This study tackles a significant deficiency in climate action by creating an advanced forecasting method for carbon dioxide (CO2) transport emissions in the Middle East and North Africa (MENA) area, where growing urbanization and reliance on fossil fuels are propelling unsustainable emission increases (projected +133% by 2050). This study aims to examine the constraints of machine learning (ML) methodologies in managing the data scarcity, volatility and institutional inconsistencies in the MENA region, which presently hinder the implementation of successful decarbonization programs.
Dynamic autoregressive integrated moving average (ARIMA) component-switching models integrate auto-adjusting parameters to effectively capture sudden policy changes and economic shocks, surpassing static ML models such as LSTM and extreme gradient boosting. Attaining a root mean square error of 0.050 MtCO2 per year in Morocco, compared to 14.805 MtCO2 per year using LSBoost, despite limited data sets. Using ARIMA performance indicators (R², BIC) to discern institutional data-quality deficiencies, facilitating focused enhancements.
Morocco (ARIMA 0,1,2) and Oman (ARIMA 1,1,1) attained R² = 0.999 due to consistent rules and a strong data infrastructure. Algeria and Sudan (BIC > 0) demonstrated that irregular reporting hinders forecasting and underscores governance difficulties. Unregulated emissions are projected to climb by 22%–133% by 2050, with Oman’s oil-dependent transportation sector anticipating a 53.85% jump.
Although ARIMA performs well in data-limited situations, its precision is contingent upon the stability of governance. Future endeavors will include satellite-derived activity data to improve models for underperforming nations, thereby solving existing data deficiencies.
Carbon pricing in Oman (elasticity: −1.2) and the expansion of Morocco’s electric rail system yield the highest return on investment according to model-derived sensitivity. ARIMA measures act as indicators of data infrastructure quality, directing focused investments by international organizations.
This study redefines emission forecasting by demonstrating ARIMA’s superiority over ML in fragmented-data contexts (in contrast to ML-focused studies), delivering the first MENA-specific policy toolkit based on emission elasticity analysis and introducing component switching as a scalable solution for volatile time-series forecasting in developing economies.
