Analysis of Land Use Change in the N'ZI Watershed of Côte d'Ivoire Using Landsat Satellite Images

Uncontrolled population growth and strong industrialization are enormous pressures on natural resources. To monitor land use changes in the N'ZI watershed, remote sensing and geographic information systems (GIS) were used from three satellite images: Landsat 4 and 5 (1986), Landsat 7 (2000), and Landsat 8 (2020). The method of supervised classification and calculation of spectral indices was used in this work. The results of the analysis identified six land use classes that changed over the 1986–2000, 2000–2020, and 1986–2020 periods. Water surfaces showed a decrease from −64.95% to −52.47% over the period (2000–2020 and 1986–2020). The forest class showed an average decrease of −86.93%. The savannah class had a reduction of −3.97% and the agricultural class of −9.30% between 1986 and 2020. All this desertification has been to the advantage of the great increase of dwellings‐bare soil of 373.63% and of the weakly covered grounds of 10.60%. These results allowed the detection of different spatial and temporal changes in the N'ZI watershed, creating an awareness to control land occupancy.

cover and land cover changes in Syria from 2010 to 2018 (Dadashpoor & Nateghi, 2017). They were used to monitor land use and human actions in the West Béqaa (Lebanon) (Hassan et al., 2019) between 1962 and 2018. Landsat satellite images were useful in detecting areas of degradation and regeneration of vegetation cover in southern Senegal from 2000 to 2018 (Solly et al., 2021).
Through the method of image overlay analysis, multi-date classification, as well as satellite data processing techniques (radiometric correction, geometric correction, shape classification and extraction) and change detection techniques (image subtraction, correlation, spatial rectification, point operations (pixel a pixel) or global) on digital images were implemented to quickly obtain information on land use and land cover and update the land use map using remote sensing and GIS. Methods for identifying changes in land use and land cover can be divided into three main groupings (Andon et al., 2022;Hoang et al., 2018;Houphlet et al., 2023;Mas, 2000): pre-classificatory methods, the changes are highlighted by creating new images through point operations (pixel-pixel) or by taking into account the whole image. The second group is the set of multi-date classification methods where a multi-date image is classified visually or numerically and the third group contains the post-classification methods consisting of comparing images independently at different dates.
Most remote sensing methods of change are based on radiometric variations between two images acquired at different times. Multi-date data are transformed to make information on changes easier to extract. Image enhancement or pre-classification methods use pixel-to-pixel operations in which the value of the pixel in the final image depends only on the value of the same pixel in the initial image. The first classification method (introduced in the 1970s) is the one based on pixel classification. This method allows to group the pixels independently according to their radiometric characteristic «values of one or more spectral bands». Several methods on this principle have developed over the years, for example, those using distances or the maximum likelihood algorithm such as K-Means, Isodata cluster methods (Forozan et al., 2020;Yang et al., 2022).
This tool allows to measure the different forms landscape change according to the level of land occupancy and the temporal dynamics of these changes (Dietzel, 2022). However, the quantitative analysis of landscape change depends fundamentally on the geographical data available, especially spatial data (cartographic, photographic or satellite) and the spatial resolution of the data used. For example, mapping land use and land use in the N'ZI catchment area, using satellite data, could be a key means of understanding land use changes and decision-making, to the extent that there are expected benefits.
The evolution of the methods and criteria for classifying the soil surface varies from one author to another. The discrimination of image classes according to the algorithm of suitable classifiers such as maximum likelihood classification, support vector machine and decision tree is important for supervised and unsupervised classification (Deilmai et al., 2014;Hu et al., 2018;Zhen et al., 2019). Today, the main classification methods used are supervised and not supervised methods (Achbun et al., 2011).
In this work, the aim is to use the supervised classification method to characterize the dynamics of land use and quantify the changes in land use, to identify their evolution in the watershed of N'ZI. Figure 1 shows the geographic location of the study area. This area is the sub-watershed of Bandama. It is drained by the N'ZI River and its tributaries, located between longitudes 3°49′-5°22′ West and latitudes 6°00′-9°26′ North. It covers an area of 35076.4 km 2 , about 10.88% of the national territory. N'ZI river runs about 600 km long today. The watershed of the N'ZI is composed of 80 sub-prefectures and is limited to the north by the department of Ferkessédougou, to the south by the departments of Tiassalé and Divo, to the east by the department of Daoukro, and to the west by the department of Béoumi. N'ZI watershed is generally dominated by a fairly monotonous relief (Avenard et al., 1971), its altitude h varies from 668 m in the north to less than 130 m in the south. The main soil types are medium-saturated (North) and strongly saturated (Central and South) ferritic soils (Koudou et al., 2016;Monnet, 1972). Due to its elongated geographical configuration, the N'ZI watershed ( Figure 1) is representative of major climatic groups of Côte d'Ivoire. The central part of basin is characterized by a humid tropical regime. The south of basin is characterized by an equatorial regime.

Data Used
Three multi-date satellite images were used in this work to study the evolution of the basin's land use. These are images from Landsat 4 Thematic Mapper (TM) from 1986, Landsat 7 ETM+ (Enhanced TM plus) from 2000, and Landsat 8 Operational Land Imager (OLI) from 2020. The study area was covered by scenes that were: 196-054, 196-055, 196-056, 197-053, 197-054, and 197-055. These scenes were downloaded on http://earthexplorer.usgs.gov, under 30 m spatial resolution. Cartographic data on the evolution of the vegetation of Côte d'Ivoire was taken at a scale of 1: 200,000, from the Remote Sensing Mapping Center of BNETD (national technical study and development office), were used.

Methods
The approach used to study changes in land use of the N'ZI watershed is illustrated in Figure 2. The method adopted was divided into four phases. The first Phase consisted of downloading the data. The second was devoted to pretreatments. The third was concerned with the treatment of satellite images. The fourth was dedicated to the analysis of the changes and the discussion.

Image Preprocessing (Pretreatment)
The image pre-processing in this work consisted of applying radiometric correction, and atmospheric correction under the QUAC and FLAASH modules to get good readings of these spectral images (Caloz et al., 2003;Chaima et al., 2019;Diallo et al., 2011;Girard & Girard, 1999;Jofack-Sokeng et al., 2016;Tra Bi, 2013). The geometric correction was applied while downloading, edited with ArcGIS 10.4.1 software (downloading on: desktop/arcgis-desktop/arcmap/10-4-1/online), and projected into projection coordinate system: WGS_1984_ UTM_Zone_30°N. Then, the assembly is applied to the strips to obtain a single multi-spectral image. Mosaicking of the scenes was done to form a larger image covering study area. Finally, the study area was extracted from mosaic scenes using ENVI software (downloading on: https://idl-and-envi.software.informer.com/5.1/) to carry out the color composition of bands. The combination of bands is as follows: bands 2, 3, and 5 for the 1986 image, bands 6, 5, and 4 or 4, 3, and 2 for the 2000 image, and bands 5, 6, and 7 for the 2020 image. These operations facilitated interpretation and discrimination regions of interest.

Image Treatment
The processing of images started with the calculation of spectral indices on ENVI software. The indices were chosen to describe and interpret the state of evolution of land use of N'ZI watershed. This approach offers a way to analyze the bio-Geo-physical processes of landscape occupation and the degradation of the surface condition of area. These indices described the state of a phenomenon and were based on an empirical approach using experimental data (Caloz et al., 2003;Diallo et al., 2011;Grimene et al., 2019;Tra Bi, 2013). There are many indices, but this work has used indices such as the NDVI (Normalized Difference Vegetation Index), IB (Gloss Index), IR (Soil Redness Index), IC (Coloration Index), IBI (Built Sector Based Index) ( Table 1).
The visual interpretation of colored compositions as well as visualization of the maps of the calculated indices allowed a good reading of information sought (degradation or change of surface state). The land use classes discriminated in this study were validated by reconnaissance trips carried out in the study area to check the control points for the 2020 images (17, 18, 26, and 27 December 2020;31 March 2021;1, 2, and 8 April 2021). This justified the choice of supervised classification method. For the 1986 and 2000 images, their visualization on Google Earth served as a photo interpretation to attest to the information. Finally, based on all this information, the areas of interest have made it possible to distinguish the different land use classes to opt for supervised classification. Therefore, the Maximum Likelihood algorithm has been chosen to visualize land use maps.

Quantification of Changes in the Land Use of N'ZI Watershed
Quantification of changes in land use classes over the study period from 1986 to 2020 was used to detect changes in the study area using multi-date satellite images as follows: • Change (Δ) in area of unit concerned between 1986, 2000, and 2020 ∆1, the difference in the area of the units concerned between 1986 and 2000; ∆2, the difference in the area of the units concerned between 2000 and 2020; ∆3, the difference in the area of the units concerned between 1986 and 2020. If ∆1 or ∆2 or ∆3 = 0, the area of the unit is stable in time and space; If ∆1 or ∆2 or ∆3 > 0, the area of the unit concerned is increasing; Si ∆1 or ∆2 or ∆3 < 0, the area of the unit concerned is said to be in regression. • Change in the overall rate of change (Tg) in area between 1986, 2000, and 2020 The overall rate of change in area of land use classes between years is estimated from following equation, proposed by FAO (1996), which is commonly used to measure the growth of macroeconomic aggregates between two given periods (Alexis et al., 2018;Kpedenou et al., 2017;Salomon et al., 2021;Soro et al., 2014).
Finally, the change in land cover of N'ZI basin between the selected periods is certified by the difference in validation by analyzing confusion matrix through the global precision and Kappa coefficient (Congalton, 1991;Skupinski et al., 2009). To assess the degree of degradation of soil conditions in the basin, methods already developed upstream and previously were used. Figure 3 shows the results of the radiometric and atmospheric corrections applied to the 1986, 2000, and 2020 image scenes respectively.

Image Mosaicking
It should be noted that the fusion of the bands follows the mosaics of the images. Fusion consisted of putting all the bands to produce a new image that retains the information contained in each of the original images. The results of the different treatments have led to the mosaic stage presented by the maps (1986, 2000, and 2020) in Figure 4.
The different scene mosaics are intended to create a new satellite image by combining several image scenes that cover the N'ZI watershed. This section concluded with the extraction of the study area shown on the maps in Figure 4. Table 2 shows the different land use classes in the study area. These choices were made based on the missions carried out in the study area. Land-use classes have been determined according to their importance, isolate classes  (2007)  10.1029/2022EA002744 6 of 16 of secondary importance or even delete unnecessary classes to avoid overloading the information to be extracted. Land uses are grouped into six (6) main types of land use in the study area (Table 2).

Mapping of Land Use Classes
The pretreatment and treatment methods contributed to 1986, 2000, and 2020 land use maps of N'ZI watershed ( Figure 5). The validation of the classification by the confusion matrix evolves significantly with Kappa values increasing slightly between 1986, 2000, and 2020 by 0.86, 0.90, and 0.92 respectively, and overall accuracy of 94.07%, 95.81%, and 94.39%. These values mean that more than 90% of the pixels in the three images were correctly classified according to the ground-truth data. The analysis of these images made it possible to define six (6) land use classes. These were the forest class; savannah class; low cover soil class; agroforestry and fallow plantation class; bare dwelling class; and waterbody class.

• Variation of different classes
The analysis of the evolution of land use classes in a GIS environment is illustrated by the histograms in Figure 6. It shows a strong regression of the Forest class (18.01%; 9.42%; 3.52%) to the advantages of the Bare-Floor Housing class which rose from 8.89% in 1986 to 9.45% in 2000 and then to 18.08% in 2020. Forest decline also benefited the savannah class, which rose from 10.85% in 1986 to 13.30% in 2000 and 20.62% in 2020. While the Low Covered Soil class shows a particular increase of 15.35% over the period 2000-2020.
There is an average increase of about 6% in the Plantations-agroforestry and fallow land class from 1986 to 2000 and a sharp decrease of about 14% from 2000 to 2020. In addition, the Water Bodies class increased between 1986 and 2000 and decreased between 2000 and 2020 ( Figure 6).

Calculation of Omission and Commission Errors
Errors of commission and errors of omission are summarized in Table 3.
Omission errors, the highest rates are recorded in the low cover class on the 2000 and 2020 images and the 1986 image in the Bare dwellings class.
Commission errors, the highest rates are also observed in the low coverage soil class for 1986, 2000, and 2020 images (Table 3). Error rates are very low, below 15% in general, except at the level of soil classes poorly covered in 2020.

Quantification of Changes in Land Use
The histograms in Figure 7 show two trends in land cover classes, the periods 1986-2000, 2000-2020, and 1986-2020 have revealed the changes which have occurred. Those pointing down indicate a loss in the land use class. This means a rate of change from these classes to others. While the upwardly oriented histograms highlight a gain. This means a rate of change from these classes to others. While the upwardly oriented histograms highlight a gain (Figure 7). This is a change of landscape from the study area to other landscapes and vice versa. The results show a regression of forests (Fo), water bodies (PE), and plantations-agroforestry and fallow (Pl-Agro-Ja). Except for the period 1986-2000, there is a progression of Water Bodies, and Plantations-agroforestry and fallow. There was a large increase in Savannah, low-cover soils, and bare dwellings over the periods of 1986-2000, 2000-2020, and 1986-2020 (Figure 7).  1986-2000, 1986-2020, and 2000-2020. • Land use unchanged in general Figure 8 provides information on the different variations in the overall change of forest, savannah, low cover, plantation-agroforestry and fallow, dwelling-bare land; and water body classes over periods of 1986-2000, 2000-2020, and 1986-2020 that did not change. • Changes in different land use classes Figure 9 shows the changes in different land use classes considered during the different calibration periods of 1986-2000, 2000-2020, and 1986-2020. There have been specific changes at level of each land use unit in study area. • Variability of individual class changes Figure 10 clearly illustrates the different changes in the percentage and area of the classes considered during the periods studied : 1986-2000, 2000-2020, and 1986-2020. The forest, low soil cover, and water classes have decreased significantly over three periods, except for an increase in water from 1986 to 2000. On other hand, there was a strong increase in the savannah, bare dwelling and agroforestry, and fallow classes over the periods 1986-2000 and 1986-2020. Except between 2000 and 2020 that there has been a decline in plantations-agroforestry and fallow land.

Discussion
The classification models and the different remote sensing images of 30 m spatial resolution were used to generate the land use classes as well as the detection of changes generally in N'ZI watershed. These means of studying land use satellite data, several researchers have made use of them in recent work by showing the contribution of Landsat images in the discrimination of main categories of land use (Alexis et al., 2018;Ju et al., 2021;Lienou, 2009;Mellor et al., 2013;Noho et al., 2018;Pelletier, 2017;Salomon et al., 2021;Useni Sikuzani et al., 2020). Land use classification resulting from the analysis of the 1986 TM images, the 2000 ETM + images, and the 2020 OLI images gave overall accuracies (94.07%, 95.81%, and 94.39%) and Kappa coefficients of 0.86; 0.90 and 0.92 respectively for the years 1986, 2000 and 2020. According to Landis and Koch (1977) and Pontius (2000) classifications are statistically good qualities and can be used wisely for the Plantations-agroforestry and fallow land; (Ch/Ja) Agricultural area, plantations, food, cereals, vegetables, meadows, and agroforestry Dwelling-bare soil; (Ha/SN) Houses, squares, vegetable gardens, markets and any other, Construction, urban areas, quarries, roads, land, outcropping rocks, and sandy soils Waterbody (PE) Wetland, rivers, water-covered surfaces on the date of acquisition of the hydrographic network image, permanent or not remainder of study. Indeed, for an image analysis whose Kappa value is higher than 0.50, the results are good and exploitable. But, if the results are acceptable, they must not lose sight of the constraints encountered when analyzing the images used. Indeed, errors were recorded during the analysis of the images from several sources when discriminating against the classes mentioned. These errors can be explained by the confusion between classes with generally similar spectral signatures. In addition, the difficulties of visual observation make the distinction of the elements of land use delicate by the structure of the landscape offering homogeneous characters. This can be explained by the fact that Savannah, forest, and low-covered soils are very heterogeneous in terms of temporal evolution. They are developed, with rubber and cocoa plantations in the south and toward  the center-north with cashew plantations; Also included are agricultural clearings and areas used for food crops in the previous year after the harvest (fallow). Also, cocoa plantations are mainly grown in forests. This can make it difficult to discriminate between the different formations. The observation same was made by J.-L. Kouassi (2019) andN'Guessan (2020). Bare-ground dwellings blend with the savannah to agricultural grip and savannah. Analysis of the results evolution of natural environment of N'ZI watershed is regressive between the three dates. Looking at land use maps, it appears that this degradation appears to be general and affects almost the entire basin. The overall rates of change over the 1986-2000, 2000-2020, and 1986-2020 series show regressions for the forest, savannah, plantation-agroforestry, and fallow and dwelling-bare land classes. Natural formations (forests, savannas) are increasingly replaced by anthropogenic occupation classes (fields and fallows, plantations and agroforestry, dwellings-bare ground). The same finding is made by Mama et al. (2013) on territory of Benin and Togo (Koumoi et al., 2013;Kpedenou et al., 2017). This situation of degradation of natural landscape in N'ZI watershed is also noted by Noho et al. (2018) and in department of Katiola by Agouale et al. (2017), and in other parts of Côte d'Ivoire by studies by Aké et al. (2012) and Alexis et al. (2018). Environmental degradation, reflected in the regression of natural formations, is essentially linked to the development of socio-economic activities (e.g., slash-and-burn, timber and fuelwood, mining); as observed during the field missions ( Figure 11)  tion. Savannas burned and left in same state remain in same places while increasing their surface area following the degradation of certain fallow areas. In addition to wildfires, livestock farming appears to be an activity that increases land degradation in the region ( Figure 11). The study area is characterized by overgrazing aggravated by internal transhumance. Indeed, overgrazing leads to a reduction in the natural regeneration of woody plants, a decrease in herbaceous cover, and the stripping of the soil and its hardening according to Agouale et al. (2017).

Conclusions
This study shows that N'ZI watershed has rapidly changed the environment. The discrimination of the regions of interest allowed to define of six (6) groups of land use (Forests; Savannas, Soils-low-cover; Plantations-agroforestry and fallow; Dwellings-bare soils; Water body) by the treatment of satellite images multi-dates (1986, 2000, and 2020). The confusion matrix of different dates with Kappa values increases slightly between 1986, 2000, and 2020 respectively by 0.86, 0.90, and 0.92 testify to the classification of images was well done, when discriminating regions of interest. The detection of change using the area calculation, the overall rates of change, and the evolution statistics of the different class units indicate a degradation of the forest, plantation, and fallow classes, of the water areas in general. On the other hand, increase in bare land (67.02%) and agglomerations (103.35%) to the detriment of forests (−80.45%) and water bodies (−52.47%) are noticed. A regression of approximately (−14.87%) between 2000 and 2020 of plantations-agroforestry and fallow land to the advantage of savannas (89.98%). In the end, the basin is undergoing a significant change in all categories of its landscape. The most affected classes are groups of forests and water bodies that are renewed according to the rainfall fluctuations of the area.

Data Availability Statement
The data used in this manuscript are available free of charge from the USGS public data archive (https://www. usgs.gov). Remote sensing images from the Landsat satellite can be accessed and downloaded from http://earthexplorer.usgs.gov. Acquisition of Landsat remote sensing images of your study area requires prior registration. These data have been processed with Envi 5.1 and ArcGIS 10.4.1 software, which can be freely accessed and downloaded from the respective links: https://idl-and-envi.software.informer.com/5.1/ and https://support.esri. com/en/downloads/patches-servicepacks/list/productid/160.