A framework for decision making under deep uncertainty in hard-to-abate industries: An application case for investment in a German steel plant

Decarbonising hard-to-abate industries is necessary for climate goals, yet investment choices are hindered by deep uncertainty. This study develops an ambiguity-aware investment evaluation framework that accounts for parameter uncertainty, probability ambiguity, and downside risk. Scenario-level net present values are computed under alternative electricity and hydrogen system narratives and evaluated using a loss law-invariant model-set approach. Probability ambiguity is represented by an ambiguity set of admissible scenario-probability vectors defined through per-scenario bounds implied by multiple probability views, and performance is assessed using worst-case mean and worst-case expected shortfall over this set. The framework is applied to a primary steel investment in Germany, comparing Natural Gas-DRI-EAF, imported-hydrogen DRI–EAF, and on-site electrolysis based DRI–EAF configurations. Electricity procurement prices come from an agent based market simulation for 2035, while fuel and policy inputs vary across 18 discrete scenarios. Natural Gas–DRI–EAF is most robust with a worst-case mean NPV of €10.83 bn and a 4.7% reduction from the Base expected value. Electrolyser-DRI-EAF with lower CAPEX and tariff relief approaches this benchmark with a worst-case mean NPV of €10.53 bn, and tariff relief increases worst-case mean performance by about €4.9 bn relative to no discounts. Imported-hydrogen DRI–EAF is ambiguity-fragile, with Base expected NPV of €7.82 bn falling to €4.57 bn under worst-case mean. For industrial decision makers, the framework indicates whether a pathway remains acceptable under pessimistic probability weightings and tail protection, and it identifies which parameters drive downside exposure. For policy and regulatory analysts, the remaining robustness gap even under tariff relief indicates where additional risk-sharing or cost-relief instruments are needed to improve performance in the adverse states that determine worst-case and tail outcomes. A sensitivity analysis with non-stationary second-decade price regimes confirms that the main conclusions do not rely on stationary annual cash flows.