Land Use and Land Cover Change for Resilient Environment and Sustainable Development in the Ethiopian Rift Valley Region

Different techniques were used to detect change and algorithms on LULC via remotely sensed data [Lu et al., 2004]. The geographic information system and remote sensing produce essential sources to obtain data used to compare the historical and current status of LULCC. Also, it is fairly accurate in representing the trends of land cover dynamics than other indicators [Lambin et al., 2006]. In order to attain the intended research objectives of the study, four time period satellite image data were obtained from Landsat 5 (™) and Landsat 8 (OLI) from 168/056 path and row (Table 1). The four time periods for the Landsat images are 1986, 2000, 2011, and 2019, downloaded from (https://glovis.usgs.gov/).

The criteria to select the year for Landsat images were based on historical events, which significantly influenced the LULC change. For instance, the 1986 image was purposefully selected as a baseline because of the quality of the Landsat image and because it represents the effect of formal watershed-based natural resource conservation and the development programme of Ethiopia in the 1970s and 1980s. Though the Federal government came to power in 1991, the new regime has spent time establishing itself in the country after a long period of civil war, formulation, or reconstitution of offices and changing administrative boundaries, which slowed down the work of an environmental conservation project and resulted in delays. Therefore, the year 2000 image was used to reflect the effect of the extensive watershed-based conservation development effort made by the government and non-governmental bodies of the military regime from 1974 to 1991.

The 2011 image was used to represent the effect of the federal environmental policy that was formally approved on 2 April 1997 and the community-based participatory watershed development program of 2005. The fourth image for 2019 was used to assess the effect of the second community-based watershed managed practices of 2010–2011. This period was known for the 30 days of free labour and massive social movements on watershed management. Also, the 2019 image presented a cut point for the overall effect of the federal government conservation movement from 1991 to 2019.

The Landsat 5 Thematic Mapper (TM) sensor was overlaid on Landsat 4 and Landsat 5 to create images consisting of six spectral bands with a spatial resolution of 30 meters for Bands 1–5 and 7 one thermal band (Band 6). While the Landsat 8 satellite image consists of two science instruments, the Operational Land Imager (OLI) and the Thermal Infrared Sensor (TIRS), these two sensors provide seasonal coverage of the global landmass at a spatial resolution of 30 meters (visible, NIR, SWIR); 100 meters (thermal); and 15 meters (panchromatic) (Table 1).

The satellite data selection has been fixed to the dry season, which has a clear sky for the specific years with no cloud cover. Selections of the lowest cloud cover period minimise the cloud effect and associated reflectance [Jensen, 2009]. The downloaded images were stacked with all the band combinations in the dataset. Since the Landsat images were geometrically corrected, their integration with other datasets was checked by overlaying them with the GPS data collected from the field and Google earth imagery. These data are essential in order to identify the different land use and land cover change dynamics.

Using ENVI software, the radiometric correction was carried out using the dark-object subtraction method. This method combines the sun and view angle effects and the sensor calibration with the atmospheric correction. Then image classification was carried out using both supervised and unsupervised classification. The unsupervised classification was done only to get the spectral characteristics of features on the image, while the supervised classification method was widely implemented using the decision rule of maximum likelihood classification algorithm on the ERDAS imagines 2014 software.

The maximum likelihood algorithm provided a consistent approach to parameter estimation and was developed from many estimations using training samples. Then all satellite data were studied by assigning per-pixel signatures and distinguishing the land area into seven classes based on the specific Digital Number (DN) value of diverse landscape elements [Im and Jensen, 2005]. The major delineating classes of the study were cropland, forest land, settlement, grazing land, agroforestry, bush/shrubland, and bare land (Table 3). For each of the predetermined classes of the study site, we used a minimum of 10 training samples for settlement and 15 training samples for bare land. Then, 30–75 training samples were randomly selected for cropland, forest land, grazing land agroforestry, and bush/shrub land.

A satisfactory spectral signature ensures that there is minimal confusion among the land covers to be mapped [Gao and Liu, 2010]. For this research, pixels enclosed by these polygons have been used to record the spectral signatures for the respective land cover types derived from the satellite imagery. As stated by Foody, [2002] there are better results with large samples representing each land cover classes. However, the required number of training samples to train classifiers depends on the classifier going to be used, the number of features considered, and the landscape structure of the classes considered. Since the landscape structure of the present study area consists of little area for settlement and bare land considered in other classes, a few points were randomly selected during an assessment of class labeling, which might be equal to the weighed proportion of the landscape.

Accordingly, through the use of ground-truthing sample polygons and online Google earth imagery data visualization and interpretation, the spectral signatures for the respective land cover types were collected. Then each class was given a unique identity and assigned a particular class testing the spectral distinctness of the training area. Moreover, in order to enhance signature repairability between the Forest and Agroforestry classes and minimise uncertainties of classification, more GPS-based ground-truthing data of natural forest were georeferenced, coupled with analysis of these feature classes using the human brain (texture, tone, size, shadows, and shape) from images.

The accuracy assessment of the images for this research was carried out using ArcGIS 10.4 software, which identifies the truth on the ground and is represented on the corresponding classified image. A simple random sampling method was used to collect a total of 280 reference data points to ensure that all seven LULCC classes were adequately represented depending on the proportional area of each class.

The confusion matrix of LULCC was estimated from Landsat 8 of 2019 and Landsat 5 of 2011, corresponding to the reference data from Google earth imagery (GEI) and GPS collected ground-truthing data. However, the accuracy assessment for the years 1986 and 2000 was not conducted because of the limitation of obtaining ground-truthing data, from aerial photo and historical Google Earth imagery. The accuracy assessment was determined using the Kappa coefficient, overall accuracy, producers’ and users’ accuracies derived from the confusion (error) matrix adopting the methodology described in Congalton and Green [2009] and Jensen and Cowen [1999].

According to Table 2, the diagonal matrix indicates sample point classes, which are correctly classified, while off-diagonal elements represent commission or omission errors [Foody, 2002]. The error matrix was computed for the overall accuracy of seven (7) land use classes individually and collectively. The overall accuracy was calculated by summing the number of correctly classified values and dividing by the total number of values. The total number of values is the number of values in either the truth or predicted-value arrays. Therefore, the overall accuracy result is 87%, which suggests a strong agreement between classification and truth. The kappa coefficient measures the agreement between classification and truth values. A kappa value of 1 represents perfect agreement, while a value of 0 represents no agreement. The kappa coefficient value of the study is 0.85; this value is closer to 1, which represents perfect agreement between classification and truth. K^=N*∑i=1kxii−∑i=1k(xi+*x+i)N2−∑i=1k(xi+*x+i) \hat K = {{N*\sum\limits_{i = 1}^k {{x_{ii}}} - \sum\limits_{i = 1}^k {\left( {{x_{i + }}*{x_{ + i}}} \right)} } \over {{N^2} - \sum\limits_{i = 1}^k {\left( {{x_{i + }}*{x_{ + i}}} \right)} }}

i = the class number

Focus group participants in the study area were purposefully selected from the selected sample kebeles (the lower administrative divisions) with a maximum of five to six participants. Two kebeles were selected from highland and lowland agroecology based on population density while three kebeles were selected in the midland area. In the seven kebeles sampled, about 14 group discussions were carried out, two group discussions for each kebele. The FGD participants included model farmers, experts, administrators, and development agents from the highland, midland, and lowland agroecology. Selection of FGD participants was carried out based on local experience, expertise on agriculture, or natural resource conservation, and acceptance among the community, targeted to understand their perceptions, knowledge, and ideas on the dynamics and drivers of LULCC.

Participants for key informant interviews were purposefully selected from seven kebeles found at different agroecology/altitudinal belts. For the interview, 9 to 11 key respondents were selected from each kebele considering gender, age over 45, local experience of LULCC, resource conservation work, and social recognition. For instance, in the lowlands and highlands, two kebeles for each were selected. In contrast, three kebeles were selected in the midlands, because the midlands area has a higher population density than the lowlands or the highlands.

According to this arrangement, 70 respondents were interviewed in the whole agroecology. The interview was mainly focused on the perception and experience of the community on driving forces and characteristics of LULCC of the study area for 33 years (Table 10). The theoretical framework is given by Turner et al., [1994]; Geist and Lambin [2002] was used to categorise the participants’ response about the forces as underlying and proximate drivers of LULCC in the three agro-ecological zones (Table 10).

Initially, a reconnaissance survey was conducted across different agroecology. Based on the survey results, the LULC, such as agroforestry, degraded land, shrub/woodland, forest land, farmland, grassland, and settlement land cover, were identified to be studied.

Even though all LULC classes have undergone changes in the study area, the magnitude and rate of changes were inherently different. According to Tables 4 and 5 in the initial year (1986), agroforestry and cropland alone accounted for about 71.2%. After 33 years, the size of agroforestry and cropland is about 84.4% of the total LULC of the study area. The dynamics of LULC indicate that for the last 33 years, the study area lost annually about 321.2 ha, 197.1 ha and 31.6 ha of shrub/woodland, grassland, and cropland, respectively.

On the contrary, the land with agroforestry, settlement, forest and bare land showed an annual increment of about 443.7 ha, 65 ha, 33.6 ha, and 4.5 ha, respectively. For instance, in the past 33 years, the highest rate of land cover increment (14%) was recorded on agroforestry and the highest rate of loss (10.3%) occurred on shrub/woodland (Tables 4 and 5 and Figure 4). The image analysis also showed that consistent increment was observed on agroforestry and settlement land classes, whereas grassland covers were consistently reduced in the study period. In the other land use class inconsistent change was recorded, for instance, crop land cover increased from 1986 to 2000, but it continuously decreased from 2011 to 2019 (Table 4).

Figure 4

LULCC trajectory refers to the possibility of land use and land cover change from one land use class to another taking place in a specific period. The LULCC trajectory of the study area was classified in four categories that showed the net loss or gain. The first category of the LULCC classes is the period from 1986 to 2000. In those 14 years, the net loss occurring in the classes was the following: on shrub/woodland, about 8.6%; grassland, about 1.3%; and forest land, about 1.3%. The net gain of LULCC observed was the following: agroforestry, about 4.73%; cropland, about 3.8%; and on bare land, about 0.19%. According to the LULCC trajectories, about 6.2% of the shrub/woodland was converted to agroforestry, about 6.3% of the grassland was converted to cropland, and about 1.2% of the forest land was converted to cropland cover. In this period, the highest loss (8.6%) and gain (4.73%) were observed on shrub/wood and agroforestry land, respectively (Table 6). The second category of LULC trajectory refers to the period from 2000 to 2011. In this period, about 1.8% of cropland, 2% grassland, and 0.5% of shrub/woodland cover showed a reduction. One of the critical changes observed in this period is cropland. Though cropland has increased in the years from 1986 to 2000, it shows a reduction in the years from 2000 to 2019. The change in the cropland implies that the high population growth causes the farmer to shift from extensive monocropping to a multiple cropping or agroforestry system. For instance, about 2.9% of cropland, 2% of grassland, and 1.7% of forest land were converted to agroforestry. The highest gain of about 3.8% was detected in agroforestry, and the highest loss, about 2%, was recorded on grassland land cover (Table 7).

The third category of the LULCC trajectory refers to the period from 2011 to 2019. In this period, about 5% of agroforestry, 2.2% of forest land, and 1.4% of settlement land have shown expansion. On the other hand, about 4.3% of grassland, 3% of cropland, and 2.1% of shrub/woodland showed a reduction. This category's land use and land cover change trajectory showed that about 4.9% of cropland and 1.37% of shrub/woodland converted to agroforestry. About 1.4 % of grassland was converted to forest and agroforestry land (Table 8). The fourth category of LULCC trajectory refers to the period from 1986 to 2019. In those 33 years, about 10.3% of shrub/woodland, about 6.3% of grassland, and about 1% of cropland showed a remarkable reduction. Conversely, about 14.27% of agroforestry, about 2% of settlement land, about 1.2% of forest land, and about 0.15% of bare land showed expansion. The LULC transition of the fourth category showed that about 8.8% of shrub/woodland and about 6.35% of cropland were added to agroforestry; also, about 1.4% of shrub/woodland and 1.3% of cropland were converted to settlement and forest land, respectively (Table 9).

LULCC have several implications at the local, regional and global scales of socio-economic and environmental conditions. The result on LULCC of the Ethiopian rift valley region of the Gidabo subbasin showed that about 58% of the river basin has been exposed to different LULCC in the past 33 years. About 46.7% of the land was covered by the agroforestry system, which is the largest land size among different land classes from the total land use and land cover size. The result is in agreement with the reports by Degife et al. [2019] and Kippie [2002] on the expansion of agroforestry and settlement land in south-eastern Ethiopia and by Meshesha et al. [2014] on forest land expansion in the Northern Central Highlands of Ethiopia.

According to the participants in group discussions and interviews, agroforestry has multiple advantages; for instance, it is used to increase vegetable yields, extend the harvesting season, improve the quality of product, increase the income of rural communities, improve food self-sufficiency, improve quality of life, improve the utilisation of forest products, regulate local climate, reduce soil erosion, and increase biodiversity in a densely populated area. The local community perception indicated that the agroforestry system is helpful to the resilience of the environment and sustainable agricultural production. The report by Brown [2018] suggests that LULCC that encourages agroforestry can improve the agricultural systems and mitigate the impact of climate change.

Though the theory of a resilient environment and sustainable development are independent, several links between the two concepts can be found in the literature [Lew et al., 2016; Derissen et al., 2011]. According to Redman [2014], a resilient environment and sustainable development can be considered complementary approaches because a resilient environment is important to long-term sustainability of livelihood outcomes [Harris et al., 2020; Espiner, et al., 2017]. In this regard, the agroforestry system of the study area is the best practice for a resilient environment, which has to be taken into account when one is designing policies for the sustainable development of ecological-economic systems and vice versa.

Nevertheless, the LULCC such as reducing shrubland and grassland and the expansion of bare land, have threatened to adversely affect the efforts on a resilient environment and sustainable development effort of the community. Therefore, the driving factors that contribute to the changes in shrubland, grassland, and bare land should be considered in future studies of LULCC. From this point of view, all the dynamic changes observed in the land cover of the study area were mainly related to such factors as demographic, economic, cultural, natural, institutional (policy), and agricultural intensification technology (Table 10).

The highlands are from 2,300 to 3,200 m a.s.l, the midlands are from 1,500 to 2,300, and the lowlands are from 500–1,500.

3.2.1.1

Demographic factors as drivers of LULCC

Among the identified drivers of LULCC, a demographic factor took the top rank in all agroecology (the altitudinal belt). About 90% of participants in the interview and the group discussion participants perceived that population growth had increased the demand for more land for cultivation, livestock production, tree plantation, and settlement. The high population growth of the study area also increased land fragmentation in holding size. Thus, the farmers started to shift the monocropping system to a mixed cropping or agroforestry system. As confirmed by satellite image analysis, informant interviews, and group discussions, land scarcity increased the conversion of grassland and cropland to the agroforestry system (Figure 5).

Figure 5

According to the report by Temesgen et al. [2018], zonal agricultural office [2020] and the central statistical report of Ethiopia [2013], the population density of the highland and lowland area ranges from 300 to 450 persons per sq.k.m, while in the midland it ranges between 774 to 900 persons per sq.k.m. This record is the highest population density in the country. According to Bilsborrow and Ogendo [1992], land scarcity exacerbates the conversion of wildlands to agriculture and other land uses.

]Therefore, the reductions of grassland and shrubland from 1986 to 2019, forest land from 1986 to 2011, and cropland from 2000 to 2019 were mainly related to the expansion of the agroforestry system, which supported the rising demand of large human populations (Figure 5). This result agrees with the report by Ketema et al. [2020] and Temesgen et al. [2018]. Population growth was positively correlated with the expansion of agroforestry land that provides year-round income to support large family sizes (Figure 5).

3.2.1.2

Natural factors as drivers of LULCC

Natural factors such as frequent drought, high-intensity rainfall, steep slope, and soil erosion highly determine the LULCC [Zondag and Borsboom, 2009]. According to Table 10, for about 80% of focus group participants and 85% of interview participants, natural factors such as climate variability, steep slope, and topographic variations ranked second in determining the LULCC of the study area. The land users said that the climate is not as regular as it used to be: it has frequently been erratic and variable.

The climate variability increases the farmers’ insecurity about cultivating a similar type of crop (monocropping), fruits, or vegetables. As a result, they favoured applying multiple types of cultivation, that is, an agroforestry system. The farmers’ insecurity feeling for climate change forced them to consider agroforestry as the best option to adapt climate change. This perception was confirmed by the satellite image analysis that showed the expansion of agroforestry land cover from 1986 to 2019. The result agrees with the findings of Temesgen et al. [2018], in which drought/RF variability was perceived as a driver of LULCC in the study area.

The other implication of natural factors on land cover change was observed in the highland agroecology of the study area. Though the land cover in the highland area is gradually converting to an agroforestry system, the satellite image analysis and key informant interview confirmed that comparatively extensive land in the highland area was left for grazing land due to the acidic the soil that is not suitable for crop cultivation (Figures 4 and 6). According to Zhang [2013], high rainfall causes essential soluble salts such as Ca, Mg, K, and Na to be leached away and the insoluble acidic residues (composed of oxides and silicates of iron, silicon, aluminum to be left with a high amount that not suitable for agriculture.

Figure 6

3.2.1.3

Economy as drivers of LULCC

The Ethiopian economy is highly dependent on agriculture; due to its reliance on rain, it has faced less productivity and fewer unsustainable products. In the study area, economic drivers are the third-highest factors that determine the LULCC. According to Table 10, about 90% of the participants in the midlands, 84.5% in the lowlands, and 66.5% in the highland areas have reported that small farm size caused frequent cultivation in the crop land and reduced productivity due to soil loss. According to the report by MOFED [2012] the study area absolute poverty lines is higher compared to the national level of poverty lines in 2011, because the high population growth, diminishing land-holdings, landlessness and a lack of on-farm technological innovation significantly contributed to the decline in productivity per household [Dorosh and Rashid, 2013].

In the study area many youth farmers who had formal education favored engaging in different off-farm activities, such as daily labouring, riding a motorbike, shoeshine, charcoaling, and commercial farming (Figure 7). According to the DA the off-farm works adversely affected the farmers land management activity, which adversely affected the LULCC of the area. Consequently, farmers are unable to feed their families. In this regard, many farmers are trying to get fertile soil from grassland, shrub/woodland, and other marginal lands, which, in return, contributed to land degradation /soil erosion (See Figure 8). This finding agrees with the report by Ketema et al. [2020], which reported that the economy dependent on agricultural land expansions was the proximate driver of rapid LULC dynamics in Ethiopia.

Figure 7

Figure 8

3.2.1.5

Cultural factors as drivers of LULCC

According to Shiferaw and Singh [2011], culture is a determinant factor in the issue of LULCC. Table 10 shows that about 77% of focus group participants and 81% of the interview key informants agreed that the culture of tree planting, inside the Enset and Coffee plantations and rotating one's own house in the Gedeo and Sidama communities has an important impact on LULCC of the area. Indigenous knowledge of the tree conservation (Songo and Baabbo) and soil treatment (Mona) system were highly accountable for expanding the agroforestry system and forest land. There are cultural institutions named ‘Songo,’ ‘Baabbo,’ and ‘Mona’ in the Gedeo zone. Songo is a sacred place where trees are reserved for ritual and cultural purposes, and Baabbo is a traditional way of tree conservation for various socio-economic purposes [Alambo et al., 2019; Maru et al., 2019].

“Mona” is a traditional acid soil treatment approach with animal dung using a circular fence constructed from bamboo plants to herd domestic animals (particularly for cattle and horses). The farmers are applying several agricultural practices in the treated soil such as production of enset, cereal crops and vegetables (potato, ginger, taro, chives) and livestock husbandry (sheep, cattle, and horses), side to side of eucalyptus trees, which is contributing for the expansion of agroforestry system (Figures 9 and 10). The result was compatible with the reports by Kura [2013] and Kippie [2002] about the cultural role for the expansion of the agroforestry system of the study area.

Figure 9

Figure 10

3.2.1.6

Institutional factors as drivers of LULCC

Institutional arrangements influence land-use decision making. The government policies are part of institutions that play an important role in LULCC [Briassoulis, 2009]. In the last 33 years, the government of Ethiopia formulated different policies on natural resource conservation. About 65% of FGD participants and 64% of the interview participants confirmed that the government policies impact LULCC of the study area (Table 10). For instance, the 1980s policy of SWC, agricultural intensification (resettlement, land privatization and access of agricultural inputs), the 2005 community-based SWC program and the 2011 green economy initiatives considerably determined the LULCC of the study area. The government policy that is related with urbanization and infrastructure development causes LULC change through conversion of land use classes, such as farmland and forests to built-up areas and infrastructures [Anteneh, et al., 2018]. The report by Yeshaneh et al. [2013] showed that the 1975 land proclamation of Ethiopia take away all lands from the landlords and distributed to the landless, which in turn caused a large deforestation for the purpose of agricultural expansion. The tenure of all rural and urban lands by the state led to a lack of belongingness to natural resources by the individual farmers, which in turn triggered vast deforestation.

The information obtained from key informant and experts revealed that the 2005 and 2014 community based SWC conservation program contributed to the improvement of shrub/woodland from 2000 to 2011, forest land from 2011 to 2019 and agroforestry from 1986 to 2019 (Table 9, Figure 11). Farmers also indicated that during the 1970s and 1990s law enforcing institutions were weakened following the social unrest and many trees were cut for construction, charcoal production and fuel wood. The interviewees also indicated some factors that were not mentioned by farmers, such as lack of lack of institutional coordination, poor infrastructure, illiteracy and poverty. Similar result report by, Alemayehu, et al. [2019] and Temesgen et al. [2018] showed that cultivated land has expanded from 1986 to 2000 in southwestern and southeastern Ethiopia, respectively, mainly related to different land-use policies of the government.

Figure 11