license: cc-by-4.0
task_categories:
- tabular-regression
tags:
- climate
- tabular
size_categories:
- n<1K
🌍 X-MethaneWet: A Cross-Scale Global Wetland Methane Benchmark Dataset
X-MethaneWet is a first-of-its-kind cross-scale global wetland methane benchmark dataset designed to support research on methane (CH₄) emissions across temporal and spatial scales. It integrates physics-based model simulations and real-world observations to enable data-driven climate science and AI-powered modeling of wetland CH₄ fluxes.
This dataset integrates two complementary sources:
- FLUXNET-CH₄: Real-world, site-level methane flux observations across diverse wetland ecosystems.
- TEM-MDM: Physics-based, gridded global simulations of CH₄ fluxes and related variables generated by the Terrestrial Ecosystem Model with Methane Dynamics Module.
By synthesizing observational and model-generated data, X-MethaneWet provides a unique platform to explore the spatial and temporal variability of methane emissions across different scales and environments.
📁 Dataset Structure
The X-MethaneWet dataset integrates both simulated and observed CH₄ data across different spatial and temporal scales. The folder structure is organized as follows:
FLUXNET-CH4/
├── FLUXNET_CH4_2024.csv
└── FLUXNET_T1_DD.csv
TEM-MDM/
├── phh2o.nc
├── topsoil_bulk_density.nc
├── clelev.nc
├── clfaotxt.nc
├── cltveg.nc
├── vegetation_type_11.nc
├── wetlandtype.nc
├── climatetype.nc
├── ch4-1979-2018.txt
├── kco21979-2018.txt
├── monthly_NPP_{1979–2018}.nc
├── daily_ecmwf_PREC_{1979–2018}.nc
├── daily_ecmwf_SOLR_{1979–2018}.nc
├── daily_ecmwf_TAIR_{1979–2018}.nc
├── daily_ecmwf_VAPR_{1979–2018}.nc
└── CH4_emission_intensity_{1979–2018}.nc
Simulated Data: TEM-MDM/
The TEM-MDM (Terrestrial Ecosystem Model with Methane Dynamics Module) simulates methane-related processes including production, oxidation, and transport, enhanced by modeling permafrost and vegetation carbon dynamics. We simulate global daily CH₄ fluxes from 1979 to 2018 on a 0.5° spatial resolution grid.
Input drivers include:
- Static Features: Elevation (
clelev) [1], vegetation type (cltveg[2],vegetation_type_11[3]), wetland type (wetlandtype) [4], soil pH (phh2o) [5], bulk density (topsoil_bulk_density) [6], soil texture (clfaotxt) [7] - Yearly Variables: Global annual CO₂ concentrations (
kco2) and CH₄ (ch4) concentrations - Monthly Variable: Net Primary Productivity (
NPP) [8] - Daily Variables: Climate inputs from ERA-Interim, including precipitation (
PREC) [9], air temperature (TAIR) [9], solar radiation (SOLR) [9], and vapor pressure (VAPR) [9]
Each grid location contains 40 years of daily data, cut into 365-day yearly sequences. The processed input has shape:
N_input = 62,470 (locations) × 40 (years) × 365 (days) × 15 (features)
The output is a single target: daily CH₄ flux intensity [8], resulting in:
N_output = 62,470 × 40 × 365
Observed Data: FLUXNET-CH4/
The FLUXNET-CH₄ dataset provides site-level CH₄ flux measurements collected using eddy covariance towers. We selected 30 wetland sites spanning diverse types (bog, fen, marsh, swamp, salt marsh, tundra) and climate zones (Arctic-boreal, temperate, subtropical). [10]
Data characteristics:
- Time span: ~2006 to 2018 (varies by site)
- Temporal resolution: Daily CH₄ fluxes
- Features: Air temperature, precipitation, vapor pressure, elevation, IGBP land cover classification
We retain only wetland sites (based on IGBP classification) and use the WAD2M wetland map to annotate site types. To create consistent inputs, we:
- Fill missing features using the nearest TEM-MDM grid cell
- Align feature space with the simulated data (15 dimensions)
FLUXNET CH₄ emissions typically range from -100 to 900 mg C m⁻² d⁻¹.
📊 Model Benchmarking
We provide baseline performance evaluations using various sequential deep learning models (e.g., LSTM, Transformer) and explore transfer learning strategies to enhance generalization from simulated (TEM-MDM) to real (FLUXNET-CH₄) data.
Detailed descriptions of the dataset and benchmarking experiments are available in our paper, and the code implementation for all baseline models can be found in our GitHub repository.
Citation
If you use this dataset, please cite the following paper:
@article{sun2025x,
title={X-MethaneWet: A Cross-scale Global Wetland Methane Emission Benchmark Dataset for Advancing Science Discovery with AI},
author={Sun, Yiming and Chen, Shuo and Chen, Shengyu and Qiu, Chonghao and Liu, Licheng and Oh, Youmi and Malone, Sparkle L and McNicol, Gavin and Zhuang, Qianlai and Smith, Chris and Xie, Yiqun and Jia, Xiaowei},
journal={arXiv preprint arXiv:2505.18355},
year={2025}
}
References
[1] Thierry Toutin. 2002. Three-dimensional topographic mapping with ASTER stereo data in rugged topography. IEEE Transactions on geoscience and remote sensing 40, 10 (2002), 2241–2247.
[2] Jerry M Melillo, A David McGuire, David W Kicklighter, Berrien Moore, Charles J Vorosmarty, and Annette L Schloss. 1993. Global climate change and terrestrial net primary production. Nature 363, 6426 (1993), 234–240.
[3] John A Matthews. 1999. ON RECENTLY-DEGLACIATED SUBSTRATES. Ecosys- tems of disturbed ground 16 (1999), 17.
[4] Elaine Matthews and Inez Fung. 1987. Methane emission from natural wetlands: Global distribution, area, and environmental characteristics of sources. Global biogeochemical cycles 1, 1 (1987), 61–86.
[5] AJ Carter and RJ Scholes. 1999. Generating a global database of soil properties. Environmentek CSIR 10 (1999).
[6] Freddy Nachtergaele, Harrij van Velthuizen, Luc Verelst, Niels Batjes, Koos Dijk-shoorn, Vincent van Engelen, Guenther Fischer, Arwyn Jones, Luca Montanarella, Monica Petri, et al. 2010. Harmonized world soil database. In Proceedings of the 19th world congress of soil science, soil solutions for a changing world, Brisbane, Australia, Vol. 2010. International Union of Soil Sciences, 34–37.
[7] Q Zhuang, AD McGuire, JM Melillo, JS Clein, RJ Dargaville, DW Kicklighter, RB Myneni, J Dong, VE Romanovsky, J Harden, et al. 2003. Carbon cycling in extratropical terrestrial ecosystems of the Northern Hemisphere during the 20th Century: A modeling analysis of the influences of soil thermal dynamics, Tellus, 55 B, 751-776, 2003. REPORT SERIESof the MIT Joint Program on the Science and Policy of Global Change (2003).
[8] Q. Zhuang, J. M. Melillo, D. W. Kicklighter, R. G. Prinn, A. D. McGuire, P. A. Steudler, B. S. Felzer, and S. Hu. 2004. Methane fluxes between terrestrial ecosystems and the atmosphere at northern high latitudes during the past century: A retrospective analysis with a process-based biogeochemistry model. Global Biogeochemical Cycles 18, 3 (2004). doi:10.1029/2004GB002239
[9] DP Dee, SM Uppala, AJ Simmons, P Berrisford, P Poli, S Kobayashi, U Andrae, MA Balmaseda, G Balsamo, d P Bauer, et al. 2011. The ERA-Interim reanalysis: Configuration and performance of the data assimilation system. Quarterly Journal of the royal meteorological society 137, 656 (2011), 553–597.
[10] Kyle B Delwiche, Sara Helen Knox, Avni Malhotra, Etienne Fluet-Chouinard, Gavin McNicol, Sarah Feron, Zutao Ouyang, Dario Papale, Carlo Trotta, Eleonora Canfora, et al. 2021. FLUXNET-CH4: A global, multi-ecosystem dataset and analysis of methane seasonality from freshwater wetlands. Earth System Science Data Discussions 2021 (2021), 1–111.