Community Coexistence for Endangered Ethiopian Wolves in the Bale Mountans National Park, with AI-driven Settlement Mapping

---

Methodology 

Data

The study employed a multi-stage methodology integrating remote sensing, machine learning and community engagement to analyse human encroachment in Bale Mountain National Park (BMNP). Satellite data acquisition began with high-resolution imagery from Pleiades 1A (55.6 cm resolution) and Pléiades Neo (30 cm resolution) for 2019 and 2024, respectively.

---

Analysis

Settlement detection data analysis

A 10 km buffer zone around Web Valley was defined using Arc GIS Pro to focus analysis on the Ethiopian Wolf’s critical habitat, while minimising computational load. For settlement detection, a Mask R-CNN model was trained on manually annotated satellite imagery, where rooftops were labelled as polygons. The model’s backbone (ResNet-50) was fine-tuned with focal loss to address class imbalance caused by sparse settlements. Hyperparameters, including learning rate (0.001) and anchor scales, were optimised for small structures (<50 m²), while post-processing removed false positives (e.g., rock formations mimicking rooftops) through size filtering and shape analysis.

Farmland delineation data analysis 

Farmland delineation utilised the Segment Anything Model (SAM) adapted with SamLoRA architecture. The pre-trained SAM image encoder was frozen, and Low-Rank Adaptation (LoRA) layers (rank=4) were inserted into transformer blocks to enable efficient domain-specific learning with minimal parameters. The prompt encoder was bypassed, relying on default positional embeddings to capture generalised textures, while the mask decoder was reconfigured to output class-indexed masks for multi-season farmland segmentation. Training leveraged manually digitised farmland polygons validated against ground-truth data from aerial surveys and field observations. 

Model performance was evaluated using precision, recall and F1 scores, with cross-validation ensuring robustness across varying topographies. All datasets were integrated in Google Earth Engine and ArcGIS Pro for spatiotemporal analysis, enabling quantification of encroachment rates and identification of high-risk zones. This approach combined technical precision with visual insights, ensuring actionable outcomes for conservation planning.

---

Change detection results

Human settlement 

A temporal comparison of maps from 2019 and 2024 reveals a dramatic proliferation of human settlements within and surrounding the Bale Mountains National Park (BMNP). The analysis, which focused on the park, the Web Valley, and a 10 km buffer zone, documents an alarming rate of expansion over these five years. 

The number of settlements within the BMNP boundary surged from approximately 388 to 3,506 — a staggering ~10-times increase (Figure 3 below). This trend was mirrored within critical zones: settlements in the Web Valley itself grew from 11 to 165, while the surrounding 10 km buffer zone saw an increase from 903 to 6,180 structures.

Figure 3: Settlements within Bale Mountain National Park increased nearly tenfold from 387 in 2019 to 3,506 in 2024, with most remaining standalone and minimally overlapping with farmland

Spatially, this expansion is most concentrated in the northeastern sections, with settlement clusters intensifying and encroaching deeper into the park's interior (Figure 4). This pattern indicates a significant and growing human footprint, likely driven by population pressure and a lack of enforcement, which poses a severe threat to the park's conservation objectives.

Figure 4: Settlement expansion is most concentrated in the northeast, with clusters spreading deeper into the park

Human settlements are growing inside and around the park, and their impacts are directly tied to the two main types of livelihoods: agriculturalists (farmers) and pastoralists (herders): 

Agriculturalists (farmers)
In areas close to existing villages, new settlements often appear alongside cleared farmland. This pattern suggests the settlers are agriculturalists. Their practice of clearing land for crops destroys the habitat of the giant mole-rat, which is the primary prey of the endangered Ethiopian wolf.

Pastoralists (herders)
In contrast, settlements deeper within the park and in the Web Valley are usually enclosed by fences, a sign that they belong to pastoralist communities. While these settlements do not significantly overlap with farmland boundaries, they pose another critical threat: diseases such as rabies and canine distemper carried by domestic dogs can spread to Ethiopian wolves. 

Combined threats
Together, the impacts of habitat loss from farming and disease risks from herding create a serious and urgent challenge to the long-term survival of the park’s unique biodiversity.

Figure 5: Snapshots of encroachment across Bale Mountains National Park

Changes in farmland encroachment

Even though a smaller percentage of settlements were classified as agriculturalists (5% in 2019 and 11% in 2024), their impact is greater.

Farmland encroachment around these agricultural settlements expanded substantially, growing from 10.69 km² in 2019 to 15.16 km² in 2024, a 42% rise, dominated by smallholder cultivation of staple crops. SamLoRA demonstrated strong performance in farmland delineation, achieving a precision of 0.77 and recall of 0.89, excelling at capturing linear field patterns typical of agricultural plots. However, commission errors arose in areas where shrublands exhibited spectral similarities to farmland, highlighting limitations in texture differentiation. The 10 km buffer zone around Web Valley emerged as the most affected area, with escalating habitat fragmentation from farmland encroachment threatening Ethiopian Wolf corridors.

Figure 6: Farmland encroachment rose 42% (2019–2024), highlighting growing habitat fragmentation and conservation concerns.

This encroachment suggests growing pressure on Wolf habitat from agricultural practices, likely driven by human population growth. The increased presence of houses, particularly along the fringes of the buffer zone and within close proximity to core conservation areas, further underscores the urgency to enforce the park’s integrity.

---

Map output and community engagement 

These findings informed strategies for community workshops, targeted education initiatives, and community-led, land use planning to improve coexistence and mitigate further degradation of this ecologically sensitive landscape.

Partners launched an ambitious community engagement program called Peaceful Co-existence. The initiative targeted five key villages (kebeles) and involved more than 200 residents through participatory workshops designed to build trust and strengthen relationships with community leaders.

Residents took part in interactive mapping sessions that used infographics to visualise patterns of land encroachment and habitat loss. At the same time, structured surveys measured local awareness of conservation challenges and coexistence policies.

A central focus of the campaign was two-way dialogue. Researchers spoke with both farming (agriculturalist) and herding (pastoralist) communities to better understand local perspectives on the challenges they face and how these intersect with the survival of the Ethiopian wolf. A recurring concern voiced by residents was the spread of disease from domestic and stray dogs, particularly rabies, which mirrors the study’s finding of increased pastoralist settlements bringing wolves into closer contact with dogs.

The sessions also provided practical conservation education. Residents learned about biodiversity, the ecological importance of BMNP, current threats to wildlife and hands-on methods for protecting species. This program has built on existing voluntary park monitoring committees across the kebeles, which already meet every 15 days to review their activities, strengthening community-led conservation efforts.

Figure 7: Bridging knowledge and communities - researchers and farmers join forces to protect Ethiopia’s wolves through dialogue and community action.