Spatio-Temporal Dynamics and Evolution of Land Use Land Cover Using Remote Sensing and GIS in Sebou Estuary, Morocco

Land use and land cover (LULC) represent the ongoing challenge of environmental variation. The understanding of the level and process of its change is the basis for any environmental planning and management. In Morocco, as everywhere in the world, human population densities are constantly increas-ing on the coastal zones. This results in a continuous and rapid acceleration of the use of coastal space and an increase in pressures on ecosystems and the different species they contain. The purpose of this study is the analysis of the changes in LULC from 1985 to 2017 in the coastal area of Sebou estuary, situated in the Northwest of the Moroccan Atlantic coast. The changes were identified and assessed after classifying a series of Landsat images taken during 1985, 2002 and 2017. The algorithm used for the classification is the Support Vector Machine (SVM), which yielded results with accuracy higher than 85%. The results of the land use land cover change describe phenomenal urbanization and deforestation, as well as an evolution of the agricultural sector, indicating the impact of anthropization in this vulnerable environment.


Introduction
Land use and land cover (LULC) is one of the fundamental topics in the study of environmental changes. During the 20 th century, the world experienced a phe-Journal of Geographic Information System Mediterranean climate with mild, moderate and rainy Atlantic influence during winter, humid and temperate weather during summer with hot wind coming from the East. The rainfall average between 1973 and 2011 is about 537 mm/year, and the average temperature is about 15˚C. However, the temperature is characterized by an apparent variability (minimum temperature of 4˚C during winter and maximum 40˚C during summer) [21] [22].
The Kenitra city being the capital of the El Gharb region, with a concentration of 90% of the region's industrial factories, has become in a few decades a territory with strong anthropogenic attraction, causing a subsequent expansion of the urban area. The Kenitra and Mehdia agglomerations have been established on an exceptional natural site, represented by two wide loops of the Estuary Sebou's meanders. The left bank is occupied by unsuitable constructions for this vulnerable environment, namely two port areas, a military air base, a penitentiary complex and an industrial area, while the right bank is occupied specially by agricultural lands.

Material and Methods
In this research, three Landsat images were selected to map and assess the LULC changes over the last thirty years. The scenes are available with a spatial resolution of 30 m and acquired during the summer period for years 1985, 2002 and 2017. After the processing of Landsat images, LULC maps and LULC changes maps were produced and assessed. The flowchart in Figure 2 summarizes the methodology used for this study. Journal of Geographic Information System

Preprocessing
Images acquired by Landsat sensors are subject to the perturbation due to the effects of sensors, sun, atmosphere and topography. The step of image preprocessing attempts to minimize these effects to the extent required for a particular application [23]. First, Digital Number (DN) values were converted to radiance values. Later, the atmospheric correction was performed using FLAASH model (Fast Line of sight Atmospheric Analysis of Spectral Hypercubes) which incorporates a radiative transfer code based on MODerate resolution atmospheric TRANs mission (MODTRAN4) [24] [25].

LULC Mapping
LULC maps of the Sebou Estuary were produced for the years 1985, 2002 and 2017, using a supervised classification algorithm SVM (Support Vector Machine), following a meticulous selection of pixel samples, based on the spectral variation of each class. The generalized images were reclassified into 7 categories (Table 1).
Generally, a supervised classification requires learning samples as well as the definition of the size and number of learning samples to achieve a specific result, which is one of the most critical problems of the supervised classification [26].
Previous studies show that SVM classifier (Support Vector Machine) is not related to sample's size, and it has been improved to give better results with quality and limited quantity learning samples [27]. Although SVM is less known than other classifiers for LULC mapping, it has been shown to achieve very good performance and has the advantage of dealing well with small ground training samples, which is of particular interest for this case study [27] [28] [29].

Accuracy Assessment
The accuracy assessment was based on confusion matrices between classification maps and training samples. For this purpose, two indices were evaluated which are the overall accuracy and Kappa index.
The overall accuracy is calculated by summing the number of correctly classified values and dividing by the total number of values according to the equation below [30]: where: OA is the overall classification accuracy. C is the number of correct points. A is the total number of reference points.
The kappa coefficient measures the agreement between classification and real values. A kappa value of 1 represents perfect agreement, while a value of 0 represents no agreement [30]. The kappa coefficient is computed as follows: where: i is the class number. N is the total number of classified values compared to truth values. m i,i is the number of values belonging to the truth class i that have also been classified as class i (i.e., values found along the diagonal of the confusion matrix). C i is the total number of predicted values belonging to class i. G i is the total number of truth values belonging to class i.

Change Detection Analysis
Change detection analysis describes and quantifies the differences between images of the same scene at different times. This analysis was performed with ENVI software, providing a simple approach to measure changes between a pair of images that represent an initial state and final state.

LULC Classification and Accuracy Assessment
Based on the accuracy assessment, it can be shown that the accuracy of the three produced maps is very similar (higher than 84%). The overall accuracies and kappa indices are, respectively, 88% and 84% for 1985, whereas years 2002 and 2017 show an overall accuracy of 93% and 88.4%, while the kappa indices are around 90.6% and 85%, respectively. Table 2 shows the producer and user accuracy for the related years. The choice of the SVM classifier made this task easier, especially for 1985 and 2002 images, where there is a lack of the data needed to establish the different samples classes ROIs (Regions of Interest). As explained above, this classification is based on the quality, not on the number of samples. Therefore, we have been able to use topographic maps, field data and data col- The Landsat images used in this study were taken in the same month of the year. All images belong to the summer season, the choice of season is based on  our needs to avoid seasonal plants and is to highlight the degradation or evolution of shrubs and forests in the study area. The corresponding classifications have been presented in the form of maps and graphs, which are illustrated in Figure 3 and Figure 4. Figure 4 shows the area corresponding to each LULC category and its percentage of the total area. Accordingly, a significant increase in built-up areas from 6.7 km 2 (0.43%) in 1985 to 49.97 km 2 (3.45%) in 2017 can be noted. On the other hand, it can be observed that the class occupying the majority of the study area since 1985 is the agricultural land which has been known a general increase from 333.1 km 2 (23%) to 468.6 (32.37%) with a peak in 2002, expressed by a surface of 496.3 km 2 (34.31%). Since 1985 Shrubs and Forest land class has known a phenomenal drop from 310.6 km 2 (21.45%) to 74.2 km 2 (5.42%) in 2017. The bare land, water body, beaches dunes sand and sparse vegetation classes do not show significant changes representing a variation of 1 to 3% over the last 32 years.

Change Detection
The

Change in Built-Up
At a rapid rate, LULC has changed in the cities of Kenitra and Mehdia over the

Change in Shrub and Forest Land
The forest domain of the study area shows enormous damage, with a loss of 75% of its area between 1985 and 2017 illustrated in the map below (Figure 7). The results obtained from the change detection statistics show that the greatest damage was recorded in the period 1985-2004 with a loss of 192.6 km 2 while during the period between 2004 and 2017 the losses were not significant (43.8 km 2 ) ( Figure 5). According to the map of change of the shrub and forest land category, it can be seen that the major part of the class is located in the south of the Sebou River. This area is the northwestern limit of the Mamora forest considered to be the largest continuous plain cork oak forests in the world. It covered 133,000 ha at the beginning of the 20th century [33] [34]. The main forest species are cork oaks, eucalyptus, acacias and pines according to the High Commissariat of Water and Forests. Morocco experienced between 1993 and 2004 the most severe periods of drought [35], hence, cork exploitation exploded during this period as a result of the increased demand for wood for socio-economic needs (the delimbing). Consequently, a loss of more than two thirds of the shrubs and forest land classes was reported between 1985 and 2017.
Despite the efforts made by the state to preserve the forest, it still suffers from a severe dysfunction. According to the technical report of the Food and Agriculture Organization of the United Nations FAO carried out in 2015 [36], the agents and causes of deforestation in the Mamora forest are: livestock farming, the creation of agricultural land, domestic and commercial firewood harvesting, coal production, urbanization, industrial and road infrastructure, and forest fires.

Change in Agricultural Land
The agricultural sector is one of the major assets of the economic activity in the study area. It benefits from different local factors such as a diverse range of high-quality soils, abundant water resources, availability of labor and proximity to Europe and major consumption centers. All these factors make agriculture the leading activity in the region. The

Change Detection Map
In order to identify anthropogenic action in the study area, we have clustered the LULC change trajectories together, into six clusters (Figure 9), which are urbanization, cultivation, deforestation, afforestation, abandonment and no change. The urbanization cluster includes all classes transformed into built-up. The cultivation cluster includes all classes transformed into agricultural land. The deforestation cluster includes changes from the shrub and forest land class to the other classes, while the afforestation cluster includes all classes transformed into shrub and forest land. The abandonment cluster includes all classes transformed Journal of Geographic Information System It is clear that the study area has undergone enormous changes in LULC, those changes due mainly to human activity given the increase in the population, which implies land needs for the expansion of the city in wood and building materials as well as an expansion and search for new agricultural lands [34]. This has perfectly led to an imbalance in the fragile ecosystems of the Sebou estuary. This evolution implies, in addition to changes in LULC, chemical pollution of the estuary's water and soil [37] [38].

Conclusions
The understanding of the LULC changes and the environmental assessment are the basis for developing an appropriate urban planning policy with a focus on sustainable development. In this study, we produced LULC maps based on multi-temporal Landsat images from 1985, 2002 and 2017. They were used to produce LULC maps and assess the impact of land use change over the last three decades in the coastal area of Sebou Estuary, Northwest of the Moroccan Atlantic coast. The present study provided a piece of evidence of significant changes in LULC, especially at the level of agricultural land, shrub and forest land and built-up categories. The shrub and forest have decreased dramatically from 310.6 km 2 (21.5%) to 74.2 km 2 (5.4%). The built-up area increased from 6.7 km 2 (0.4%) in 1985 to 50 km 2 (3.5%) in 2017. The agricultural land increased from 333.9 km 2 (23%) to 468.6 (32.4%).
The phenomenal evolution of the built-up areas, the degradation of the forest as well as the green spaces in the study area, the expansion and creation of agricultural land are leading to a degradation of environmental quality throughout the study area.
Over the last 32 years, Kenitra city and its surroundings have become more and more a large urban region. We recognize that urban development may be beneficial for a variety of social and economic reasons in a region, such as Kenitra. However, it seems necessary to take into consideration the environmental and landscape concerns more seriously in order to achieve sustainable development.
LULC mapping and documentation may not provide the ultimate explanation for all the problems associated with environmental degradation. They can provide a descriptive overview of the evolution of the causes of anthropogenic expansion, but the impact of this evolution cannot be assessed without several biogeochemical studies on sediments and estuarine waters. However, it is one of the most important steps for better understanding of trends and possible causes of this degradation.