Landscape composition and configuration in the central highlands of Ethiopia

Abstract Landscape dynamics are common phenomenon in the human‐dominated environments whereby it can be observed that the composition and configuration between landscape elements change over time. This dynamism brings about habitat loss and fragmentation that can greatly alter ecosystem services at patch, class, and landscape levels. We conducted a study to examine composition and configuration of forested landscape in the central highlands of Ethiopia using satellite images of over a period of four decades, and FRAGSTAT raster dataset was used to analyze fragmentation. Our result showed five land use/land cover (LULC) types in the study area. Cultivated land and settlement land increased at the expense of forestland, shrubland, and grassland. Fragmentation analysis showed the number of patches increased for all LULC types, indicating the level of fragmentation and interspersion. Juxtaposition increased for shrubland, grassland, and cultivated lands and decreased for settlement and forestland resulting in the fragmentation and isolation of patches. The study of LULC along with fragmentation at the landscape level can help improve our understanding of the pace at which conversion of landscape elements is happening and the impacts on ecosystem services as studies of LULC are courser in nature and would not show how each land use is reducing in size, proximity and shape among other things that determine ecosystem services. Such type of studies in rural landscapes are very vital to consider appropriate land management policies for the landscape level by taking into account the interaction between each element for sustainable development. We recommend land managers, conservationists, and land owners for observing the roles of each patch in the matrix to maximize the benefits than focusing on a single element.

Such changes in the composition and configuration are the results of anthropogenic activities (Echeverria, Coomes, Hall, & Newton, 2008;Wu, 2013). Understanding the extent, spatial character and distribution of forest patches within the mosaics of landscapes modified by human activities represent one dimension of theories from island biogeography although fragmentation studies have advanced much more than island biogeography through metapopulation theory and landscape ecology (for example, Erik & Priya, 2003;Laurance, 2008;MacArthur & Wilson, 1967;McGarigal, Cushman, & Ene, 2012).
Habitat loss and fragmentation are nonrandom processes where the conversion of forestland use to agricultural, settlement, and grazing land use is undertaken intentionally by farmers based upon potential productivity of the land for crop production, proximity to roads and urban centers, topography, and drainage (Laurance, 2008). Landscape pattern metrics provide a relative measure of fragmentation, facilitating comparisons between different geographic areas, as well as multitemporal analysis within the same area. The study of fragmentation also depends on the results of change in land cover, and it employs vector or raster data for fragmentation analysis of a given matrix where modified landscapes exist. The degree of fragmentation has been described as a function of patch, class, and landscape metrics described in the Methods section because the best way to quantify the relative importance of habitat loss and fragmentation is to conduct comparative analyses at the landscape scale because a combination of both landscape-and patch-scale variables determines structure and function of ecosystems (Laurance, 2008;Santos-Filho, Peres, da Silva, & Sanaiotti, 2012). In Ethiopia, several studies have been conducted to understand LULC particularly with reference to deforestation (Feoli, Vuerich, & Zerihun, 2002;Gebrehiwot, Bewket, Gardenas, & Bishop, 2014;Reid et al., 2000;Tsegaye, Moe, Vedeld, & Aynekulu, 2010;Wondrade, Dick, & Tveite, 2014;Zeleke & Hurni, 2001) but to the best knowledge of the authors, no investigation has been carried out on fragmentation analysis combined with LULC.
In this study, we explore landscape composition and configuration and its implication on landscape structure in Jibat Forest because it has been used to infer the spatial and temporal integrity of ecological processes. This assessment identifies how land use changes and fragmentation varies within a rural landscape particularly within a forest boundary that affects patch dynamics and connectivity. We compared land use and land cover changes and fragmentation processes in the study area on temporal scales. The study of landscape composition along with landscape configuration is particularly very useful as fragmentation provides detailed analysis of the changes over time.
We selected Jibat Forest as our study site because it is one of the few remnant forests in a highly modified landscapes in the central highlands of Ethiopia and it is one of the centers of diversity for plant and animal species (Tamrat, 1994;Tesfaye, Fashing, Bekele, Mekonnen, & Atickem, 2013). It is also one of the remnant moist afromontane forests of the country found in the western escarpments servicing as headwater for Gibe River which is serving the country as major source of hydroelectric power generation. In addition, although studies related to vegetation characterization (Tamrat, 1993(Tamrat, , 1994 and feeding behavior of Boutourlini's blue monkeys (Cercopithecus mitis boutourlinii) were conducted (Tesfaye et al., 2013), analysis of landscape composition and configuration was not conducted and hence, we believe that studying landscape structure is an important aspect to consider in order to understand how the changes in landscape structure can have implication for appropriate management of the resources.

| Study area
Jibat Forest is located on the mountain chains of the central highlands of Ethiopia (37°15′-37°30′E; 8°35′-8°50′N) (Figures 1 and 2). It F I G U R E 1 Partial view of forest fragment of the study area extends from the west to southern portion of the lower altitudes (2,000-3,000 masl) where the forest takes a form of mosaics of landscapes interspersed with farmlands. The forest has been heavily exploited for commercial timber production, agricultural land expansion, and logging by the communities for selling of the wood for small-scale wood industries (Tamrat, 1993). As a result, the forest did not reach its climax state, rather the forest is regarded as secondary as it is highly disturbed. Hagenia and Rapanea species which are the characteristics of high-altitude forests are growing, and the forest is regarded as humid Afromontane forest type (Tamrat, 1993(Tamrat, , 1994 Forest conversion to agriculture and grazing lands began in the 1970s and has resulted in ongoing fragmentation around the edges of the forest; pioneer tree species such as Bersama abyssinica and Clausena anisata are the most dominant species (Tesfaye et al., 2013).
Livestock currently graze legally both on grasses in recently cleared grazing land near the edge of the forest and on forest shrubs in some peripheral portions of the forest. Illegal tree cutting occurs throughout the forest, although it reaches its highest intensity along the forest edge and in fragmented areas of the forest. The current total area of the forest is estimated to be 38,461 ha (Tesfaye et al., 2013). The total area considered for our study is much higher than the current amount specified because the demarcation processes over the study period were not consistent across the years and the boundary use to change inwards as the forestland was converted to cultivated land through clearing of the vegetation as this method is one way by which farmers obtain the right to cultivated land in Ethiopia. The total area of the forest reported does not include forest patches isolated from the continuous forest due to land use conversion. In this particular study, we tried to incorporate that landscape matrix to take into account fragmentation and land use change processes on spatial and temporal scales as our analysis goes beyond the current boundary of the forest.
Mean annual rainfall was 1,768 mm and mean monthly low and high temperatures were 7.8 and 23.6°C, respectively, for the area between 1997 and 2006. The rainy season occurs from March to October with a peak in rainfall between June and September, and the dry season occurs from November to February (Tesfaye et al., 2013).

| Satellite data preprocessing and land use classification
In this study, time series datasets of LULC were produced from multispectral Landsat imagery, which were acquired on four separate years: 1973, 1984, 2000, and 2015. All of the images were clear and nearly free of cloud as it was taken during dry season (Table 1). Prior F I G U R E 2 Location of the study area to interpretation, atmospheric correction and geometrical rectification were performed. The dates selected for processing of LULC were mainly dependent on the availability of the image, important dates in the change of government, and policies related to rural land and agriculture.
The image processing and data manipulation were within ArcGIS software. Five land use types classified as settlement land, cropland, shrubland, grassland, and forestland were identified within the study area. Land use/land cover classification signature was prepared for each class; historical training site, image interpretation and personal experience, knowledge of the watershed physical topography were used. Each signature was evaluated by separability test between and within the signature; finally, signature recorded good separability kept; and others are redefined until the signature separability was within an acceptable range.
Ground control points were collected to compute accuracy assessment for the classification year of 2015. The number of GPS points collected was determined using the classification area proportion of the LULC map of 2015. The overall producer's accuracy, overall user's accuracy, and overall Kappa statistics were 85.4%, 89.2%, and 0.84, respectively. These met the recommended values suggested by Janssen and Vander Wel (1994). Thus, these data were available for further study on the level of fragmentation. There are various change detection methods that can be used in ERADS Imagine and other remote sensing software. In this study, the postclassification comparison was employed using separately classified Landsat images and then, three comparisons were made : 1973-1986, 1986-2001, and 2001-2015. To study the changes in land use/land cover between the above year intervals, conversion matrix models are applied. The direction of land use/land cover change between classified images used ERADS Imagine matrixes. During characterization of land cover about five land cover types were identified (Table 2).

| Measurement of landscape fragmentation
In order to assess landscape fragmentation, we adopted McGarigal et al. (2012) and Smiraglia, Ceccarelli, Bajocco, Perini, and Salvati (2015) landscape metrics (see Table 3). To measure land fragmentation under different land use, fragmentation metrics at landscape levels were selected (Wang & Yang, 2012) depending on the metrics we desire to address because each classes have their own unique characteristics to be described in landscape structure studies. Class-level metrics such as number of patches, percentage of landscape, edge density, largest patch index, mean patch size, area-weighted mean shape index, mean Euclidean nearest neighbor distance, interspersion, and juxtaposition and aggregation index were employed to measure the average fragmentation (McGarigal et al., 2012). These fragmentation matrices are the best methods to compare the level of fragmentation of land uses over temporal scales.
In this study, computation of the above landscape metrics was performed with FRAGSTATS 4.2.1 (McGarigal et al., 2012) on raster datasets as an input because of its accuracy for calculating fragmentation metrics and the ease of the use of the program on raster datasets (MacLean & Congalton, 2015). The raster datasets were processed in ArcGIS software for use by FRAGSTATS. Shrubland Land with >20% bush or shrub cover with <20% tree cover (<5 m in height).

| Land use/land cover change
Grazing Land/Grassland Land under grass cover but highly influenced by grazing and browsing of domestic animals.

Forestland
Land dominated by trees with greater than 80% canopy cover Of the natural vegetation cover types, forestland and shrubland experienced the lowest persistence, whereas grassland was the most persistent cover type (Table 5). The net change-to-persistence ratio was large for forestland (negative), cultivated land (positive), shrubland (negative), settlement land (positive), and grassland (positive), indicating the most dominant trends in the changing landscape (Table 5).
Overall, 35910.4 ha (i.e., sum of diagonal elements) of the total landscape remains unchanged (Table 5).

| Analysis of the dynamics of landscape metrics
During the study period, there have been changes in the size, number, distance, and spatial distribution of fragments, with different patterns for different land uses at the class level. Cultivated land is the predominant landscape matrix with significant increases for the entire studied variable and study period. One very important fragmentation metric which is vital to note for cultivated land is the number of patches.
Across the study period, the number of patches increased which is an area of concern where a piece of land is further divided into several smaller patches by the family. Deforestation was found to be pronounced, and the landscapes dominated by forests became fragmented at the first stage followed by the removal of the fragments through time. Each fragmentation variable indicated below showed a decline in the forest and shrubland followed by an increase in cultivation land and settlement land. The shaded figure is the sum of diagonals and represents the overall persistence (i.e., the landscape that did not change). b Net change = gain-loss. c Np refers to net change-to-persistence ratio (i.e., net change/diagonals of each class).
F I G U R E 3 Land use/land cover for four decades  | 7415 Tolessa eT al.

| Shrubland
Shrubland is essentially found along with forestlands. It is highly influenced by human activities next to forestland where some forest products are regularly collected and livestock graze. It serves as a buffer zone, and secondary growth is taking place but there is also conversion to cultivated land as it is very difficult to obtain land for cultivation by farmers. Number of patches, mean patch size, and percentage of landscape are very dynamic and changed both ways, indicating its fluctuation in the different metrics.

| Grassland
Grasslands are decreasing in their size within the study period that can be an indication of the regular conversion of the land use to cultivation. Grasslands are the smallest in mean patch size next to settlement as common grazing areas are further partitioned by communities for cultivation and the recent land certification for use by farmers which grant an exclusive use rights and serving as a base for government tax collection from rural land use.

| Forestland
Forest cover decreased progressively throughout the study period with lower AREA_MN and LPI and more complex and less aggregated patches (high AI and lower SHAPE_ AM). The number T A B L E 6 Representation of spatial pattern at class level for five land uses in four periods of the study (1973, 1986, 2001, and 2015), based on nine landscape metrics for the central highlands of Ethiopia in 2015 which shows that it is not only the overall size and number of patches decreased but each patch is also affected by edge with nearly no interior forest. The relatively large-sized forests are found on the rugged slopes, low soil fertility areas, and mountain chains of Jibat (Table 6).  (Table 6).

| Landscape composition and configuration
The land use/land cover analysis showed five classes in the study area. Each of the metric used to describe fragmentation processes at the landscape determines connectivity between patch types, thereby affecting ecosystem services. So, the percentage landscape is very critical to show the overall impacts of each land use benefit at the landscape level. In our study, it can be generally deduced that the overall benefits of each landscape metric are highly dominated by cultivated land as the largest part of the landscape is occupied.

| DISCUSSION
The current study integrated LULC and landscape metrics to try to The ecosystem services such as regulating and supporting services are highly influenced by changes in land uses. Several studies conducted in different parts of the world confirmed that land use changes especially deforestation impacted ecosystem services (Costanza et al., 1997(Costanza et al., , 2014de Groot et al., 2012;MA, 2005).

| Land use/land cover analysis
Subsequent reduction in forestland and shrubland was observed across the study period. This has been evidenced by many studies in Ethiopia (Fetene et al., 2016;Meshesha, Tsunekawa, Tsubo, Ali, & Haregeweyn, 2014;Reid et al., 2000;Tsegaye et al., 2010;Wondrade et al., 2014) and many tropical countries (Lira, Tambosi, Ewers, & Metzger, 2012;Nahuelhual, Carmona, Aguayo, & Echeverria, 2014;Putz et al., 2014). Such conversion of forestland, shrubland, and grasslands was anthropogenic in nature conditioned by socioeconomic, political, and institutional factors (Echeverria et al., 2008;Temesgen et al., 2013). The conversion of one land use to the other is dynamic, and it did not follow a linear pattern over a 40-year period in the study area. This means that certain land use types such as grassland showed increase and decrease patterns. The rate of conversion of forestland tended to reduce over time particularly in the year 2001-2015. A small amount of increase in forestland was recorded, but the overall reduction is recorded for the whole study period. The small positive increase was also reported by local communities when discussion was made. They indicated that the increment was due to law enforcement by the Oromia forest and wildlife enterprise which took over the forest and started to harvest, plant, and protect the remaining forests.
The conversion of forestland, shrubland, and grassland into settlement and cultivated land can be attributed to population increase, policy incentives of the governments, and market failure to value the ecosystem services of the land uses converted and world economic order which encourages farmers to grow exportable crops to the world market. In the last four decades, the population of the study area and the country increased at higher rate than the economic growth of the country (Meshesha et al., 2014). The population of the study area increased through migration from other parts of the country in search of fertile land to cultivate and higher fertility rate of the indigenous population which is similar to the country as a whole (CSA, 2008;Jacob et al., 2015).
From the number of patches of settlement land, it can be observed that the population had increased in the study area which can be con- Government policies on food self-sufficiency were also another bottleneck for resource conservation as farmers need to produce more, and they are forced to expand their farms to other land uses due to the low level of technology delivered to farmers as well as the economic capacity of the farmer to afford these technologies. Marketoriented crop production, a policy of the EPRDF government, is also contributing to land use changes in Ethiopia.
Market failure to value the ecosystem services of forests, shrubs, and grassland relative to cultivated land is another factor resulting in deforestation and fragmentation. In this perspective, natural resource assets will inevitably be misused or exploited until realistic long-term social and environmental costs are internalized and reflected in market prices. The effort made so far to properly value the services given is

| Analysis of the dynamics of landscape metrics
The  (FAO, 2010;Teketay, 2001;Zeleke & Hurni, 2001). Furthermore, the temporal changes of matrix as described in the fragmentation analysis above cannot continue to have similar ecosystem services, rather it changes in accordance with the composition of each matrix within the landscape. The quality of the landscape matrix is very vital in determining ecosystem services at the landscape scale, and each one of the patches within affects the overall benefits (Putz et al., 2014;Santos-Filho et al., 2012).

| Shrubland
Shrublands are at the forefront for conversion to cultivated land and used for cattle grazing as an alternative to grazing land when there is scarcity of grasses especially during dry season. The number of patches and largest patch index show increasing rates of fragmentation, and interspersion indicates isolation of patches. In 2001, the mean patch size is higher than any of the year and this may be due the conversion of forest land to secondary forests through selective cutting, fire, and intensive grazing that can reduce the canopy of the forests. This is one strategy by which farmers gradually covert forests into cultivated land.

| Settlement
Settlement patches increased by 89% and percentage of landscape increased by 82.6% from 1973 to 2015. It indicated how many households increased within the study period. Although residential area increased to some extent, the absolute area was small and therefore, it was not the main category to impact ecosystem change in the study area (Li et al., 2007). The small-sized settlements as depicted with mean patch size are a clear indication of the nature and quality of each house built in rural areas as they are not large enough to accommodate large-sized families of rural Ethiopia. Each family with the house is overcrowded in a very small room reflecting the economic, social, and environmental conditions that must be modified to improve housing.

| Grassland
The

| Forestland
Forestland experienced a series of changes on spatial and temporal scale. As observed in the field and studies such as Munsi, Areendran, Ghosh, and Joshi (2010)   share from parents. In general, landscape metrics showed that the composition and configuration of arable land in the landscape was further driven by population increase which was more pronounced during the study period. This situation is seriously undermining productivity of land, labor, and other production factors. The lack of land for cultivation forces family labor especially the young who are economically active to move to other livelihood options such as forest product collection, offfarm activities, and migration to urban centers.

| Land use/land cover analysis and landscape metrics
Combining LULC data with fragmentation analysis improves understanding of the level of landscape transformation, the nature of such changes, and how each land use types did aggregated or dispersed from each other (Gillanders, Coops, Wulder, Gergel, & Nelson, 2008;Riitters et al., 2016;Smiraglia et al., 2015;Uuemaa et al., 2013). The analysis of fragmentation at class level provided detailed information in relation to size of each patch, number of patches, percentage of land use within the landscape, and other important variables that can be useful to understand how the different land uses could be used for optimizing ecosystem services such as biodiversity conservation, pollination, erosion control, protecting cultural landscapes, and hydrological cycles at a landscape level than at patch level which is often very difficult on modified landscapes similar to the study area.
So, although this study is not the first to be carried out in this way, it is the first attempt to try to understand landscape metrics along with LULC analysis and its implication in land management practices in the central parts of Ethiopia where land and land-related resources are becoming fragmented so that it is difficult to develop the land through the use of soil and water conservation practices (Bekele & Drake, 2003).

| CONCLUSION
Land use/land cover change detection and fragmentation analysis are useful tools to address the amount and location of change and also provided the ability to compare matrix boundary change over the stated study period by the use of ArcGIS and FRAGSTA software. The interspersion, isolation, and connectivity affect (positive and negative depending on the service required) ecosystem service delivery of patches in our study. The forest boundary has been under a continuous change through habitat loss and fragmentation over time. The forests in the landscape not only are becoming increasingly smaller patches, but increasingly isolated producing both environmental and social implications.
Remote sensing and patch analysis methods can be advantageous in efficiently observing and monitoring land cover changes and fragmentation processes that occur in the landscape boundary across multiple dates at multiple locales. Such studies provide an insight into the processes of changes of land uses between themselves and show the need for appropriate land management policies at landscape level than patch level. Hence, shifting our view of fragmented landscapes toward the full inclusion of landscape matrix in our study is very important to understand how landscapes are composed of different land uses, the dynamics of each matrix on spatial and temporal scale, and the implication of such changes for ecosystem service provision.
In this study, we first analyzed LULC dynamics over four decades and found that settlement land, cultivated land, and grassland increased over four decades. Settlement land consistently increased in the study period, while cultivated land showed an increase  and then decreased afterward. Grassland showed a decrease in cover for the year 1973-1986 drastically than any other land use identified but then increased in the subsequent years. On the other hand, forestland and shrubland decreased by 47.4% and 26.3%, respectively.
Our study indicated the dynamic shifts of land uses across spatial and temporal scales. We found that forestlands, shrublands, and grasslands are at the forefront for conversion to the two major land uses particularly to cultivated lands due to economic, demographic, and policy changes in the study area. In addition, fragmentation of all land uses is a common phenomenon indicating the nature of rural small-scale framing practices in Ethiopia, which is very difficult to properly manage the land.
We also found that fragmentation of the different land uses was evident, indicating the partitioning of each landscape into smaller patches. One interesting finding we observed in our analysis is the fact that the number of patches for forestland did not increase dramatically. This is mainly because isolated forest patches are slowly converted into cultivated land which is a typical case in our study area.
But, as compared to forestland, the number of patches for other land uses increased over the study period. Our finding provided up-to-date changes in the level of fragmentation processes. The ongoing land use changes and further fragmentation thereof can have significant effects on the net provision of ecosystem services at the landscape.

ACKNOWLEDGMENTS
This work was supported by Addis Ababa University research grant for thematic area related to forest fragmentation research. We also appreciate the anonymous reviewers of the manuscript for their useful and constructive comments. The first author also thanks Ambo University for granting scholarship and Addis Ababa University for funding the research.

FUNDING INFORMATION
No funding information provided.