Population growth and agricultural expansion in northern Ethiopia are intensifying land resource pressures, driving notable land-use and land cover (LULC) changes. Yet, existing LULC prediction models, particularly in complex highlands like the Gubalafto district, face methodological limitations and fail to capture fine‑scale heterogeneity, reducing projection reliability. This study (i) evaluates the performance of Weight of Evidence (WoE), Logistic Regression (LR), and Multi-Layer Perceptron-Neural Network (MLPNN) models within the MOLUSCE QGIS plugin, (ii) develops and validates an integrated WoE‑MLPNN‑Cellular Automata (CA) model using LULC drivers, and (iii) generates LULC projections for 2029 and 2032, aligning with SDGs Agenda 2030 goals. Modeling inputs included Sentinel‑2 LULC maps (2016, 2019, 2024) and drivers, such as slope, temperature, precipitation, and proximity to roads and water. Combining WoE, MLPNN, and CA reduces the limitations of individual models, enhances prediction reliability, and improves the spatial realism of future LULC projections. Validation against the 2024 LULC map revealed that the WoE‑MLPNN‑CA model outperformed the standalone approaches, achieving 82.07% correctness and a kappa value of 0.74, indicating substantial agreement. Future LULC projections indicate that agriculture and woodland will be the dominant LULC in 2029 and 2032. Agricultural land is projected to change from 46.76% in 2024 to 44.01% in 2029 and then rise to 49.09% by 2032. Forest cover is projected to continuously decline from 11.41% in 2024 to 11.09% in 2029 and further to 7.90% in 2032, while woodland expands before slightly declining. The WoE‑MLPNN‑CA framework improves predictive performance and provides insights for prioritizing forest conservation and sustainable agricultural planning.
Tefera et al. (2026) studied this question.