X-MethaneWet: A Cross-scale Global Wetland Methane Emission Benchmark Dataset for Advancing Science Discovery with AI

TL;DR

The paper addresses the need for a standardized, high-temporal-resolution benchmark for global wetland methane emissions by constructing X-MethaneWet, which combines TEM-MDM physics-based simulations with FLUXNET-CH$_4$ observations at daily scale. It establishes a comprehensive evaluation framework and benchmarks multiple sequential models (LSTM, EA-LSTM, TCN, Transformer variants, Pyraformer) across temporal and spatial extrapolation tasks, while exploring transfer learning from simulations to real observations. The results show that pretraining on simulated data and subsequent fine-tuning generally improves generalization, particularly in data-sparse scenarios, highlighting the potential of integrating physics-based models with AI for improved methane flux predictions. Overall, this work provides a valuable resource and methodological blueprint for AI-driven climate modeling and methane mitigation planning through knowledge-guided learning.

Abstract

Methane (CH$_4$) is the second most powerful greenhouse gas after carbon dioxide and plays a crucial role in climate change due to its high global warming potential. Accurately modeling CH$_4$ fluxes across the globe and at fine temporal scales is essential for understanding its spatial and temporal variability and developing effective mitigation strategies. In this work, we introduce the first-of-its-kind cross-scale global wetland methane benchmark dataset (X-MethaneWet), which synthesizes physics-based model simulation data from TEM-MDM and the real-world observation data from FLUXNET-CH$_4$. This dataset can offer opportunities for improving global wetland CH$_4$ modeling and science discovery with new AI algorithms. To set up AI model baselines for methane flux prediction, we evaluate the performance of various sequential deep learning models on X-MethaneWet. Furthermore, we explore four different transfer learning techniques to leverage simulated data from TEM-MDM to improve the generalization of deep learning models on real-world FLUXNET-CH$_4$ observations. Our extensive experiments demonstrate the effectiveness of these approaches, highlighting their potential for advancing methane emission modeling and contributing to the development of more accurate and scalable AI-driven climate models.

Figures (8)

• Figure 1: Temporal variations in air temperature, precipitation, and $\text{CH}_4$ flux across 2014 to 2016 in US.BZB site.
• Figure 2: Train-test split of FLUXNET-$\text{CH}_4$ dataset for temporal and spatial extrapolation experiments.
• Figure 3: Label and LSTM prediction performance in the TEM-MDM dataset in 2018.
• Figure 4: Comparison of LSTM predictions and labels in the TEM-MDM dataset across climate zones in 2018.
• Figure 5: Examples showing performance improvements through LSTM fine-tuning in both temporal and spatial extrapolation.