The study was carried out in the Kadzel area, located in the southeast of the Manga countryside in the Diffa region between the towns of Maïné-Soroa to the southwest and N’Guigmi to the northeast. To the south, the plain descends slightly to the valley of the Komadougou Yobé river. To the east of Kadzel are the ancient shore of Lake Chad and the desert valley to the north. The inhospitable and sparsely populated plain extends over an area of approximately 7500 km² ( Figure 1 ). The climate is Sahelian, the annual averages of minimum and maximum temperatures are respectively 22˚C ± 28.61˚C and 39˚C ± 43˚C, and the cumulative annual precipitation is 200 to 400 mm. Relative humidity varies between 19% and 70% [8] . The dry period extends from the beginning of June to the end of September, and the wet period in the Sahel is in August ( Figure 2 ). It is the rainiest month, where the average optimum rainfall reaches 135.23 ± 32.25 mm. This corresponded to a period of strong vegetation growth. Most plants grew and developed during this period, with the exception of Faidherbia albida. The region’s vegetation consists mainly of steppes (covered with shrubs, grasses and trees) and gallery forests along watercourses. The hydrographic network of

the study area is mainly composed of Komadougou Yobé, Lake Chad, and temporary pools during the wintering period in certain lowlands.

Satellite images from Thematic Mapper (TM5) sensors for 1992, Landsat 7 ETM+ for 2002 and Landsat 8 OLI for 2012 and 2022. Scenes 186/050, 186/051 and 185/050 for the four different dates were downloaded from Google Earth Engine ( Table 1 ). These data were used to produce maps of land use dynamics in the study area for the years 1992, 2002, 2012 and 2022 respectively. The choice of the dates of the images was guided by the major climatic events that occurred (periods of flooding), the evolution of agricultural practices and deforestation (the felling of trees for firewood and the manufacture of charcoal).

The study area was covered by three (3) Landsat image scenes. These were scenes 186/050 and 186/051. Before processing, images were mosaicked. Mosaicing involves juxtaposing identical strips of scenes to form one larger strip [9] . First, the images were enhanced to standardize the hues of their spectral bands of different images [10] . Furthermore, although images from satellites differ, it is not uncommon for two adjacent images (scenes) to not have the same brightness, which causes a strong contrast to appear. To address the issue of brightness variations across images, we performed local contrast enhancement in two stages [11] . First, we homogenized the radiometric distribution for every image using radiometric stretching, also known as histogram equalization. Second, we performed a dynamic adjustment between multiple images. Radiometry of the corrected image was calibrated to match the reference images.

The study area was delimited from mosaic images, and the administrative boundaries of the municipality were extracted using the Google Earth Aegina software. The spectral characteristics of objects allow the creation of color visuals through colored composition. According to [12] , its main application is to differentiate various objects in images to aid interpretation. In this treatment, the true color compositions of bands 4/3/2 of Landsat (TM5) for 1992 and the false color compositions of bands 3/2/1 of Landsat ETM+ for 2002 and 7/5/4 of Landsat 8 OLI for 2012 and 2022, respectively, were used to discriminate the different land cover units of the study area ( Figure 3 ).

When using supervised classification, the computer algorithm specifies the numerical descriptors of the different types of land cover present in the scenery, allowing the image analyst to oversee the pixel categorization process. Thus, representative samples of known sites (training plots) were used to establish a key numerical characteristic that best described the spectral attributes of each class type. In this case, the chosen parametric algorithm is the maximum likelihood.

Each landscape unit’s spectral signature could be identified as the training region that was typical of the classes’ numerical attributes. To prevent the

inclusion of mixed pixels (i.e., pixels that could be divided into two different classes), the training regions were marked off from the transition zones [12] . The training areas were evenly distributed across the study area and were traced to the closest pixels. Each class of landscape unit was represented by these diverse training areas. As the classes became more diverse, there were more training areas [12] , the dimensions of the training area must be both smaller than the object to be recognized and larger than the localization error [13] , the estimate was calculated as ( Figure 3 ).

$A=P\left(1+2L\right)$(1)

where A is the surface area of the training area, P is the pixel dimension in meters, and L is the location accuracy in meters.

The validation of this classification using ground truth only concerned 2022 maps. Thus, to check the reliability of the provisional maps, control points were taken in each occupancy unit of the maps produced and compared with reality on the ground using GPS. This made it possible to verify their agreement and calculate the overall precision of each map in the field [14] . This index, between 0 and 1, also results from a confusion matrix based on the projection of land use units from provisional maps onto the ground. Thus, 100 points were checked in Chetimari, 123 in Diffa, 34 in Guskérou, 98 in Bosso, 58 in Tumor, and 119 in Kabléwa, resulting in a total of 474 control points in the study area.

Thus, for each classification, fairly common and widely used measures such as overall precision, Kappa coefficient, and overall field precision [15] were calculated.

In addition, overall precision and 95% confidence intervals (CI) were calculated.

$I_c=\sqrt{a}\frac{1-a}{N}\ast 1.96$(2)

CI: 95% confidence interval; a: overall precision; N = Number of observations.

$K=\frac{a-b}{1-a}\text{avec}a=\frac{1}{N}{\displaystyle \underset{i=1}{\overset{N_c}{\sum }}{x}_{ii}}\text{et}b=\frac{1}{N^2}{\displaystyle \underset{i=1}{\overset{N_c}{\sum }}\left(x_{+i}\cdot x_{i+}\right)}$(3)

N_c: Total number of classes; N: Total number of observation points; x_ii: number of observation points in column i and row i corresponding to the diagonal of the matrix; x_i+ = total number of observation points in column i (total number of observation points in the matrix); x_+i is the total number of observation points in row i (total at the bottom of the matrix). In total, ten land use classes were identified. These included gallery forest (forest_G), degraded shrubby tree steppe (Stp_Aa_D), shrubby tree steppe (SteppAa_D), fallow-pasture complex (Complex_J_P), degraded grassy steppe (Stp_H_D), irrigated cultivation zone (Z_C_I), agglomerations (Agglom), field-fallow complex (ComplexC_J), bare soil, and water bodies.

Vectorization constitutes the final stage of the image processing. It consisted of converting the classified images from raster mode to vector mode (polygons) to facilitate their management in the GIS analysis software ArcGIS 10.8 [16] . The cartographic drafting consisted of producing land-use maps for 1992, 2002, 2012, and 2022 ( Figure 3 ).

1) Transition Matrix

The transition matrix makes it possible to highlight the different forms of conversion that landscape units have undergone over two years. The areas of these different classes of landscape units were calculated from the intersection of land use maps (interset) from 1992 to 2002, 2002 to 2012, and 2012 to 2022 using the ArcGIS extension. 10.8. [17] ( Figure 3 ).

2) Conversion rate

According to [17] , a land-use class’s conversion rate reflects the level of change it experiences when changing to different classes. The evolution of land use was assessed by comparing the surface areas of land use units in different years. This rate makes it possible to quantify the changes obtained at the level of a land-use unit between the two years. It is calculated from the transition matrix using the following formula:

$T_c=\frac{\left(S_{it}-S_{is}\right)}{S}\ast 100$(4)

where T_c is the conversion rate, S_it: Area of landscape unit i on the initial date t, and S_is: Area of the same unit remaining stable on date t₁.

3) Degradation rate

The rate of degradation for each of the two periods, 1992-2002, 2002-2012, and 2012-2022, was calculated from the surface area data for the land use classes. The formula described by [18] . Was used [19] . The rate of degradation, Td, is expressed as a percentage (%) of the surface area per year using the following formula:

$Td=\left[1-\left(1-\left(\frac{A_1-A_2}{A_1}\right)\ast \frac{1}{t}\right)\right]\ast 100$(5)

where Td is the degradation rate; A₁ and A₂ are the initial and final areas of the land cover units, respectively; and t is the time interval in which we want to evaluate changes in land cover.

1) Choice of model

For the simulation of land cover, we chose the CA_Markov model for its performance, multi-scale potential, spatially explicit procedure based on raster data, and the fact that it has been successfully applied multiple times in tropical regions [20] . This model combines Markov chains, Evidential Markov chains (EMC), and cellular automata [21] . Markov chain analysis predicts future land-use patterns based on the knowledge of past and present land-use patterns. To better spatialize and comprehend the change, it is further enhanced by the use of cellular automata and a multi-criteria evaluation of a specific number of change criteria [21] .

2) Development of criteria integrated into the model

For the EMC, environmental variables that are likely to affect the dynamics of land use were identified and weighted. The purpose of identifying and weighing these variables was to produce suitability maps for each land use class.

3) Identification of criteria

The number of explanatory variables to be integrated into the model is limited by their availability, their spatialization and their influence on the location and evolution of land use types. The number of factors presented and integrated is limited compared to the range of potentially explanatory environmental, socio-edaphic, political-economic, biophysical variables (soil and ecosystem degradation, silting of Lake Chad and Komadougou, erosion of biodiversity , legislation on natural resources, poverty, non-compliance with the texts in force and population density) listed by [22] . The choice of different capacities for each land use class was based on the work of [22] .

4) Standardization of factors

The factors used had different minima and maxima values. To use them in EMC, they must be standardized at 256 Gy levels corresponding to the scale 0 - 255 (from least fit to most fit). This operation makes the factors comparable. The factors were standardized using fuzzy functions and weighted using the Saaty matrix, which returned the eigenvector of each factor.

To simulate the dynamics of land use on a later date (2052), the model must first be calibrated using the known data. The 2022 image was the most recent and was the subject of the first test simulation calibrated using two previous dates (2002 and 2012). Images from 2002 and 2012 served as the basis for extrapolating the quantity of future land use. This was a linear extrapolation, because the simulation was based on two points in time to calibrate the model. Some authors define calibration as the process of estimating and modifying the parameters and constraints of a model to enhance the fit between the model’s outputs and data. This stage is crucial because proper model parameterization determines the calibration of the findings that are produced [23] .

For validation, the results of the 2052 land cover simulation were compared with those of the 2052 land cover map resulting from the classification. After visual comparison, a more in-depth analysis was performed by calculating the kappa indices to evaluate the quality of the prediction in terms of location. This evaluation produces Kappa agreement indices: 1) Kappa for the location of the grid-cell level (K location) and 2) K standard for determining the overall success rate [23] . Location indicates the location of the grid cells in the landscape. The standard indicates overall agreement.

Analysis of variance was performed to determine the rate of change in occupancy units per 30-year period. To do this, Shapiro-Wilk and Levene tests were performed to check the normality and homogeneity of the variance, respectively. Parametric ANOVA was performed when these conditions were met. R-studio software was used for all statistical analyses.

The mapped area covers an area of 5704.71 km². Ten land use types were mapped ( Figure 4 , Figure 5 ). These types were represented by the gallery forest (forest_G), degraded shrubby tree steppe (Stp_Aa_D), shrubby tree steppe (SteppAa_D), fallow-pasture complex (Complex_J_P), degraded grassy steppe (Stp_H_D), irrigated cultivation zone (Z_C_I), urban areas (Agglom), field-fallow complex (ComplexC_J), bare soil, and water bodies.

The discrimination of these different types of land-use units was significant, despite some confusion for all the defined classes (the pixels of certain land-use units of the fallow-pasture complex were confused with others). The results of the reliability analysis showed an overall accuracy of 86.2%, with a confidence interval of ±3.48% and a kappa index of 0.80. This reflects an almost perfect agreement in the quality of the image interpretation ( Table 2 ). All land cover units had higher accuracy rates (>80%).

The levels of bodies of water (2.14% to 16.90%), gallery forests (2.00% to 4.83%),

urban areas (9.43% to 24.79%), field-fallow complexes (15.86% to 16.28%), degraded shrubby tree steppes (3.33% to 3.76%), and bare soils (0.03% to 0.25%). In contrast, irrigated crop areas, shrubby tree steppes, fallow-pasture complexes, and degraded grassy steppes have been declining. The highest rates of conversion and degradation were observed in the tree-shrub steppes and fallow-pasture complexes ( Table 3 ).

Legend: Uos: land use units, P_E: bodies of water, Forest_G: Gallery forest, Stp_Aa_D: degraded shrubby tree steppe, SteppAa_D: shrubby tree steppe, Complex_J_P: Fallow-pasture complex, Stp_H_D: degraded grassy steppe, Z_C_I: irrigated cultivation area, Agglom: urban areas, ComplexC_J: Field-fallow complex, TC: Conversion rate, TD: Degradation rate et So_n: Bare floors.

Analysis of variance for the rates of change from 1992 to 2022 showed that there was a statistically significant difference between the occupied areas at the level of all land-use units, except for gallery forests and cultivation areas. Irrigated; that is, these two units did not show statistically significant changes from 1992 to 2022 ( Table 4 ).

The changes in land-use units between 1992 and 2022 are revealed by the transition matrix presented in Table 5 . The transition matrix established between 1992 and 2022 made it possible to better characterize the changes in land use over 30 years. Indeed, 76.86% of the water body remained unchanged, with 23.86% benefiting from other units. 23.33 Of the gallery forests, 23.33% and 49.34% of the irrigated crop areas were transformed into water bodies; 6.86% of the towns were transformed into water bodies; 14.26% into gallery forests; 81.85% of shrubby tree steppe; 85.03% of fallow-pasture complex; and 87.62% of degraded grassy steppes were transformed into field-fallow complexes. 82.32 Of the field-fallow complex, 82.32% was transformed into urban areas, 91.09% of the degraded tree-shrub steppe became a field-fallow complex, and 56.90% of bare soil was transformed into a field-fallow complex ( Table 5 ).

Analysis of the prediction for 2052 horizon indicated the dominance of urban areas, field-fallow complexes, and bare soils. Gallery forests (forest_G), tree-shrub

Legend: Uos: land use units, P_E: bodies of water, Forest_G: Gallery forest, Stp_Aa_D: degraded shrubby tree steppe, SteppAa_D: shrubby tree steppe, Complex_J_P: Fallow-pasture complex, Stp_H_D: degraded grassy steppe, Z_C_I: irrigated cultivation area, Agglom: urban areas, ComplexC_J: Field-fallow complex, So_n: Bare floors, M: mean et σ: Standard deviation.

Legend: P_E: bodies of water, Forest_G: Gallery forest, Stp_Aa_D: degraded shrubby tree steppe, SteppAa_D: shrubby tree steppe, Complex_J_P: Fallow-pasture complex, Stp_H_D: degraded grassy steppe, Z_C_I: irrigated cultivation area, Agglom: urban areas, ComplexC_J: Field-fallow complex, So_n: Bare floors.

steppes (SteppAa_D), fallow-pasture complexes (Complex_J_P), degraded grassy steppes (Stp_H_D), irrigated cultivation areas (Z_C_I), and waterbodies regressed over time. The highest rates of conversion and degradation were observed at the level of the shrubby tree steppes, gallery forest (forest_G) of the fallow-pasture complex (Complex_J_P), degraded grassy steppe (Stp_H_D), and irrigated cultivation zone (Z_C_I) ( Table 6 ).

Legend: Uos: land use units, P_E: bodies of water, Forest_G: Gallery forest, Stp_Aa_D: degraded shrubby tree steppe, SteppAa_D: shrubby tree steppe, Complex_J_P: Fallow-pasture complex, Stp_H_D: degraded grassy steppe, Z_C_I: irrigated cultivation area, Agglom: urban areas, ComplexC_J: Field-fallow complex,TC: Conversion rate, TD: Degradation rate et So_n: Bare floors.

1) Quantifying changes between 2022 and 2052

A comparative analysis of the rate of change in land-use units in horizon 2052 showed a statistically significant difference between the areas at the level of all land-use units: water bodies, urban areas, tree-shrub steppe, fallow-pasture complex, field-fallow complex, degraded tree-shrub steppe, degraded grassy steppe, gallery forest, irrigated crop areas, and bare soils. A significant difference between these units indicates that they will be followed by changes in the next period until 2052 ( Table 7 ).

2) Quantifying future changes

The transition matrix established between 2022 and 2052 made it possible to better characterize land use changes over the next 30-year period. Indeed, 46.12% of the water body remained unchanged compared to 53.88% for the benefit of the other units. A total of 46.12% of gallery forests will be transformed into irrigated crop zones, 56.36% of irrigated crop zones will be transformed into towns, 40.93% of shrubby tree steppe, 41.51% of fallow-pasture complex, and 42.22% of degraded grassy steppe will be transformed into field-fallow complexes. 82.92 Of the field-fallow complex, 82.92% was transformed into urban areas, 74.87% of the degraded tree-shrub steppe became a field-fallow complex, and 31.19% of bare soil was transformed into a field-fallow complex ( Table 8 ).

The mapped area covers an area of 5704.71 km². Thus, ten land use units were

Legend: Uos: land use units, P_E: bodies of water, Forest_G: Gallery forest, Stp_Aa_D: degraded shrubby tree steppe, SteppAa_D: shrubby tree steppe, Complex_J_P: Fallow-pasture complex, Stp_H_D: degraded grassy steppe, Z_C_I: irrigated cultivation area, Agglom: urban areas, ComplexC_J: Field-fallow complex, So_n: Bare floors, M: mean et σ: Standard deviation.

Legend: P_E: bodies of water, Forest_G: Gallery forest, Stp_Aa_D: degraded shrubby tree steppe, SteppAa_D: shrubby tree steppe, Complex_J_P: Fallow-pasture complex, Stp_H_D: degraded grassy steppe, Z_C_I: irrigated cultivation area, Agglom: urban areas, ComplexC_J: Field-fallow complex, So_n: Bare floors.

defined with significant statistical precision because the Kappa index was 0.80. It can be concluded that the results of this analysis were statistically acceptable. An analysis of the dynamics of land use in the Kadzel zone shows the different forms of change that have occurred over time. Thus, the results reveal, from 1992 to 2022, a regression in the irrigated cultivation area, shrubby tree steppes, fallow-pasture complexes, and degraded grassy steppe in favor of camp-fallow complexes, agglomerations, urban areas and water bodies. These results are consistent with those of [24] and [25] who showed continuous regression of vegetation in favor of fallow fields, bare soils, and urban areas based on a diachronic study. The creation of new fields is detrimental to the forest ecosystems [26] . In addition, the increase in agricultural land to feed an ever-increasing population justifies the reduction in the duration of fallows and the annual increase in anthropized spaces [27] . Plant formation is influenced by climate change and human activities [28] . The former are considered inevitable on a scale of a century [29] , and their effects on species extinction are generally limited [30] . However, the impact of human activities is likely to cause changes in plant succession and environmental degradation of the environment [31] .

The highest rates of degradation were found in the tree-shrub steppe and fallow-pasture complexes. This justifies the extension of the cultivated areas to the detriment of plant growth. This expansion of agricultural land leads to the deforestation of natural plant formations by producers in search of fertile land, probably due to the drop in crop yields but also due to population growth whose vital needs must be met to improve incomes [31] . This also corresponds to the periods of space occupation by refugees fleeing the abuse of Boko Haram in the beds of Lake Chad. We noted evolutionary trends (gains to the detriment of another type of land occupation) and regressive trends (loss of surface area to the benefit of another type of land use). Apart from agricultural activities, the felling of trees for firewood and the production of charcoal as a source of energy also constitute direct actions to destroy natural plant formations. Furthermore, the cutting of woody trees activates wind and water erosion, resulting in drainage of the arable part of the cultivated land. This situation is also one of the causes of the decline in fertility, which negatively affects crop yields and justifies the expansion of registered cultivated lands. This transformation was the result of human actions, such as deforestation and overgrazing, following a rapid increase in the population in the area, which recorded a massive flow of displaced people and refugees. With the Boko Haram crisis, the population increased from 40,923 in 2013 to 123,995 in 2017. In addition, the settlement of displaced people has contributed to the expansion of host villages. This has resulted in an increase in residential areas [31] .

The prediction for 2052 indicated the dominance of urban areas, field-fallow complexes, and bare soil. The highest rates of conversion and degradation were observed at the level of the shrubby tree steppes, gallery forest (forest_G) of the fallow-pasture complex (Complex_J_P), degraded grassy steppe (Stp_H_D), and irrigated cultivation area (Z_C_I). By others found in the same area, a progressive dynamic in terms of an increase was observed in agglomerations, field-fallow complexes, and bare soils [31] . The simulation of land use contributes to the understanding of the functioning of landscaped spaces. According to [32] , the importance of predictive modeling is not to precisely determine the future of a territory but to lead decision-makers to be proactive in anticipating the instantaneous degradation of natural resources.