More people, more trees: A reversal of deforestation trends in Southern Ethiopia

Despite global commitments to forest restoration, evidence of the pathways through which restoration creates social and ecological benefits remains limited. The objective of this paper is to provide empirical evidence to generate insights on the relationship between forest cover change and key provisioning ecosystem services and reforestation pathways. In Southern Ethiopia, three zones along a gradient of decreasing land cover complexity and tree cover were examined. The land cover change was assessed using satellite remote sensing and complemented ground‐based tree inventory. Perceptions of land cover and ecosystem services change and farmer responses were evaluated through three Participatory Rural Appraisals and eight Focus Group Discussions. Since the 1970s, a landscape shift from a forest‐grassland to a cropland mosaic was associated with increased food production, improved food security, and higher incomes. However, this shift also coincided with reductions in livestock, construction materials, fuelwood and water availability, prompting reforestation efforts designed to recover some of these lost ecosystem services. In particular, some households established Eucalyptus woodlots and encouraged natural regeneration. Natural trees, Eucalyptus woodlots, Ensete plantations (a type of plantain), and grasslands were positively associated with homestead proximity; thus, homestead establishment resulting from population increase in this predominately agricultural landscape appeared to foster a viable forest restoration pathway—that is, 'more people, more trees'. This is a reforestation pathway not previously described in the literature. A return to a more diverse agricultural landscape mosaic provided more secure and diversified income sources along with better provisioning of construction materials, fuelwood, and higher livestock numbers.

current biodiversity extinction crisis (Ceballos et al., 2015). Biodiversity loss impacts people's livelihoods in many ways. For example, in landscapes composed of agricultural-forest mosaics, forests and trees provide fuelwood, fodder for livestock, construction materials, and nutrients for crops (e.g., Baudron et al., 2017). Forest products are also an important source of income as well as food, particularly for the poor, and particularly during times of crises (Arnold et al., 2011;Beck & Nesmith, 2001;Wunder et al., 2014). Similarly, because on-farm trees may provide provisioning and other ecosystem services-that is, products and other benefits from the ecosystem to humans-critical for agricultural production, they are often integrated within mixed crop-livestock systems (Reed et al., 2017). Therefore, deforestation along with general loss of trees tends to have negative consequences for rural livelihoods.
Despite claims to the contrary (Boyd & Slaymaker, 2000;Gibbs et al., 2010), reforestation can occur in conjunction with a growing human population and also be associated with increased agricultural productivity (Tiffen et al., 1994). For example, farmer-managed natural regeneration dramatically increased tree density and household income in Niger (Haglund et al., 2011). In a heavily degraded and densely populated region of Northern Ethiopia, rural communities have restored tree cover and improved soil protection over the last 140 years, largely through the support of environmental recovery programs during the 1980s (Nyssen et al., 2009). In the Ethiopian Central Rift Valley region, farmers have also recovered tree cover, ground vegetation and related ecosystem services by implementing livestock exclosures (Baudron et al., 2015). In Southern Ethiopia, farmers actively manage their trees to promote ecosystem services and reduce disservices (Ango et al., 2014).
The forest transition literature describes several, sometimes interacting, reforestation pathways which lead to contrasting environmental and societal outcomes (Lambin & Meyfroidt, 2010;Meyfroidt et al., 2018). Understanding these transitions is key to guide interventions for improved sustainability of land systems while avoiding negative outcomes (Meyfroidt et al., 2018). Indeed, global forest and landscape restoration efforts are often assumed to lead to improved livelihoods and wellbeing, yet evidence on this remains limited. Our understanding of the direct and indirect pathways by which restoration creates social benefits is also limited (Erbaugh & Oldekop, 2018).
Examining restoration activities undertaken and led by communities and farmers can provide a useful perspective and potentially, actual evidence of such benefits.
The objectives of this paper are to explore empirical evidence that: (a) generates insights on the relationship between forest cover change and key provisioning ecosystem services; and (b) helps identify forest transition pathways leading to overall forest gain and its socioecological outcomes. For this, three agricultural landscapes spanning a gradient of tree cover and land cover complexity in Southern Ethiopia were examined. A suite of integrated methods-from participatory surveys to remote sensing to biodiversity inventories-designed to capture spatial and temporal dynamics of ecosystem services and human wellbeing was used .

| Study area
The study landscape is located between the Munesa State Forest and the town of Arsi-Negele-between 38 42.14 0 and 38 49.92 0 East and 7 15.05 0 and 7 22.57 0 North-in Oromia Region, Ethiopia (Figure 1).
Munesa forest is considered a dry Afromontane forest, it is owned and regulated by the Ethiopian State and it can be accessed and used by the neighboring communities with certain limitations. Mixed croplivestock farming is the main livelihood strategy with the main crops being maize, wheat, potato and Enset (Ensete ventricosum), an endemic perennial from the Musaceae (banana) family used to produce starchy food. Unharvested annual crop residues become a communal resource available for communal grazing after grain harvest.
Three zones, each with two villages, were identified along a gradient of decreasing tree cover and land cover complexity (i.e., increasing agricultural specialization; Figure 1). The zones were otherwise similar in social context (land tenure, farm area, ethnicity, local and religious institutions, transport services, population density) and environmental conditions (soil type, elevation, agroecological zone). The first zone borders the State forest of Munesa, which residents can access for grazing and collection of firewood, but live trees cannot be brought down; this zone has a high tree density and is referred to as 'complex' (in land uses and farming system composition) hereafter. The second zone is located 5.5 km away from Munesa Forest and its residents do not have access to the forest due to distance and local regulations. However, they have access to a large communal grazing area; this zone has a relatively high tree density and is referred to as 'intermediate'. The third zone is located about 11 km away from Munesa Forest and its residents do not have access to it or any other common land; it has a lower tree density and is referred to as 'simple'. The complex, intermediate and simple zones are located about 16, 11.5 and 6.5 km away from the main market of Arsi Negele Town, and their total area is 100.1, 164.9 and 110.5 ha, respectively. More information on the context and household and farm variables can be found in Baudron et al. (2017), Duriaux Chavarría et al., (2018 and Duriaux and Baudron (2016). Site selection occurred before the remote sensing analysis based on a coarse aerial assessment using Google Earth, consultation with Ethiopian scientists and field expeditions, to select a study set up as described by Sunderland et al. (2017). 2.2 | Land cover assessment

| Remote sensing
To explore historical changes in land cover, satellite imagery (LANDSAT MSS, TM and OLI) captured in January of 1972January of , 1986January of , 1999 and 2013 were used to classify dense forest (dense tree cover; native forest), medium dense vegetation (degraded open native forest), woodlots (monocrop forests for timber production), cropland (annual crops), and grassland (pastures). Medium dense vegetation refers to pixels with high tree presence, but less than the dense forest class; for example, degraded native forest or non-forest land uses with high tree cover. To distinguish among these classes, a maximum likelihood supervised classification was performed. This included the steps of selecting training sites-representative georeferenced samples of each land use-estimating the spectral signature of each land use based on the training sites, and for each pixel, assigning the class with the highest probability. Training sites were verified using a mix of Google Earth imagery, RAPIDEYE imagery, historical aerial imagery and ground verification. The minimum mapping unit (MMU) was set to the size of the spatial resolution of the imagery (60 m for 1972 and 30 m for subsequent dates). The specific years were selected based on the best available imagery (cloud free during the right dry season when annual crops are harvested).
To explore differences in landscape composition, the contemporary land cover was assessed at high spatial resolution using RAPIDEYE 3A imagery (5 m resolution) from January 2015. The F I G U R E 1 Land cover maps for 1972for , 1986for , 1999for and 2013for using imagery from LANDSAT 1 (1972 and LANDSAT 5 satellites. The borders of the studied zones are marked in black lines, the border of the Munesa State Forest is marked by the red line [Colour figure can be viewed at wileyonlinelibrary.com] landscape was classified into cropland and bare soil, Enset, grassland, natural forest and tree cover and woodlots using a combination of object-based and maximum likelihood supervised classification as well as manual delineation of Enset homegardens and woodlots. Training sites were verified using a mix of high spatial resolution imagery from Google Earth and ground verification.

| Tree inventorying basal area
Between May and September 2015, tree basal area and tree numbers were assessed in the three zones. In each zone, 24 sampling points were randomly selected from a grid (of 150 m sized cells) across the zone (Baudron, Schultner, Duriaux, Gergel, & Sunderland, 2019).
Within a 50 m radius of each point, all trees >10 cm of diameter at breast height (DBH; at 135 cm from ground level) were identified to species level and their DBH were recorded. For each circular plot, the relative basal area was determined using the cumulative basal area of all trees divided by the total ground surface within the circular plot, and then represented as m 2 /ha. Tree density was calculated by dividing the total tree count by the total ground surface of the circular plot and then represented as trees ha −1 .

| Participatory data collection
A participatory rural appraisal (PRA) was held in each zone with 50 to 60 community members of different households and of diverse age, gender and wealth, recruited with the help of the village leaders (Baudron et al., 2011). The PRA goal was to understand the current and historical context through the generation of a timeline of historical events, natural resources and land use maps, diagrams of access to major resources, Venn diagram of major institutions, value chains of main commodities and household typologies based on selfcategorization (Duriaux & Baudron, 2016).
Two Focus Group Discussions (FGDs) were held in each zone with groups of 12 to 14 elders. In the first set of FGDs, trend lines were drawn and discussed by participants describing changes over the last 40 years (Geilfus, 2008). During the second set of FGDs, historical diagrams of land-use change (Geilfus, 2008) for five points in time during the last 40 years were created (see Figure 4).
Another FGD was held with 33 elders from the three zones to identify the main land use and land cover (LULC) change events, discuss their impact on the most important wellbeing elements as considered by the group and identify differences between the zones.
Results from this FGD were similar across all zones and therefore are presented as one compiled table. Finally, a last FGD was held with adults of different ages from the three zones to present key results to the community and receive feedback which was used to improve the results. PRA and FGDs were held between October 2014 and December 2015; equal proportions of male and female participants were invited to the events, having similar proportions in attendance.

| Qualitative data analysis
Land use classification and spatial analysis were undertaken using ENVI 5.0 and ArcGIS 10.3. Change statistics were calculated by comparing the values of the dataset in one period with another (1972-1986, 1986-1999, 1999-2013) For each zone over three periods, annual net forest change was calculated as the change in tree cover area as a percentage of the total surveyed area, using the formula X = ((A 2 − A 1 )/A * 100)/(t 2 − t 1 ), Where: A represents the total surveyed area, while A 1 and A 2 and t 1 and t 2 represent the tree cover areas and the year for the first and second period, respectively (Puyravaud, 2003). For tree basal area data, the Kruskal-Wallis test was used to compare the medians among zones, while Chi-squared contingency tables were used to compare proportions of tree cover. Both were conducted using R software.  The area covered by perennial vegetation-that is, trees, grassland, Ensete and woodlots-tended to increase with increasing proximity to the homesteads (Figure 4). Home gardens generally include Enset and trees, and together with grasslands, are often found near the homestead. Eucalyptus woodlots in the simple zone showed to be an exception, as their area increased with increasing distance to the homestead.

| Tree species and community composition
Of the 50 tree species inventoried, two are endemic to Ethiopia and four are considered of high conservation value by the Ethiopian government. While seven non-native species were encountered, only Eucalyptus species were particularly widespread and often found in woodlots. Tree basal area was highest in the complex zone, followed by the intermediate, then the simple zone (Table 1). The proportion of Eucalyptus from the total number of sampled trees was higher in the simple zone than in the other zones, but there were no significant differences in the proportion of the total basal area or the total tree density represented by Eucalyptus.  1972-1986 1986-1999 1999-2013 Annual net tree conver change (% of toatal area yr -1 )

Complex
Intermediate Simple F I G U R E 3 Annual net tree cover change in three zones along a gradient of decreasing land cover complexity and tree cover. The annual net changeis the annualized percentage that forested areas increased or decreased relative to the total area of a zone. Through reforestation (Figure 3), farmers recovered some of the services lost during land conversion ( Although the LULC change that took place in the study area since the 1970s resulted in increased income and improved food securitythat is, adequate household food provisioning described during FGDs as "...enough food throughout the year for all household members"-they also resulted in a decrease in water availability in rivers, livestock numbers, construction materials and fuelwood availability ( Table 2). As a response to the decreasing availability of these resources-highlighted as key to wellbeing during FGDs-farmers started a reforestation process through farmer-managed natural regeneration and Eucalyptus establishment (Table 1) that the complex zone might represent an optimal landscape in terms of outcomes for the farm and household such as livestock ownership and productivity, fuelwood availability, diversity of income sources, equality, and dietary diversity. During the FGDs, the key elements for rural livelihoods (Table 2) were compared among the three zones with the goal of elucidating which mix of land cover types (e.g., Figure 6a, b) is likely to provide the most desirable bundle of ecosystem services.
The simple zone was perceived as having higher crop and overall food productivity compared to the other zones, but fewer crop types such as Ensete, lower dietary diversity, and reduced livestock products. The complex zone was perceived as having more livestock, livestock productivity and higher dietary diversity . Duriaux Chavarría et al. (2018) found no differences in overall farm or crop productivity between the zones as suggested by farmers, but their measurements agree with the perceptions of larger livestock herds and greater livestock productivity in the complex zone. Baudron et al. (2017) agree in that dietary diversity was greatest in the complex zone due to higher livestock density and more homegardens. Still, promoting the abovementioned benefits might prove difficult to reproduce in other zones without access to common areas and especially forests, which were mentioned as a fundamental source of the services.

| More people, more perennial vegetation: A reforestation pathway
It is commonly believed that population increase leads directly to deforestation (Boyd & Slaymaker, 2000). However, our study demonstrated that population may grow in parallel with increased cover of perennial vegetation. The proportion of trees, grassland, Enset and woodlot cover tends to increase closer to the homesteads (Figure 4).
Based on the association between trees, grassland, Enset, woodlots and the proximity to the homestead, it can be hypothesized that a population increase will lead to an increase in cover by perennial vegetation types. Indeed, it was mentioned during the FGDs that an increase in grazing area and homegardens-which include trees and Enset (Figure 6c)-is expected as younger families establish new homesteads. The aforementioned land uses are associated with daily tasks-fuelwood collection, tending to livestock, milking, collection of manure and cultivation of Enset fields-and this spatial arrangement might be a strategy to centralize assets to guard livestock against theft/wildlife and reduce labour burden by reducing the travel time to carry out daily tasks. Tiffen et al. (1994) in their book "More people, less erosion" described the processes that lead to a natural regeneration and improved environment due to population increase and their actions.
The same phenomenon has been observed in Guinea, where local communities were blamed for a shift in vegetation from pristine humid forest to savannah. It was later demonstrated that population increases actually promoted reforestation in locations where edaphic conditions would otherwise not allow natural establishment and regeneration of humid forest (Fairhead et al., 1995). Nyssen et al. (2009) showed that in Tigray region of Ethiopia, land rehabilitation occurred through an increase of vegetation cover-mainly Eucalyptus-in parallel with a tenfold increase in population. Desalegn et al. (2014) found that in another area of Ethiopia, Eucalyptus cover increased with settlement expansion, resulting in decreased burden of fuelwood transport over long distances-especially for women-and an increase in income as well as provisioning of other ecosystem services such as wind breaks.
Another example of restoration associated with population growth is farmer managed natural regeneration (FMNR). FMNR is based on the regeneration of native trees (Haglund et al., 2011), and although the planting of Eucalyptus was observed as the main mechanism for vegetation regeneration, natural regeneration of native trees also played an important role according to FGDs (Figure 6a, b and d).
Many individual stems of native tree species had small basal areas, an indication they were not remnants of the original forest but likely the result of regeneration. FMNR is more effective when benefits to local communities are the main driver, used in combination with easy and accessible technology to promote diffusion through the community (Rinaudo, 2007). Similarly, in Ethiopia Eucalyptus was established largely without extension efforts because of its economic value and attractive traits; highlighting the 'demonstration effect' in others' farms (Ango, 2010).
Our results demonstrate a compelling juxtaposition with some of the literature on forest transitions. Often, forest transitions occur through interacting reforestation pathways described in the literature as 'forest scarcity', 'state forest policy', 'economic development', 'globalization' and 'smallholder intensification' (Lambin & Meyfroidt, 2010;Meyfroidt et al., 2018). Although in the study area the 'smallholder tree-based intensification pathway' occurred as described in the literature: 'under the influence of smallholder land use systems that actively manage the multifunctionality of ecosystems' (Lambin & Meyfroidt, 2010), restoration did not occur on abandoned land but on productive land. It did not evolve over millennia but occurred within decades. In addition, its main objective was not risk reduction through diversification but the recovery of key provisioning services.

| Farmer-led reforestation, novel ecosystems and biodiversity trade-offs
The simple zone, with the lowest proportion of tree cover, had the highest proportion of Eucalyptus, pointing to the importance of Eucalyptus for tree cover regeneration in the most deforested conditions.
Eucalyptus and Ensete were introduced to the area by local inhabitants during the period under investigation (1970 to 2014). The LULC change process-shift from a highly forested area to an area dominated by cropland with a high proportion of exotic tree species-has led to what could be considered a 'novel ecosystem'. A novel ecosystem differs in composition and/or function from present and past ecosystems and is often linked to LULC change and agriculture (Hobbs et al., 2009). There is much discussion about the negative implications of novel ecosystems in general (Miller & Bestelmeyer, 2016) and Eucalyptus in particular (Poore & Fries, 1985;Sunder, 1993). On-the-otherhand, in Ethiopia Eucalyptus establishment has been found to have an indirect positive impact on soil and water conservation mainly through the reduction of soil runoff (Mhiret et al., 2019;Nyssen et al., 2009 increasing frequency of drought that affects crop and livestock production" (Jenbere et al., 2012). Similar results were found in a study by Milkias et al. (2014), in which farmers ranked construction materials, income and fuelwood provision as the main reasons for planting Eucalyptus. Indeed, the products of Eucalyptus not only recover the diminishing availability of construction materials and fuelwood due to previous deforestation, but also reduce the pressure on forest and native trees allowing conservation and regeneration. According to FGD participants, Eucalyptus were planted primarily in waterlogged grasslands, followed by cropland areas with low agricultural potential.
Future research should quantify the possible biodiversity trade-offs resulting from the plantation of exotic tree species like Eucalyptus, but also compare scenarios with or without Eucalyptus reforestation to understand its overall effect on biodiversity and the bundle of ecosystem services it provides.

| CONCLUSIONS
This study provides empirical evidence of restoration pathways where farmer-led reforestation leads to the recovery of key provisioning services, with simultaneous population increase. LULC changes are affected by the changing need for ecosystem services of landscape users. After two decades of deforestation and cropland expansion, residents of the study area shifted to reforestation through Eucalyptus planting and farmer managed natural regeneration. This was as a response to the growing scarcity of fuelwood and construction materials resulting from deforestation. In a landscape dominated by cropland, the benefit of such reforestation appears disproportionately high for people-for example, greater livestock numbers, improved resilience and diversification of income sourcesbut also for biodiversity .
Population increase promoted reforestation ('more people, more trees'), as the multiplication of homesteads also led to an increase of the number of natural trees, woodlots and Ensete plantations, which are all traditionally maintained around the homestead to provide key services for the household. This could represent a forest transition pathway (Meyfroidt et al., 2018) applicable to other landscapes with similar situations. However, it is incorrect to assume that an increase in population will always lead to environmental regeneration. We Previous studies in the area identified the importance of common-access forest, grazing areas and trees for sustainable intensification of agriculture (Duriaux Chavarría et al., 2018) and dietary diversity . However, it is important to identify the threshold of extractions before resource degradation occurs as well as the rules and institutions (local and state governance) that would allow this. Further research is needed in the study site to identify this knowledge gap and ensure the sustainability of the system.
Finally, while regeneration may occur in ways not always approved by conservationists-e.g., Eucalyptus-these 'novel ecosystems' might be a better option than simpler agricultural configurations with scant trees and perennial land uses (grassland, Ensete). Interdisciplinary landscape approaches could be a framework guiding regeneration in agricultural landscapes that help identify the land use balance that brings the most positive impact to both livelihoods and biodiversity. Further research using this approach could make use of 'natural experiments' to provide better scientific information to guide policy making and landscape re-designs to maximize the parallel development goals of humankind.

ACKNOWLEDGMENTS
We appreciate the help of researchers, field assistants, community