Policy Robustness & Uncertainty in Model-based Decision Support for the Energy Transition

Technical Summary: Policy Robustness & Uncertainty in Model-based Decision Support for the Energy Transition

Problem Statement
Climate policy modelling is essential for assessing mitigation strategies in complex systems, yet it is inherently plagued by uncertainty. While uncertainty analysis is common in climate policy, the field often relies on limited scenario or local sensitivity analyses that fail to capture the full scope of parametric and structural uncertainties. Furthermore, the application of rigorous Uncertainty Quantification (UQ) methods—specifically those utilizing emulators to overcome computational constraints—is rare in energy policy modelling. This gap limits the ability to design "robust" policies, defined as those that retain functionality and achieve objectives across a wide range of plausible future contexts, rather than just under narrow assumptions. The authors argue that without comprehensive uncertainty analysis, policymakers cannot fully assess the risks of policy failure or the resilience of strategies against techno-economic and political variability.

Methodology
The authors propose a general methodology based on UQ to identify key uncertainties and evaluate policy robustness, applied to the FTT:Power (Future Technological Transformations) model. FTT:Power is a 71-region global model using evolutionary dynamics (replicator equations) to simulate technology diffusion based on levelized costs of electricity (LCOE).

The workflow consists of four stages:

1. Parameterisation: The study identifies 30 key input parameters, split between 15 techno-economic factors (e.g., learning exponents, lifetimes, lead times, discount rates, cannibalisation rates, fuel prices) and 15 policy instruments (e.g., phase-outs, feed-in tariffs, subsidies, carbon pricing). These inputs are mapped across regions (China, US, India, Global North/South) and technologies.
2. Simulation: To generate training data, 500 runs of FTT:Power are executed using Latin Hypercube Sampling to ensure comprehensive space-filling across the 30-dimensional parameter space.
3. Emulation: Gaussian Process (GP) emulators are trained on 80% of the simulation data to approximate the model's outputs (solar PV capacity, onshore wind capacity, power sector emissions, renewable shares, and weighted generation costs). The emulators provide fast predictions and quantify prediction uncertainty, validated against the remaining 20% of data and via leave-one-out cross-validation.
4. Prediction and Analysis: The trained emulators are used to perform extensive uncertainty analyses:
   • One-at-a-Time (OAAT) Sensitivity Analysis: To rank inputs by their influence on output variability.
   • Uncertainty Propagation: Sampling 20,000 times from input distributions under fixed policy scenarios to assess the probability of achieving specific targets (e.g., emission reductions, capacity goals) under uncertainty.

Key Contributions

• Methodological Advancement: The paper introduces a UQ-based framework using emulators to climate policy modelling, enabling the exploration of thousands of scenarios that would be computationally prohibitive using the original model alone.
• Policy Robustness Framework: It demonstrates how to systematically evaluate the robustness of policy packages against a wide range of techno-economic and political uncertainties, moving beyond simple effectiveness assessments.
• Application to FTT:Power: The study applies this framework to a global model, explicitly incorporating "cannibalisation" (the reduction in renewable value due to high market penetration) and lead times as critical uncertainty drivers, areas often under-represented in global models.

Results

• Global Sensitivity: At the global level, the primary drivers of uncertainty in transition outcomes are cannibalisation rates (for solar and wind), construction lead times, and grid connection lead times. These factors outweigh regional price policies, including potential policy reversals in the US. Solar PV is found to be resilient to many uncertainties due to low costs but remains sensitive to infrastructure constraints and cannibalisation. Onshore wind is exposed to a broader range of uncertainties, including demand and discount rates.
• Global Emissions: Stronger policy ambition reduces median emissions and narrows the distribution of outcomes (reducing uncertainty). However, high cannibalisation and slow lead times significantly increase median emissions. Even with high ambition, slow lead times can lead to higher emissions in the short term.
• India Case Study:
  • Capacity Targets: Policy packages combining partial fossil fuel phase-outs with subsidies or carbon pricing demonstrate greater robustness in achieving capacity targets (e.g., 393 GW by 2030) compared to purely market-based approaches (subsidies/carbon pricing alone). Pure market-based policies only succeed in a narrow set of favorable conditions.
  • Emissions and Multi-Goal Robustness: Partial phase-outs substantially reduce the variability of emissions outcomes. While market-based policies struggle to meet multiple simultaneous targets (emissions, renewable share, and price) under uncertainty, packages including phase-outs are more robust to variations in demand, discount rates, and cannibalisation.
  • Critical Bottlenecks: Regardless of the policy mix, long lead times remain a critical barrier that jeopardizes policy goals. Even robust policy packages fail to meet targets if lead times are slow.

Significance and Claims
The authors claim that their approach reveals much larger uncertainties in transition outcomes than commonly represented in standard climate policy modelling. They argue that while policy design can be shaped to mitigate these uncertainties, the current reliance on narrow scenario analysis is insufficient.

The study concludes that:

1. Regulatory intervention is critical: Enabling policies and regulating fossil fuels (via phase-outs) are essential for robust power sector transitions, as market-based instruments alone are often insufficient under deep uncertainty.
2. Addressing root causes: Policy efforts must target the root causes of uncertainty, specifically cannibalisation (through storage, flexibility, and market design) and lead times (through infrastructure and permitting reform).
3. Resilience of Solar vs. Wind: Solar PV's diffusion appears resilient to some policy uncertainties (like US rollbacks) due to cost momentum, whereas onshore wind is more vulnerable to a wider array of factors.
4. Methodological Utility: The emulator-based workflow provides a novel, efficient way to analyze uncertainty and evaluate policy robustness, offering a template for future research that can be applied to different models and policy domains to better contextualize findings and address structural biases.

The paper maintains a modest tone regarding structural uncertainties, acknowledging that the FTT:Power model relies on LCOE comparisons which may not fully capture investor profit-maximization dynamics. However, the authors assert that their method allows for the easy replication of results under different assumptions and model structures, thereby enhancing the transparency and reliability of policy advice in the face of deep uncertainty.