Calibrated Uncertainty Quantification for Sectoral Exergy Analysis of Transport Using Probabilistic Machine Learning

Sectoral exergy studies of transport usually end with a single projected path and no stated uncertainty. In the current study, we develop a calibrated framework to address this and apply it to the Nigerian transport sector. First, we show that exergy-based indicators reduce to one parameter, the subsector exergy efficiency, so that they agree by construction. Next, we forecast fuel demand with an equal weight combination of a Gaussian process trend model and a drift model. Rolling origin backtests show that drift remains the best point forecaster, while the combination gives the best interval score. One scale factor calibrates the 90% intervals to 90.3% coverage, 88.3% on held-out origins, with no widening needed at long horizons. Independent national statistics for 2020 to 2023, never used in fitting, fall inside the 90% intervals in all four years. After updating with these data, Monte Carlo propagation gives 81 Mt of CO2 in 2040 (90% interval 34 to 123 Mt) and an improvement potential of 695 PJ (294 to 1028 PJ). This potential is a thermodynamic upper bound, not an achievable saving. Demand drives almost all CO2 uncertainty. Correlated demand, interactions and changing efficiencies alter them only modestly.