Analyzing Land Use/Land Cover Changes Using Google Earth Engine and Random Forest Algorithm and Their Implications to the Management of Land Degradation in the Upper Tekeze Basin, Ethiopia

Land use and land cover change (LULCC) without appropriate management practices has been identified as a major factor contributing to land degradation, with significant impacts on ecosystem services and climate change and hence on human livelihoods. Therefore, up-to-date and accurate LULCC data and maps at different spatial scales are significant for regular monitoring of existing ecosystems, proper planning of natural resource management, and promotion of sustainable regional development. This study investigates the temporal and spatial dynamics of land use land cover (LULC) changes over 31 years (1990–2021) in the upper Tekeze River basin, Ethiopia, utilizing advanced remote sensing techniques such as Google Earth Engine (GEE) and the Random Forest (RF) algorithm. Landsat surface reflectance images from Landsat Thematic Mapper (TM) (1990, 2000, and 2010) and Landsat 8 Operational land imager (OLI) sensors (2021) were used. Besides, auxiliary data were utilized to improve the classification of LULC classes. LULC was classified using the Random Forest (RF) classification algorithm in the Google Earth Engine (GEE). The OpenLand R package was used to map the LULC transition and intensity of changes across the study period. Despite the complexity of the topographic and climatic features of the study area, the RF algorithm achieved high accuracy with 0.83 and 0.75 overall accuracy and Kappa values, respectively. The LULC change results from 1990 to 2021 showed that forest, bushland, shrubland, and bareland decreased by 12.2, 24.8, 1.2, and 15.4%, respectively. Bareland has changed to farmland, settlement, and dry riverbed and stream channels. Expansion of dry stream channels and sandy land surfaces has been observed from 1990 to 2021. Bushland has shown an increment by 17.2% from 1900 to 2010 but decreased by 19.5% from 2010 to 2021. Throughout the study period, water, farmland, dry stream channels and riverbeds, and urban settlements showed positive net gains of 484, 8.7, 82, and 26778.5%, respectively. However, forest, bush, shrub, and bareland experienced 12.17, 24.8, 1.2, and 15.37% losses. The observed changes showed the existing land degradation and the future vulnerability of the basin which would serve as an evidence to mitigate land degradation by avoiding the future conversion of forest, bushland, and shrubland to farmland, on the one hand, and by scaling up sustainable farmland management, and afforestation practices on degraded and vulnerable areas, on the other hand.


Introduction
Land use and land cover change (LULCC) without appropriate management practices has been identifed as a major factor contributing to global land degradation, with signifcant impacts on ecosystem services and climate change, which in turn afect human vulnerability [1,2].Te term LULCC expresses modifcation of the earth's terrestrial surface, mainly the results of interaction between natural and anthropogenic processes [3,4].It afects earth's system functioning [1,5,6] and its capacity to support human needs and increases the vulnerability of places and people to climatic and economic shocks [6,7].At the local scale, changes in the use of land and its cover afect watershed runof, microclimatic resources, groundwater, land degradation processes, and landscape-level biodiversity [6,8,9].Climate-driven land-cover modifcations and human activities mediated by institutional factors, markets, policies, and global forces drive land-cover changes [10,11].In general, LULCC is a critical driver of global environmental change, signifcantly impacting ecosystems, climate dynamics, and human vulnerability.Understanding these changes is essential for developing efective land management and conservation strategies.
Terefore, up-to-date and accurate LULC maps at different spatial scales are signifcant for the regular monitoring of existing ecosystems, proper planning of natural resource management, and promotion of sustainable regional development [12][13][14].LULCC has been observed across the world [6,[15][16][17].Many studies have also been conducted at the continental [13], regional [18,19], and local levels [6,9,[20][21][22][23][24][25][26][27][28][29][30][31].Many researchers have studied LULC changes in diferent parts of Ethiopia using GIS and remote-sensing techniques [4,9,[20][21][22][23][24][25][26][27][28][29][30][31].Tese local studies have yielded mixed results.An earlier study by Zeleke and Hurni [32] reported expanding cultivated land at the expense of forest and grass lands from 1957 to 1995 in Dembecha areas.Bewket and Abebe [9] found a reduction of natural vegetation cover and expansion of open grassland, cultivated areas, and settlements from 1957 to 2001 in the upper Blue Nile basin.In the same basin, Gashaw et al. [23] found an increase in cultivated land and built-up area at the expense of forests, shrubland, and grassland from 1985 to 2015.Mekonnen et al. [33] reported an increment in agriculture and settlement areas along with shrinks in forests and woodland cover in the Central Rift Valley of Ethiopia.In the Awash and Afar-Danakil basins, Ayele et al. [22] found the conversion of grazing land into plantation trees and area closure.In the Gelan watershed, Birhan Asmame and Assefa Abegaz [30] found a decrease in forest and wetland and an increase in shrubland, cultivated land, grassland, bareland, and settlement area.Most of these studies concluded that, like the trends on the global scale, the most remarkable change in Ethiopia is towards agricultural land, built-up/settlement areas [3,34,35].Nevertheless, few other studies have shown forest recovery under protected areas.For example, Alemayehu et al. [36] found a considerable increase in dense forest in semiarid Eastern Tigray from 1965 to 2005.Similarly, Nyssen et al. [37] reported an increase in forest cover in the Bella-Welleh watershed in the Waghimra zone of Ethiopia.
Geographically, most of the studies in Ethiopia are concentrated in the Rift valley [3,38], Blue Nile basin [9,32,35,[39][40][41], central Ethiopia, and the Awash basin [4,21,42].However, such types of studies are lacking in the northwestern and eastern parts of the country [34] generally, and in the arid and semiarid catchments of the upper Tekeze basin in the Waghimra administrative zone, in particular, except for a few studies on the Tekeza catchments in Tigray [36,43,44].Te upper Tekeze basin is the most degraded and drought-prone area with erratic rainfall.Due to their nature of rigid topography, most of the land is either degraded or highly vulnerable to feature degradation.Due to the severity of land degradation, government intervention in soil and water conservation has been introduced decades ago.Te ecological restoration programs under sustainable land management (SLM) practices, such as terraces and bunds, and establishing exclosures on communal grazing lands are among others.Following this program from 2010 to 2015, about 15 million people have contributed unpaid labour each year and more than 12 million hectares of land have been rehabilitated through implementing physical and biological conservation measures [45,46].However, as the level and cause of the previous degradation, drivers of degradation and LULCC and the efectiveness of conservation practices are contextual; the current understanding of LULCC and its implication to current and future degradation is not well studied.Terefore, this study was conducted in the data scarce region of Northern Ethiopia and can fll the gap by providing up-to-date LULC information and its implications for land degradation management.
Methodologically, several studies have improved landcover change measurements in recent decades, and scientists' ability to monitor changes in LULC has improved due to the application of GIS and remote-sensing technology [1,6].However, accurate LULCC assessment is still a vital study for evidence-based policy advising and strategies to achieve future conservation measures [3,8,22,47].In contrast to traditional remote-sensing methods, Google Earth Engine (GEE) ofers cutting-edge technologies and unrestricted access to a broader range of remote sensing data, fostering the formation of transformative land-change research issues [48].GEE has been used for a variety of applications and at various scales of analysis [49], including drought assessment [50,51], normalize diference vegetation index (NDVI) mapping [50][51][52][53], and land cover classifcation [52,53].Researchers have used this platform for land cover classifcation [52][53][54][55][56][57], and it has been suggested that GEE is very useful in spatial data management because of its fast and easy computation.
GEE also provides various pixel-based machine learning classifcation algorithms.Random forest (RF) support vector machines, K-nearest neighbour, artifcial neural network, and classifcation and regression tree can improve classifcation accuracy by reducing collinearity and overftting [53].Among these, support vector machine (SVM) and RF are two widely used algorithms [56,58] for a variety of earth science applications, including modelling forest cover, LULC, and object-oriented mapping [59][60][61].However, RF has gained great popularity and become one of the best classifcation algorithms in LULC mapping due to its better efciency and higher accuracy and the need for a few parameters [58].While several studies of developed regions used RF on the GEE platform for LULC change mapping, fewer studies have used it for less developed regions [62].Although several studies have been conducted on LULCC in Ethiopia, none of them used GEE and RF for data access and classifcation.Unlike previous LULCC studies that use ArcGIS and maximum likelihood classifcation, this study uses GEE for data access, preparation, and computation.Moreover, the RF algorithm has been used for the classifcation of LULC.

2
Te Scientifc World Journal Terefore, the overall objective of the study was to analyse the temporal and spatial dynamics of LULC in the last 31 years (1990-2021] and its implications for sustainable land management practices in the upper basin of the Tekeze River in the Waghimra Administrative Zone of Ethiopia.

Study Area.
Te study was conducted in the upper Tekeze basin in the Waghimra administrative zone of Amhara Regional State, Northern Ethiopia.It covers an area of 4593 km 2 and geographically extends from 12.11 °to 13.13 °N latitude and 38.40 °to 39.30 °E longitude (Figure 1).Te area's elevation ranges from 1060 to 3880 m above sea level (Figure 1).Te slope of the area ranges from fat plains to steep escarpments.Te mean slope of the zone is 19.17%.Undulated topography with poor vegetation cover characterizes most of the Waghimra administrative zone, particularly the south west and the northern part of the study area.
Tis zone is one of the most drought-afected areas in the past.Land degradation, frequent drought, and the high population densities characterize it [64,65].Migration for short-term employment has been a feature of the livelihoods of the poorest in Waghimra.Most young people consider migration for work to be a temporary response to an inadequate cash fow [66].Waghimra is part of Ethiopia's northern highlands, which are highly vulnerable to climate change, including extreme climate events such as droughts.High vulnerability to drought and famine, growing population pressure, and land degradation have resulted in reduced yields.People residing in Waghimra are frequently in immediate need of food aid.Although NGOs and governmental organizations could encounter basic needs, the root cause of food insecurity is related to the region's degraded ecosystem and not yet tackled.
In almost all parts of the Tekeze basin, the summer (Kiremt) rain starts in June and ceases around the end of August [67].Specifcally, the mid-and high-altitude areas receive rainfall from late June to early September, while in the lowland parts of the basin, it extends from early July to mid of August.Te primary crop production system is rain fed during the summer season.About 63% of the annual rainfall occurs in July and August.Te area's annual minimum and maximum temperatures are 12.02 °C and 30.4 °C, respectively (Figure 2).

Data.
Tis study used level 2 surface refectance (SR) products from Landsat 5-TM and landsat 8 OLI sensors available in the GEE.Te U.S. Geological Survey (USGS) provides the level 2 SR data from Landsat 8 generated using the Land Surface Refectance Code (LaSRC) algorithm.For Landsat 4 to 7, SRs are derived with the Landsat Ecosystem Disturbance Adaptive Processing System (LEDAPS) algorithm [68].Landsat collection 2 level 2 data have undergone a series of atmospheric and geometric corrections and are science products with values added and highly processed Landsat products [69,70].Tese SR collections ensure that multidate images are comparable and allow more stable and reliable land change analyses [52].SR is generally more appropriate for measuring and monitoring vegetation and other land cover types at the land surface [69,71,72].
Four sets of digital satellite imagery from TM and OLI sensors for the years 1990, 2000, 2010, and 2021 were used to examine LULC dynamics (Table 1).Years of analysis (1990, 2000, 2010, and 2021) were selected purposely to align with signifcant sociopolitical events in Ethiopia and Northern Ethiopia.Accordingly, the 1990 image indicates the land and environmental conditions during the Derg regime.Te year 2000 represents the aftermath of the fall of the Derge regime and the early period of Ethiopian People Revolutionary Democratic Front (EPRDF) regime.During these periods, environmental management and protection were not the government's top priorities.Te year 2010 represents the efforts of soil and water conservation (SWC) and SLM programs in diferent parts of Ethiopia, including the study area.In 2010, the government of Ethiopia launched nationwide ecological restoration and area exclosure programs [73,74].Finally, the 2021 image represents the current biophysical status and observed changes after 2010 area exclosure and degraded land restoration policy.In order to minimize the seasonality efects, images were mainly selected from January to March.

Auxiliary Data.
Previous studies have shown that diferent remote sensing indices are sensitive to diferent types of LULC and enhance classifcation accuracy [54,55,57,75,76].To improve classifcation accuracy, spectral indices retrieved from the original bands, such as NDVI, Normalized Diference Water Index (NDWI), Normalized Diference Moisture Index (NDMI), and Normalized Diference Built-Up Index (NDBI), were used as additional bands in this study.Furthermore, the tasselled cap ratio was derived and incorporated in the classifcation bands.Te tasselled cap component is widely applied to characterize vegetation conditions.Tese indices measure the presence and density of green vegetation, total refectance, and soil moisture content [57].After visually inspecting the pattern and accurateness of the index, by overlaying on known sample points, the derived indices were evaluated and added to the classifcation bands.In addition, topographic variables, such as elevation and slope, were used as auxiliary variables for the classifcation.

Data Processing and Classifcation
Overview.An overview of the methodological framework applied in this study is shown in Figure 3. Te majority of the image processing and analysis for the study was implemented through GEE.All data sets used in this study are freely available on GEE.We have fltered cloud-free, dry season (January to February) SR images clipped to the study area.Indices were also derived from these cloud-free images and added as an independent band to the original Landsat images.Indices and other axillary data were generated using the GEE API.After we had obtained all the required bands, we overlaid the training and validation points of each LULC on the fnal image.Te sample points were selected using feld data, Google Earth imageries, and NDVI thresholds.Ten, tuning Te Scientifc World Journal  Te Scientifc World Journal machine learning classifer hyperparameters, generating classifcation maps, and assessing the accuracies of classifed maps were performed.Finally, the total area and the change of each LULC class and the temporal pixel transition matrix from one LULC category to the other were calculated to assess and analyse LULCC patterns.

LULC Classifcation Schemes.
As described in Table 2, eight LULC types were identifed in the study area.All training and validation samples were collected based on feld survey data and manual visual interpretation of highresolution images from Google Earth and the GEE base map.Tis method is widely applied in the literature [53,55,54].Multitemporal Google Earth aerial imageries were used to select suitable training sites for the eight LULC types from the acquired Landsat images.For the earlier study periods (1990 and 2000), the available Google Earth imageries corresponding to the study area are less precise than those from 2010 to 2021.However, the false colour interpretation of the acquired satellite images was employed to identify sample points from all land covers.Tis photo interpretation technique was also enhanced by incorporating axillary data in the classifcation algorithm.For example, the NDVI threshold was used to identify vegetation.Dry and high-refecting bareland was also easily identifed with the help of the tasselled caped dryness index.In addition, the temporal stability principle was considered to select sample points.For example, church forests are among the convenient sample points in the study area due to their stability for extended periods.Riverbeds and urban settlements were also easily identifable features and have been stable since their existence.A total of 1721 ground control points (GCP) were collected and used for the LULC classifcation.

Classifcation Method and Process in GEE Using RF
Algorithm.Machine learning-based classifers help identify complex patterns while at the same time minimizing the problem of data dimensionality [53].Te RF consists of many individual decision trees.Each decision tree has several nodes, and the majority vote determines the result.Te algorithm not only randomly selects subsamples from the input variables but also randomly selects the best feature through a voting process to establish the splits in the nodes of trees.According to a systematic review from 2010 to 2019, the RF algorithm is one of the most frequently used classifcation algorithms.Hence, this study applied the RF classifcation algorithm for LULC classifcation.Among the 1721 sample points, 70% were used for training the RF classifer, and the remaining 30% were used for validation.Te number of decision trees was set to 50, which was found to be the optimum number producing a good accuracy level.Te accuracy evaluation was performed using indices including the user's accuracy (UA), producer's accuracy (PA), overall accuracy (OA), and kappa coefcient [77][78][79].Tese accuracy indices are calculated by constructing a confusion matrix using the GEE syntax.Te GEE inbuilt code for construction of error matrix and calculation of LULC accuracy is based on Stehman [80].

Accuracy Assessment of the LULC Maps.
After classifying the images using the collected samples, accuracy was assessed.Table 3 shows the details of the accuracy evaluation of the RF classifcation (user's accuracy (UA), producer's accuracy (PA), and overall accuracy (OA)) and kappa classifcation.Te accuracy evaluation for each classifed image was calculated by constructing a confusion matrix.
Te RF classifer produced good overall accuracies, with overall accuracy assessments of 0.83, 0.86, 0.88, and 0.88 for 1990, 2000, 2010, and 202, respectively, and the kappa accuracies were 0.75 (1990), 0.79 (2000), 0.83 (2010), and 0.82 (2021).Water had high classifcation accuracies each year because its refectance is easier to distinguish from other categories, whereas bareland and urban settlements had low accuracies.It was also a challenge to diferentiate the spectral characteristics of bushland and forest areas due to the narrow spectral signature diference among these land cover types.Although the area coverage of forest was low in the study area, it was essential not to merge with bushlands.We think it is crucial to quantify and identify the church forests, riverine trees, and plantations that are sparsely located across the study area.Compared with surface water and vegetation, the built-up and barren areas showed a relatively low accuracy.Te aridity of the northern part of the study area challenged the identifcation of barren land from urban settlements, dry stream channels, and sandy fooded and sparse shrubs.However, the incorporation of the axillary data, elevation, slope, and tasselled cap indices, helped improve the identifcation of these resembling spectral characteristics.Note that urban settlement in 1900 was almost absent because it was very small during this period, and most of the settlement was a rural settlement with a similar refectance value with farmland and bareland.Figure 4 shows the land cover classifcation maps for 1990, 2000, 2010, and 2021.

LULC Change in the Study Area.
In 1990, shrubs covered 55.93% (256866 ha) followed by agriculture, which covered 26.84% (123250 ha), and bareland, which covered 9.11% (39408 ha) of the study area (Figure 4(a) and Table 4).Water bodies, forests, and bushlands each accounted for 0.1, 0.5, and 7.05% of the study area, respectively.Urban settlement and residential areas covered only 1.66 ha of the study area.
In 2000, shrubland and farmland remained the most predominant land cover types, accounting for 242284 ha (52.76%) and (132724.28ha) (28.90%) of the study area, respectively (Figure 4(b) and Table 4).Forest cover increased by 160 ha, accounting for 0.51% of the entire study area.In 2000, bushland was 42334 ha, increased by 30.7% (9953 ha) from the 1990s coverage.Bareland and stream channels combined covered 8.45% of the study area in the 2000 LULC map.
On 2010 LULC map, the study area was covered by 51.30% shrubland, 30.04% farmland, and 8.33% bushland (Figure 4(c) and Table 4).Other land use types covered the remaining 10.28% of the area.Shrubland, bushland, and Te Scientifc World Journal forest cover decreased by 6691, 4083, and 406 ha, respectively, from 2000 to 2010.Tese reductions were due to the expansion of 5212 ha farmland, 14 ha bareland, and 2364 ha riverbed and stream channels.Compared to the preceding study period, the water body has increased dramatically by 1529 ha.Tis was due to the accumulation of water in the Tekeze River's channels due to the Tekeze Hydroelectric Dam's construction.Forest cover has decreased by 4066 ha compared with the previous study period.
Except for what has been observed in bushland decline and shrubland increase, LULC distribution in 2021 is comparable with that in 2010.Te study area was still dominated by shrubland, followed by farmland, as they have been for the last two decades.Only water bodies, shrublands, and urban     Te Scientifc World Journal settlements expanded.Unlike the past 20 years of the study period, bush lands have experienced the most signifcant decline in this period.Conversion to shrubland and encroachment of riverbeds and fooded sandy stream channels are the prime causes of bushland decline.Te general trend in the last three decades (1990-2021) showed that water, farmland, stream and riverbed, and urban settlements and residential areas were increasing.Te remaining land covered classes, forest, bush land, shrubland, and bareland, decreased by 12.17, 24.79, 1.18, and 15.37%, respectively.Settlement, water bodies, riverbed, and stream channels have also expanded, indicating the expansion of dry stream channels and sandy land surfaces in the study area.Generally, the water body has increased by 81.17 ha every year from 1900 to 2021.In contrast, forest land increased by 12.17% from 1990 to 2021, with a rate of 8.84 ha yearly.Bushland has shown a continuous increment in the frst two consecutive decades (1900-2010) by about 17.24%.However, it experienced a decline of 19.56% in the next decade from 2010 to 2021.Troughout the study period, water, farmland, stream channels, riverbed, and urban settlements showed positive net gains of 483.97, 8.70, 82.14, and 26778.47%,respectively.At the same time, forest, bush, shrub, and bare LULC classes experienced losses of 12.17, 24.79, 1.18, and 15.37%, respectively.Figure 5 shows that the rate of LULC change is not stationary during the studied time intervals.Te length of bars on the right indicates that the annual change in the three periods was not identical.Left of Figure 5 shows that the size of the change during the frst-time interval (1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998)(1999)(2000) was the largest.Te next two periods experienced decreasing change.On the right side of Figure 5, the frst interval (1990 to 2000) was with the fastest change intensity in terms of the annual rate of LULC change, and it decreased in the period between 2000 and 2010 and then in the latest period.

LULC Transition Dynamics.
Te land-use transition matrix between 1990 and 2000, 2000-2010, and 2010-2021 was derived using JavaScript code in the GEE environment.Multistep (Figure 6) and one step (Figure 7) transition graphs were produced in the OpenLand package in R software [81].Te purpose of generating a transition matrix was to identify the magnitude of the transition of pixels under a specifc land cover class to another land cover category.Tis matrix shows the magnitude and direction of LULC changes within the study area.All LULC types experienced changes, and the intensity of change difered among the seven LULC classes.(147.37 ha or 29.2%) was changed to dry riverbeds and stream channels.Despite the conversion of water bodies to dry riverbeds and stream channels in 2000, 67.3 ha of land covered by a dry riverbed in 1900 also changed to a water body in 2000.Nevertheless, from 1990 to 2000, other land use types also changed to a water body.Te remarkable conversion was from a shrub in which 199 ha of its cover changed to a water body.Despite the net areal extent increment of frost from 1900 to 2000, 809 and 502.42 ha of forest were changed to shrubland and bushland, respectively.However, another 613.55 and 552.61 ha of land covered by bushland and shrub in 1990 have changed to forestland.Te appearance of trees and agroforestry in irrigated areas might be reasons for the addition of 345.34 ha to forests.During 1900-2000, 56.47% of bushlands remained unchanged, while 38, 3.33, and 1.9% were converted to shrubland, farmland, and forest.Te remaining small portion has changed to water, bareland, stream channels, and riverbeds.From 256610 ha of land covered with shrubland in 1900, 73.78% of it remains unchanged.Te other 3588, 20411, and 7885 ha of shrubland changed to farmland, bush land, and bareland, respectively.Te reaming 1.5% of shrub bland changed to water, forest, riverbed, and stream channel.Bareland experienced the greatest instability; only 46% of its coverage in 1900 remained unchanged from 1990 to 2000.Te primary conversion was towards shrubland, farmland, and dry riverbeds.In this transition period, an increase in the areal coverage of dry riverbed and stream channels has been observed.Tis increment was due to conversion from shrubland, farmland, and water bodies.Tis conversion of shrubs and farmland into dry riverbeds indicates severe drought that leads to dry, fooded farmlands and sandy and highly dispersed shrubs.

LULC Transition from 2000 to 2010.
Water has experienced the most tremendous increase in this period, gaining 1579.11ha of land.While 19.23% of its cover comes from 2000, 18.41%, 27.63%, and 27.67% were transformed from shrubland, bareland, riverbeds, and stream channels.Forest cover has decreased from 2807 ha to 1935 ha due to the encroachment of shrubs and bushlands.Bush and woodland have gained a net of 3754 ha of land from 2000 to 2010.Almost 2000 ha of the increment come from shrubland.About 400 ha of land were also converted from farmland and forest.During 2000-2010, shrubland and farmland showed the highest stability, in which 79% and 78.45% of the pixels were unchanged.On the other hand, forest, bush, and bare and dry riverbeds have experienced the greatest transition in which only 40, 58.57, 59.22, and 53% of their coverage, respectively, were unchanged from 2000 to 2010.Although there was a bidirectional transition, the transition of pixels of farmland and shrubland to bareland was marked.Similarly, land that was covered by bareland and shrubland has also changed to farmland in this period.Te areal coverage of forest has shown a reduction because of the conversion to bush and shrublands.Both conversions to and from stream channels and riverbed were observed during this period.Te net increase in the area coverage of this land cover type was primarily due to the conversion of shrubs and barelands to stream channels and riverbeds.Unlike the previous study period, urban settlement has considerably increased in this period.Te LULC of the upper Tekeze basin has experienced irregular gain and loss; however, high intensity of change occurred between the study periods.Figure 8 shows the frequency of change of each pixel in the study time interval and the areal percentage of these frequencies of changes.

LULC Transition
Te Scientifc World Journal Almost half of the land in the overall basin experienced LULC change from 1990 to 2021.51.01% of the basin was unchanged throughout the study period.However, 23.2, 19.7, and 6.1% of the land experienced 1, 2, and 3 times change, respectively, during the study period.

LULC Distributions and Changes along Diferent Slope
Classes and Agroecological Zones.In the study area in 1990, 65.54% of the land cover lay below 20% slope.In addition, 32.66% were between 20% and 50% slope gradients (Figure 9(a)).Only 1.76% of the land cover was found above

38.6°E
Frequancy of Changes 0 Change (51.02%)  12 Te Scientifc World Journal maps in 1900, 2000, 2010, and 2021.In all study periods, more than 26% of the lands cover lies between 20 and 50% slope gradient.A very small proportion of the total land cover, less than (1.76%) was found above the 50% slope.35% of the farmland lies between the slope classes 10-15.
Regarding the distribution of the diferent LULC categories across the diferent slope gradients, Figure 9 shows the percentage of the total land cover of each class that lies on each slope class.Looking at the recent LULC distribution among diferent slope classes, the forest was distributed throughout all slope gradients.Almost all the gentle and higher slope gradients were nearly equal in forest coverage (Figure 9(b)).
Farmland was distributed across all the slope classes.Only 15% of the farmland was found below a slope of 5%.While 60% of the total farmland is found between 5 and 20% slope, there is still a signifcant proportion of farmland (14.47%) between 20 and 30% slope gradient.Te study area has a long history of crop cultivation where most of the cultivable land is already occupied by croplands.Recent expansion of farmlands is on high slope areas where critical care is needed in order to protect soil and water conservation.A similar trend of increased cultivation of steep slopes has been observed in other parts of Ethiopia [31].In this agroecology, the farmland lies in high slope classes.Tis aggravates land degradation through water erosion processes and washes away soil nutrients.Te highland agroecology which most of its part is distributed across a high slope gradient and elevation constitute Dega agroecology.Tis agroecology is good for agricultural production.Particularly, legume plants, wheat, and barley are produced on fragmented plots.Tis high slope area cultivation in this zone requires continuous eforts to conserve soil and water so that the productivity will sustain in the future.
Although we have found forests, bushes, and shrubs fairly distributed from the lower to the higher slopes, their proportion is considerably high on the higher slope gradient.Forest, bush, and shrub percentages on the extremely high slop area (i.e., above 50%) are 4.4%, 2.57%, and 2.36%, respectively.Tis may be due to the growth of vegetation in areas with less human intervention.
Land use is the primary driver of land degradation, when compared with soil type, terrain, or climatic factors.Te general pattern of LULC change shows that the study area is either degraded or vulnerable to degradation.In addition to this, the trend of LULC change indicates no sign of improvement in vegetation cover.Te observed increase in farmland, stream channels, and riverbeds, conversion of bushland to shrubs and bareland, and decrease in forest cover are indications of land degradation in the study area.For instance, as farmland land was expanded at the expense of other land use and land cover units, bushland and forestland declined, increasing the vulnerable area to soil erosion.Te increase in dry riverbeds and stream channels is also at the expense of shrubland.Te increase in dry riverbeds and stream channels alone is an indication of degraded land.Te combined efect of erratic rainfall, poor vegetation cover, and high human and livestock population pressure on land resources may have contributed to an increase in dry riverbeds and stream channels and further expansion of bareland.
From the agroecological perspective, the high and midaltitudes are better in terms of forest and bushland distribution.Bareland and dry riverbeds are less available in this agroecological zone.For example, Gazigibla district, whose major part lies at mid-and high latitudes, holds 65.2, 44.87, 54.42, and 49.8% of the total forest coverage in 1990, 2000, 2010, and 2021.Te lowland part of the study area, which has a semiarid and arid climate, was found to be more vulnerable to land degradation due to its LULC status.Te fnding showed that this part of the agroecological zone has greater area coverage of bareland, river beds, and stream channels.Due to this and a steep slope and ridged topography, the majority of the land is vulnerable to soil erosion.For instance, the lowland part of the study area, Zquala district, has shown signifcant degradation as expressed in vegetation cover and expanded bareland.Riverbeds and stream channels are also abundant, which are indications of gully formation.

Implication of the LULC Dynamics to the Management of
Land Degradation in the Study Area.Land cover is one of the factors that determine the rate and status of land degradation [30].Reduction in vegetation cover is the major cause of soil erosion particularly in mountainous ecosystems.After vegetation cover is removed, factors such as the steepness, length, and shape of a slope become important accelerators of erosion [32,82].Tis process increased sheet, rill, and gully erosion by reducing the protection of soil cover.Tis study identifes land use transition from and to vegetation cover to assess the degradation level and trend.Te vegetation cover at the base year (1990) was minimal; particularly forest was only 0.47% of the total basin and bushland was also 7.5% of the basin in 1990.Tis indicates that the LULC of the study area was already degraded and was potentially exposed to soil erosion.Shrubland that covers 56% of the study area is characterised by very spare small tress and bush, and the ground consists of exposed rock and soil.Te land covered by this LULC category is also prone to soil erosion.Considering the topographic and LULC diversity of the basin, dense bushes and forests are helpful for erosion management and reduction of land degradation.Te study assessed the level of aforestation or reforestation and forest conservation: by studying the persistence of or change to vegetation from 1990 to 2021.Conversion of forest, bushland, and shrubland to other LULC types (urban settlements, bareland, riverbed and stream channel, and farmland) was identifed as vegetation degradation, while the reverse change is gain of vegetation.Figure 10 shows the distribution of vegetation gain and loss and persistence of farmland and shrubs from 1990 to 2021.
Te overall trend observed in the study area is the decline of forest cover, bushland, and shrubs and an increase in farmland (including rural settlement), dry riverbeds and stream channels, and urban settlement.Even forestland was and is very small in area coverage; no aforestation or Te Scientifc World Journal reforestation through plantation was observed over the study period (Figure 10).Bushland has decreased, but was increasing from 2000 to 2010 due to area closure introduction; however, it has decreased from 2010 to 2021.Bareland has changed to farmland and shrubs, washed away by fooding, and continuously changed to dry riverbeds and stream channels.Tese LULC types are highly susceptible to erosion.Moreover, slope, elevation, and high surface temperature aggravate the rate of soil degradation unless an urgent measure is taken.In the Ethiopian mountains, soil degradation due to water erosion is a major threat to agricultural production [82].Te observed expansion of dry stream channels is an indication and aggravating factor of soil erosion.
A land cover change interacts with the hydrological cycle.Infltration, runof production, and soil erosion are determined by the nature of ground cover.Low level of vegetation cover and expansion of barelands' area are responsible for high surface runof and hence soil erosion.Te observed LULC dynamic of the basin showed that there is poor vegetation cover and extensive land of barelands.Furthermore, there is increased areal extent of dry riverbeds and stream channels which are the results of high surface erosion during intense rainfall.Particularly the ridged topography of the northern part of the basin there is expansion of increased dry river stream channels.Te major rivers and their tributaries form deep canyons and gorges with steep and narrow river valleys.Along the course of these valleys there are many springs that emanate along the contacts of the diferent rocks [83] as they draw downward to the lower part of the basin; they have developed a wide food plain and eroded bare grounds, expanding dry river beds.Due to the low retention capacity of the soil, low vegetation cover, and steep slope, the erratic rainfall has produced wide and many irregular dry streams.Tese many stream channels bring fow to the farmlands on the hillside and lowland areas, destroying vegetation and crops during the rainy season.
In addition, the constructed hydroelectric dam has changed the hydrological regime of the basin by increasing water at the back of the dam as well as by widening the Tekeze river channel.Te sideways of the river have been washed away, shrubs have been reduced and bare, and dry riverbeds have expanded.In the same basin, Welde and Gebremariam [82] reported that the mean annual stream fow and annual sediment yield of the Tekeze dam watershed show an increase in average annual stream fow.Similarly, sediment yield change shows an increment.Implementation 14 Te Scientifc World Journal of soil and water conservation measures and construction of hydropower dams and microdams for irrigation are predominantly found in the upper catchments.Te land use dynamics observed in the study area would have a clear efect on the erosion rate, sediment yield, and expansion of many irregular dry riverbeds and stream channels.Tis has implications for the management of land degradation in the basin, requiring integrative efort to manage the land change system in the way it benefts the soil and hydrological regimes.

Discussion
Te LULC change studies provide useful information for a better understanding of previous practices, current LULC patterns, and future LULC trajectory [83].A change in LULC is one of the major causes of changes in Earth's system functioning [84].Te overall accuracy and kappa coefcient of this study are comparable with [12,84].
Te fndings of this study show that there is a decline in vegetation cover and no marked improvement in forest cover and other vegetation as expected following the continuous land management programs.Tis fnding is in line with [3,40,[85][86][87] who reported a substantial decrease in forest cover, grasslands, and bush-shrub-woodland.However, the fnding is against reports by [88] who reported an increase in woody savannas, deciduous broadleaf, grasslands, permanent wetlands, and mixed forest areas and reductions in croplands and water bodies in the Baro-Akobo River Basin of Western Ethiopia.Similarly, in recent LULC dynamics of the Blue Nile River, an increment in plantation has been observed in the North Gajjam sub-basin [41].However, in this study, no signifcant increment was observed in plantations such as eucalyptus trees.
In the same basin, Nyssen et al. [37] reported an increase in vegetation cover in the Bella-Welleh watershed in the Waghimra zone.Because our study covers a large areal extent and was not limited to a single watershed, the fndings are not comparable with those of this study.However, Welde and Gebremariam [89] reported a reduction of shrubland and grassland due to the expansion of the agricultural practice in the area in the same basin.In another part of Ethiopia, but with the similar environmental conditions, Shiferaw et al. [90] reported a decrease in bush-shrub-woodland and natural forests in the Afar region.In the Blue Nile basin, Gashaw et al. [91] reported decreased coverage of shrub/bush LULC for 30 years .Similar decreasing trends were reported by [4,21].Te continuous decline of forest, shrub, and woodland cover was primarily due to the expansion of the urban built-up areas and cultivated and rural settlement areas.Te decrease in shrub/ bushland use and land cover implies that the land is vulnerable to soil erosion and fooding, afecting farmland productivity in the areas.Despite the loss of vegetation cover in the study area, the areal extent of conversion to bareland was not found to be a marked fnding.Nevertheless, there was a decline in bareland cover due to conversion to farmland, riverbeds, and stream channels.Unlike other studies such as [30,41], water body has expanded in the study area during the recent decades of the study period.Surface water accumulation has increased along the river channel of the Tekeze River, particularly at the back of the Tekeze hydroelectric power dam.Tis increment has positive implications for the surrounding ecosystem and livelihood of communities residing in the study area.
Te pattern of temporal LULC changes was highly complex, with a multidirectional transition from one LULC to the other.It does not have a clear trend except for the overall transition towards the LULC categories that have gained from decade to decade and throughout the study period.Due to climatological efects, LULC has undergone a series of transitions, for example, from farmland to eroded/ fooded bareland, from bareland to dry stream channels and riverbeds, and to fooded, sandy shrubland.Te probability of transitions to forestland and bushland was very low from decade to decade.Instead, a higher probability of transitions was recorded from all land use types to farmland, shrubland, bareland, and dry riverbeds and stream channels.
All the changes observed in the study area showed that there is unhindered land degradation as manifested by the decline of the existing forest cover and no sign of aforestation.Furthermore, the degradation of bushes and shrubs, expansion of dry stream channels, and barelands are causes and indications of degradation.Many studies have shown that surface erosion is minimal in areas where the soil is covered by vegetation [4].
What is specifc to this study is that the expansion of farmland is less pronounced than other studies such as Mekonnen et al. [3] and Moisa et al. [86].Farmland has increased only by 8.70 from 1990 to 2021; there was a slow transition of other land cover types to farmland every decade.Te possible reason is that almost no land was left for further expansion.Topographic, hydroclimatic, and soil degradation are challenges to farmland expansion as well as crop production.However, the recorded increase in cultivated land was at the expense of bush and shrubland.Tis aggravates the vulnerability of soil to erosion.
LULC driving factors in Ethiopia include demographic, socioeconomic, and institutional factors [31].Like the rest of the Ethiopian basins, these factors apply to the upper Tekeze basin.Nevertheless, physical and climatic elements that trigger signifcant LULC [5] are among the driving forces in the arid and semiarid land of the upper Tekeze Basin.As Te Scientifc World Journal observed in the LULC change across agroecologies, the drier and rigid topography of the lower basin was a hotspot of vegetation degradation and hence is susceptible to soil erosion by water.Tis study demonstrated that although there is not signifcant farmland expansion, urban and infrastructural induced LULC change in this marginal arid and semiarid basins, and topographic, agroecological, and climatic factors such as aridity and frequent drought could cause land cover change and challenge land degradation management.
Although a lot of implications can be associated and discussed based on the observed LULC change magnitude and direction, future research should focus on understanding of the impact of LULC change on the water resources, ecosystem productivity, and livelihood security in the study area.

Conclusion
Tis study focused on the status and trends of LULC changes in the semiarid and arid land catchment of Tekeze River in the Waghemra Administrative zone of Amhara region, Ethiopia.LULC classifcation was performed using the RF algorithm on the GEE computing platform based on sample datasets with sufcient auxiliary data.Te methodology combined seven axillary data and took the fast processing and variable controlling ability of both GEE and RF classifcation algorithms.Although the landscape of the study area is so complex and hence hinders accurate identifcation of pixels of diferent land covers, the classifcation accuracies and visual inspection of classifed images ensured that the RF classifer yielded consistent and accurate maps for the study area.
Te study demonstrates that the arid and semiarid land of the upper Tekeze basin in Waghimra Administrative Zone has experienced land use/land cover change over the last thirty-one years.Te fnding revealed that the water body has expanded 4.8 times from its 1990 coverage by increasing on average 81.17 ha every year 1900 to 2021.Te expansion of the water body was primarily due to the accumulation of water behind the Tekeze hydroelectric dam.Forest increased by 7.36% in the frst 10-year interval and decreased by 17.37 and 0.98% in the next two decades from 2000 to 2021.Bushland has shown continuous increment during the frst decade (from 1990 to 2000) by about 30.74% and decreased by 9.64 and 36.34%from 2000 to 2010 and 2010 to 2021, respectively.Te study revealed that most of vegetation degradation was happened during from 2000 to 2021 despite the installation of conservation and restoration government policies.Positive vegetation changes were limited to small pocket area exclosures and managed watersheds.Overall, throughout the whole study period, 2435.04,10647.44,3825.67 44, and 444.27 ha net gain was recorded on water bodies, shrubland, dry riverbeds, and stream channels, respectively.Te water body has per year from 1990 to 2021.Dry riverbeds and stream channels and farmland have increased by 82.14 and 8.64%, respectively, during the same period.On the other hand, 8028.69,6056.25,3043.15, and 265.12 ha loss was recorded on bushland, bareland, shrubland, and forest cover, respectively.
Te study area's Woynadega and Dega agroecological districts are better regarding vegetation cover.Furthermore, barelands and dry riverbeds are less available in these agroecological zones.Te fnding showed that this part of the agroecological zone has greater area coverage of bareland, riverbeds, and stream channels.However, the semiarid and arid low land part of the study area was found to be more vulnerable to land degradation because of its LULC status.Combined with a steep slope and ridged topography, the majority of the land is vulnerable to land degradation, particularly soil erosion.For instance, the lowland (Kolla agroecology) part of the study area has shown remarkable degradation expressed by the increment of bareland, riverbeds, and stream channels.Furthermore, the areal coverage of forest and bushland is minimal compared with the Dega and Weyna-dega agroecology and the upper part of the basin.
Te general trend observed in the study area is the decline of forest cover, bushland, and shrubs and an increase in farmland and rural settlement, dry riverbeds and stream channels, and urban settlement.Forestland was and is very small in area coverage; no aforestation or reforestation through plantation was observed over the study period.Shrubland in the basin is characterised as sparse, short shrubs over a sandy, rocky landscape.Coupled with steep slopes, rigid topography, and high surface aridity, the observed LULC change aggravates soil erosion, biodiversity loss, disturbance of hydroclimatological balance in particular and land degradation in general in the study area.
Results from this study provide an important input to decision makers in their eforts towards sustainable land use planning and management.Te study highlights the need for implementation of sustainable LULC practices such as largescale reforestation, area exclosure, and prevention of bareland through restoration of degraded areas, conservation of forest and bushland, and limiting the expansion of cultivation areas in high slope regions.Despite being one of the most severe hotspots of land degradation due to its history of signifcant degradation and drought, the land restoration programs initiated decades ago have not signifcantly infuenced the LULC dynamics.Terefore, it is crucial to evaluate the past restoration approaches and adopt new policy frameworks towards sustainable land management.In general, the study suggests that urgent measures must be taken to mitigate the observed land degradation by avoiding the future conversion of forest, bushland, and shrubland to farmland while scaling up sustainable land management, reforestation, and aforestation practices on degraded and vulnerable areas.

Figure 3 :
Figure 3: Schematic overview of the methodological framework of the study.

3. 3 . 1 .
LULC Transition from 1990 to 2000.Multistep LULCC transitions from 1990 to 2000, 2000 to 2010, and 2010 to 2021 are presented in Figure 6.About 60.5% of land covered by water in 1900 transformed to 2000 without change.Te remaining portion of land covered by water in 1990

Figure 5 :
Figure 5: LULC change intensity showing the size and speed of change across the three studied time intervals (1990-2000, 2000-2010, 2010-2021).Bars that extend to the left of zero show the percentage of change during the corresponding interval, and bars that extend to the right of zero show the percentage of change per year within each time interval.Uniform rate (U) assumes if the annual changes were spread evenly over the study period.

Figure 6 :
Figure 6: Multistep LULC transition between 1990 and 2021 in the upper basin of the Tekeze River; size of colored bars and transition links are displayed proportionately to area in km 2 (note: W � water body, F � forest, BL � bushland, Shl � shrubland, FrL � farmland, BrL � bareland, RbSc � dry river beds and stream channels, and UrS � urban settlements).

Figure 7 :
Figure 7: One-step LULC transition from 1990 to 2021 in the upper Tekeze Basin.Te left bars represent LULC categories and their proportion in 1990; the right side bars represent the LULC categories and their proportional size in 2021.Links are displayed proportionately to area in km 2 (note: W � water body, F � forest, BL � bushland, Shl � shrubland, FrL � farmland, BrL � bareland, RbSc � dry river beds and stream channels, and UrS � urban settlements).

Figure 9 :
Figure 9: Distribution LULC types across diferent slope gradients in the upper Tekeze basin in 1990 (a) and 2021 (b).

Figure 10 :
Figure 10: Changes/conversion to and persistence of LULC in the upper Tekeze basin from 1990 to 2021.

Table 1 :
Details of satellite images and other datasets used in the study.

Table 2 :
Description of LULC classes used in the study area.

Table 3 :
Accuracy of the RF classifcation algorithm.

Table 4 :
Magnitude and pattern of LULC change in the upper Tekeze basin (1990-2021).
Te expansion of agricultural land, for example, occurs at the expense of shrub and other vegetated lands which fnally causes soil erosion, sedimentation, and loss of biodiversity.Land cover changes are usually caused by human activity such as urbanization, agricultural expansion, and deforestation.Tis study shows the LULC changes dynamics in the upper basin in Tekeze in the Waghimra Zone of Northern Ethiopia for the last three decades.Advanced techniques of LULC classifcation aided by the GEE cloud computing platform and RF algorithm were used to study the LULC dynamics of the study area.