Climate change mitigation requires balancing economic growth against environmental targets across diverse regions with competing priorities. Integrated Assessment Models (IAMs) are widely used to inform climate policy, yet existing Reinforcement Learning (RL) extensions rely on the homogeneity assumption — treating world regions as identical decision-makers. This misrepresents the diversity of economic conditions and climate vulnerabilities that characterise real-world global systems. This paper extends the JUSTICE Multi-Objective Multi-Agent RL (MOMARL) framework by introducing heterogeneous agents, enabling each region to maintain an independent policy. We propose two algorithms — MOHAPPO and MOHASAC — and evaluate them against the homogeneous baseline MOMAPPO. Heterogeneous algorithms achieve higher Hypervolume and Expected Utility, produce more diverse Pareto fronts, and identify more aggressive emission mitigation strategies. Regional analysis reveals that the distribution of mitigation burdens varies across algorithms and policy types, with some configurations placing disproportionate mitigation effort on vulnerable regions. We demonstrate how explicitly encoding equity in the reward structure — via utilitarian versus prioritarian welfare objectives — can shift the distribution of burdens across regions, reinforcing the need for mindful optimisation design. These findings underscore that
algorithmic design choices in RL-based climate modelling are not value-neutral, and argue for heterogeneous, equity-aware frameworks as essential in this domain. ... Read more

New global climate policies must be deemed just in order to be effective. However, global climate policymaking is complex and subject to normative uncertainties, especially in relation to the distribution of resources, risks, and consequences. These diverging views are dependent on what is distributed and are ultimately based on moral rules and principles that prescribe when a distribution is morally just; distributive justice principles. Enhanced understanding of these distributive preferences is necessary to account for them in both policymaking and modelling. A bottom-up evaluation of stakeholder views in negotiations is a time-consuming but useful method to add to this understanding. This research evaluates the potential of using GPT-4o to perform this task, identifying distributive justice principles in High-Level Segment (HLS) speeches from UNFCCC COP. By identifying distributive justice principles—egalitarianism, utilitarianism, prioritarianism, sufficientarianism, and libertarianism—this research examines the moral foundations of climate policy decisions. Manual annotations of 51 HLS speeches created a ground truth dataset, revealing complexities and class imbalances in principles, with prioritarianism being most dominant. GPT-4o’s performance in identifying relevant sentences and distributive justice principles showed promise but struggled with consistency compared to human annotations. Despite limitations, the model demonstrated efficiency, highlighting its potential for pre-processing and classification tasks. The study underscores the importance of a nuanced, bottom-up understanding of distributive justice in climate negotiations, contributing to climate justice and IAM by offering theoretical insights and practical implications. ... Read more

Climate Change continues to pose a considerable threat to the well-being of people and economies. Today, to avoid catastrophic and irreversible damage, decision-makers and policy advisers need to explore possible scenarios and enact mitigation and adaptation policies to curb the rise of global temperatures within the thresholds set by the Paris Agreement. However, avoiding a 1.5-degree warming seems already out of hand, and the last Conference of Parties in Glasgow (COP26) sparked a contentious debate surrounding the role of coal. Representatives rely on climate reports and models to understand the problem, including integrated assessment models that aim to encompass the whole process straightforwardly and transparently. One example is the RICE-2010 model developed by the 2018 Nobel Prize winner in economics William Nordhaus, used by the IPCC and known for its simplicity. However, the model does not include an explicit formulation of energy. This renders it hard to explore scenarios and policy questions directly tied to the diversification of the energy mix, a topic that has gained considerable attention with the Energy Crisis sparked by the Russian Invasion of Ukraine. Therefore, this thesis attempts to introduce energy intensity and carbon intensity to the model by decomposing the Emission Output Ratio. These parameters will allow the user to explore the drivers behind decarbonisation, whether it is related to an improvement in the energy efficiency of processes or a greener energy mix. The selected approach yielded surprising insights, such as the poor documentation and data quality of the RICE model, the over-simplistic design choices behind emissions and decarbonisation, and the under-representation of carbon intensity. These outcomes have highlighted potential, underestimations of future temperature rise, limited policy testing potential and a lack of transparency in data, methodology, and reproducibility. ... Read more