Modeling land use dynamics in Borkena watershed, Awash River basin, Ethiopia: Implications for sustainable land management

ABSTRACT

The land use and land cover (LULC) change has far-reaching repercussions on both natural ecosystems and socioeconomic systems worldwide. The objective of this study is to investigate the dynamics of LULC, identify its causes, and evaluate its magnitude over three distinct time intervals (1993–2003, 2003–2013, and 2013–2023) in the Borkena watershed. Various methodologies, including remote sensing techniques, field observations, and focus group discussions, were employed to analyze the changes in LULC. Additionally, community perceptions were assessed through Focus Group Discussions (FGDs) and Key Informants Interviews (KIIs). Four LULC classification maps were generated and utilized to analyze land use changes, while image classification was conducted using the maximum likelihood method. The six primary land use and land cover categories considered were forest, cultivated land, shrubland, water bodies, built-up areas, and bare land. The results indicated an increase in settlements (77%), cultivated land (12%), and bare land (7%) between 1993 and 2023, while a decrease in forest cover (10%), shrubland (150%), and water bodies (−101%). These changes were predominantly observed in areas with gentle slopes (0–8%) and low altitudes (0–500 m). The findings suggest a significant expansion of agricultural and urbanization activities within the watershed. Anthropogenic activities such as agriculture and settlement had a more pronounced impact on low-lying and gently sloping areas compared to high-altitude and steep slope regions. Key informants and focus group discussions highlighted rapid population growth, rainfall variability, soil fertility decline, and a scarcity of fuelwood as the primary factors contributing to these changes.

KEYWORDS:

• Land use
• land cover dynamics
• land degradation
• drivers and causes of LULCC
• Awash River basin
• remote sensing

Introduction

The highland of Ethiopia is currently experiencing an unprecedented rate and scale of change in land use and land cover (LULC). This trend poses a significant challenge to the natural ecosystem and increases vulnerability to environmental hazards (Bekele et al., 2019). It is worth noting that deforestation is a global concern, with approximately 16 million hectares of forest land being cleared annually, and this trend shows no signs of decline (Chala et al., 2022). Furthermore, the Millennium Ecosystem Assessment Panel has observed African grassland, woodland, and other vegetated areas are being converted into cropland in recent times (Tessema et al., 2020). These changes in land use and land cover have negative implications for the environment and natural ecosystems, leading to the emergence of momentous environmental hazards (Mariye et al., 2022). Understanding the spatiotemporal dynamics of LULC change and its drivers is instrumental in synthesizing knowledge for informed natural resource management, planning, and associated decisions (Gashaw et al., 2014). The natural processes, socioeconomic aspects, and land exploitation by human activities in time and space determine the land cover pattern (Mariye et al., 2022). Driving forces which are similarly referred to as factors can be categorized as natural and human-induced (Yericho & Mulugeta, 2019). The major drivers of LULC changes in Ethiopia emanated mainly from population growth, which is manifested mostly through the expansion of cultivated lands, even in areas where cultivation is almost impossible, and urban expansions (Gashaw et al., 2019).

In Ethiopia, the Awash River Basin (ARB) is highly developed and intensively utilized and provides water to large and small irrigation schemes, industries, hydropower plants, and household use. However, this important basin is vulnerable to human-induced factors such as land degradation and high population intensity due to its strategic location, access to roads, available land and water resources. Earlier findings showed that cropland increased by 12% between 1988 and 2002, and by 2018 it had increased by 15%. Similarly, the built-up area expanded by 52 km² (184%) between 1988 and 2002, and by 2018 it had reached 225%. The analysis showed that the cropland and built-up area expanded at the expense of forest and shrubland, with shrubland and forest reducing by 4% and 25%, respectively over the 30 year (Tadese et al., 2020).

It is widely accepted that LULC change is a result of the complex interactions between driving factors (Akdeniz et al., 2022). Population increase, intensive and extensive agricultural practices, urbanization, industrialization, and economic development are among the forces that cause changes in LULC which lead to severe environmental problems such as floods, drought, and landslides (Chala et al., 2022).

In the past, recognition of the importance of the natural environment for human well-being has been less influential in maintaining sustainable development and poverty alleviation strategies (Gashaw et al., 2019). In recent years, the study of LULC change has become an important topic of research. Studies have shown that LULC change has been intense in the highlands of Ethiopia. In particular, the expansion of intensive agriculture, urbanization, and extraction of forest products are accelerating over time to meet the requirements of an increasing population (Assefa et al., 2021). However, some studies have reported different trends of LULC changes. For example, Akdeniz et al. (2022) reported that agricultural land declined while the grassland increased in the Somodo watershed, in southwestern Ethiopia. Gashaw et al. (2018) also reported the expansion of grassland and shrublands in the Gelan sub-watershed, northern highlands.

The Borkena Watershed has a large share of the Awash River basin supply. However, there has been limited multidisciplinary and independent research in the catchment compared to other areas. Leta et al. (2021) studied how land use and management practices affect surface runoff and sediment yield based on hypothetical scenarios without considering the real spatio-temporal LULC change. Alemu et al., (2023) assessed the state of the climate change impacts in the Borkena watershed by considering hydrology. However, there is no information on the driving factors, spatial-temporal LULC change, and the implications on the entire Borkena watershed.

This research aims to investigate the LULC dynamics, its cause, and its extent over the period from 1993–2003, 2003–2013, and 2013–2023. In addition, it aims to assess the driving factors and causes of LULC change through the integration of remotely sensing techniques with socioeconomic information. Several researches in the same way link census or survey-based socio-economic data to remotely sensed land use data (Tessema et al., 2020). The study will provide quantitative information for planners and decision-makers to design sustainable land management and resource use planning.

Material and methods

Description of the study area

The Borkena River starts in Kutaber Woreda, where it marks the boundary between two river basins, Abay and Awash. It flows from the chain of mountains and escarpments, in the plateau, to the Afar Rift. The river then heads southeast. Eventually meets the Jara River near the Cheffa swamp before joining the Awash River. On the other hand, the Berberie River originates from northeast of Kombolcha town. After winding its way through the town it merges with the Borkena River in Kombolcha town (Makwinja et al., 2021). Geographically, the study area is bounded by 10°40′0″–11°20′0 ″N latitude and 39°27′0″–39°58′30″ E longitudes and the total area of the sub-basin was estimated to be 126,243 km² (Figure 1). The total distance of the flow path towards the outlet of Borkena River is 85 km. The elevation of Borkena varies between 1412 and 3463 m.a.s.l (Figure 2).The climate of the study area varies between Dega and Weyna Dega. The mean annual rainfall of the catchments was 1028 mm. The mean monthly temperature of the area varies between 16.1 and 22.1 °C which corresponds to December and June, respectively (Kifle, 2022).

Figure 1. Location of the study area.

Figure 2. Digital elevation model map of the study area.

Data collection and image processing

The images with 30 m resolution and cloud-free were collected from the US Geological Survey (USGS) Earth Explorer (https://earthexplorer.usgs.gov). For 1993 and 2003 the Landsat 5 (TM) data while for 2013 and 2023, the Landsat 8 (OLI/TIRS) data were downloaded for the same season of each study year, i.e. March or April months. The spatial resolution of Landsat TM data was 30 m and OLI/TIRS data was 15 m (after incorporating the panchromatic band). The details of the satellite data used in this study are given in Table 1. Ground validations were done using samples collected through Google Earth Pro (for the period from 1993 to 2023) and a field survey for (2023).

Table 1. Details of the satellite data sources

No	Satellite	Sensor	Spectral Bands	Month/year of Acquisition	Av. Cloud cover
1	Landsat 5	TM	7	Mar-Apr 1993	3%
2	Landsat 5	TM	7	Mar-Apr 2003	5%
3	Landsat 7	OLI/TIRS	7	Mar-Apr 2013	7%
4	Landsat 8	OLI/TIRS	7	Mar-Apr 2023	7%

To accomplish classification of image in multi-temporal approach and for mapping purposes, ArcGIS 10.8 software were used. For each LULC, as many as possible training samples were selected throughout the entire image, based on the composite images, as well as Google Earth images. Initially images were converted into Universal Transfer Mercator and geo-referenced to WGS-84 datum. For classification, verification, and validation of the classified images, the training data were used. A bisect field study was conducted between May 2023 and July 2023 utilizing draft-classified maps derived from satellite images with reference years, topographic maps, and Google Earth as guides. Besides, a supervised digital image classification technique was employed, complemented by field surveys that provided ground information regarding the study site. The supervised classification method for land use and land cover dynamics analyses with spatiotemporal changes were monitored by analyzing multi-temporal remotely sensed Landsat images. A combined procedure was developed to analyze, map, quantify and interpret the collected data (Table 2). The LULC types in the study sites were grouped into six that were forest, cultivated land, shrubland, waterbody, builtup, and bare land (Table 3).

Table 2. Data type and sources

No	Data types	Sources	Spatial resolution (m)
1	Landsat image (Landsat 5TM (1993), Landsat 7 ETM +(2002) & Landsat 8 OLI/TIRS (2019)	USGS	30
2	Digital Elevation Model	ASTER	30
3	GPS Data	Field survey
4	Socio-economic data	HH survey

Table 3. Description of LULC classes identified in Borkena Watershed

No	LULC type	Description
1	Forest	Areas covered by trees forming closed or nearly closed canopies; forest; plantation forest
2	Cultivated Land	Area covered with perennial and annual crops and crops uncovered agricultural land
3	Shrubland	Land covered by small trees, bushes, and shrubs, in some cases mixed with grasses; less dense than forests
4	Waterbody	The area is covered by a small dammed lake fed by rainfall small streams and a sandy river path
5	Built-Up	dispersed rural settlements, Urban and asphalts, and roads
6	Bare Land	The land surface which is mainly covered by bare soil and exposed rocks over gentle and steep mountain slopes

Socio-economic data

Between May and July, 2023, a socio-economic survey was conducted. While the methods for socio-economic surveys vary depending on the goal of the research, in this study the key informant interview (KII) and focus group discussion (FGD) were conducted. For the FGD and KII, four sub-watersheds were chosen based on the agro-ecological settings. Awash Kombolcha Sub-Basin has 12 woreda and 91 kebeles including Kombolcha, Kemisie, Albuko, Dessie, Chaffie golana dawe, Kutaber, Ancharo, GisheRabel, Tehulederie Anstokian Gemeza, Artuma, and Harbu/Kalu (Bokke & Bayu, 2022). Household surveys were carried out in four places namely Kutaber (Upper watershed), Kombollcha, Harbu and Kemisse (lower watershed). The selected places represent the higher, middle, and lower watershed parts.

In total, 180 farm households, 45 from each selected place, were interviewed using a structured survey questionnaire on a range of issues about land use, climate changes, farming, soil fertility, and policy changes related to land and markets. Discussions were also held with groups of 134 male and 46 female farmers of various ages. In addition, farmers aged over 50 years were interviewed to obtain historical information on land use characteristics, socio-economics and policy changes, and LULC change. Open-ended questions were used concerning the major shifts in LULC. All the statistical analyses were performed using SPSS statistical software, version 25.0.

During the discussion and interviews, the main focus was to get sufficient information on the trends of LULCC and identify the underlying driving factors of the changes and implications of LULC change on the socio-economic activity of the community and the environment. To better understand the major observed problems of the catchment and resource management practice, transect walks, field walks, and informal talks with people in their farms/fields were used. Farmers were asked to explain what parts of the landscape were changed and explained why the change had occurred. The farmers were also asked to describe the consequences of the changes in their livelihood, surroundings, and environment. Furthermore, farmers were asked to explain how their socio-economic activity contributes to the land-use change. The sample size was determined following Yamane by Mulugeta (Dega et al., 2022), and respondents were selected using a simple random sampling technique

where n =is the sample size, N is the study population, and e = is the level of precision. By using the formula, the sample size was determined:

$\mathit{n}=\frac{N}{\left(1+N\left(e^2\right)\right)}$

$\mathit{n}=\frac{1,1850,150\mathit{\hspace{0.17em}}}{\left(1+1,1850,150\mathit{\hspace{0.17em}}\left({0.075}^2\right)\right)}$

Where n = 180

Classification accuracy assessment

The accuracy assessment reflects the real difference between classification and the reference map or data (Akdeniz et al., 2022). If the reference data is highly inaccurate, the assessment might indicate that classification results are poor. Therefore, classification accuracy assessment is an important step in evaluating image classification. Random sampling was applied to select the classified pixels. About 180, 150, 160, and 100 test pixels from the classified image of 1993, 2003, 2013, and 2023 were selected to assess image classification results. The classified image were verified using the visual interpretation of the image and the reference data gathered from the digital class cover map. Researchers most frequently utilize an error matrix to evaluate the categorization accuracy of remotely sensed data (Akdeniz et al., 2022) which sometimes referred to as confusion matrix or contingency table. An error matrix is a square array of numbers set out in rows and columns that represent the number of sample units (i.e. pixels, clusters of pixels, or polygons) assigned to a particular category relative to the actual category as indicated by reference data (Bekele et al., 2019). In this study, standard criteria of accuracy assessment of the classification such as producer’s accuracy, user’s accuracy, overall accuracy, and the kappa coefficient were computed from the error matrix (Akdeniz et al., 2022).

Results and discussion

Spatial analysis of the 1993, 2003, 2013, and 2023 classified image maps (Figure 3) showed that various major changes had occurred in the watershed (Table 4). In each period, the land category with the largest proportion of land use and cover was cultivated land. The built-up areas was land use and cover category which was increased rapidly between 1993 and 2023. Areas of shrubland, forest, and water decreased by 150, 101, and 10 percent, respectively, whereas intensive cultivation, built-up, and bare land increased by 12, 77, and 7 percent, respectively (Table 4). LULC varied significantly during the study period. The tendency is diminishing for both shrubland and forests. In contrast, the remaining land uses which include cultivated land, waterbodies, and built-up areas showed an increase in area coverage. In 1993, the highest extent of the study area was covered with cultivated land 72,666 ha (58%), while dense forest, shrubland, waterbody, built-up areas, and bare land were 8,336 ha (7%), 29214 ha (23%), 1,950 ha (2%), 2,533 ha (2%), and 11,544 ha (9%), respectively. By the year 2003, the area coverage of cultivated land and built up increased to 77,716 ha (61%), and 3,534 ha (3%), respectively. The forest land was mainly located on the steeper slopes of the mountains which is unsuitable for crop cultivation. The extent of the forest, however, continuously shrunk from 8,336 ha (7%) in 1993 to 6,336 ha (5%) in 2013 and slight improvement 7,610 ha (6%) by the year 2023. According to the LULCC analysis, during the last 30 years, cultivated land has significantly expanded at the expense of forest land and other land uses. (Table 4).

Figure 3. Land cover changes detected by post-classification methods.

Table 4. LULC types of Borkena Watershed (1993–2023)

Class name	1993		2003		2013		2023
	Ha	%	Ha	%	Ha	%	Ha	%
Forest	8,336	7%	6,336	5%	6,566	5%	7,610	6%
Crop Land	72,666	57%	77,716	61%	80,142	63%	82,486	65%
Shrub Land	29,214	23%	27,214	21%	11,869	9%	11,709	9%
Water body	1,950	2%	1,899	2%	698	1%	932	1%
Built Up	2,533	2%	3,534	3%	8,319	7%	11,106	9%
Bare Land	11,544	9%	9,544	8%	18,649	15%	12,364	10%
Total	126,243	100%	126,243	100%	126,243	100%	126,243	100%

Accuracy assessment

In this study, a random sampling method was applied to evaluate the accuracy of the classified images. About 215, 211, 500, and 100 sample pixels from 1993, 2003, 2013, 2023 were selected, respectively. The results show that overall accuracies for 1993, 2003, 2013, and 2023 classified images were 87.40%, 93.11%, 97.11%, and 87.01% with Kappa coefficients of 0.79, 0.89, 0.95, and 0.77, respectively (Tables 6, 7 , 8 , and 9). The Kappa coefficients demonstrate that the classified images of 1993 and 2023 had moderate classification performance, while the classified images of 2003 and 2013 had good classification performance.

The results of the confusion matrix generated to evaluate the accuracy of the classification are presented in Tables 6, 7 , 8 and 9. The overall accuracy was 87%, and the kappa statistic was 0.77. This shows that 87% of the LULC classes were properly classified. The user accuracy of individual classes ranges from 100% for the water body to 91% for the shrubland. Producer accuracy also ranged from 95% for cultivated land to 75% for bare land (Table 10). The overall accuracies obtained from the images were greater than the 85% minimum threshold set by (Sikorski et al., 2021) for effective LULC change analysis.

Table 6. Confusion matrix of classification accuracies for Borkena River watershed 2023

Class Value	FL	CL	SL	WB	BU	BL	Total	User accuracy (%)
Forest	5	0	0	0	0	0	5	1
Crop Land	0	58	7	0	0	2	67	0.87
shrubland	0	1	10	0	0	0	11	0.91
Waterbody	0	0	0	3	0	0	3	1
Built-Up	0	1	1	0	5	0	7	0.71
Bare Land	0	1	0	0	0	6	7	0.86
Total	5	61	18	3	5	8	100	Overall Accuracy 87%
Producer accuracy (%)	1	0.95	0.56	1	1	0.75	0	Kappa Coefficient 0.77
Kappa

Table 7. Error matrix of the classified image 2013

Class Value	FL	CL	SL	WB	BU	BL	Total	User accuracy (%)
Forest	28	0	2	0	0	0	30	0.93
Crop Land	0	322	0	0	0	2	324	0.99
shrubland	3	2	43	0	0	0	48	0.90
Waterbody	0	0	0	2	0	0	2	1
Built-Up	0	0	0	0	31	0	31	1
Bare Land	0	3	1	0	0	61	65	0.94
Total	31	327	46	2	31	63	500	0
Producer accuracy (%)	0.90	0.98	0.93	1	1	0.97	0	Overall Accuracy 97%
Kappa	0	0	0	0	0	0	0	Kappa Coefficient 0.95

Table 8. Error matrix of the classified image 2003

Class Value	FL	CL	SL	WB	BU	BL	Total	User accuracy (%)
Forest	10	0	0	0	0	0	10	1
Crop Land	1	111	6	2	0	1	121	0.92
shrubland	1	4	39	0	0	0	44	0.89
Waterbody	0	0	0	10	0	0	10	1
Built-Up	0	0	0	0	10	0	10	1
Bare Land	0	0	0	0	0	16	16	1
Total	12	115	45	12	10	17	211	0
Producer accuracy (%)	0.83	0.97	0.87	0.83	1	0.94	0	Overall Accuracy 0.93
Kappa	0	0	0	0	0	0	0	Kappa Coefficient 0.89

Table 9. Error matrix of the classified image 1993

Class Value	FL	CL	SL	WB	BU	BL	Total	User accuracy (%)
Forest	10	0	4	0	0	0	14	71%
Crop Land	0	117	0	1	0	12	130	0.9
shrubland	6	0	32	0	0	0	38	84%
Waterbody	0	0	0	3	0	0	3	1
Built-Up	0	0	0	0	15	0	15	1
Bare Land	0	5	0	0	0	10	15	67%
Total	16	122	36	4	15	22	215	0
Producer accuracy (%)	0.63	0.96	0.89	0.75	1	0.45	0	Overall Accuracy 87%
Kappa	0	0	0	0	0	0	0	Kappa Coefficient 0.79

Table 10. Summary of accuracy assessment from 1993–2023 (%)

Class Value	1993		2003		2013		2023
	Producer’s Accuracy	User’s Accuracy	Producer’s Accuracy	User’s Accuracy	Producer’s Accuracy	User’s Accuracy	Producer’s Accuracy	User’s Accuracy
Forest	63	71	83	1	90	93	1	1
Crop Land	96	90	97	92	98	99	95	87
shrubland	89	84	87	89	93	90	56	91
Waterbody	75	100	83	1	1	1	1	1
Built-Up	100	100	1	1	1	1	1	71
Bare Land	45	67	94	1	97	0.94	75	86
Overall accuracy	87		93		97		87
Kappa coefficient	79		89		95		77

Land use land cover dynamics

The major LULC types shown on the maps of 1993, 2003, 2013, and 2023 include forest, cultivated land, built-up, bare land, shrubland, and waterbody (Table 4). Cultivated land and built-up areas had substantially increased in the watershed. In 1993, about 72,666 ha (58%) of the watershed area was cultivated land, and by 2003, this had increased to 77,716 ha (61%), followed by a rise to 82,486 ha (63.5%) in 2023 (Table 4). This resulted in the conversion of 9,820 ha (12%) of land from 1993 to 2023 (Table 5). The major land use transitions to cultivated land between 1993 and 2023 were mostly from bare land 8,366 ha (29.5%), followed by shrubland 3,409 ha (2.8%), and forest area 584 ha (1.9%) (Table 5).

Table 5. LULC types of Borkena Watershed (variation) expressed as area (ha) and percentage of total

Class name	Change (1993–2003)		Change (2003–2013)		Change (2013–2023)		Change (1993–2023)
	Ha	%	Ha	%	Ha	%	Ha	%
Forest	−2000	−2%	230	0%	1044	1%	−726	−10%
Crop Land	5050	4%	2426	2%	2344	2%	9820	12%
Shrub Land	−2000	−2%	−15345	−12%	−160	0%	−17505	−150%
Water body	−51	0%	−1201	−1%	270	0%	−982	−101%
Built Up	1001	1%	4785	4%	2787	2%	8573	77%
Bare Land	−2000	−2%	9105	7%	−6285	−5%	820	7%

The shrubland, which includes the grass and bushland, progressively declined (Table 5). In 1993, 29214 ha (23%) of the total land area was under shrubland; this had shrunk to 11,869 ha (9%) in 2023. This indicates 17,505 ha (12%) of land converted at an average rate of 583 ha per year from 1993 to 2023. The major land use transitions between 1993 and 2023 were from shrubland to cultivated land (3409 ha), built-up land (523 ha, or 23.1%), and bare land (662 ha, or 1.6%) (Table 4). According to Mariye et al. (2022), the expansion of cultivable land was mostly caused by population growth and farmers’ inability to pay for modern agricultural inputs. The fast-growing population pressure and diminishing soil productivity were also mentioned by Abebe (2020) as contributing causes to the cultivated land expansion.

According to Bekele et al. (2019), the expansion of cultivated land and the collection of firewood could have contributed to the decline of shrubland. Growing population pressure is another issue that may be contributing to the loss of natural vegetation, according to (Esa et al., 2018). Mostly at the cost of shrubland, bare land expanded into the steeper portion of the watershed. About 11,544 ha (9%) of the entire watershed area was bare land in 1993. This percentage rose to 18,649 ha (15%) in 2013 and fell to 12,364 ha (10%) in 2011 (Table 4). From 1993 to 2023, there was an average annual increase of 413 ha. Next to cultivated land and settlement, the bare land exhibited the second fastest change. Between 1993 and 2023, cultivated land accounted for 3,336 ha (4%) of the major land use transitions to bare land, followed by shrubland on 1,097 ha (Table 5). Expanding and intensifying cultivation in marginal areas exacerbates the process of land degradation and increases the amount of bare land (Esa et al., 2018; Zeleke & Hurni, 2001).

Forest cover slightly increased in the watershed area during the study period. From 1993 to 2003, forest coverage showed a decrease in the watershed. However, 6566 ha (5%) of forestland appeared in 2013, followed by an increase to 7610 ha (6%) in 2023 (Table 4). The major land use transitions to the forest between 1993 and 2023 were mostly from cultivating land (2,117 ha) and shrubland (1,137 ha). The conversion of cultivated land to forest is due to the replantation of trees on the degraded land. Replanting trees on previously used agricultural land is necessary since the land has degraded.

The built-up area increased during the study period. In 1993, the area was 2,533 ha (2%) and 1,1106 ha (9%) in 2023. The total increment showed 8,573 ha (77%) from 1993 to 2023, with an annual average increase of 286 ha.

According to Shiferaw and Singh (2011), an important aspect of change detection is to determine what is actually changing to what category of LULC type. Knowing the change ‘from to’ process information for each category over a specific period is crucial to the LULC change detection study. There is a requirement for information regarding initial and final land cover, types, and uses, the ‘“from to”’ analysis (Hassan et al., 2016). The actual changes in the area, including their size, location, and type, could be better understood with the help of this information. To enhance comprehension of the variations in ‘from-to’ procedures from 1993 to 2023, comprehensive data was provided and shown in (Figure 3).

Detected changes by post-classification

A post-classification examination of land cover change showed that thirty different types of land conversion occurred throughout the study period (1993–2023). According to Table 9 change detection, during the past 30 years, the amount of forest cover has dropped by 5672.8 ha, 613.3 ha, and 283.9 ha, respectively, due to land expansions for cultivation, bare land, and built-up areas. The high rate of forest conversion to agricultural land was a sign that the majority of the local population depended on agriculture for their livelihood.

The study area experienced deterioration and population growth, resulting in the conversion of about 17,035 ha of cultivable land in 1993 into bare and built-up areas in 2023 (Figure 4). This occurred in the lowest section of the watershed. About 8859 hectares of shrubland were converted from shrubland to cultivated land between 1993 and 2023 in the upper watershed due to population expansion and the need for agriculture; in the lower watershed, this occurred because pastoralists needed to feed their livestock. The enhancement of vegetation cover through plantation on the shrubland areas converted to forest cover of 1138.5 ha from 1993 to 2023, according to the Ethiopian government’s afforestation projects from the Million Development Goal. About 5,039 ha and 242 ha of barren and degraded land in the watershed region have been restored and turned into agriculture and shrubland, respectively. This is a result of the increased need for animal feeding and agriculture.

Figure 4. Change detection map of Borkena Watershed.

Drivers and impacts of land use and land cover change

According to the data obtained from the Ethiopia Central static agency, the current population of the Woreda in the study area is estimated at 1,734,366 of which 544,111 are urban and the remaining 1,190,255 are rural, which is 54.3% of the total population in this area (Abebe, 2020). The population was estimated to be 1,1850,150 in 2023 (Board, 2007). Resettlement, immigration, and natural population increases are the causes of population growth, according to KIIs and FGDs. Field observation, KII, and FGD data demonstrated that both natural and human factors contributed to the LULC change. However, anthropogenic activities were discovered to be a more direct cause of LULC change than the natural processes. Respondents identified seven human-related activities as the primary causes of LULCC in the study area out of a variety of other drivers (Table 11). The ranks were determined by calculating the frequency with which the respondents chose the factors.

Table 11. Characteristics of respondents

Attribute	Frequency	Percentage
Age
45–50	70	39%
51–70	71	39%
>70	39	22%
Total	180	100%
Educational Status
Illiterate	98	54%
1–8	42	23%
9–12	25	14%
Above Grade 12	15	8%
Total	180	100%
Sex
Male	134	74%
Female	46	26%
Marital Status
Single	15	8%
Married	138	77%
Divorced	12	7%
Widowed	15	8%
Total	180	100%

Table 12. Drivers of LULC change as perceived by households (%) (n = 180)

Perception about LULC Drivers
Drivers of LULC Changes	Agree		Undecided		Disagree
	no	percent	no	percent	no	percent
Population Growth	165	91.7%	5	2.8%	10	5.6%
Expansion of Farmlands	171	95.0%	0	0	9	5.0%
Fuelwood Collection	169	94%	0	0	11	6%
Settlement	158	88%	6	3%	16	9%
Topography	111	62%	21	12%	48	26%
Drought	94	52%	36	20%	50	28%
Flood	89	49%	44	24%	47	27%

According to the respondents, there have been notable changes in LULC during the past thirty years. For example, almost 90% of respondents said they had seen a decrease in water bodies, dense forests, shrublands, while cultivated land and settlements were increased. Accordingly, the most important and highly recognized drivers of LULCC found in the Borkena watershed (Table 12) were population growth, expansion of crop land, settlement, fuelwood collection, topography, flooding, and drought.

Expansion of crop land

Population increase has occasionally resulted in a decline in the number of rural farmland holdings and sustained farming, according to data from households (95%), focus groups, and key informants. The growing demand for more cultivable lands in turn spurred the development of cultivating land, which is another factor fueling LULC change (Table 12).

Fuelwood collection

Next to the cultivated land, 94% of respondents named fuelwood exploitation as the primary proximal source of LULC change (Table 12). Focus groups discussions and key informants linked the depletion of acacia woodland forests in the lower watershed, which were once the predominant land cover class, to the rural area’s lack of electricity as well as the production charcoal. Potential causes of LULC change in the Awash basin include fuelwood exploitation, charcoal production, and rural electrification. These factors are often combined with limited access to alternate energy sources. Acacia trees are the most popular trees for producing charcoal in Ethiopia because of their high calorific content (Aredehey et al., 2020). Furthermore, rural farmers use the market for firewood and charcoal as a substitute source of income owing to the ongoing drought, which has resulted in poor income and issues with food insecurity in recent years. This was seen as an adaptation strategy to get through such a difficult era (Mlotha, 2018).

Population growth

A high population increase was mentioned by almost 92% of the farmers surveyed as the primary underlying cause of LULCC (Table 12). According to Kegunaan et al. (2019), Ethiopia’s population has grown by more than 235% since 1986, with an average annual growth rate of 3% (Ethiopia CSA, 2007). This suggests that during the past three decades, the population of the study area has expanded by a factor of more than three. The population expanded by around 4.5 times its value from 1965 to 2008 (Meshesha et al., 2012). In addition, information gathered from district administrative officials indicates that, as a result, there is currently a significant level of rural-urban migration in the area.

Infrastructure and settlement expansion

Since the 1993, the growth of infrastructure including urban and rural settlements and the road network, such as road from Desse to Addis Ababa has increased by 77% at the expense of other LULC units in the Borkena watershed, according to information from change detection analysis using remotely sensed data (Table 1). A little over 88% of the respondents said that the LULC shift in the Borkena watershed was caused by the growth of built-up areas. Previous research Elias et al. (2019) showed that improved market and road infrastructure availability were the primary causes of LULC modifications. In this regard, 85% of respondents asserted that improved market prospects for pole, fuel, and charcoal, as well as various agricultural outputs and road accessibility, are also influencing factors for the shift in land cover. The industrial area has increased due to industrialization, which creates environmental pollution. In Kombolcha town, the river is highly polluted because of industrialization.

Drought, flood, and topography

Recurrent drought and flood are the other factors perceived by about 52% and 49% of respondents, respectively, as immediate causes of LULC change (Table 12). Key informants and focus group discussions also mentioned that they observed an increase in diurnal temperature and a more erratic rainfall distribution (late onset, early set-off, frequent dry spells, and unusual floods due to increased intensity in a few periods) in the past ten years. Flooding is another significant force behind LULC change, particularly in the lower section of the watershed (Harbu and Kemisse), where it contributes to the drying out of swampy areas and water bodies and the degradation of the land. However, for a greater portion of the watershed, topography is blamed for the degradation. Kutaber district respondents claimed that hillside farming results in lower crop yield and soil erosion.

The results have shown that extensive LULC changes occurred for the last 30 years (1993–2023) in Borkena watershed. The finding of this research indicated that the area that was covered with cultivated and settlement in 1993 has increased in 2023. Whereas, the forest and shrub land had decreased (Table 5). Accordingly, it is evident that forest cover and shrub land are most at risk of undergoing LULC changes. Comparable LULC changes dynamics were reported by previous studies in different parts of Ethiopia. For example, Zeleke and Hurni (2001) evaluated the decline of natural forest cover due to agricultural land expansion in the northwest Ethiopian highlands between 1957 and 1995. Gashaw et al. (2014) also revealed that the highest gain of crop land was obtained from grazing and shrubland from 1986 to 2017 in the upper Blue Nile.

The study area is now more populated, which has an adverse impact on the ecosystem. In reference to this pattern, interviews with key informants and FDG participants revealed that the primary factor contributing to the deterioration of natural resources in the Borkena Watershed is population growth and expansion of farm land. As a result, the parents’ farmland is constantly divided among their children, resulting in continued land fragmentation despite the land’s significantly decreasing size. This has led to a shortage of natural resources and promoted the area’s natural resource degradation. The livelihood of many poor people relies on the sale of firewood, charcoal and dung cake. According to the KIIs and FGDs participants, in recent years, firewood and charcoal have become the most commercialized energy sources for both the rural and urban poor in the Harbu and Kemise Districts. This finding is similar to the finding of Kuma et al. (2022), recently conducted in Southern Ethiopia. Generally, the main causes of LULC changes in the study area include population pressure, cultivate land and built-up expansions, increasing wood demand for fuel, collection of farm implement and construction wood, charcoal production and livestock grazing.

Conclusions and recommendations

This study showed significant LULC changes in the area between 1993 and 2023. We investigated the long-term dynamics of LULC change, driving factors, and magnitude of change in the Borkena River watershed by integrating remote sensing with sociological studies. The most prominent changes were a big rise in population growth, cultivating land, settlement, and construction, a sizable decrease in forest and shrubland, and a modest increase in plantations.

Lower elevations, gentle slopes, less populated areas, locations near markets or towns, and locations farther from roads around Harbu and Kemisse increase the likelihood of shrubland-to-cultivated land. Farmers are inclined to convert shrubland rather than natural forest for cultivation. Therefore, the widely held view that the expansion of agriculture is the primary cause of the loss of natural woody vegetation in Ethiopia was found to hold true in the case of the Borkena watershed. Forest’s location near rivers and on wet soil (i.e. riverine forests) seem to have a higher susceptibility to deforestation than the other forest types. Roadside woods are more likely to be cleared for development since they yield higher returns due to reduced transportation expenses. Thus, the loss of forest could be largely due to increased demand for fuelwood, construction, farm implements, and other uses.

Generally, floods, population pressure, unemployment, rainfall variability, declining soil fertility, a lack of fuel, and a lack of arable land were the main drivers and causes of LULC change in the watershed. Therefore, more initiatives beyond the current sustainable land management campaign are required to give young people new career options in industries other than agriculture. In addition, better community-based land resource management policy and implementation will be required to ensure sustainable rural livelihoods by focusing on bare land restoration, shrub and bush land conservation, making grazing lands available through restoration of degraded land and regulating further expansion of areas under cultivation.

Acknowledgements

I would like to thank the editors and anonymous reviewers for your comments to improve the quality of the paper. I express my appreciation to Dr. Sileshi Degefa for his assistance without him this research might not happened.

The authors would like to thank the Ethiopian National Meteorological Agency and Ethiopian Ministry of Water Resources, Irrigation and Electricity for providing us with the meteorological and daily flow data respectively.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are available upon a requirement and reasonable request.

Additional information

Notes on contributors

Engdaye Mersha

Engdaye Mersha is a researcher at Center for Environmental Science, Addis Ababa University, Ethiopia. His research lies in the field of sustainable natural resource, remote sensing and GIS, ecosystem and habitat quality and soil and water conservation.

Sileshi Degefa

Sileshi Degefa (PhD) is an associate professor at Center for Environmental Science, Addis Ababa University having experience in project evaluation and Land Use and Land Cover (LULC) dynamics, as a researcher, guiding postgraduate students and evaluating government interventions in LULC impacts

Wondimagegn Mengist

Mekuria Argaw (Professor) is a professor at Center for Environmental Science, Addis Ababa University having a lot of experience in climate change, Land Use and Land Cover (LULC) dynamics and guiding postgraduate students and evaluating government interventions in LULC impacts