TerraNova: A Foundation Model for the Anthropocene
A foundation model trained on 1,024 physical and societal records, coupling gridded Earth-system fields with national indicators.
Publications
24 SELECTED WORKS · 2021—2026FILTER:
2026
A harmonised dataset for Earth system foundation models
A harmonised global dataset aligning hundreds of environmental and socioeconomic variables to a standardised 0.25° grid.
Understanding electricity consumption behaviour through Inverse Reinforcement Learning
Uses Inverse Reinforcement Learning to recover household reward functions and track shifts in Italian cooling behaviour, 2021-2023.
EU-ETS under attack? The impact of carbon price suppression on the decarbonization of the power sector
Uses a multi-agent RL model of the Italian power system (MARLEY) to test how suppressing the EU ETS price affects costs and emissions.
Carbon sink and methane source: the net role of wetland restoration in CDR portfolio
Extends the WITCH integrated assessment model to size wetland restoration's carbon dioxide removal potential against its methane trade-off.
A Blueprint for Integrated Climate Intelligence
A perspective proposing AI as an integration layer linking climate forecasts, impact assessment and institutional decision-making.
Artificial intelligence to support cross-disciplinary climate change research
A commentary in Nature Climate Change on using artificial intelligence to support cross-disciplinary climate change research.
Fast Voxelization and Level of Detail for Microgeometry Rendering
Introduces a fast parallel voxelization method and a hierarchical SGGX clustering representation for microgeometry level of detail.
Towards ecologically meaningful foundation models
A perspective arguing that ecological foundation models, trained on multimodal ecological data, could help unify fragmented data and theory.
Generative AI for Climate Governance and Acceptability-Constrained Policy Design
Proposes using LLMs as cultural world models in agent-based simulations to co-optimize climate policy for efficacy and social legitimacy.
A dataset of harmonized global air quality monitoring metadata
A Nature Scientific Data paper presenting a harmonized dataset of global air quality monitoring station metadata.
Neural Conditional Transport Maps
A hypernetwork-based framework for learning conditional optimal transport maps, applied to global sensitivity analysis of black-box models.
2025
Assessing long-term electricity market design for ambitious decarbonization targets using multi-agent reinforcement learning
Applies MARLEY, a multi-agent reinforcement learning framework, to assess long-term electricity market design under decarbonization targets.
Single-image Reflectance and Transmittance Estimation from Any Flatbed Scanner
Estimates material reflectance, opacity and transmittance from a single flatbed-scanner image using a cycle-consistency formulation.
2024
TexTile: A Differentiable Metric for Texture Tileability
A differentiable metric, trained as a classifier, that quantifies texture tileability and can guide tileable texture synthesis methods.
2023
Neural Networks for Digital Materials and Radiance Encoding
PhD thesis on neural network methods for digital materials and radiance encoding: material capture, texture synthesis and neural rendering.
NeuBTF: Neural Fields for BTF Encoding and Transfer
A neural BTF representation that uses a guidance image to jointly address BTF compression, tiling and extrapolation.
NEnv: Neural Environment Maps for Global Illumination
Combines a normalizing flow and an implicit neural representation to compress environment maps and sample them for global illumination.
Towards Material Digitization with a Dual-scale Optical System
Presents a dual-scale optical capture system for material digitization, published in ACM Transactions on Graphics (Proc. SIGGRAPH).
UMat: Uncertainty-Aware Single Image High Resolution Material Capture
Recovers normals, specularity and roughness from a single diffuse image via a generative network, with test-time uncertainty quantification.
How Will It Drape Like? Capturing Fabric Mechanics from Depth Images
Estimates fabric mechanical parameters from a casual depth-camera capture using a sim-to-real learning framework.
2022
SeamlessGAN: Self-Supervised Synthesis of Tileable Texture Maps
Generates tileable texture maps from a single exemplar by tiling the latent space of an adversarially trained generative network.
A Survey on Intrinsic Images: Delving Deep into Lambert and Beyond
Surveys deep learning approaches to intrinsic image decomposition, classifying methods by decomposition type, priors and architecture.
2021
Neural Photometry-guided Visual Attribute Transfer
Propagates spatially varying material attributes, such as texture maps or stylizations, to larger samples via image-to-image translation.