Leveraging a Fully Differentiable Integrated Assessment Model for RL and Inference

Koen Ponse, Kai-Hendrik Cohrs, Phillip Wozny, Andrew Robert Williams, Tianyu Zhang, Erman Acar, Yoshua Bengio, Aske Plaat, Thomas M. Moerland, Pierre Gentine, Gustau Camps-Valls

Keywords: Differentiable modeling, Integrated Assessment Models, Climate–economic simulation, Multi-agent reinforcement learning, Hybrid modeling, Gradient-based optimization

TL;DR: We present a fully differentiable JAX-based RICE-N model enabling gradient-based calibration, uncertainty analysis and policy optimization, transforming IAMs into a learnable, data-driven framework for climate–economy modeling.

Abstract: Integrated Assessment Models (IAMs) such as RICE have long provided a foundation for studying the coupled dynamics of the global economy and climate system. Traditionally, these models have been used in a forward-simulation mode, with parameters hand-calibrated and dynamics treated as fixed. In this work, we introduce RICE-N-JAX, a fully differentiable implementation of the multi-region RICE-N model in JAX. Beyond significantly accelerating the training of multi-agent reinforcement learning (MARL) agents, differentiability opens new research directions, including automated calibration to historical data, recovery of latent regional behavioral parameters (e.g., risk aversion and time preferences) and sensitivity analysis of economic and technological assumptions. Moreover, it allows us to treat policy design and international negotiation mechanisms as learnable parameters within a gradient-based optimization framework. We outline new research opportunities that arise when an IAM becomes a differentiable environment and discuss implications for climate–economics modeling, machine learning for climate policy and the fusion of data- and theory-driven approaches.

Email Sharing: We authorize the sharing of all author emails with Program Chairs.

Data Release: We authorize the release of our submission and author names to the public in the event of acceptance.

Submission Number: 29