CarbonBench: A Global Benchmark for Upscaling of Carbon Fluxes Using Zero-Shot Learning

2026-03-10

AI Summary

Existing carbon flux models exhibit limited generalization in regions with sparse observations and high ecological diversity, and lack a unified benchmark to evaluate their zero-shot spatial transfer performance across diverse geographic and climatic conditions. To address this gap, this work establishes the first zero-shot spatial transfer learning benchmark for carbon flux upscaling, integrating over 1.3 million daily observations from 567 global eddy covariance flux tower sites. A hierarchical evaluation protocol is designed to disentangle spatial transfer effects from temporal autocorrelation, accompanied by standardized remote sensing and meteorological features alongside multiple baseline models. The benchmark explicitly distinguishes generalization challenges arising from vegetation types and climate zones, enabling systematic comparison of transfer learning approaches and providing a reproducible, reliable testbed for next-generation Earth system models in carbon flux estimation.

Technology Category

• Machine Learning: Transfer, Domain Adaptation, Multi-Task Learning
• Computer Vision: Remote Sensing / Geospatial AI
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Sustainability and carbon-aware systems for Web, mobile, and WoT
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics

Abstract

Accurately quantifying terrestrial carbon exchange is essential for climate policy and carbon accounting, yet models must generalize to ecosystems underrepresented in sparse eddy covariance observations. Despite this challenge being a natural instance of zero-shot spatial transfer learning for time series regression, no standardized benchmark exists to rigorously evaluate model performance across geographically distinct locations with different climate regimes and vegetation types. We introduce CarbonBench, the first benchmark for zero-shot spatial transfer in carbon flux upscaling. CarbonBench comprises over 1.3 million daily observations from 567 flux tower sites globally (2000-2024). It provides: (1) stratified evaluation protocols that explicitly test generalization across unseen vegetation types and climate regimes, separating spatial transfer from temporal autocorrelation; (2) a harmonized set of remote sensing and meteorological features to enable flexible architecture design; and (3) baselines ranging from tree-based methods to domain-generalization architectures. By bridging machine learning methodologies and Earth system science, CarbonBench aims to enable systematic comparison of transfer learning methods, serves as a testbed for regression under distribution shift, and contributes to the next-generation climate modeling efforts.

Problem

• carbon flux upscaling
• zero-shot learning
• spatial transfer learning
• eddy covariance
• distribution shift

Innovation

• zero-shot learning
• carbon flux upscaling
• spatial transfer learning
• benchmark dataset
• distribution shift