A Validation Framework for Machine Learning Emulation of Integrated Assessment Models

Prevailing modes of societal epistemic orientation are undergoing transition, driven in considerable part by the proliferation of generative artificial intelligence. Large language models (LLMs), in particular, regularly produce verisimilar outputs nonetheless lacking evidential grounding. These ‘hallucinations’ have foregrounded a challenge in how humans, and the scientific community, assess and calibrate what constitutes sufficient warrant for truth. Climate mitigation modelling presents a parallel. As a domain, it has long given rise to epistemological debates between the systems of production and verification enshrining scientific knowledge and the criteria by which that knowledge is publicly received and trusted. Thus, crucially important is establishing validation standards for the emerging intersection of artificial intelligence and climate mitigation modelling. To date, we know of two published works have attempted to use machine learning to emulate climate mitigation simulations (namely Integrated Assessment Models (IAMs)). Such emulation efforts represent a potentially transformative shift in how climate futures are constructed, interrogated, and communicated. Despite this change, and calls for a validation framework for these emulated scenarios, no systematic framework for validation has been presented in the literature. This paper addresses that gap, considering the epistemological difficulties inherent in producing a statistical representation of a mechanistic model, with particular attention to the inadequacy of current approaches which focus narrowly on output interrogation. Drawing on literature spanning philosophy of science, integrated assessment modelling, earth system modelling, and simulation validation theory, the paper develops a multi-pronged validation framework organised across distinct axes of model/justification and domain plausibility/statistical fidelity. Subsequently, this framework is applied to both existing IAM emulation studies to offer demonstration of its utility, a comparative assessment of each approach's respective merits, and a principled evaluative structure against which future emulation efforts may be measured.

Authors

Institutions

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-12
DOI
https://doi.org/10.5281/zenodo.24157785
Primary Topic
Climate Change and Geoengineering
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Validation Framework for Machine Learning Emulation of Integrated Assessment Models

Joeri Rogelj, Hamish Beath, Rossella Arcucci, Georgia Ray
Zenodo (CERN European Organization for Nuclear Research)
Climate Change and Geoengineering
article

A Validation Framework for Machine Learning Emulation of Integrated Assessment Models

Joeri Rogelj, Hamish Beath, Rossella Arcucci, Georgia Ray
article en

Abstract

Prevailing modes of societal epistemic orientation are undergoing transition, driven in considerable part by the proliferation of generative artificial intelligence. Large language models (LLMs), in particular, regularly produce verisimilar outputs nonetheless lacking evidential grounding. These ‘hallucinations’ have foregrounded a challenge in how humans, and the scientific community, assess and calibrate what constitutes sufficient warrant for truth. Climate mitigation modelling presents a parallel. As a domain, it has long given rise to epistemological debates between the systems of production and verification enshrining scientific knowledge and the criteria by which that knowledge is publicly received and trusted. Thus, crucially important is establishing validation standards for the emerging intersection of artificial intelligence and climate mitigation modelling. To date, we know of two published works have attempted to use machine learning to emulate climate mitigation simulations (namely Integrated Assessment Models (IAMs)). Such emulation efforts represent a potentially transformative shift in how climate futures are constructed, interrogated, and communicated. Despite this change, and calls for a validation framework for these emulated scenarios, no systematic framework for validation has been presented in the literature. This paper addresses that gap, considering the epistemological difficulties inherent in producing a statistical representation of a mechanistic model, with particular attention to the inadequacy of current approaches which focus narrowly on output interrogation. Drawing on literature spanning philosophy of science, integrated assessment modelling, earth system modelling, and simulation validation theory, the paper develops a multi-pronged validation framework organised across distinct axes of model/justification and domain plausibility/statistical fidelity. Subsequently, this framework is applied to both existing IAM emulation studies to offer demonstration of its utility, a comparative assessment of each approach's respective merits, and a principled evaluative structure against which future emulation efforts may be measured.

Zenodo (CERN European Organization for Nuclear Research)
International Institute for Applied Systems Analysis (AT), Imperial College London (GB)
Climate action
Climate Change and Geoengineering
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon