Evaluation of land use land cover changes using remote sensing Landsat images and pastoralists ’ perceptions on range cover changes in Borana rangelands , Southern Ethiopia

1 Department of Animal and Range Sciences, Madda Walabu University, Ethiopia. 2 Department of Plant Production and Soil Science, University of Pretoria, P. Bag X20 Hatfield, Pretoria, 0028, South Africa. 3 Department of Geography, Geo informatics and Meteorology, University of Pretoria, Pretoria, South Africa. 4 Deparment of Animal Science and Wilde life, University of Pretoria, Pretoria, South Africa. 5 School of Animal and Range Sciences, Hawassa University, Awasa, Ethiopia. 6 Department of Natural Resource Management, Arsi University, Rustavi, Georgia, Ethiopia.


INTRODUCTION
In the tropics, it is well acknowledged that human activities such as land cultivation, grazing, mining and settlements represent the main drivers of land use land cover (LULC) changes and land dynamic (Meyer, 1995).Such human activities including direct or indirect anthropogenic pressures are sources of continuous threat on territorial land on one hand and impact negatively the livelihood security of people who depend on such land ecosystem on the other hand.In the case of land ecosystem such as rangelands, pastoralists are among the people who are mostly affected by such human activities.
Elsewhere, rapid human population growth, increased frequency of droughts and expansion of settlement in arid and semi-arid environment are among the well-known drivers of land use changes (Vitousek et al., 1997).On the contrary, in arid and semi-arid areas for example, vegetation changes are triggered by both anthropogenic and natural factors such as soil moisture (White et al., 2008), erratic and uneven distribution of rainfall (Fensham et al., 2005) and grazing (Rahlao et al., 2008).However, up to date remote sensing tools used were mostly directed to assess land vegetation cover changes because plants are key indicators of the healthiness of ecosystem and its ecological dynamics (Jensen, 1996).
Knowing that plants are part of arid and semi-arid ecosystem and that such ecosystems are highly susceptible and vulnerable to natural and anthropogenic perturbations that often affect the livelihoods of security of pastoralists, therefore it is essential to identify the driving forces (for example biotic and abiotic) of land use changes, especially for appropriate management of ecosystem and ensuring the livelihoods security of pastoralists.This implies gathering data on the type of ecosystem and its land use dynamics.As a result, remote sensing and geographic information systems are one of the key tools available to land use planner.In Judean Mountain and the Judean Desert in the Mediterranean and Arid ecosystems, Shoshany et al. (1994) has successfully monitored vegetation cover change through remote sensing technique.
Multi temporal satellite image plays an important role in monitoring of vegetation cover changes in rangeland through prepared LULC map (Palmer and Fortescue, 2003) and Normalized Difference Vegetation Index (NDVI) (Eklundh and Olsson, 2003).Using multiple satellite images acquired at different dates over the same study area provides an opportunity to planners to monitor changes in land cover by using proxy parameters such as vegetation index differencing (Purevdoria et al., 1998).For land use planners, Landsat series [that is the multispectral scanner (MSS), Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+)] provide with multiple satellite image data that plays a major role in detecting changes land use land cover change (De Fries and Belward, 2000).Schmidt and Gitelson (2000) used Advanced Very High Resolution Radiometer (AVHRR) satellite image for monitoring temporal and spatial vegetation cover changes in Israeli transition zone.
However, in arid and semi-arid environment, such multi-temporal satellite imagery have been poorly applied for monitoring and planning for land use changes, such as Borana rangelands, Southern Ethiopia.As a result, carrying out research on the dynamics of LULC changes through multi-temporal satellite approach could therefore play an important role in providing valuable information to planners on the way to successfully plan changes and map changes over time (Moshen, 1999).
The importance of traditional knowledge is widely acknowledged as valuable tools for natural resources management and monitoring changes over time.In East Africa, the Borana rangelands was known as one of the best remaining pastoral lands however, increasing rangeland use has driven vegetation decline and threaten traditional knowledge of pastoralists with regards to access, use and management of natural resources of that area.Pastoralists' approach over management of the Borana rangelands consist of mobility of herds during wet and dry season, strong community norms and regulations on range and water resource use (Coppock, 1994).
Rangelands use by shift from grazing land to crop land and other livelihoods activities has driven changes in Borana areas and in pastoralists while contributing to erode traditional knowledge of pastoralists on livestock grazing practices (Desta and Cppock, 2004).For appropriate land use planning, monitoring and safeguarding, it is crucial to document pastoralists' traditional knowledge on rangeland uses and their perceptions on land use/land cover changes.
This study aimed at improving rangelands (grasslands, woodlands and cultivated lands) management by detecting land use/land cover changes over time  in the Borana rangelands of Ethiopia.The specific objectives were two folds: 1) map LULC document, the major driving forces of such changes and patterns of change over time period of 1973 to 2003 using satellite (Landsat) image data analysis, and 2) assess pastoralists' perceptions on rangeland cover changes in *Corresponding author.E-mail: dreyaman@gmail.com.Tel: +251-913870962.
Author(s) agree that this article remains permanently open access under the terms of the Creative Commons Attribution License 4.0 International License the study area.

Study area
The study area Borana rangelands, covers approximately 63,939 km 2 (Figure 1), and hold the largest pastoral population in Oromia Regional State of Ethiopia.The Borana rangeland is located between 4° to 6°N of latitude and 36° to 42° E of longitude and its altitude ranges from 1000 to 1600 (Coppock, 1994).The area is characterized by arid and semi-arid climate with annual rainfall ranging on average between 500 and 700 mm (Helland, 1982).Rainfall is bimodal with 60% of the annual high precipitation encountered during March-May (main rainy season) and the remaining is encountered during September-November (short rainy season).The long dry season is from late November to early March.The mean annual temperature varies from 15 to 24°C and shows little variation across seasons.The vegetation cover of Borana rangelands is mainly evergreen and semi evergreen bush land, with shrubby Acacia and Commiphora.The rangeland is dominated by alien genera, and dwarf shrub grassland (Gemedo, 2004).The geology of the area is dominated by 40% quaternary deposits, 38% basement complex formation and 20% volcanic (Coppock, 1994).

The satellite images
For this study, Landsat image of Multi Spectral Scanner (MSS), Thematic Mapper (TM) and Enhanced Thematic Mapper Plus (ETM+) for the year 1973, 1986 and 2003 obtained from Global Land Cover Facility (https://www.landcover.org)were used, respectively (Table 1).In order to cover the study area, eight satellite images were mosaic and thereafter extracted by using the study area boundary.LULC classification was done based on classification criteria drawn for East African rangelands by Pratt and Gwynne (1977) (Table 2).
Prior to image classification and NDVI analysis the acquired satellite images were geometrically corrected and radiometrically normalized then registered image to image.These applications were carried out using ENVI; satellite image analysis software.

Image classification and data analysis
For vegetation cover changes analysis, a post classification comparison of change detection method was used.The most widely used method of change detection is a comparative analysis of spectral classifications for series of time produced independently (Singn, 1989).Landsat images of three years were independently classified using supervised classification method.Since class change precision depends on spectral separation, the interest class group was created and used by maximum likelihood classifier and smoothed with filter to reduce the misclassified pixels (Erdas, 1999).
Overall, our analysis was based on hybrid classification approach (Odindi et al., 2012).In this approach, the ISODATA unsupervised classification algorithm was used to categorize the natural classes.Separability parameters were used for the reliability test of the methodology.Classes that show low degree of separability were regrouped until seven dominant classes were obtained.By combining the visual images inspection of false color composite and unsupervised classified image, two sets of samples each with 25 points were created namely: the training set and postclassification set.The training set was used as Regions of Interest (ROIs) in the maximum likelihood of supervised classification while the post-classification ROIs was used for accuracy assessment using the confusion matrix (Congalton, 1991).Supervised classification was repeated three times and the derived outputs for the 2 nd and 3 rd land cover classifications yielded similar results which showed level of accuracy and consistence in the classification.Consequently, the LULC classes were sized down from seven classes to four classes comprising of grassland, woodland, bareland, and cultivated/built up area.
Once the classification was made for 1973, 1986 and 2003, the percentage of LULC changes was calculated as follows: % trend change = (observed change/sum of change) x 100 The annual rate of change was obtained by dividing percentage change by the number of the study years : 1973-1986 (13 years) and 1973-2003 (30 years).The greenness of the study years were evaluated by calculating NDVI values as follows: The prediction of LULC class changes to another class in the future based on past changes was done using Markov chain model (Aavikson, 1995).This is done by computing transition matrix of pixels in each class for two time periods.The matrix contains unchanged pixels diagonally, while other cells contain pixels that have changed the probabilities of change between classes computed by dividing each cell value by its row total.
On the other hand, the spatial distribution of occurrences within each land use category was interpreted using land consumption rate (LCR) and land absorption coefficient (LAC) formula as follows: Where, A = areal extent of the rangeland in hectares, P = population.A1 and A2 are the area extends (in hectares) for the early and later years, and P1 and P2 are population numbers for the early and later years, respectively.LCR is a measure of compactness and indicates a progressive spatial expansion of settlers and LAC is a measure of change in consumption of new state of land by each unit increase in human population.The 1973The , 1986 and 2003 population statistics were obtained from the censuses of Ethiopian Central Statistics Authority (CSA) (CSA, 2008) following a recommended 2% growth rate of pastoral area.The estimated population figure was calculated using the following formula: n = r/100 * Po (1) Where, Pn and Po are the estimated population and base year population, respectively, r = growth rate (2%), n = annual population growth, and t = number of years projecting for the given period.

Pastoralist perceptions on the driving forces of changes of vegetation and land cover
The major driving forces of change of vegetation and land cover in the study area were investigated using 200 key knowledgeable pastoral respondents.A group discussion was carried out with local key informants, community development practitioners, and with local and regional relevant professionals.Semi-structured interviews with the key informants were also used to generate information to identify the major causes driving changes of vegetation or land cover, and to understand the perception of Borana pastoralists on these changes.Scores were given for the identified drivers of change and these were put in order of priority.Secondary data was used for comparison of the present and the past status of vegetation and land cover changes and to identify the scale and degree of these changes.The correlations and strength of the relationships between independent (pastoralist perceptions in vegetation and land cover change) and dependent variables (Landsat image data) were computed using Stata version 10.

Classification accuracy assessment
Highest producer's accuracy (71%) and low omission error (28%) was recorded for grassland of the year 1986 Landsat image.In contrast, the producer's accuracy (66%) was low for the cultivated/built-up area for the same year.For bare land, the producer's accuracy (73%) was higher in 2003 than (71%) 1973.The overall classification accuracy of 1973 was 67.3 and 69.1% and 69.5% for the year 1986 and 2003, respectively.

Change detection and LULC
In the Borana rangelands, grassland shows significance decrease over the year that is by 24.16% from 1973 to  3 and Figure 2).Over the 30 years period, the changes associated with woody, grass, bareland and cultivated land cover per year were estimated at 1.27, 0.87, 0.50 and 0.1%, respectively.LULC map (Figure 2) shows an increment of woody plant in 2003 and decrease of bareland compared to the years 1973 and 1986.The decrease could be attributed to canopy cover damage when remote sensing images were taken at that period of time.Since no remote sensing images were available for few parts of Southern Borana, therefore these areas were excluded in the present analysis, especially those for the year 1973.
The NDVI value also revealed the reduction of vegetation cover or biomass from the 1973 to 2003.In  1973, 46.14% of the total area was bareland, whereas, 53.86% was covered by vegetation (Table 4 and Figure 3).The minimum negative NDVI value was -0.93, while the bareland portion and the maximum NDVI value was 0.93, especially for the dense vegetation parts of the study area.For the year 2003, the non-vegetation covered area was 57.20%, (with a minimum NDVI value of -0.65) whereas the vegetation covered area was 42.80% (maximum value of NDVI 0.83).

LULC change prediction
After using the Markov model to detect land cover change, a 4 by 4 matrix table of land cover categories for the year 1973 and 2003 was constructed to predict the probability of LULC class changes over the years.As shown in Table 4, woodland had 0.66 probability of remaining woodland and a 0.36 of changing to grassland in 2003.On the other hand, a 0.34 probability of change from grassland to woodland showed that there might be a high level of instability in grassland during this period.
The 1973 grassland cover had a 0.28 probability of remaining grassland in 2003.Bareland also had a 0.18 probability of the remaining bareland in 2003.Cultivated land had a 0.13 probability of the remaining as cultivated land.Moreover, cultivated land had a 0.29 probability of changing to grassland and a 0.19 probability of changing to bareland.
The corresponding human population values for the recorded time period are also presented in Table 5.From 1973 to 1986, the LAC increased with population but declined sharply as population increased.

Driving forces of land-use/cover change and their correlation
According to Borana pastoralists', eight forces driving changes in LULC changes were identified.Among them recurrent drought, policy issue and increment of livestock population are mentioned as the prime factors (Table 6).
From group discussion, it is understood that continuous heavy grazing pressures have contributed to changes of land-cover from grassland to woodland.Human population growth is partly due to the infiltration by the highland farmers in the area.In addition, the recent expansion of public services like water points, school and health posts might have contributed to the expansion of settlements in the study area.From a policy perspective, sedentarization of pastoralists is a government rural development strategy.This has widely affected pastoral land-use patterns over the years.Moreover, the opportunistic farming activities also instigate the expansion of settlements in order to manage the farm plot as mentioned by our respondents.The steady increase of human population attracts the boom of livestock population, which pressurizes the changes of land-use.Overall, people stated that the changes in landuse/cover are mainly caused by frequent droughts and increasing numbers of dry years among other drivers, which highly affected their livelihoods.Further, the correlation and strength of their relationship among different driving forces and dependent variables (rainfall and year time series) are presented in Tables 7 and 8.

DISCUSSION
The observed decline of bareland cover in 2003 compared to 1973 might be explained by the poor capture of satellite sensing image of woody plant canopy.The larger canopy cover observed in Southern Ethiopia might have been interpreted as vegetation cover instead of bareland (Haile et al., 2010).However, data collected through on ground observation, local community responses and NDVI values tend to reveal that the bareland cover increased in recent years compared to the 1973.The expansion of bareland in the fragile and environment is often explained by cultivation practices leading to pronounced period of soil erosion even during small rain events and wind effects.In the Afar low-land of north eastern Ethiopia (Tsegaye et al., 2010) and in southern Ethiopia (Mintesnot, 2009) similar changes have also been reported in arid environments (Elmore et al., 2000).In addition, Solomon et al. (2007) have all reported that the increasing expansion of woodland cover around patch of resource areas was attributed to the increase of grazing pressures around these areas.In areas where there was low grazing pressure, high density of woody plants was reported (Brown and Archer, 1999).
NDVI is known as a measure of a photo that is from synthetically active vegetation in a given area (Asrar et al., 1985;Myneni et al., 1995;Weiss et al., 2004).The NDVI had strong correlations with biomass, vegetation phenotype, density of leaf and canopy cover (Petter and Eve, 1995;Tucker et al., 1985;Weiss et al., 2004).The NDVI analysis across the Borana rangeland showed, the highest positive and negative NDVI values were largely observed from the 1973 MSS imagery.For the inherent biases in reflectance measurements of the MSS sensors and the atmospheric cloud cover, high positive value of NDVI could be associated with the high percentage of grassland cover (Peters and Eve, 1995).On the other hand, the low value of NDVI value in year 2003 might be due to the high proportion of woody plant cover (Nemani et al., 1996).The other probable reasons for the low value of NDVI obtained in year 2003 were driven by canopy coverage and the extent of bareland use.They all have certainly contributed to reduce the NDVI value.Overall, NDVI values are useful measurement in understanding the vegetation cover of the Borana rangelands and can be used in rangeland monitoring and management approaches.
Over the years, between 1973 and 2003, Borana rangelands has demonstrated persistent LULC changes and this might be associated to different factors sure as rainfall variability and heavy grazing pressure (Oba and Kotile, 2001) shift in the traditional management practices and poor government regulations mainly ban of rangeland fire and developing communal rangelands (Oba et al., 2000).According to Angassa and Oba (2008) and McCarthy et al. (2002), banning of range fire facilitates the propagation of weeding species and inhibits woody plant growth.Consequentially, the survival of desirable herbaceous is undermines by dominated weeding species (Singn, 1989).Gemedo et al. (2006) noted that woody vegetation clearing out for home construction purposes, enclosure management along with the pastoralist's involvement in agricultural cropping activities through land clearance also contributes to woody plant decline.
The observed increase of cultivated/built in Borana rangelands might be explained by the farming activities pastoralists used as an alternative for lowering drought risk and rainfall uncertainty (Campbell et al., 2005).Frequently, occurring drought (1972/1973, 1984/85 and 1999/2000) results to food insecurity and livestock loss in Borana rangelands.In addition, infiltration by large groups of farmers from the neighboring areas also contributed to the expansion of cropping in the study area.Similar to East African pastoral land, the expansion of cropland in the Borana rangeland has significantly contributed to the change of grassland management practice to cultivate land practice (Reid et al., 2004).
Despite pastoralists resistance (Chatty, 2007), voluntary crop farming and settlements have been established in the Borana areas for generating income and livelihood diversification (Reid et al., 2004).Pastoralists resist the expansion of cultivation and sedentarization and mainly this reduces the size of their rangelands and mobility to utilize the unevenly distributed resources in this unpredictable environment (Chatty, 2007).In semi-arid areas of East Africa, the changes of landuse/cover have been accelerated by government policies since the 1973 (Omiti et al., 1999;Reid et al., 2004).This is also true in the case of Ethiopia.Recurrent drought is among the prominent natural catastrophes that have caused changes of land cover (Ndikumana et al., 2001).As an alternative, pastoralist had shifted their land uses from grassland to woodland and other land use forms (Coppock, 1994;Gemedo, 2004).Among other drivers of shift in the rangeland cover included population increase of human and livestock, conversions of rangelands to crop land in valley bottoms (dry season grazing areas) and increasing pressure on grazing due to multiple (ranch and private enclosures) end uses and competition for land from other tribes (Gari and Somali) in Borana grazing area (Oba and Kotile, 2001).Although dry land cropping represents an opportunity for pastoralists to increase income without damaging the land (Campbell et al., 2005), however, other scholars indicated that it represents a way of increasing risks and threats to the livelihood of pastoralists (Little et al., 2008).This has to do with the fact that increasing land cropping in semi-arid area contribute to aggravate desertification and Teka et al. 9 expansion of bareness lands Traditionally, the Borana pastoralists used to regulate their rangelands by traditional by-laws.However, these by-laws have been violated since the 1973 by inappropriate intervention policies of the government.As a result, such inappropriate intervention policies might have contributed to explain the current impacts on rangeland use and cover in the Borana area.
Earlier studies (Angassa and Oba, 2008;Coppock;1994;Oba et al., 2000) have shown that an increase in the abundance of woody vegetation cover could have been driven by the ban of fire and continuously high grazing pressures.On the other hand, there is expansion of woody plant cover in the areas where there is little grazing due to aggressiveness of some species (Brown and Archer, 1999).In the present study, encroachment of woody cover across the Borana rangeland is highly pronounced so that the shifts from grassland to woodland would be approximately five-times greater in just three decades.These changes might have significant implications on pure pastoralism practices that rely on livestock production for the most part.The recent demand for land use forms for farming purpose has certainly led to a push for more use of land for large scale mechanized farming along with the expansion of large private farming activities in semi-arid areas.In those areas, climate variability plays a major role in regulating ecosystem function.Since drought represents one of the major climatic factors affecting negatively LULC changes therefore it is more likely that traditional land-use practices might be influenced by drought in Borana.This means that the observed LULC changes in the Borana rangelands might adversely affect ecosystem dynamics.The latter may have a negative impact on livelihoods of pastoralists in turn.As a result, to sustain pastoralists' livelihoods support systems and masteries normal ecosystem functioning in the Borana rangelands the following important issues should be addressed: a) careful implementation of initiatives related to pastoralist driven land use policies that are ecologically sound, and b) regulating population growth in the rangelands.

Conclusions
Providing key information on LULC changes is vital for understanding land use dynamics and monitoring resources over time.Gathering such information could contribute to policy makers with insights to make informed decision over land use planning and enhancing pastoralists' livelihoods through proper support.This study has demonstrated the usefulness of satellite remote sensing image used in producing land-use/cover maps and changes in the Borana rangelands for the past 30 years.It is evidenced that: i) the proportion of woodland cover increased in the year 2000 compared to the 1973 whereas, the grassland proportion declined for the same period; ii) the increase in cultivated land/built up area represents a recent phenomenon related pastoralist grassing systems; iii) bareland increased affecting therefore the productivity of the rangeland and eventually the livelihoods of the pastoral communities in southern Ethiopia; and iv) drought, population pressure, inappropriate government policy and mismanagement of rangelands are the major driving forces of changes in land-use/cover in Borana rangeland areas.
In the case of Ethiopian rangelands, drought, human and livestock's population increase, conversions of rangelands to crop land in valley bottoms represent the main drivers' forces of land cover changes.Such changes have impacted pastoralists' livelihoods.In order to survive, pastoralists have shifted their land uses from grassland to woodland and other land use forms (Coppock, 1994;Gemedo, 2004).In the long run, pursuing such land uses shift might not be sustainable approach because additional rangelands will continue to be degraded.Appropriate response to the issue of further degradation of rangelands calls for providing a careful implementation of pastoralists' land use policies that are not destructive of the ecosystem and its functioning.
Further studies should focus on promoting lower stocking rates and reducing human and livestock population's growth through educational campaigns and raising public awareness on the necessity of conserving the ecological functioning of the ecosystem and sustainable livelihood of pastoralists in the Borana rangelands of southern Ethiopia at the same time

Figure 1 .
Figure 1.Location of the study area in Borana rangeland in Southern Ethiopia.
1986 and by 26.01%from 1973 to 2003.In the contrary, woodland exhibits an increase of 28.01%from 1973 to 1986 and 37.92% from 1973 to 2003.The other land use did not have such a big change (Table

Figure 3 .
Figure 3. Trend of NDVI changes in the Borana between the 1973 and 2003.

Table 1 .
Satellite images acquired for three different times in the Borana rangelands.

Table 2 .
LULC classes used and their brief definitions in Borana rangelands.

Table 3 .
Classification of land-use /cover changes in the Borana rangelands of southern Ethiopia.

Table 4 .
Land-use/covers changes following Transition probability matrix from the year 1973 to 2003 in Borana, Southern Ethiopia.

Table 5 .
Population, land consumption rate and absorption coefficient changes in Borana Southern Ethiopia.

Table 6 .
Pastoralists' perception towards rangeland cover changes in the study area.

Table 7 .
Correlations between the different driving forces with dependent variables such rainfall.

Table 8 .
Correlation between the different driving forces with dependent variables such year time serious.