Geospatial Assessment of Three Decades of Shoreline Shifts and Two Decades of Vegetation Change in the Grand Saloum Transboundary Wetland Complex, Senegal-The Gambia

International Journal of Current Science Research and Review ISSN: 2581-8341 Volume 08 Issue 10 October 2025 DOI: 10.47191/ijcsrr/V8-i10-18, Impact Factor: 8.048 IJCSRR @ 2025 www.ijcsrr.org 5048 *Corresponding Author: Ousmane Badji Volume 08 Issue 10 October 2025 Available at: www.ijcsrr.org Page No. 5048-5061 Geospatial Assessment of Three Decades of Shoreline Shifts and Two Decades of Vegetation Change in the Grand Saloum Transboundary Wetland Complex, Senegal-The Gambia Ousmane Badji1*, Adam Ceesay2, Kwame Oppong Hackman3 1WASCAL Graduate Research Programme on Climate Change and Land Use, Department of Civil Engineering, Kwame Nkrumah University of Science and Technology, Kumasi, Ghana 2Institut des Sciences de l’Environnement, Cheikh Anta Diop University (UCAD) 3Competence Center, West African Science Service Center on Climate Change and Adapted Land Use (WASCAL), Ouagadougou, Burkina Faso ABSTRACT: Coastal wetlands at the land–sea interface are on the frontline of climate change, yet integrated evidence on geomorphic and ecological responses remains limited in West Africa. We quantified shoreline trajectories (1990–2020) and landcover dynamics (2000–2020) across the transboundary Grand Saloum complex (Senegal–The Gambia) using Landsat surfacereflectance time series, spectral indices (NDVI, NDWI, NDBI), and the Digital Shoreline Analysis System (DSAS). Shorelines were extracted from NDWI-based water masks, filtered and vectorized, then analyzed in DSAS with End Point Rate statistics. Vegetation was mapped in Google Earth Engine with a Random Forest classifier (mangrove, other vegetation, built/bare, water). The coastline is dominated by erosion (mean −2.44 m·yr⁻¹) interspersed with localized accretion (mean +1.84 m·yr⁻¹). Erosion hotspots concentrate in central sectors, whereas mixed erosion–accretion patterns occur near the northern and southern mouths. Concurrently, mangrove cover expanded from 57,867.61 ha in 2000 to 66,840.17 ha in 2020 (~+15.5%), while other vegetation declined from 23,483.18 ha to 16,146.11 ha (~−31.3%). Within a 1-km coastal buffer, mangroves remained broadly stable to slightly increasing (16.43%→16.81%). These findings depict a dynamic yet resilient system where mangrove gains coexist with heterogeneous shoreline retreat and conversion of non-mangrove covers to bare substrates and water. Management should safeguard landward migration corridors, target erosion-prone reaches with nature-based measures, and institutionalize a transboundary monitoring, reporting, and verification framework that updates DSAS and satellite products at 2–3-year intervals while integrating in-situ elevation, salinity, and sediment data. Our workflow provides transferable, decision-relevant evidence for coastal adaptation and blue-carbon planning in data-limited deltas and policy design. KEYWORDS: Coastal erosion, Shoreline change, Mangroves, Remote sensing, DSAS, Landsat, Google Earth Engine, Grand Saloum (Senegal–The Gambia). 1. INTRODUCTION Coastal wetlands at the land–sea interface are on the frontline of climate change. Rising seas, driven by ocean thermal expansion and accelerating land-ice loss, are altering shoreline position, inundation regimes, and salinity gradients that structure wetland ecosystems (Church et al., 2001). These physical shifts cascade into ecological and livelihood impacts for communities dependent on fisheries, agriculture, and coastal protection; conversely, slower relative sea-level rise can expand the window for adaptation in deltas and low-lying coasts ((IPCC, 2007, 2023). Mangrove ecosystems are both climate sentinels and buffers: they sequester carbon, attenuate waves, and sustain coastal economies, yet they are sensitive to changes in hydrodynamics and salinity (Ellison, 2014). Prolonged or more frequent inundation and salinity shifts can exceed species-specific tolerance thresholds, triggering dieback or community reassembly unless sediment accretion and landward migration keep pace (Friess et al., 2012). These responses vary regionally, underscoring the need for sitespecific, decadal monitoring (Ellison, 2014; Friess et al., 2012). In West Africa, and Senegal in particular, coastal risks have intensified, threatening settlements and critical habitats. Along sectors of the Senegalese coast, multi-year analyses report shoreline retreats commonly on the order of 1–2 m·yr⁻¹ (Diop et al., International Journal of Current Science Research and Review ISSN: 2581-8341 Volume 08 Issue 10 October 2025 DOI: 10.47191/ijcsrr/V8-i10-18, Impact Factor: 8.048 IJCSRR @ 2025 www.ijcsrr.org 5049 *Corresponding Author: Ousmane Badji Volume 08 Issue 10 October 2025 Available at: www.ijcsrr.org Page No. 5048-5061 2014; P. W. Bakhoum et al., 2017). A notable morphological shift in the Saloum system was the 1987 storm-breach of the Sangomar spit, which reconfigured connectivity and sediment dynamics (MEPN, 2006). Scenario analyses further suggest substantial potential loss of low-lying areas under sea-level rise in the Saloum estuary and significant erosion pressure along the Senegal–Gambia border (Jallow et al., 1996; Niang et al., 2010). The Grand Saloum, encompassing Senegal’s Saloum Delta Biosphere Reserve and The Gambia’s Niumi National Park, forms a transboundary Ramsar wetland complex of high ecological and economic value. Although numerous studies have addressed threats, pressures, and climate–mangrove linkages in parts of the system (Drame & Sambou, 2013; SIDIBE, 2010; Sow & Ba, 2019), an integrated, geospatially explicit assessment that couples three decades of shoreline dynamics with two decades of vegetation change remains limited. This study addresses that gap by leveraging satellite remote sensing and GIS to quantify shoreline trajectories (1990–2020) and vegetation dynamics with emphasis on mangroves (2000–2020), providing a decision-relevant geospatial evidence base for coastal management in the Grand Saloum (Ceesay et al., 2017). 2. METHODOLOGY 2.1. Study area The Grand Saloum transboundary complex (Figure 1) covers an area of 83,758 ha. It is composed of: - The Saloum Delta National Park (PNDS) in Senegal, which is located between latitudes 13.583333 and 13.916667, and longitudes 16.466667 and 16.083333. This site was erected by a Decree under the Senegalese law N°76 577 on 28th March 1976 and covered a total of 76,000 ha. - The Niumi National Park (NNP) in The Gambia, which is located between latitudes 13.516667 and 13.983333 and longitudes 16.933333 and 16.083333, is a coastal strip of 7758 ha erected as a National Park in 1986 and a RAMSAR Site in October 2008. It is the natural southern extension of the Saloum National Park (PNDS) (WOW, 2015). 2.2. Climate The Grand Saloum transboundary complex is marked by a Sudano-Sahelian climate type characterised by rainfall values between 400 and 800 mm with an average temperature of 29° C. The rainfall is generally less in the northern part of the complex (Saloum) and greater in the southern region (Niumi). The Canary current coastal influence is much more prominent on the Senegalese section of the complex. Two main seasons characterise the climate: - A dry season (cold from November to March, hot from March to June), where the prevailing winds are maritime trade winds, fresh (in a north to north-west direction) - A dry continental winds (in an east to north-east direction, known as Harmattan). - A hot, humid rainy season from July to October, dominated by monsoon winds (direction: West and southwest). Annual rainfall in the Saloum Delta has declined from a range of 600-900 mm for the period 19311960 to less than 400-600 mm today. There is a total of 50-60 days of rain per year, with maximum rainfall in August. Recently, in Niumi, there have been reports of increased annual average rainfall from 2000 to 2010, and this certainly might be the same at the whole complex level. Average annual temperatures vary between 26 and 31° C (WOW, 2015). International Journal of Current Science Research and Review ISSN: 2581-8341 Volume 08 Issue 10 October 2025 DOI: 10.47191/ijcsrr/V8-i10-18, Impact Factor: 8.048 IJCSRR @ 2025 www.ijcsrr.org 5050 *Corresponding Author: Ousmane Badji Volume 08 Issue 10 October 2025 Available at: www.ijcsrr.org Page No. 5048-5061 Figure 1: Map of the study area (Saloum-Nuimi Transboundary Ramsar Complex) 2.3. Remote sensing of the shoreline dynamic 2.3.1. Satellite images Satellite images with different spatial resolutions processed with various change analysis methods are effective for quantifying changes in the wetland (Toure et al., 2018). Accordingly, surface reflectance images from Landsat 5, 7, and 8 (Table 1) between 1990 and 2020 were accessed and processed in ENVI. Table 1: Landsat images properties Dates and time of acquisition Paths and Rows Cloud Cover Sensors Data Provider: Bands 1990-12-21 10:47:02 (PATH: 205, ROW: 50) & (PATH: 205, ROW: 51) 0 Landsat 5 USGS Blue, Green, Red, NIR, SWIR-1, SWIR-2, NDVI, NDBI, NDWI 2000-12-08 11:17:48 0 Landsat 7 International Journal of Current Science Research and Review ISSN: 2581-8341 Volume 08 Issue 10 October 2025 DOI: 10.47191/ijcsrr/V8-i10-18, Impact Factor: 8.048 IJCSRR @ 2025 www.ijcsrr.org 5051 *Corresponding Author: Ousmane Badji Volume 08 Issue 10 October 2025 Available at: www.ijcsrr.org Page No. 5048-5061 2.3.2. Image pre-processing The bands in the surface reflectance images were atmospherically corrected and orthorectified. At any pixel location, the value recorded on a remotely sensed image does not refer to the true ground-leaving radiance at that particular point. One part of the brightness is due to the target of interest reflectance and the remainder from the atmosphere itself. Their contributions are not known a priori, so the objective of atmospheric correction was to quantify these two components in order to use correct target reflectance (Themistocleous et al., 2008). The orthorectification is necessary because of deformations mainly due to camera distortions and acquisition geometry. The terrain-related geometric distortions that were removed during the orthorecfication stage are related to the image formation process (error tracking), such as distortions caused by the platform, and mainly related to the variation of the elliptic movement around the Earth, instantaneous field of view, topographic relief changes, etc. (Chmiel et al., 2004). 2.3.3. Data analysis and processing Spectral indices, also known as band transformations, were obtained from the Landsat 5, 7, and 8 surface reflectance images by the following equations (Table 2). Table 2: Formulas for the NDVI, NDWI, and NDBI calculation Index Used Equations NDVI 𝑵𝑫𝑽𝑰 = 𝝆𝑵𝑰𝑹 − 𝝆𝑹𝑬𝑫 𝝆𝑵𝑰𝑹 + 𝝆𝑹𝑬𝑫 NDWI 𝑵𝑫𝑾𝑰 = 𝝆𝑮𝒓𝒆𝒆𝒏 − 𝝆𝑵𝑰𝑹 𝝆𝑮𝒓𝒆𝒆𝒏 + 𝝆𝑵𝑰𝑹 NDBI 𝑵𝑫𝑩𝑰 = 𝝆𝑺𝑾𝑰𝑹𝟏 − 𝝆𝑵𝑰𝑹 𝝆𝑺𝑾𝑰𝑹𝟏 + 𝝆𝑵𝑰𝑹 With: ρ_Green=ToA reflectance of green band, ρ_NIR=ToA reflectance of near infrared band. ρ_NIR=ToA reflectance of near infrared band ρ_SWIR1= short-wave infrared 2.4. Shoreline Detection and analysis 2.4.1. Shoreline processing DSAS is one of the most efficient and effective as well as less time-consuming tools in shoreline change analysis compared with the many traditional tools and methods and produces results of better accuracy (Sekovski et al., 2014). It relies on input data such as the date and year and a digitized geometry (in shapefile format) of the shoreline. A series of processes were carried out to analyse the changes in the shoreline, as given in Figure 2. ➔Segregation of water and non-water feature using a spectral index NDWI, as defined mathematically in Table 2, was used to determine the water and non-water features. NDWI value ranges from −1 to +1. The NDWI image typically provides positive results for water features and negative for non-water features (McFEETERS, 1996). Only water and non-water features are required to delineate the separation line as a shoreline, and therefore a binary image classification, i.e., 0 and 1, was performed for depicting non-water and water features (Ji et al., 2009). 2010-12-28 11:17:22 0 Landsat 5 2020-12-07 11:27:49 0 Landsat 8 International Journal of Current Science Research and Review ISSN: 2581-8341 Volume 08 Issue 10 October 2025 DOI: 10.47191/ijcsrr/V8-i10-18, Impact Factor: 8.048 IJCSRR @ 2025 www.ijcsrr.org 5052 *Corresponding Author: Ousmane Badji Volume 08 Issue 10 October 2025 Available at: www.ijcsrr.org Page No. 5048-5061 ➔Post-processing of binary raster image A 3 × 3 mode filter was applied for the post-processing operation that substituted the isolated pixels to the most common neighboring class (either water class or non-water class) to decompose the scattered and isolated pixels (Bartuś, 2014). The jagged boundaries of the water and non-water classes were smoothened by using QGIS clean tool. The shoreline vector was then produced using a raster binary image, and the abutting line of water and non-water class was traced to extract the final shoreline. Figure 2: Shoreline detection and analysis process ➔Shoreline generation After that, the different time periods shoreline data was fed to the DSAS for further computation of shoreline change for 30 years from 1990 to 2020. In the DSAS tool, shorelines positions are compiled with five attribute fields which include Object ID (a unique number assigned to each), shape (polygon), date (original survey year), and shape length, and uncertainty values. Shorelines of different years were merged as a single feature, which creates a single shapefile of the multiple shorelines. The baseline was generated for calculating the shoreline change by closely digitizing the direction and shape of the outer shoreline. From this process, the rates of shoreline change were generated. ➔Shoreline change statistics The calculation of the shoreline change was done in the form of End point Rate (EPR). The final decision matrix was prepared on the basis of the results and output. EPR formula (equation 1) was used to present the computational results. The DSAS tool itself chooses the shoreline transects, gives them dependent and independent variables, and automatically calculates (EPR) the rates of erosion and deposition. The accuracy level would be as high as when more years satellite data set has been incorporated (Sekovski et al. 2014). For example, 4 years of satellite images were chosen for the shoreline change analysis. A ± 5 m uncertainty and 95% confidence interval was set as default parameter to calculate the statistics. 𝐄𝐏𝐑 =𝐃𝐢𝐬𝐭𝐚𝐧𝐜𝐞 𝐢𝐧 𝐦𝐞𝐭𝐫𝐞𝐬 (𝐦) 𝐭𝐢𝐦𝐞 𝐛𝐞𝐭𝐰𝐞𝐞𝐧 𝐨𝐥𝐝𝐞𝐬𝐭 𝐚𝐧𝐝 𝐦𝐨𝐬𝐭 𝐫𝐞𝐜𝐞𝐧𝐭 𝐬𝐡𝐨𝐫𝐞𝐥𝐢𝐧𝐞(𝐘𝐞𝐚𝐫) Eq. (1) The EPR values can either be positive or negative, where a positive value represents seaward or offshore movement, and a negative value represents landward movement. Landsat (5,7,8) NDWI Shoreline extraction Shoreline Digitization (DSAS) Baseline (Buffering) Transect (50 m interval) Shoreline change statistics Vectorisation of raster data EPR Final Decision Matrix International Journal of Current Science Research and Review ISSN: 2581-8341 Volume 08 Issue 10 October 2025 DOI: 10.47191/ijcsrr/V8-i10-18, Impact Factor: 8.048 IJCSRR @ 2025 www.ijcsrr.org 5053 *Corresponding Author: Ousmane Badji Volume 08 Issue 10 October 2025 Available at: www.ijcsrr.org Page No. 5048-5061 3.4. Remote sensing of the vegetation dynamic 3.4.1. Image classification features Due to the long record of continuous observation and high spatial resolution, the Landsat series of satellite images are one of the most useful data for biodiversity assessment (Hackman et al., 2017) and widely used in wetland change assessments (Ajaj et al., 2017; Ceesay et al., 2017). The Tier 1 surface reflectance images from the Landsat series of satellites available in GEE were used because surface reflectance gives the most accurate information about the surface characteristics. In addition, three spectral indices (NDWI, NDVI, and NDBI) obtained from the Landsat 5, 7, and 8 surface reflectance images (see Table 2) were used as features. Because the study area is a wetland, the 30m spatial resolution digital elevation model (DEM) from the NASA Shuttle Radar Topography Mission (SRTM) was added to the feature space to distinguish mangrove from other vegetation. Thus, in all the feature space was a 10-band image stack made up of six surface reflectance bands (Blue, Green, Red, Near infrared, SWIR-1, and SWIR2), three spectral indices, and the DEM. 3.4.2. Image pre-processing Prior to their ingestion in GEE, the surface reflectance images from the three Landsat sensors were atmospherically corrected using the Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS) or the Land Surface Reflectance Code (LaSRC). Also, the visible bands were bands processed to orthorectified surface reflectance. The bands from Landsat 8 were renamed to match those in Landsat 5 and 7. It was impossible to get cloud-free Landsat images for the study area. As a result, the clouds in all available images were masked. Finally, for each year, the complete collection of images from the Landsat sensors was merged using the median filter. In this way, clean Landsat composites were obtained for each year from 2000 to 2020 for use as inputs to the image classification work. 3.4.3. Training and testing sample collection Training and testing samples were manually collected using the high-resolution orthophotos on Google Earth (GE). The sample collection protocol was used as the following: - Generate simple random points within the study area. - Visually inspect the land use at all points with at least 30m radius homogeneous neighborhood, and accept/reject based on local knowledge. - Split samples into training and testing sets. 3.4.4. Image classification A supervised classifier (Random Forest) was used for the land-cover classification on a pixel-by-pixel basis. Apart from its availability in Google Earth Engine, this classifier was selected because they are widely used in land-cover classification (Jia et al., 2014; Yu et al., 2013). The classification workflow is provided in Figure 3 below. In order to make the map of the land-cover classification of the Grand Saloum the classified maps have been exported from GEE to ArcGIS 10.4. Four (4) classes have been taken into accounts such as mangrove, other vegetation, built and bare sand, and water. To access the vegetation close to the shoreline, a buffer has been manually created for a distance of 1km from the shoreline. Zooming of the classified map along the shoreline has been done to detect areas of great change. 3.4.5. Accuracy assessment The accuracy was tested using an independent set of samples that were randomly selected from the training and testing samples and computed the confusion matrix for each classified map. The classification procedure was done in Google Earth Engine while testing procedure were carried out in ArcGIS 10.4. For accuracy, 65% of sampling points were used for training and 35% for testing. The accuracy was calculated using the following formula: 𝐀𝐜𝐜𝐮𝐫𝐚𝐜𝐲 (%) = 𝐓𝐨𝐭𝐚𝐥 𝐓𝐫𝐮𝐞 𝐕𝐚𝐥𝐮𝐞 𝐏𝐢𝐱𝐞𝐥𝐬 𝐓𝐨𝐭𝐚𝐥 𝐒𝐚𝐦𝐩𝐥𝐞 𝐕𝐚𝐥𝐮𝐞 𝐒𝐚𝐦𝐩𝐥𝐞𝐬 x 100 Eq(2) International Journal of Current Science Research and Review ISSN: 2581-8341 Volume 08 Issue 10 October 2025 DOI: 10.47191/ijcsrr/V8-i10-18, Impact Factor: 8.048 IJCSRR @ 2025 www.ijcsrr.org 5054 *Corresponding Author: Ousmane Badji Volume 08 Issue 10 October 2025 Available at: www.ijcsrr.org Page No. 5048-5061 Figure 3: Landsat image classification process 4. RESULTS 4.1. Shoreline dynamic 4.1.1. General observation in the shoreline changes The studied segment includes the Djiffer coast (Sector E) and goes as far as Dionewar (Sector D), Niodior (Sector C), Bettenty (Sector B), and Djinack Bara and Jinack Kajata Island northern coast of the Gambia (Sector A). Between 1990 and 2020, erosion and accretion occurred in some places and the ecosystem is highly dominated by erosion. Figure 4 highlights five main sections highly dynamic. Sector A and E show Moderate to High erosion and accretion. The sector B, C, and D are characterised by moderate to high erosion at some points. For this purpose, an annual average erosion rate of 2.44 m is observed and an average accretion rate of 1.84 m. Figure 1: Point of erosion (Red) and accretion (Dark Green) along the Grand Saloum Shoreline Landsat 5, 7 & 8 (SR) TS NASA SRTM 30m DEM NDWI, NDBI, & NDVI 10-band image composite Training and testing samples Google Earth Classification (Random Forest) Accuracy Assessment Classified Maps Common training samples (65%) Common testing samples (35%) International Journal of Current Science Research and Review ISSN: 2581-8341 Volume 08 Issue 10 October 2025 DOI: 10.47191/ijcsrr/V8-i10-18, Impact Factor: 8.048 IJCSRR @ 2025 www.ijcsrr.org 5055 *Corresponding Author: Ousmane Badji Volume 08 Issue 10 October 2025 Available at: www.ijcsrr.org Page No. 5048-5061 4.1.2. Sectorial Analysis Table 3 shows information related to the rate of change in the shoreline occurring in each section. Sections A and E showed a balance erosion of 4.13 ±0.47 and 1.62 ±0.47 respectively and accretion of 2.82 ±0.47 for both sections. Sections C and D are characterised by High rate of erosion with an average of 2.39 ±0.47 and 2.63 ±0.47, respectively. The average accretion for sections C and D range between 1.45 ±0.47 and 1.018 ±0.47, respectively. Section B doesn’t show so much dynamic with an average erosion and accretion of 1.41 ±0.47 and 1.12 ±0.47. Table 3: Parameters of shoreline dynamics calculated in each transect 4.2. Vegetation dynamic 4.2.1. Analysis of the changes in the whole transboundary wetlands Figures 5 and 6 show that from 2000 to 2020 the whole Grand Saloum wetlands experienced an increase in mangrove vegetation and a decrease in the other vegetation. The figures show estimated mangrove coverages of 57867.61 ha and 66840.17 ha in 2000 and 2020 respectively. The coverage of the other vegetation has reduced from 2000 to 2020 with an estimated coverage of 23483.18 ha to 16146.11 ha respectively. The accuracies of the classification vary between 97.51 % and 99.37 % . Figure 5: Map of the mangrove and other vegetation for 2000 and 2020 Region A B C D E Transect 1-481 482-914 9151276 1277 -1391 1392 - 1490 Number of transect 481 433 362 114 98 Average Accretion (m/yr) 2.82 ±0.47 1.12 ±0.47 1.45 ±0.47 1.02 ±0.47 2.82 ±0.47 Average Erosion (m/yr) -4.13 ±0.47 -1.41 ±0.47 -2.39 ±0.47 -2.63 ±0.47 -1.62 ±0.47 Max. accretion (m/yr) (transect) 14.52 ±0.47 2.7 ±0.47 2.8 ±0.47 2 ±0.47 4.98 ±0.47 Max. erosion (m/yr) (transect) -47.28 ±0.47 -4.53 ±0.47 -12.02 ±0.47 -9.09 ±0.47 -4.02 ±0.47 International Journal of Current Science Research and Review ISSN: 2581-8341 Volume 08 Issue 10 October 2025 DOI: 10.47191/ijcsrr/V8-i10-18, Impact Factor: 8.048 IJCSRR @ 2025 www.ijcsrr.org 5056 *Corresponding Author: Ousmane Badji Volume 08 Issue 10 October 2025 Available at: www.ijcsrr.org Page No. 5048-5061 Figure 6: vegetation dynamics of the Grand Saloum The resulting map (Figure 7) from the change detection analysis shows an increase of mangrove northward and a decrease of the other vegetation southward. Figure 7: Change detection analysis of the vegetation from 2000 to 2020 57868 58365 64097 63161 66840 23483 17931 18386 16580 16146 0 10000 20000 30000 40000 50000 60000 70000 80000 2000 2005 2010 2015 2020 AREA (HECTARE) YEARS Mangrove Other Vegetations