methods for the extraction of the shoreline on NDWI index. The topographic morphology such as linear and non-linear coastal surfaces have been considered. The estimation of the rates of change of the shoreline was performed using the statistical linear regression method (LRR). The results revealed that the SVM Supervised Classification method showed good performance on linear and non-linear coastal surface than the other methods. For the kinematics of the shoreline, the southwest of the Togolese coast has an average erosion rate ranging from 2.49 to 5.07 m per year. The results obtained will serve as decision-making support tools for the design and implementation of appropriate adaptations How to cite this paper: Konko, Y., Okhimambe, A., Nimon, P., Asaana, J., Rudant, J.P. and Kokou, K. (2020) Coastline Change Modelling Induced by Climate Change Using Geospatial Techniques in Togo (West Africa). Advances in Remote Sensing, 9, 85-100. https://doi.org/10.4236/ars.2020.92005 Received: March 18, 2020 Accepted: June 5, 2020 Published: June 8, 2020 Copyright © 2020 by author(s) and Scientific Research Publishing Inc. This work is licensed under the Creative Commons Attribution International License (CC BY 4.0). http://creativecommons.org/licenses/by/4.0/


Introduction
Climate change and its effects are of global concern and has become an active area of research for scientists [1]. One of the consequences of climate change is global warming causing melting glaciers, rising sea levels and shoreline regression [2]. A shoreline is the line of contact between land and a sea [3]. The regression of shoreline leads to coastal erosion with significant damage on socio-economic infrastructure and natural ecosystem [4] [5].
The West African coast is one of the areas severely affected by coastal erosion phenomena [6]. Of particular concern is the Togolese coast. Due to its configuration, the Togolese coast faces many challenges such as urban expansion and coastal erosion. The recent research of   [2] in the East of the Togo coast showed that, the shoreline experienced an average recession ranging from 1.66 to 5.25 m per year while the urban area increases from 7.84 ha per year. Predictions based on those averages rate showed that the barrier beach with its residential areas whose the local population estimated at 82.724 inhabitants are exposed to a continual risk of inundation by shoreline regression which may affect more than 7% of its surface area by the 2070s. The inundation of the barrier beach could increase damage on socio-economics structures, human beings and rural exodus phenomenon.
Although the recent research of   [2] links coastal erosion phenomenon to climate change and provides decision-making support tools for the implementation of appropriate adaptation plans in the East of the Togo coast, the data on the shoreline regression in the west part of the coast are lacking. According to Blivi and Adjoussi (2004) [7], the western part of the Togolese coast experienced coastal erosion phenomenon leading important damages. The lack of recent statistics data on the evolution of coastal erosion in this part of the coast does not allow predictions as well as the implementation of adaptation and mitigation measures.
Monitoring of shoreline in the whole of the Togolese coast is fundamental for planning, defining of hazard zones, and predicting of erosion. The present study, based on remote sensing techniques, is part of this perspective and focuses on the western part of the Togolese coast.
Remote sensing provides new technological advantages in costal monitoring research [8] and gives a big opportunity to obtain reliable shoreline information.
Tools including satellite images are the preferred means for discrimination and Although the water index of satellite images allow a discrimination of terrestrial and aquatic areas, the major scientific problem concerns the precision of the different classification algorithms used for the extraction of the shoreline from these water index.
The techniques of image processing can provide reliable solutions for extraction shoreline issues. Some methods have been proposed. Support Vector Machine (SVM) [2] [11], Iterative Self Organized Data Analysis (ISODATA) [12], thresholding and morphological filtering [13], Random forest method [14], object-oriented fuzzy classification approaches [15], mean-shift segmentation [16] and genetic algorithm based methods [17] can be counted as mainly used methods for shoreline extraction.
In order to use the most suitable method for shoreline extraction from NDWI index commonly used, this study based in the west coast part of Togo, firstly, assesses the performance of Otsu threshold segmentation method, Iso Cluster Unsupervised Classification method and SVM Supervised Classification method for the extraction of the shoreline on linear and non-linear coastal surfaces (topographic morphology). Secondly, the study focused on the kinematics of the shoreline from 1988 to 2018.

Study Area
This study was conducted in the west of the Togolese coast, located in west part of the transboundary biosphere reserve of the Mono [18]. It is bounded between the latitudes 6˚5'55.045" and 6˚11'34.668" North and the longitudes 1˚11'33.347" and 1˚23'4.636" East ( Figure 1).

Data Used
The data used in this study was downloaded from Landsat (https://landsat.usgs.gov/) and Sentinel (https://sentinel.esa.int/) platforms.  Sheet NB-31-XIV-1a Lome), aerial photograph acquired on January 2018 and the fields data had been used as reference data.

Images Pre-Processing
All the optical images were pre-processed using the ENVI software. Firstly, the images were geo-referenced using GPS field surveys and JICA topographic data, Various water index methods to improve the discrimination of aquatic and terrestrial areas from satellite imagery have been developed: Normalized difference water index [22], modified normalized difference water index [21], etc. In this study, the spectral water index based on normalized difference water index commonly used, namely, NDWI is used because of its efficiency and convenience [21]. NDWI is computed according to the Equation (1)

Shoreline Extraction Methods
The shoreline is the line of contact between land and a sea [3]. Various methods for shoreline extraction and change detection from satellite images have developed: Manual method, image enhancement, write function memory insertion, density  [25]. In addition, several image-processing algorithms such as pre-segmentation, segmentation and post-segmentation have been used for automatic extraction of the shoreline from satellite images [26].
The NDWI indices of satellite images allow a discrimination of terrestrial and aquatic areas but the major scientific problem concerns the performance or precision of the different classification methods used for the extraction of the shoreline from these water index. For that, this study explores the performances of three distinct methods commonly used in order to define the most suitable method for shoreline extraction on NDWI indices. The three methods are: Otsu threshold segmentation, Iso Cluster Unsupervised Classification and SVM supervised Classification methods.
The JICA data, aerial photograph acquired on January 2018 and the fields data had been used as reference data for validation and comparison of the result of three algorithms (methods).

Otsu Threshold Segmentation Method
Otsu's method proposed by Otsu (1975) [27], is a dynamic method that has been successfully used to partition the input raster image into homogeneous land and water regions through minimizing the intra-class variance [28] [29] [30].
The Otsu threshold segmentation method is powerful and provides dynamic and variational thresholds based on different regions in different sectors [31]. It still remains one of the most referenced thresholding methods [32] [33]. The thresholding method sets the threshold value automatically according to the local characteristics to achieve a good separation between the land and sea water.
It is based on discriminate analysis and uses the zeroth and the first-order cumulative moments of the histogram for calculating the value of the thresholding level.

Iso Cluster Unsupervised Classification Method
The Iso Cluster Unsupervised Classification method has used to separate terre-

The SVM Supervised Classification Method
The SVM Supervised Classification method has been successfully used to separate terrestrial and aquatic areas [2]. SVM is a supervised non-parametric statistical learning technique. In general, it is based on the search for a hyperplane making it possible to separate observations while maximizing the margin (the distance from the point closest to the hyperplane) [36]. Vapnik (1982) developed the original formulation of the algorithm [37]. This algorithm has the ability to provide good classification results even when the number of training samples for region of interest (ROI) is low, which is a fairly recurrent constraint in applications of remote sensing. SVM also has the ability to minimize the risk of classification error even without prior knowledge on data distribution. Its performance of separating pixel is often higher than other traditional classifiers such as the maximum likelihood algorithm [36].

Shoreline Kinematics
In order to allow a better knowledge of the human habitats affected by coastal erosion phenomenon, the coast was subdivided into eight (8)   Shoreline kinematics is evaluated by the dedicated Digital Shoreline Analysis System (DSAS) extension of ArcGIS [38]. DSAS extension requires the integration of shoreline shapefile within a geodatabase and the digitization of an imaginary baseline from which DSAS creates perpendicular transects to the lines to be compared. For this study, 590 transects were created. A conventional step of 50 m measurements is used between 500 m long transects.
The estimation of the rates of variation of the shoreline is performed by the linear regression (LRR) method. This method was used because of the multi-temporal image data. A LRR method of change statistic is determined by fitting a least-squares regression line to all shoreline points for a particular transect. It allows highlighting the evolutionary trend over time [2]

Comparison of the Performance of the Three Methods on Linear Surfaces
The performances of Otsu threshold segmentation method, Iso Cluster Unsupervised Classification method and SVM Supervised Classification method for the extraction of the shoreline from the NDWI indices are shown in Figure 3. As shown on the NDWI indices of the MSI images, the three methods show good performance for the extraction of the shoreline (Figures 3(b)-(d)). On the other hand, the NDWI indices of the ETM + images, Otsu threshold segmentation shows better performances followed by SVM Supervised Classification and Iso Cluster Unsupervised Classification (Figures 3(f)-(h)). For the NDWI indices of TM images, although Otsu threshold segmentation and SVM Supervised Classification show better performance, Iso Cluster Unsupervised Classification has a tendency which is substantially the same (Figure 3(j)-(l)). Some research using Otsu threshold segmentation and SVM Supervised Classification have also concluded that they are efficient [2] [21].

Comparison of the Performance of the Three Methods on Non-Linear Surfaces
The performances of Otsu threshold segmentation, Iso Cluster Unsupervised Classification and SVM Supervised Classification methods for the extraction of the shoreline on non-linear surfaces from NDWI indices are represented in Figure 4. It can be observed on the NDWI indices of the MSI images, that SVM Supervised Classification (Figure 4(d)) and Iso Cluster Unsupervised Classification (Figure 4(c)) show good performance compared to Otsu threshold segmentation (Figure 4(b)).
The same observations were experienced on the NDWI indices of the ETM + and TM images (Figures 4(f)-(h); Figures 4(j)-(l)). However, with regard to the comparison of SVM Supervised Classification and Iso Cluster Unsupervised Classification methods, the interpretation does not allow any conclusion. Additional statistical analyzes could help to judge the performance of these two algorithms on non-linear surfaces.

Method Used for Shoreline Extraction
In the extraction of the shoreline from the NDWI indices of the MSI images which have good spatial resolution (10 m), the three methods showed good results on the linear surfaces. On the other hand, on the low spatial resolution images (ETM + and TM), SVM Supervised Classification and Otsu threshold seg-

Kinematics of the Shoreline
The analysis indicates that over time, the shoreline changed substantially by sector ( Figure 5). These variations are expressed in the first place by erosion phenomena. An in-depth analysis of the changes using the LRR method allowed to draw Table 1. A positive evolution rate means an accretion whereas a negative rate represents an erosion. It follows from Table 1 and Figure 5 that the shoreline had generally experienced erosion with sector C and D had experiencing more accretion phenomena. The average erosion rate is ranging from 2.49 to 5.07 m/year. The lowest rate is observed in sector A towards Kodjoviakopé locality while the highest rate is observed in the sector H towards Kpogan locality.
However, this regression observed on the Togolese shoreline is lower than the results of Blivi and Adjoussi (2004) [7], which showed a regression of the shoreline ranging from 5 to 10 m/year. This discrepancy could be explained by the difference in the methodological approachs. In this study, the shoreline is extracted from satellite images (Landsat TM, ETM+ and Sentinel-2A MSI) by using SVM supervised classification method, and the kinematics are evaluated using the statistical LRR method. In a previous study Blivi and Adjoussi (2004) [7], using Landsat image analysis (MSS and TM), the methods of shoreline extraction and calculation of the kinematics were not clearly defined.
Although the lowest erosion rate is observed in sector A, the erosion process is very advanced in this sector and the sea is near to asphalt road. Appropriate measures of adaptation must be taken in this sector to avoid the immersion of the road by sea. This will have negative consequences on human habitats. Indeed, this road plays an important role in the transit of merchandises between Lome, the Togolese capital, and Ghana the neighboring country.

Conclusions
This study focused on coastal erosion monitoring in Togo from 1988 to 2018.
Based on spatial remote sensing approach, this study firstly assesses the performance of Otsu threshold segmentation method, Iso Cluster Unsupervised Classification method and SVM Supervised Classification method for the extraction of the shoreline from the NDWI indices on linear and non-linear coastal surfaces. Secondly, this study focused on the kinematics of the shoreline.
The results reveal that the SVM Supervised Classification method showed good performance on linear and non-linear coastal surface than the other methods.