Periodic Reporting for period 1 - EUNICE (Debiasing the uncertainties of climate stabilization ensembles)
Reporting period: 2022-12-01 to 2025-05-31
EUNICE: Tackling Climate Uncertainty for Smarter Decision-Making
Predicting the future of our climate and its economic and social consequences is not straightforward. Scientists use complex models to estimate everything from rising temperatures to economic damages and the effectiveness of policies. However, these models carry uncertainties—about how extreme climate events will unfold, how societies will respond, and which technologies will drive emission reductions. Making the right decisions in the face of such uncertainty is difficult. EUNICE aims to improve how we assess climate uncertainty by developing cutting-edge methods that integrate multiple perspectives—from climate science to economics and policy. By doing so, the project helps decision-makers navigate complex climate risks and design more effective policies for a resilient future.
What EUNICE has achieved in the first 2 years of existence:
The project is structured around three key pillars:
1⃣ Climate Extremes & Their Impacts
EUNICE analyzes how climate extremes (heatwaves, droughts, floods) will evolve and impact societies.
A global climate database has been created, covering 1881-2100, using machine learning and climate simulations to identify risk hotspots.
Researchers assess the economic costs of extreme events on GDP, human development, and inequality.
2⃣ Policy & Mitigation Strategies
EUNICE develops AI-driven models to predict greenhouse gas emissions and evaluate mitigation strategies like Carbon Dioxide Removal (CDR).
A multi-agent learning framework is being designed to explore different electricity market structures and their resilience to climate uncertainty.
The project examines how policies can be designed to minimize economic risks while ensuring climate goals are met.
3⃣ Better Decision-Making Under Uncertainty
EUNICE applies deep learning, econometrics, and robust control techniques to improve climate and economic forecasting.
It develops climate emulators—simplified yet powerful models that help policymakers explore different scenarios quickly and efficiently.
The project also pioneers sensitivity analysis methods, helping researchers understand which assumptions in climate models matter most
Predicting the future of our climate and its economic and social consequences is not straightforward. Scientists use complex models to estimate everything from rising temperatures to economic damages and the effectiveness of policies. However, these models carry uncertainties—about how extreme climate events will unfold, how societies will respond, and which technologies will drive emission reductions. Making the right decisions in the face of such uncertainty is difficult. EUNICE aims to improve how we assess climate uncertainty by developing cutting-edge methods that integrate multiple perspectives—from climate science to economics and policy. By doing so, the project helps decision-makers navigate complex climate risks and design more effective policies for a resilient future.
What EUNICE has achieved in the first 2 years of existence:
The project is structured around three key pillars:
1⃣ Climate Extremes & Their Impacts
EUNICE analyzes how climate extremes (heatwaves, droughts, floods) will evolve and impact societies.
A global climate database has been created, covering 1881-2100, using machine learning and climate simulations to identify risk hotspots.
Researchers assess the economic costs of extreme events on GDP, human development, and inequality.
2⃣ Policy & Mitigation Strategies
EUNICE develops AI-driven models to predict greenhouse gas emissions and evaluate mitigation strategies like Carbon Dioxide Removal (CDR).
A multi-agent learning framework is being designed to explore different electricity market structures and their resilience to climate uncertainty.
The project examines how policies can be designed to minimize economic risks while ensuring climate goals are met.
3⃣ Better Decision-Making Under Uncertainty
EUNICE applies deep learning, econometrics, and robust control techniques to improve climate and economic forecasting.
It develops climate emulators—simplified yet powerful models that help policymakers explore different scenarios quickly and efficiently.
The project also pioneers sensitivity analysis methods, helping researchers understand which assumptions in climate models matter most
1⃣ Modeling Climate Extremes and Their Impacts
Building a Global Climate Extremes Database
- Developed a high-resolution global dataset covering 1881-2100, incorporating thirteen types of extreme weather events (heatwaves, droughts, windstorms, etc.).
- Data is based on ISIMIP3b bias-adjusted projections and ERA5 reanalysis, ensuring high accuracy and spatial detail.
- The dataset enables detailed assessments of how climate extremes will affect human populations, ecosystems, and economic sectors.
AI-Based Prediction of Climate Extremes
-Designed a probabilistic deep learning framework capable of predicting heatwaves, floods, and droughts based on simple climate inputs (e.g. temperature, precipitation).
-Uses a conditional generative adversarial network (GAN) architecture, improving prediction accuracy and enabling better uncertainty quantification.
-Enhances reliability in areas of high socioeconomic exposure, allowing policymakers to identify future risk hotspots.
Economic and Social Impacts of Climate Change
-Analyzed the historical and projected impacts of climate extremes on GDP, inequality, and human development (HDI) using econometric fixed-effects models.
-Assessed regional economic damages by linking extreme weather events to life expectancy, education, and income levels.
-Provided first-of-its-kind quantifications of economic losses due to extreme climate events under different Shared Socioeconomic Pathways (SSPs).
2⃣ Climate Mitigation and Policy Evaluation
AI-Driven Emission Prediction Models
-Developed a deep learning-based system to forecast CO2 and CH₄ emissions from land-use changes, including wetlands and forestry.
-Uses Generative Adversarial Networks (GANs) and reinforcement learning techniques to simulate future emissions pathways.
-Aims to improve projections beyond traditional IAMs by incorporating data-driven uncertainty quantification.
Integrated Market and Policy Modeling
-Designed a Multi-Agent Reinforcement Learning (MARL) framework to simulate electricity markets under net-zero scenarios.
-The model helps identify regulatory gaps, evaluate different policy mechanisms, and assess market stability under different climate targets.
-Provides quantitative insights for policymakers on market designs that can accommodate large-scale renewable energy and carbon pricing.
Robust Transition Pathways for Net-Zero
-Developed a small-scale IAM with technological richness, integrating robust control methods to assess transition risks.
-Applied deep-learning solutions for high-dimensional partial differential equations (DGM-PIA method) to improve policy resilience analysis.
-Developed a global sensitivity analysis (GSA) methodology using Optimal Transport solvers, allowing better uncertainty assessments in climate-economic modeling.
3⃣ Advancing Decision-Making Under Uncertainty
Climate Emulators for Fast Policy Assessment
-Built a machine-learning-based climate emulator to generate high-resolution climate projections from simple temperature inputs.
-Enables policymakers and researchers to quickly assess regional climate risks without running computationally expensive full-scale climate models.
Uncertainty Quantification and Model Misspecification Analysis
-Developed new uncertainty quantification techniques for climate policy cost-benefit analysis.
-Investigated the impact of model misspecification and deep uncertainty on economic evaluations of climate mitigation pathways.
-Demonstrated how different assumptions about climate damages can significantly alter policy recommendations, emphasizing the need for robust decision-making.
Developing a Foundation Model for Climate-Economy Interactions
-Launched efforts to build the first deep-learning foundation model that integrates climate and socioeconomic interactions.
-Uses multimodal AI architectures, diffusion models, and explainable AI techniques to capture complex feedback loops between climate and economy.
-Provides a groundbreaking tool for forecasting climate-economic dynamics, making it a potential state-of-the-art solution for climate policy analysis.
Building a Global Climate Extremes Database
- Developed a high-resolution global dataset covering 1881-2100, incorporating thirteen types of extreme weather events (heatwaves, droughts, windstorms, etc.).
- Data is based on ISIMIP3b bias-adjusted projections and ERA5 reanalysis, ensuring high accuracy and spatial detail.
- The dataset enables detailed assessments of how climate extremes will affect human populations, ecosystems, and economic sectors.
AI-Based Prediction of Climate Extremes
-Designed a probabilistic deep learning framework capable of predicting heatwaves, floods, and droughts based on simple climate inputs (e.g. temperature, precipitation).
-Uses a conditional generative adversarial network (GAN) architecture, improving prediction accuracy and enabling better uncertainty quantification.
-Enhances reliability in areas of high socioeconomic exposure, allowing policymakers to identify future risk hotspots.
Economic and Social Impacts of Climate Change
-Analyzed the historical and projected impacts of climate extremes on GDP, inequality, and human development (HDI) using econometric fixed-effects models.
-Assessed regional economic damages by linking extreme weather events to life expectancy, education, and income levels.
-Provided first-of-its-kind quantifications of economic losses due to extreme climate events under different Shared Socioeconomic Pathways (SSPs).
2⃣ Climate Mitigation and Policy Evaluation
AI-Driven Emission Prediction Models
-Developed a deep learning-based system to forecast CO2 and CH₄ emissions from land-use changes, including wetlands and forestry.
-Uses Generative Adversarial Networks (GANs) and reinforcement learning techniques to simulate future emissions pathways.
-Aims to improve projections beyond traditional IAMs by incorporating data-driven uncertainty quantification.
Integrated Market and Policy Modeling
-Designed a Multi-Agent Reinforcement Learning (MARL) framework to simulate electricity markets under net-zero scenarios.
-The model helps identify regulatory gaps, evaluate different policy mechanisms, and assess market stability under different climate targets.
-Provides quantitative insights for policymakers on market designs that can accommodate large-scale renewable energy and carbon pricing.
Robust Transition Pathways for Net-Zero
-Developed a small-scale IAM with technological richness, integrating robust control methods to assess transition risks.
-Applied deep-learning solutions for high-dimensional partial differential equations (DGM-PIA method) to improve policy resilience analysis.
-Developed a global sensitivity analysis (GSA) methodology using Optimal Transport solvers, allowing better uncertainty assessments in climate-economic modeling.
3⃣ Advancing Decision-Making Under Uncertainty
Climate Emulators for Fast Policy Assessment
-Built a machine-learning-based climate emulator to generate high-resolution climate projections from simple temperature inputs.
-Enables policymakers and researchers to quickly assess regional climate risks without running computationally expensive full-scale climate models.
Uncertainty Quantification and Model Misspecification Analysis
-Developed new uncertainty quantification techniques for climate policy cost-benefit analysis.
-Investigated the impact of model misspecification and deep uncertainty on economic evaluations of climate mitigation pathways.
-Demonstrated how different assumptions about climate damages can significantly alter policy recommendations, emphasizing the need for robust decision-making.
Developing a Foundation Model for Climate-Economy Interactions
-Launched efforts to build the first deep-learning foundation model that integrates climate and socioeconomic interactions.
-Uses multimodal AI architectures, diffusion models, and explainable AI techniques to capture complex feedback loops between climate and economy.
-Provides a groundbreaking tool for forecasting climate-economic dynamics, making it a potential state-of-the-art solution for climate policy analysis.
- Developing a Foundation Model for Climate-Economy Interactions – A first-of-its-kind AI-driven model that learns from the complex feedback loops between climate change, economic inequality, and mitigation policies. This approach brings unprecedented predictive power to climate policy analysis.
- AI-Based Climate Extremes Prediction – While traditional climate models struggle with high-resolution extreme event forecasting, EUNICE has designed deep learning architectures (GANs, probabilistic AI, and reinforcement learning) to improve spatial and temporal precision of extreme weather projections, enabling real-time risk assessments.
- Robust Climate Policy & Economic Uncertainty Modeling – EUNICE is the first project to integrate deep uncertainty into economic decision-making using reinforcement learning and AI-enhanced integrated assessment models (IAMs). This allows policymakers to design more resilient and adaptive climate strategies that account for unknown risks.
- AI-Based Climate Extremes Prediction – While traditional climate models struggle with high-resolution extreme event forecasting, EUNICE has designed deep learning architectures (GANs, probabilistic AI, and reinforcement learning) to improve spatial and temporal precision of extreme weather projections, enabling real-time risk assessments.
- Robust Climate Policy & Economic Uncertainty Modeling – EUNICE is the first project to integrate deep uncertainty into economic decision-making using reinforcement learning and AI-enhanced integrated assessment models (IAMs). This allows policymakers to design more resilient and adaptive climate strategies that account for unknown risks.