Rethinking the measurements and predictors of environmental degradation in Ethiopia: Predicting long-term impacts using a kernel-based machine learning approach

Abstract

Given the severity of global environmental degradation, particularly in countries like Ethiopia, it is urgent to rethink its drivers and measurements for actionable policy development. The relationships among these predictors are complex, often nonlinear, non-additive, and include reverse causality, making it difficult for traditional econometric models to capture them. Conventional CO₂ metrics also tend to overlook unique emission sources in developing countries, where emissions are closely linked to energy production, unsustainable agriculture, deforestation, and land use rather than industry. To address these gaps, this study applies a kernel-based machine learning model and develops context-specific CO₂ metrics to analyze environmental degradation predictors and forecast their long-term impacts in Ethiopia using quarterly data from 2000Q1 to 2020Q4. The findings indicate that economic growth, industrialization, energy poverty, urbanization, ICT, and resource rent are significant predictors, exhibiting complex, nonlinear relationships. Long-term prediction analysis shows that energy poverty, economic growth, ICT, and urbanization initially worsen degradation but lead to stabilization over time. In contrast, industrialization and resource rent predominantly exacerbate environmental issues before leveling off. The study recommends policies to enhance energy access and efficiency through renewable energy subsidies and financial incentives, integrate green infrastructure into urban planning, incentivize clean industrial technologies, promote environmentally inclusive growth, regulate eco-friendly ICT, such as energy-efficient data centers and e-waste management, implement a resource rent tax, and use adaptive policies with real-time analytics to address degradation thresholds, balancing economic growth with resilience and sustainability. • Predictors of environmental degradation and their long-term impacts are identified. • A novel supervised machine learning model and a new degradation index are used. • Economic growth, urbanization, and industrialization predict environmental degradation. • Energy poverty, ICT, and resource rent significantly predict environmental degradation. • Degradation rises with energy poverty and resource rent but declines after thresholds.