Landsat satellite images from the Multi Spectral Scanner (MSS) for 1978, Thematic Mapper (TM) for 1990, ETM for 2006 and Operational Land Image (OLI-TIRS) for 2020 have been downloaded free of charge from https://earthexplorer.usgs.gov of the United States of Geological Survey. These data are acquired at the same time in September. This logic keeps the spectral signatures of different landscape units consistent, minimizing the effects of the atmosphere, differences in soil moisture and the difference due to variation in solar angle ( Fabre, Briottet, & Lesaignoux, 2015 ). They are ideal for ecoregion studies as they integrate both biophysical and anthropogenic landscape attributes ( Hounto et al., 2019 ).

Image pre-processing

Satellite image pre-processing is an important step in the study and detection of change using remote sensing. Its aim is to make multidate Landsat images compatible, i.e. images taken on different dates ( Jofack, Kouamé, Dibi, & Tankoano, 2016 ). This operation also produces images that are more faithful to the reality on the ground. Performed in ENVI, it involved applying atmospheric and geometric corrections to Landsat images from 1978, 1990, 2006 and 2020. The images were geometrically corrected for distortions caused by the earth’s rotation ( Gibson, 2000 ). USGS (United States Geological Survey) provides Landsat images already orthorectified and corrected to the appropriate coordinate system ( Ouatara, 2011 ). Nevertheless, we have made a further correction to the Landsat images to minimize the root mean square error (RMSE), which is generally below 1/2 pixel. They are then projected in the global WGS 1984, UTM, Zone 28N system. The images were radiometrically corrected. The detection and characterization of landscape evolution using multiple temporal satellite imagery requires atmospheric correction of the image to be used ( Ouatara, 2011 ). These conversions provide a basis for standardized comparison of data in a single scene or between images acquired on different dates or by different captures ( Chander et al., 2013 ). The image DN value was converted to radiance and then to spectral reflectance. The global contrast enhancement technique is applied to all Landsat image bands. This technique involves histogram dilation, by selecting two threshold values, Smin and Smax, which will be assigned to values 0 and 225 respectively when images are displayed on screen ( Caloz & Collet, 2001 ). Several image enhancement and transformation techniques were applied to the extracted study area. These included band combinations, principal component analysis and ratio calculations. This led to the creation of new channels, notably PCA and NDVI. The latter allows changes in vegetation productivity to be used to detect changes in forest cover ( Guo, Li, He, Xu, & Jin, 2018 ). Differences or changes in species types, the presence of disturbed vegetation in an area with a similar vegetation type, or even variations in limatic conditions, and the drivers of change can be reflected through NDVI ( Yengoh, Dent, Olsson, Tengberg, & Tucker III, 2015 ). PCA allows us to synthesize a data set initially expressed by highly correlated variables into a reduced number of new “decorrelated” variables that express the maximum variance in the raw data ( Carvalho & Gherardi, 2008 ).

These were used to obtain the best color compositions, facilitating good spectral discrimination of the major land-use classes.

Determining and understanding the different thematic classes of mangrove ecosystems on satellite images was made possible by field observations combined with those derived from the high-resolution (50 cm) Pleiade 2015 image and previous work. These observations enabled us to distinguish the different spectral signatures that characterize the study area when defining the training plots. Thus, the land-use units listed are the filao strip (closed and open strip), sand dunes, bare soil, herbaceous carpet, beach, farmland vegetation, dwellings and water surfaces. Based on the spectral characteristics of the land-use types identified, sample plots were selected, visited and validated in the field. This field mission, which coincided with the month in which the 2020 image was taken, enabled us to select training sites that were close enough to reality in the field to offer good supervised image classification. Thus, a supervised classification was applied to the different images. It relies on a priori knowledge of the terrain by the remote sensor, which identifies several areas on the image whose significance it knows ( Ndao, 2012 ). Until recently, maximum likelihood classification was the most commonly used method for supervised classification of remote sensing data ( Bhatta, 2008 ; Kumar et al., 2016 ). Moreover, this algorithm assumes that the training site statistics for each class follow a Gaussian distribution ( Duminili, 2007 ).

The maximum likelihood classification of Landsat images for the year 2020 was first carried out on the basis of training sites collected during field campaigns, with reference to areas that remained stable between 1978 and 2020. The classification of images from 1990 and 2006 was also carried out.

This was a very important step, enabling us to accurately confirm the information contained on the map. This verification is used to validate the classification result by calculating a confusion matrix ( Antoine, 2014 ). The 2020 images were used as a reference to validate the classification of land-use units from the 1978, 1990, 2006 and 2020 images. On these images, a maximum number of training zones were sampled in different land-use classes. This enhances the “overall accuracy” of reference image classification ( Idrissa et al., 2019 ). Four hundred and sixty (460) points evenly distributed over all the occupancy units resulting from the classification of the 2020 image were verified during a field campaign. Field validation provided a visual and qualitative assessment of the physiognomic characteristics of the different types of land use. This information on the state of land occupation and use was then used to validate the classification, and then to deduce the issues involved in the spatio-temporal dynamics of the land-use units in the study area. The confusion matrix was used to validate the supervised classification performed. Here are the Kappa tables from 1978 to 2020 ( Tables 1-4 ).

This matrix is obtained by comparing classification data with field verification data (reference data), which must be different from those used for classification. The estimated quality of the classification is indicated by the kappa coefficient, which varies from 0 to 1. Overall, classifications are judged to be excellent, with Kappa values above 0.90 everywhere (see Kappa tables above).

This calculation shows the evolution of land-use units between the dates 1978-1990, 1990-2006, 2006-2020 and 1978-2020. So, for understanding the evolution of each land-use unit over time and space, we used a series of ensemblistic transformations (the ensemblistic method of economist Fréderic Poulon). The relationship between the same class at two different dates allows us to extract the << stable >> areas of << regression >> and << progression >> of that class. Thus, to map the transformations that have taken place in the filao strip, we used the equation proposed by: ΔU = U2 − U1.

Let U1 be the set of pixels at date 1, U2 the set of pixels at date 2 and ΔU the variation of one pixel of any land-use unit between dates U1 and U2.

ΔU = 0, reflects steady state (no change);

ΔU < 0, indicates regression (decrease, loss... etc.);

ΔU > 0, reflects progression (extension, gain... etc.).

To identify the factors influencing land-use dynamics, we used a bibliography on the aforementioned area and, above all, field data collected through structured and semi-structured individual interviews with resource persons. These surveys revealed that the main factors are man-made, notably anarchic urbanization, lack of vigorous enforcement of legislation, extraction of sand dunes by carters, cutting of filao stumps and garbage dumps. The natural factors are linked to the intrinsic characteristics of the filao (ageing of the filao with dieback of some of its plants, difficulty of natural regeneration) and to the physico-chemical conditions of the environment, particularly the water deficit from 1968 to 1993, sea spray and topography.

All ecosystems undergo mutations from one year to the next. But these are exacerbated by natural and man-made changes that can lead to large-scale phenomena that can be irreversible. The evolution of the surface areas occupied by the various biotope units of the filao strip is manifested by an increase, a decrease or a stable state of the portion of the surface area occupied by the biotopes. We have thus identified nine land-use units: Closed filao strip, Open filao strip, Building, Sand dune, Water, Beach, Bare soil, Herbaceous carpet and Natural woody vegetation over a total area of 20,678 ha ( Figure 2 , Figure 3 ). Depending on the interest and objectives of the study, we have restricted the nine units into five, namely the closed and open filao strip, the sand and beach dunes and the built-up area ( Figure 5 ).

The evolution of the closed filao strip is characterized, on the one hand, by an extension between the periods 1978-1990 and 1990-2006 and, on the other hand, by a retreat between 2006 and 2020. The area expanded by 47.5% (122 ha) between 1978 and 1990, and by 6% (24 ha) between 1990 and 2006. This increase can be

explained by the reforestation activities undertaken by the forestry service in 1981 on new areas with the “PL 480” project financed by USAID and which continued until 1986 in the south of the Great Coast, as well as the Conservation des terroirs du Littoral (CTL) activities begun in 1988 and financed by the Canadian International Development Agency (CIDA). However, the trend is downward by 46.5% (187.9 ha) due to human activities such as urbanization, lack of vigorous enforcement of legislation, illegal filao logging and others.

The natural degradation factors of the open filao strip are ageing, rainfall deficit and sea spray ( Figure 4 ). Its dynamics are marked by an upward trend between 1978-2020. Indeed, the increase in the area of open filao strip is estimated at 94.4% (330.6 ha) between 1978-1990, 26% (88.2 ha) between 1990-2006 and 2.6% (11.2 ha) between 2006-2020. The main cause of the degradation of these filaos is due to the intrinsic characteristics of the filao on the one hand (difficulty of natural regeneration, dieback of certain plants) and on the other hand to the physico-chemical characteristics of the environment, notably water deficit, salinity and topography. From 1968 to 1993, rainfall fell by 20 to 25%, coinciding with the long Sahelian drought, during which annual rainfall trends in Dakar were irregular, especially in the 1978-1987 decade. These deficits are bound to have a negative impact on groundwater recharge and, by the same token, on the filaos trees, which cannot withstand a severe drop in the water table (below 3 m). Physiography also seems to have an influence on the decline of filao plantations: trees behave differently depending on whether they are located on depressions, ridges, slopes or flat land. Low mortality classes are more represented on depressions, while high mortality classes are represented on flat terrain. On ridges, mortality is higher the closer the observations are to the ocean.

In addition, the extension of the open filao strip is particularly dependent on anthropogenic factors (rampant and anarchic urbanization, lack of vigorous enforcement of legislation, sand extraction, illegal filao cutting and garbage dumps) which permanently shrink its surface area ( Figure 5 ).

Similarly, this regressive evolution of the open filao strip is also linked to the excessive downgrading in this strip. Thus, with decree N˚1738 of September 11, 2018. Union of Magistrates of Senegal has been granted a 4-hectare plot of land to build an administrative building in the municipality in Wakhinane Nimzatt, a 12-hectare plot of land within the reforestation perimeter was declassified for the construction of Muslim (8 ha) and Christian (4 ha) cemeteries. The cooperation of land registry agents has also been granted a 5-hectare plot of land on May 20, 2021, as has the tax union, which has been granted a 4-hectare area at Golf Sud, 4 ha at Wakhinane and 6 ha at Gadaye, without the site being declassified. Two weeks later, decree no. 2021-701 of June 4, 2021 declassified the 150.58-hectare zone as part of the Guédiawaye department’s detailed urban development plan. In the department of Pikine, 23 ha 99a 95ca were declassified by the municipality of Yeumbeul Nord, 37 ha for the resettlement of people affected by the VDN3 and Bus Rapid Transit (BRT0) and 45 ha for the Layenne family. Not far from the Gadaye traffic circle, 9 ha have been declassified and dedicated to a contractor for the construction of an Demba Ka company (EDK) station, not forgetting the 8 ha warehouse of a major Chinese construction subsidiary, China Road and Bridge Corporation (CRBC). It is in charge of the Bus Rapid Transit (BRT) project. According to our investigations, its current location will host a station for this government project. Abusive cutting of filao trees reveals the pressure on these stands ( Figure 6 ).

Sand dune dynamics show a downward trend from 1978 to 2020. Indeed, the reduction in sand dune area is estimated at 112.2% (2684.1 ha) between 1978-1990, 17.8% (362.7 ha) between 1990-2006 and 7.1% (144.1 ha) between 2006-2020. This downward trend is closely linked to sand dune extraction at the Yoff-Cambérène quarry (1945) and the Mbeubeuss quarry (1985), but above all by carters who continue to extract sand at night and discreetly ( Figure 7 ).

The evolution of the beach is characterized, on the one hand, by a decrease between the period 1978-1990 and, on the other hand, by an extension between the periods 1990-2006 and 2006 and 2020. In fact, the decline was 68.9% (304.1 ha) between 1978-1990, while the increase was 1.29% (5.7 ha) between 1990-2006 and 36.4% (162.9 ha) between 2006-2020. This decline from 1978 to 1990 is linked in particular to the extraction of beach sand and sand dunes at Yoff-Cambérène (1945) and Mbeubeuss (1985). In fact, 1978 statistics from the Regional Mining Department show 200 truck movements per day, loaded with an average of 5 to 6 m³, i.e. an estimated borrowing of over 1000 m³/day. An estimate of the volume of sand borrowed from the new Yoff quarry on January 10, 1980, showed that over 375,000 m³ had been extracted in 10 months. This decline is also due to anthropogenic activities that aggravate the sediment deficit on beaches, such as sand dune extraction and beach construction. At Mbeubeuss, 305,150 m³ of sand was extracted in 1996, 328,330 m³ in 1997 and 234,510 m³ in 1998. However, the upward trend is closely linked to the measures taken by the authorities to regulate activities on the coast, through the application of the Mining Code and the Forestry Code. Law n˚98/03 of January 08 1998 and decree n˚98/164 of February 20 1998 clearly regulate mining, including sand extraction. It is the application of this law that has made it possible to curb the abusive exploitation of sea sand in unsuitable locations, and also the fact that people are using less and less beach sand for their constructions.

Between 1990 and 2020, the built-up area underwent uncontrolled expansion. It was 50% (557.2 ha) between 1990-2006 and 59.9% (1011.3 ha) between 2006-2020. This increase is due to the galloping urbanization that began in the 1970s and has been taking on worrying proportions since 1990. After an initial test phase with Hamo 4, 5 and 6 in 1986, this phenomenon became more pronounced in the 1990s, when a number of companies and institutions decided to house their employees there. As a result, SONES, SENELEC, SOFRACO, DOUANES, COMICO, Supreme Courst, Presidence, housing estates for municipal employees, Cité Air Afrique, Cité Gadaye 2005 and “Lotissements Baba Thiam” 2003 were built. Construction, particularly along the coastline, is causing the filao strip to narrow. This phenomenon is most visible between Golf Sud and Yeumbeul Nord, where the width of the strip, which was 200 m at the start of planting, is now only around 20 m. The trend is for it to become thinner as developers encroach on the reforestation and restoration perimeter for construction purposes ( Figure 8 ). Similarly, infrastructures have been built on the reforestation perimeter, such as the sludge treatment plant at Tivaoune Peulh, the VDN3 road and the Chinese company CRBC, not to mention the human settlements on the living dunes near Lac Rose. Paradoxically, the increase in building activity is leading to a reduction in the surface area of this green cordon, which is in danger of disappearing under the helpless gaze of the local population.