Random and systematic land-cover transitions in north-eastern Wollega, Ethiopia

RANDOM AND SYSTEMATIC LAND-COVER TRANSITIONS IN NORTH-EASTERN WOLLEGA, ETHIOPIA Africa has seen significant changes in land cover at different spatial scales. Changes in Land Use and Land Cover (LULC) include deforestation and subsequent use of the land for arable cropping, conversion to grassland or urbanization. The work reported in this article was conducted to examine land cover transitions in north-eastern Wollega (Ethiopia) between 2005 and 2015. The analysis focused on land cover transitions that occurred systematically or randomly, and identified the main drivers for these changes. Landsat data from 2005 and 2015 were examined to better unders-tand the various dimensions of land cover transitions, namely: swaps, losses, gains, persistency and vulnerability. Results showed that shrubland exhibited the largest gain (22%), with a 63% gain-to-loss ratio, a 47% gain-to-persistence ratio and a positive net change-to-persistence ratio of 46%. Cropland showed the largest loss (19%) while grassland was the most stable type of land cover des-pite some fluctuation ( ≈ 10%) observed during the 10-year period. The land cover transition was dominated by systematic processes, with few random processes of change. Systematic land cover transitions such as agricultural abandonment and vegetation re-growth were attributed to regular or common processes of change. This study suggests that the implementation of practices conducive to sustainable intensification of existing agricultural land, supported by policies that promote increased diversification of Ethiopian agriculture, would mitigate pressure on forests by avoiding their future conversion to cropland.


Introduction
Studies using remote sensing technology have shown that land-cover changes are mainly due to deforestation, expansion of cropland areas, including areas used for pastures, urbanization, and increased desertification (Carmona and Nahuelhual, 2012;Belay et al., 2015).Such changes in land-cover are more significant in tropical regions compared with areas outside the tropics (Lambin et al., 2003).For example, the rate of deforestation in tropical countries was estimated to be approximately 7 million hectares per year on average between 2000 and 2010 while the rate of forest re-growth was estimated at about 1 million hectares per year over the same period (FAO, 2015).By contrast, the area used for cropping in tropical regions exhibited a gain of approximately 6 million hectares per year during the 10-year period indicated above.Globally, the expansion of agricultural land has mainly occurred at the expense of forests and natural areas of non-agricultural use, which have shrunk by approximately 16% and 5%, respectively (Brink and Eva, 2009).Underlying factors affecting conversion of forest into cropped land are: population growth, socio-economic changes, active policy intervention or lack of appropriate policy measures aimed at protecting the natural environment, and technology changes (Lambin et al., 2003).
Recent studies conducted in Ethiopia (e.g., Messay, 2011) showed that the main land-covers are: woodland (27%), shrubland (21%), cropping (19%), grassland (12%), natural forest (4%), and other land uses (17%), which includes urban and industrial areas, mining areas, and mountains.Land-cover changes such as expansion of subsistence crop production into the ecologically-fragile environments of the Ethiopian highlands and deforestation have been significant over the past few decades (Kassa et al., 2011).These changes in land-cover have contributed to increased soil erosion rates and reduced land productivity, and are mentioned to have adverse effects on climate change (Bewket and Teferi, 2009;Nyssen et al., 2015).In eastern Ethiopia, Mohammed (2013) also showed that the area of cultivated land expanded as a result of deforestation, which has resulted in increased erosion in the catchment of Lake Alemaya.Demeke (2013) quantified and mapped the extent and spatial patterns of land-cover over a 30-year period in the Borodo watershed in central Ethiopia, and showed that open lands expanded because of progressive removal of shrubs and forest.Further work by Yodit and Fekadu (2014) examined land-cover changes in the Kebribeyah district of the Somali Region of Ethiopia and reported greater settlement areas due to population growth.Other drivers of land-cover changes identified by Yodit and Fekadu (2014) were: price incentives, intervention by development agencies, change in land tenure and development of urban infrastructure, attitudinal change of pastoralists, and deforestation.Similar observations in terms of land-cover dynamics were also reported by Amanuel and Mulugeta (2014) for the Nada Asendabo region in south-western Ethiopia.A requirement for further work has been identified to determine dominant land-cover changes, when and where, and at what rate these changes take place.Therefore, this study investigated land-cover changes observed in north-eastern Wollega and discussed processes of land-cover transition associated with those changes.
Accurate information on processes driving land-cover changes may be obtained through the detection of dominant land-cover transitions (Pontius et al., 2004;Braimoh, 2006).Land-cover transitions can be classified as random or systematic based on the rate of change (Pontius et al., 2004;Braimoh, 2006).Abrupt or unique processes of change in land-cover are regarded as random (Carmona and Nahuelhual, 2012).Random land-cover changes are caused by unexpected factors such as spontaneous migration, conflicts, insecure land tenure, changes in socio-economic conditions, as well as factors influencing agricultural production (Lambin et al., 2003).Systematic transitions are characterized by permanent, stable or common processes of change.Population growth, largescale commercial agriculture and subsistence agriculture, lack of public awareness on environmental management-related issues, and changes in functions of institutions that control the right to use natural resources are often regarded as the main factors driving systematic transitions (Siren, 2007).North-eastern Wollega (Ethiopia) has a relatively rich natural environment in terms of its diversity, and has experienced significant changes in land-cover (Urgesa et al., 2016).For the area relevant to our study, there appears to be a knowledge gap in identifying systematic transitions and understanding the determinants for such transitions.Therefore, the objectives of this study were to: (i) identify systematic and random transitions and their determinants, and (ii) explore the implications of those transitions for land management and food security in Northeastern Wollega, Ethiopia.

Study area
The study area (figure 1) is located in north-eastern Wollega and within the Horo-Guduru Wollega region (Ethiopia), which covers an area of 14,979 ha that extends from 9°45' N to 10°00' N, and from 37°00' E to 37°15' E, respectively.The mean annual temperature is 27°C and the mean annual rainfall is approximately 1,900 mm.North-eastern Wollega is the source of the Hangar River, which is the main tributary of Nile River (locally known as Abay).Dominant tree species include Juniperus sp.(Tid), Podocarpus sp.(Zigba), Olea europaea (Woyera), Rosa abyssinica (Kega) and Carissa edulis (Agam).Nitosols are the dominant soil type on undulating ground and steep slopes (Adugna and Abegaz, 2016).Relatively flat river valleys have well-developed Vertisols, whereas Regosols and Cambisols are also found on steep slopes areas.The area has a population of ≈59,000, which has increased at an average rate of ≈3% per year.Agriculture is the primary form of occupation for about 90% of the population, and includes both cropping and livestock farming.The main crops are cereals, which are produced for subsistence.Livestock is usually tethered and uses the ''cut and carry'' feeding system.Cut and carry, also known as zero-grazing, is a feeding system consisting of cutting, collecting and transporting the forage to the cattle, which is maintained indoor.Horse-carts are used for carrying the harvested forages from the field.This system is commonly practiced by smallholder farmers.

Data source and land-cover sampling
Landsat images from 2005 and 2015 were used to identify land-cover transitions.The Enhanced Thematic Mapper-Plus from Landsat 5 (February 2005 and February 2015, path 168, row 55) was the source of data.February was selected because it is the driest month in Ethiopia when the sky is free from clouds obstruction and sensors are able to capture untainted surface reflectance.Using Band 3 (visible 0.63-0.69µm) and Band 4 (near-infrared 0.75-0.90µm), the images were radiometrically corrected based on sensors' calibration parameters provided by the satellite images and Equation 1 (Awdenegest and Holden, 2009): where: L is the radiance (Wm -2 sr -2 ), and DN is the digital number, subsequently used to calculate the NDVI, as follows (Equation 2): Where NDVI is the Normalized Difference Vegetation Index, NIR is the spectral reflectance measurements acquired in the Near Infrared channel, and R is s the spectral reflectance measurements acquired in the red channel.NDVI data were analysed to define cover classes.Generally, NDVI ranges from -1 to 1, where higher values reflect higher vegetation cover and lower values represent shrubland, grassland, cropland and settlement.Landsat Enhanced Thematic Mapper (ETM+)-derived NDVI data were categorized into 5 land-cover classes listed above based on the values from the literature (e.g., Smith et al., 2007) and field observations.Land-cover classification was performed using the supervised maximum likelihood algorithm classifier to allocate pixels to land-cover classes (Oudraogo et al., 2015).After the land-covers were classified, changes were calculated and maps were prepared.ArcMap geographic information system was used for analysis and preparation of the maps.
The two maps (2005 and 2015 Landsat images) were geometrically rectified into the Universal Transverse Mercator-World Geodetic System (UTM-WGS 84 Zone 37) ahead of change detection.This was performed using topographic maps and digital elevation models derived from Landsat images.A tasselled cap orthogonal transformation of the original bands was performed to improve the visual classification of landcover types.This process required a thorough ground truthing, which was conducted in 2015 to include collection of ground control points, training areas, ground-check and validation.The number of samples (n) for training areas and the validation for each of the land-cover classes was selected by applying Equation (3) from the multinomial distribution (Congalton and Green, 1999), as follows: ( ) where: B=6.63 is the upper x 100 th k α percentile of the x 2 distribution with one degree of freedom, π denotes a 0.01 level of significance, and π i (i = l…,k)is the proportion of the population in the i th category, b is the absolute precision of the sample, k is number of land-cover categories.Equation (3) is calculated for each of 'k' categories and 'n' for all categories is subsequently selected proportional to the size of the category.The analysis used a 95% confidence level, which is considered to be appropriate based on related studies, and the absolute precision was set at 0.05.For the study site, 'n' from the smallest class was chosen as the desired sample size to match Congalton and Green (1999)'s rule of thumb of a minimum of 50 samples.Stratified random sampling methods were used to collect an optimum number of sample reference polygons for the classification accuracy assessment.The population (sampling frame), land-cover classes, embraces distinct categories (separate strata): forestland, shrubland, grassland, cropland, and settlement, respectively.Each stratum is subsequently sampled as an independent sub-population from which individual elements can be randomly selected (table I).Producer's accuracy (measure of omission error) was computed by dividing the total number of correct pixels in a category by the total number of pixels of that category as derived from the reference data (the column total).This method illustrates how well a certain area can be classified or indicates the probability of a reference pixel being correctly classified.The user's accuracy, that is, a measure of commission error is computed by dividing the total number of correct pixels in a category by the total number of pixels that were classified in that category (the row total).This provides a measure of reliability and indicates the probability of a pixel classified on the map or image actually represents that category on the ground.The overall accuracy was 82% for 2005 and 86.7% for the 2015 land-cover maps.
The types of land-cover within the study area were classified into five categories, based on previous work by Lambin et al. (2003).These categories were: (1) forestland (high and dense trees with 70%-100% closed canopy), (2) shrubland (bush canopy mixed with some trees, closure areas, young forest plantations and agroforestry), (3) grassland (mainly grasses without long-fallowed land or with less than 10 trees per ha), ( 4) cropland (agricultural land with crops as well as harvested agricultural land), and ( 5) settlement (built-up areas).
Note: "C" is any conversion from one land-cover to another.LULC is land-use/land-cover.

Land cover transition matrix
Land-cover transitions were analyzed using a transition matrix obtained from cross-tabulation of the two-raster maps presented in figure 2. The approach of Pontius et al. (2004) was used to identify major signals of systematic land cover changes.The general structure of the transition matrix was shown in such a way that the columns and rows presented the proportions of the five land-cover classes in 2005 and 2015, respectively (table II).The main diagonal elements (i.e., C jj ) designate the proportion of persistent land-cover classes that showed no changes.Diagonal elements were used to calculate the gains and the losses of land-cover classes (Pontius et al., 2004).The off-diagonal elements showed the amount of the land-cover transformed from class i to class j between 2005 and 2015 as calculated using Equations (4a-b) (after Pontius et al., 2004;Ouedraogo et al., 2010) where: n is the total number of classes, C ij (where i # j) designates the proportion of the landscape that experienced a transition from class i to class j between 2005 and 2015, C i+ is the proportion of land-cover in 2005 and C +j the proportion of land-cover in 2015.
The "loss" column indicates the amount of land-cover that experienced a net loss of class i between 2005 and 2015, whereas the "gain" rows show the amount of the land-cover that experienced a gross gain of class j between the same two years (table II).Swap is the exchange between classes, and denotes concurrent gain (i.e., difference between class i and persistence) and loss (i.e., difference between class j and persistence) of a given land-cover class.Equation (5) describes the proportions of swap, which required the pairing of each pixel that loses with those pixels that gain (Braimoh, 2006): where: S j is amount of swap of land class j, 2 min is twice the minimum of the gain and loss, C i+ is the proportion of landcover in 2005, C +j the proportion of land-cover in 2015 and C ij (where i # j) designates the proportion of the landscape that experienced a transition from class i to class j between 2005 and 2015, respectively., respectively (Ouedraogo et al., 2010).The terms g, p, and l stand for gain, persistence (that is areas of a given category that remained unchanged), and loss, respectively.Values of G p and L p greater than one imply that a given land-cover class has a higher probability to change to other land-cover class than to persist in its current condition (Braimoh, 2006).If the value of N p was negative, the land-cover class would have a higher probability to lose area to other land-cover classes than to gain from them.

Detection of systematic and random transitions
Land-cover data from 2005 and 2015 were used to analyze recent land-cover transition processes observed in north-eastern Wollega (Ethiopia).The procedure employed in our study is based on approaches satisfactorily applied in earlier studies (e.g., Pontius et al., 2004;Braimoh, 2006;Ouedraogo et al., 2010;Ouedraogo et al., 2014;Ouedraogo et al., 2015).Detection of systematic and random inter-category transitions involves four interrelated procedures.The first procedure calculates the expected gain (G ij ) for each class under a random process of gain (Equation 6), as follows (Pontius et al., 2004): 6)   where: G ij is expected gain of each cover type, C +j is the proportion of land-cover in 2015, C jj is the proportion of persistent land-cover classes that showed no changes and C i+ is the proportion of land cover in 2005.Equation ( 6) distributes the gain across the off-diagonal entries of 2015 according to the relative proportions of the other classes in 2005.The assumption of this equation is that the gain for each class and the proportion of each class in 2015 remain unchanged.These expected gains represent the random process of gain.The second procedure estimates the difference between the observed (measured) and expected proportions of gain under a random process of gain.Large positive or negative differences from zero denote systematic inter-category transitions, rather than random transitions, which occur between two land-cover classes.The larger the variance (positive), the larger the area affected by systematic gain of class j from class i, and the larger a negative calculated variance, the weaker the tendency of class j to gain systematically from class i (Braimoh, 2006;Ouedraogo et al., 2014).The third procedure calculates the expected loss, L ji (loss of class i to class j) under a random process of loss using Equation ( 7) (Braimoh, 2006): where: L ij is expected loss, C i+ is the proportion of land-cover in 2005, C jj is proportion of persistent land-cover classes that showed no changes and C +j is the proportion of landcover in 2015.Equation ( 7) assumes that the loss of each land-cover class is fixed and the loss in each row across the other classes is distributed relative to their proportions in 2015.The forth Photo 2. Land-cover transition in north-eastern Wollega, Ethiopia.Photo A. Adugna.procedure calculates the difference between the observed and expected proportions of gain under a random process of loss.Likewise, large positive or negative differences from zero indicate that systematic inter-category transitions, rather than random transitions, have occurred between two land-cover classes.The larger a positive calculated variance, the higher the area affected by systematic loss of class i to class j, and the larger a negative calculated variance, the weaker the tendency of class i to systematically loss to class j (Braimoh, 2006;Ouedraogo et al., 2014).Generally, it can be concluded that the transition from class i to class j is a systematic process if class i loses systematically to class j, and class j gains systematically from class i (Pontius et al., 2004).

Land-cover transitions
In north-eastern Wollega, land-cover exhibited significant changes during the 10-year period investigated in this study, as shown in table III and figure 2. In 2005 and 2015, forestland accounted for ≈24% and ≈11% of the total area, respectively.The area of cropland decreased from ≈63% in 2005 to ≈55% in 2015 while grassland areas remained close to constant during the same period (table III).Shrubland areas increased from less than 1% to 22.5% between 2005 and 2015, respectively.The loss in area of cropland was highest representing almost 20% of the landscape.The gainto-loss ratio was highest (55.0) for shrubland and settlement (1.0), and lowest for forestland (0.2) and cropland (0.6), respectively.Changes attributable to location (swap) were highest for cropland (23.5% of the change in cropland), followed by grassland (14.5% of the change in grassland) and forestland (6.8% of the change in forestland), respectively.Net changes (quantity) were highest for shrubland (≈22%) and forestland (≈13%).

Persistence and vulnerability of land-cover
The proportion of different land-cover classes that remained unchanged (persistence) from 2005 to 2015 is presented in table III.The land used for cropping showed relatively less changes compared to settlement areas (table IV).Among the natural vegetation, forestland and shrubland areas exhibited the highest and lowest persistence, respectively (table III).The vulnerability of land-cover classes to transition exhibited some variability depending on the type of land-cover, as shown in table V.The gain-to-persistence ratio (G p ) was higher than one in all land-cover classes with the exception of forestland and cropland (table V).The G p ratio was highest for shrubland (46.9) and lowest for cropland (0.3).The loss-to-persistence ratio (L p ) for all landcovers, except for cropland and shrubland, was higher than one (table V).The L p ratio was highest for grassland (8.3) and lowest for cropland (0.4).The net change-to-persistence (N p ) was positive for shrubland and negative for the other land-cover classes (table V).The N p value was highest for shrubland (46.1), followed by settlement (-7.2).By contrast, the value of N p was lowest (-0.2) for cropland.

Systematic and random processes of change in north-eastern Wollega
The expected gains for each land-cover class under random process of gain are shown in table VIa, and the difference between observed and expected gains in table VIb.Differences between observed and expected gains for cropland-forestland and forestland-cropland transitions were 0.3% and 2.1%, respectively.This indicated that forestland randomly gained 2.4% of the landscape from cropland, whereas cropland systematically gained 4.9% of the landscape from forestland (table IV).Differences between observed and expected gains for forestland-shrubland and cropland-shrubland transitions were 4.1% and -4.4%, respectively (table VIb).Therefore, this suggested that there was a systematic exchange of forestland and cropland with shrubland.Shrubland gained systematically 9.4% of the landscape from forestland and cropland (table IV).The difference between observed and expected gain for cropland-grassland and grassland-cropland transitions were 0.7% and 3.9%, respectively.This suggested that during the 10-year period there was a systematic exchange between cropland and grassland.Grassland systematically gained 5.2% of landscape from cropland, whereas cropland systematically gained 5.1% of the landscape from grassland.Differences between observed and expected gains for forestland-settlement and cropland-settlement transitions were -0.5% and 0.6%, respectively.This indicated that settlement systematically gained 0.1% and 2.2% of the landscape from forestland and cropland, respectively (table IV).
Expected losses under a random process of loss are given in table VIIa, and the difference between the observed and expected losses in table VIIb.Relatively large differences between observed and expected loss for forestland to shrubland (9.3%) indicated that forestland systematically lost to shrubland.Large negative values of cropland to forestland transitions (-6.5%) imply that loses of cropland to forestland areas were systematically avoided.Large differences between observed and expected loss for cropland to shrubland (9.2%), cropland to grassland (4.5%) and cropland to settlement (0.8%) transitions indicate that cropland tended to lose systematically to other land-cover classes.

Discussion
Results showed that shrubland represented the least dominant form of land-cover in 2005, but this condition was reverted by 2015.The calculated gain-to-loss ratio for shrubland suggested that this type of land-cover exhibited 63 times more gain than loss, and that it will likely continue to increase.The assessment of vulnerability of land-cover to transition also showed a high Gp value for shrubland (47), implying that this land-cover is prone to gain area from other types of land-cover than it is to persist.This relates to conversion of forestland to shrubland, afforestation of land previously not occupied by forest, improvement of shrubland in protected areas, and agricultural abandonment with subsequent expansion of natural forest into areas previously under arable cropping.These processes combined are understood to have assisted the rehabilitation of degraded land in NE Wollega, also noted in earlier studies in NW Ethiopia (e.g., Bewket, 2007).The implementation of the 1997 Environmental Policy of Ethiopia (EPE, 1997) may also explain changes (increase) in the area occupied by shrubs.EPE encourages tree planting including forestry development on farms, around homesteads, and on eroded hillsides.It has also introduced a ban on illegal hunting, bushfire and free mobility of livestock so as to improve secondary forest cover in protected areas.This policy instrument assists adoption of best management practice for natural forests and woodland resources, and encourages sustainable and participatory approaches to natural resource management (Teshome et al., 2014).Loss of forestland is mainly attributed to deforestation aimed at developing new agricultural land as well as logging for firewood and charcoal production (FAO, 2015).The assessment of land-cover transition process showed that forestlands are systematically transformed to croplands implying that farmers have preferentially cleared-up forests to make space for new farmland.Historically, Ethiopian farmers have increased agricultural output through this process, but with proportionally less increases in productivity per unit area (Ndah et al., 2015).Large-scale forest conversion and expansion of livestock-based farming means that the amount of suitable land available for cropping is rather limited in north-eastern Wollega.Thus, as of 2015 cropland areas experienced a loss and tended to decrease during the past decade, but it is still the dominant land-cover in the region.This loss was mainly associated with the abandonment of cropland areas due to soil degradation in the form of erosion, also highlighted in studies in the Gerado catchment (e.g., Bahir et al., 2015), the central Ethiopian Highlands (e.g., Adimassu et al., 2013), and other tropical regions (e.g., Schneider and Geoghegan, 2006;Carmona and Nahuelhual, 2012;Angonese and Grau, 2014).
Cropland areas increased little (7.4%), which was also observed for grassland (1.6%) and settlement (0.1%) areas.A low Gp value (0.3) for cropland suggested that this class is prone to losing area to other land-cover types, and means that there were relatively large swaps resulting from concurrent gains and losses of similar order of magnitude (table III).As a result, there has been a relocation of 24% of the cropland area, which in turn reflects a process of agricultural adjustment.Agricultural adjustment is a process by which agricultural lands are relocated to relatively more fertile or productive areas (Carmona and Nahuelhual, 2012).This process plays a critical role in promoting re-establishment of native vegetation in degraded areas, but it requires some degree of resilience of that environment to enable basic ecosystem's functions to be progressively restored (Ouedraogo et al., 2014).While this process has taken place in north-eastern Wollega, other factors are also noted.For example, spontaneous abandonment of agricultural production by farmers on degraded land is considered to be one of the main drivers behind deforestation reversal and subsequent forest re-growth (Singh et al., 2015).
The findings of our study support the need of Ethiopian farmers to shift towards a process of sustainable intensification of agriculture to ensure long-term protection of the soil resource and advance food security in the region.These observations are in agreement with earlier studies (e.g., Alexandratos, 2005;Pretty et al., 2011).The opportunities for opening-up of new land from forest, wetlands, hillsides or pastures for cultivation of arable crops are largely regarded as non-sustainable practices and therefore should be avoided.At present, no steps towards sustainable intensification of agriculture in the region have been taken.By contrast, the socio-economic context and the agricultural policies in place have traditionally pointed towards extensification (land-demanding strategy).Furthermore, sustainable intensification is perceived as risky by most farmers due to the level of investment required for the provision of means needed to improve productivity.Other limiting factors are also encountered such as inadequate resource allocation, restricted access to technology, lack of qualified labor and the land tenure system (Bahir et al., 2015).Limited extension effort also means that farmers are largely unaware of alternatives practices available for mitigation of environmental degradation.
A study by Wood et al. (2004) suggested that the following is required to drive sustainable intensification of agriculture forward: agricultural products can be commercialized, input and output markets are accessible, return on investment is satisfactory, labor is available, and farmers have financial means to acquire agricultural inputs.Availability of labor in rural areas is reduced when migration from rural to urban areas is significant, and people formerly employed in the agricultural sector join non-farming related sectors (Pretty et al., 2011).Agricultural abandonment has both positive and negative consequences on the environment and rural livelihood (Izquierdo et al., 2009;Barbier et al., 2010).Negative consequences include the disappearance of traditional farming practices, food shortages and associated poverty, and potential invasion of exotic species (Zucca et al., 2015).Positive impacts are also reported and include restoration of ecosystem services and functions, but this requires some degree of intervention (Ouedraogo et al., 2010;Ouedraogo et al., 2014).In our study, agricultural abandonment was regarded as a process that involves steady conversion of the land previously used for agricultural production towards natural vegetation.Our results showed that the transition of landscapes from cropland to shrubland, grassland and settlement was mainly explained by a systematic rather than a random process of change.Based on this, it is possible to state that local farmers appear to preferentially fallow or abandon croplands to make space for pasture production and expansion of agroforestry.This also suggested that there appears to be a shift from cropping to other activities, such as cattle fattening, trade, and handicraft and service provisions.

Conclusions
The main conclusions derived from this work are: The land-cover changes observed in this study denote a trend towards abandonment of agricultural land, which is likely to assist environmental restoration through natural processes.The land-cover classes investigated showed that shrubland areas had the highest gains whereas cropland areas exhibited the highest losses.Consequently, abandonment of croplands and re-growth of natural forest were identified as the main land-cover changes in north-eastern Wollega.
Land-cover changes in north-eastern Wollega have undergone systematic transitions.The main drivers for such transitions were abandonment of cropland areas and clearing-up of forests for arable cropping and grazing.Environmental policy in Ethiopia should aim at rehabilitating degraded land and restoring ecosystems' functions.The effectiveness of such measures needs to be assessed at time intervals appropriate for the land-use and climate.
Land-cover transitions in north-eastern Wollega had positive implications for land management and food security in the region.The rapid expansion of shrubland implies that there has been afforestation of land previously not occupied by forest.Afforestation assisted the rehabilitation of degraded land and improved the establishment of secondary forest cover in protected areas.
The pattern of transition observed in the region suggested improved land-cover and therefore a gradual shift towards improved environmental quality.Extension work is needed to promote practices that preserve the natural forests, and restore the productive capacity of degraded agricultural and grazing lands.Practices such as agro-forestry or silvopasture are therefore encouraged.This approach calls for assisted intensification of agriculture so that food security is not compromised.Agricultural policies also need to be developed to mitigate pressure on forests and avoid future conversion of relatively fragile environments into cropland.

Table I .
Congalton and Green (1999)roducer's accuracy and user's accuracy assessment in North-eastern Wollega, Ethiopia.Note: For shrubland, 50 sample sizes were chosen as the desired sample size close toCongalton and Green (1999)'s rule-of-thumb of a minimum of 50 samples per class since the value was small to consider for the study using Equation (1).LULC is land-use/land-cover.

Table V .
Gain-to-persistence (G p ), Loss-to-persistence (L p ), and net change-to-persistence (N p ) ratio of the land cover classes in the north-eastern Wollega, Ethiopia
CHANGEMENTS D'UTILISATION ET DE COUVERTURE DES SOLS / LE POINT SUR…

Table VI .
Inter-category gains in north-eastern Wollega, Ethiopia: (a) expected gain, and (b) difference between observed and expected.