Change detection is a process in which aerial extent and spatial distribution of land features of two periods is analyzed [11] [39] [40] [41] . Change detection in this study was conducted for two periods. The first period range between 2000 and 2010. The second was between 2010 and 2015. The extent of change and rate of changes calculated following equation employed in [30] .

Land change Modular tab used to analysis change detections in terms of gain, loss, net change and net contribution based on land-use land-cover maps of two different periods [11] . Contribution to the net change explains the question about which class is changed to what and by how much to provide complete picture of land-use land-cover dynamics [41] [42] . Transitions detection in this study was conducted for two periods. The first period covers transitions that occurred between 2000 and 2010 whereas the second included transitions that moved from 2010 to 2015. In this study a contribution to net change analysis conducted only for a forest because the forest is the main focus of this study

The transition from the forest class to other classes like a farmland, shrubs, settlement and woodland generated based on LULC maps of 2000 and 2010 using LCM. Those four transitions were added to transition sub-model and named as the deforestation that referred to as the categorical dependent variable. The maximum number of transitions is equal to a number of classes indicated in input maps [11] [43] . However, in this study only four transitions were considered while the transition from forest to bare land was not considered because its transition found insignificant (Table 3).

Determination of deriving factors is prerequisite step to conduct future prediction

Source: Own analysis using IDRSI software.

[44] [45] because model is a tool to analyze relationship between deforestation and associated derivers [46] . Deforestation can be influenced by spatial factors such as elevation, slope, roads and soil type [47] [48] . Moreover, it has remarked that gentle slope, soil fertility, proximity to settlement and water sources are favorable conditions that attract people to convert more vegetation to farmland areas [11] [49] . A part from this, it is noted that selection of the spatial variables heavily depends on availability of reliable data and ability of variables to express past and future dynamics [35] [50] . Considering this fact, six deriving factors considered like elevation, slope, soil, road, settlement and river. Hence a digital elevation models with 30-meter resolution downloaded from the ASTER’s website using link http://gdem.ersdac.jspacesytems.or.jp. It was projected to WGP-84 and 37 North. Geo-referencing and mosaic were conducted using ENVI4.3 software. Mosaic was conducted because a single scene did not cover the whole study area. The mosaic scene was clipped using shape file of the study area with use of ENVI4.3. Finally, DEM file converted to IDRSI file format. Additionally, the slope factor calculated from DEM layer with use of IDRSI software (Figure 2). The road and river data were clipped from the data set of Ethio-GIS using ArcMap GIS 10 while settlement data of each village was accessed from GIS data base of Farm Africa -SOS Sahel Ethiopia, Bale REDD+ project (Figure 2). Those three layers imported to IDRSI file format and rasterized. Subsequently, distance from road, river and settlement were separately calculated using IDRS Software. The soil type layer was clipped from the Ethio-GIS database and imported to the IDRSI file format and rasterized to make suitable for deforestation modeling (Figure 2).

The Cramer’s V statistics employed to calculate the powers of all six factors with the use of IDRSI software. Cramer’s V statistics is correlation coefficient that ranges from zero to one [11] [46] . The value greater than 0.15 is considered as acceptable while value greater than 0.40 is more accept for modeling [46] [51] . Prior to testing power of all variables in this study, layer of soil type linearized to continuous data using Evidence Likelihood because of its categorical data that could not be accessible to MLP algorithm. Powers of all six deriving variables were tested as indicated in Table 4. However, only three factors were considered such as DEM, Soil type and proximity to road based on v Cramer’s statistics that exceed 0.40. These deriving factors are categorized as static and dynamics variables. Static variable is variable which is not changed over the time while the dynamic variable is time-dependent that continue to change over a time [52] . It is important to note that the dynamic variable requires updating in process of a modeling.

Transition potential maps use to estimate susceptibility of the pixels from one class to other by influence of deriving factors [42] [46] . The transition potential map for each pixel contains probability value that ranges from zero to one whereas a large value indicates high rate of vulnerabilities [11] [49] . Transitional potential maps from a forest to farmland, shrubs and settlement were produced for this study purpose. These transitions maps produced based on the land-cover maps of 2000 and 2010 along various factors like road, DEM and soil type using multilayer perception algorithm. This conducted achieving an accuracy of 89.16% with RMS value of 0.28 while a default value of 10,000 iterations completed for both training and testing. In general, accuracy of around 80% is acceptable in predication of LULC [43] [51] .

Multilayer perception is preferred because of its capability of modeling more than one transition at time and transforming categorical data to continuous data [46] [53] . Multilayer perception algorithm trains two sets of classes allocating 50% of samples for training and 50% to test the model. Accordingly, it trains pixels that have undergone transitions from first land category to the next classes.

Source: extracted data analysis.

Secondly, it trains set of pixels that persisted from a first time to the next without change. This algorithm adjusts a model parameters through repeated iteration for training and testing until stopping criterion is satisfied in which the RMS error decreases as weight adjusted and accuracy rate increases [53] [54] .

Deforestation predictions in this study were conducted for 2015 and 2030 based on transition potentials maps and Markov chain transition probability. Prediction for 2015 conducted for the purpose of model validation while a prediction of 2030 was conducted for purpose of analyzing a future scenario. Markov chain transition probability is used to generate probability that helps to predict future scenario based on LULC maps of two periods [50] [55] [56] .

Land change modular provides predictions maps of hard and soft predictions. The hard prediction yields a projected map at a certain year in which each pixel is assigned to a certain land class [46] [49] . The soft prediction model uses to indicate vulnerability map in which each pixel is assigned a value from zero to one [49] [51] [53] . Probability value with lower value refers to less vulnerability while the higher value indicates high vulnerability to the change [11] [49] .

Model validation is required to assess predictive ability of a model to predict what would happen in future [50] . It can be conducted comparing a projected map with a reference map [57] . Model validation for this study purpose was conducted by comparing the predicted map of 2015 with actually classified map of 2015 using IDRSI software. In the process, the actual map of 2015 was used as the reference map while a simulated map served as a comparison. The validation results are usually expressed in terms of Kappa indices which show agreement and disagreement in quantity and location between pair of two categorical maps [47] [58] .

Proximate causes are immediate human activities at a local level that directly cause deforestation or land-cover change [65] [66] [67] . At the global scale, agricultural expansion, wood extraction and expansion of an infrastructure were identified as the main proximity causes of deforestation [65] . Likewise, in this study proximity derivers were identified as farmland expansion, fuel wood extraction, settlement increase and investment land expansion.

About 52% of total 60 PRA’s participants responded that agriculture expansion was the major cause of deforestation in the case of Delo Mena district (Table 8). This finding strongly supports to the remote sensing analysis that reveals that a farmland increased by 62% between 2000 and 2010 while it increased by 35% between 2010 and 2015 (Table 6). This study finding provides empirical evidence to theoretical argument that explains agricultural expansion is the most direct driver of land-cover at the global scale [67] . On the other hand, settlement expansion was perceived as proximity cause of deforestation by 18% of PRA’s participant. Key informants added that inline to a population increase individuals were clearing the forest and woodland areas to establish own settlements that resulted in spontaneous deforestation. This supported to argument of [23] that explained a rapid population growth resulted in expansion of a farmland and settlements that threaten high conservation values of the area.

In this study area, expansion of the investment land identified as cause of deforestation by 13% of PRA’s participants (Table 9). The participants of PRA explained that many forestlands areas converted to agricultural land by investors particularly in Baraq, kelegolba and Nanga-dherakebeles. However, key informants

Source: Summarized finding based on Participatory Rural Appraisal data.

Source: Summary of scores from Participatory Rural Appraisal data.

argued that a proportion of forest area that was converted to investment land not significant as compared to expansion for subsistence farming. Besides this, 17% of PRA’s participants perceived that fuelwoods consumption particularly charcoaling was identified as one of the main causes of forest degradation (Table 8).

Furthermore, PRA’s participants explained that a charcoaling activity was commonly practiced in rural areas by poor people to generate their own incomes. This supports to argument that explains fuelwood and charcoal productions are not only source of energy but also serves as means of income generating [9] . The key informants explained that demand for charcoal in urban areas became increasing because of absence of alternative energy sources. Therefore, increasing demand for charcoal resulted in rising of price that attracted illegal charcoal traders to motivate rural people to produce more charcoals that continuously supplied to urban areas. In general, it was confirmed that there was an increasing consumption trend of fuelwood. This was strongly support to a prediction that made at national scale. The prediction indicated that a fuelwood increase by 65% between 2010 and 2030 that implied a forest degradation by removing more than 22 million tons of woody biomass [68] . In addition to this, the study finding supported to study finding of [69] that reported 15 to 29 kg of fuel-woods consumption in a week to produce meal of 13 to 26 adult. In the meantime, it was also reported that adoption of fuelwood saving stoves contributed to a reduction of fuelwood consumption by 28.6%.

Underlying forces of deforestation are indirect factors that aggravate effects of proximity causes [67] [70] . In this study, the underlying derivers identified as demographic, economic, policy and institutional factors. Of the total PRA participants 37% perceived that population increase was the key underlying deriver of deforestation (Table 8). This supports to the Neo-Malthusians school of taught which argues that population increase results in environment destruction unlike optimistic taught of Boserupians that argues population growth helps to develop technological innovation which helps efficient utilization of resources [9] . The cause why population size was increasing explained by key interview informants and PRA’s participants indicating that insufficient access to family planning and in coming of migrants from other places like from Harragae, Shawa and Gonder. According to [71] , population of Delo Mena District in 2007 was 181,537 individuals and estimated to reach 219,297 in year 2015 at annual growth rate of 2.6%. Therefore, comparing population of 2015 to the available farmland by 2015 (Table 5) implies that an average farm holding size will be 0.01 km²/person. This indicted that there was shortage of a farmland that could not support farmers to produce sufficient yield.

In this study, a market force and demand for agricultural land investment identified as main economic factors that aggravated deforestation. Accordingly, 13% of PRA’s participants perceived economic force was underlying causes of deforestation (Table 8). The key interview informants confirmed that a vegetation land was converted to a farmland for the production of cash crops like the sesame and coffee. This finding supports to a theoretical argument that explains higher market price for crops leads to an increase of deforestation [72] . This indicates that deforestation is not only occurred because of subsistence farming but it also occurs because of economic reason. Besides to this investment land expansion was indicated as one of underlying causes of deforestation. Key interview informants indicated that investment expansion particularly in the woodland areas created sever deforestation.

Many misdirected policies that poorly defined property right of forest resources led to loss of large forest areas in many developing countries which was ended up with tragedy of open access situation [65] [72] . The Tragedy of open refers to the condition in which resources do not belong to anyone where resources are indefinitely exploited [9] . In the study site, the key interview informants thoroughly explained that denial of forest customary rights put a forest under situation of open access where sever deforestation happened. This was happen because communities restrained themselves from a forest protection in light of losing a forest ownership right. On the other hand, though there was claim that a forest belong to a state, a government’s machineries could not reach every place to make sure forest protection. Therefore, 25% of the PRA’s participants responded that policy factor was one of underlying cause deforestation (Table 9).

The finding strongly supports to the finding of [73] that was reported before 1930s, the forest resource in the Delo Mena District was belong to local community and managed sustainably. Similarly, it was argued that in Ethiopia around 1980sa forest management arrangement overlooked needs and access rights to a forest [74] . In general, the finding supports to a conclusion in many countries the governments had nominal control of the forest control but were found weak in a forest protection [65] [72] .

Apart this, the key informants interviews indicated that the Participatory Forest Management was introduced in 2007 by the Farm Africa and SOS Sehal Ethiopia. As it is witnessed by the key informants through PFM process, customary rights have restored in which rights and responsibilities are clearly defined that contributes to decline in a rate of deforestation. This was also confirmed in remote sense data analysis that indicated deforestation was reduced to 1% between 2010 and 2015 against a rate of 2% between 2000 and 2010 when PFM was not either initiated or at infant stage. The finding was strongly support to contribution of PFM that reported by [75] that argued a forest cover in Ethiopia was increased up to 15.6% between 2001 and 2006 in the PFM areas while decreasing up to 16% was detected in non PFM areas.

Institutional factors refer to a government’s organizations that are responsible of law enforcement [65] [76] . Failure of institutions to deliver its responsibility can lead to deforestation. In this study, institutional factors perceived as underlying deriving forces of deforestation by 10% of PRA’s participants (Table 8). The implications of this factor are explained by key informants. Accordingly, they explained the implications from two perspectives:

Primarily, it was explained that Office of Agriculture at the District level was encouraging farmers to expand their farmland size at the expense of a forest land and woodlands under a motive of ensuring food security. This was real, the paradox to a food security strategy of Ethiopia that launched in 1996 and revised in 2002 which paid attention to environmental rehabilitation and forestry management [77] . This clearly has revealed that how sectoral institution at grass root level contradicts with the national policy. Secondly, it was explained that a forestry sector that supposed to administer forest resource was less equipped in terms of manpower and logistic including frequent restructuring that undermined its efficiency. This view strongly supports to what was reported by [74] that explained the forestry institution was gone through frequent restructuring more than 20 years both at the Federal and Regional levels that undermined institutional efficiency that led to continuous deforestation.

Forest provides variety of social and economic benefits, ranging from easily quantified economic values to ecological services [78] . However, these services became under the pressure because of increasing trend of deforestation. In this study area, climate change, loss of biodiversity, soil erosion with poor soil fertility, scarcity of fuelwoods were identified as major impacts of deforestation based on perception of local community (Figure 13). About 20% of the PRA’s participants perceived that soil erosion and soil fertility loss mentioned as clear impacts of deforestation. The key informants explained that poor soil fertility caused by soil erosion resulted in low crop production and productivity which led to food insecurity.

Besides to this, 40% of PRA’s participants perceived that scarcity of fuelwoods was the top impact of deforestation. Key informants indicated that location of remaining forests was become far away from settlement areas. Therefore, local community, particular women were obliged to walk a long distance to collect fuelwoods. This implied for more workload and physical exhaustion to women. The finding supported to argument of [79] that reported a women spent 3 - 7 hours every week to collect fuel woods that let them to loss their productive time. Moreover, it was pointed out that a price of fire wood increased much and beyond the purchasing capacity of some low income section of the society.

In this study, 13% of PRA’s participants perceived that deforestation resulted in climate change (Table 9). The Key informants explained that they observed variability of climate elements such as shortage of rain fall and persisted drought that resulted in decline of agricultural production and productivity. The remote sense data analysis revealed that between 2000 and 2015 about 21,068 ha of forest converted to other land uses (Table 5). This implied for carbon dioxide emission of ten millions of tons at a rate of 510-ton carbon dioxide/ha that was estimated for moist forest of Bale Ecoregion [80] . In terms of monitory value, it implies for a loss of USD eighty five million, if a ton of carbon dioxide emission is supposed to sell for USD five per a ton.

The forest loss of 21,068 ha had a negative implication for biodiversity of a forest resource. Key informants explained that charcoal producers used to cut Acacia species and Combertumsps to produce charcoal which resulted in depletion of those preferred tree species. Additionally, it was explained that an expansion of agricultural investment resulted in a loss of plants species particularly in a lowland areas like at Beraqkebele. It was explained that investors converted woodland areas that used to serve as habitat of wildlife and pasture for cattle grazing. The loss of biodiversity which can be described in terms of loss of grasses and tree species has reduced its potential uses to feed cattle and bees’ colony. This resulted in a low productivity of livestock and low production of forest honey including decreasing in the size of cattle holding. In light of all these, 27% of PRA’s participants perceived that deforestation resulted in biodiversity loss (Table 9).