The role of natural methane (CH₄) oxidation by microbes (i.e. methanotrophs) in upland soils, as the second-largest global sink after chemical sinks [1], remains critically under-appreciated despite its potential to influence the global CH₄ budget significantly [2,3]. The current estimate of the global CH₄ soil sink is ~30 TgCH₄ yr-1 but with a huge uncertainty (7 to >100 TgCH₄ yr-1) [1,4,5]. Soil amendment [6–10], still in its early stages and tested on a small scale, has the potential to enhance the global soil CH₄ sink, which can be realized only with precise quantification of the sink's magnitude on a large scale and a thorough understanding of how CH₄ oxidation are interconnected with climatic and soil environmental variations. Conventional approaches, such as the process-based (PB) models [11,12] or pure data-driven methods [13,14] are limited by underrepresented soil microbial processes or lack of big datasets for robust training. The growing field of knowledge-guided machine learning (KGML, or hybrid modeling) [15,16] provides a promising modeling framework that combines the advantages of process understanding, machine-learning (ML) models, and multi-scale datasets, and demonstrates the successes in physical [16–20] and biogeochemical modeling [21–24].

The guiding science question of this project is: (1) what are the driving factors to change current soil CH₄ oxidation and (2) what is the future potential of soil microbial CH₄ sink? The key challenge is estimating the global soil CH₄ sink spatial and temporal variability with limited data [13,25]. This requires a comprehensive approach integrating the recent data, scientific knowledge of CH₄ oxidation, and advanced modeling. Our team is playing a leading role in modeling CH₄ biogeochemistry. Leveraging existing KGML frameworks [22,24], we aim to refine CH₄ sink estimates, deepen the understanding of driving mechanisms, and explore future sink potentials under various scenarios, including changes in climate, CH₄ concentrations and soil properties of pH, moisture (SM), organic carbon (SOC) and inorganic nitrogen (SIN) (Fig. 1). Previous studies show that these variables are controlling soil CH₄ oxidation, with responses varying by ecosystem [5,14,26,27]. Success of this project will be measured through the KGML's predictive accuracy against historical data and its utility in guiding experimental soil amendments, as evidenced by model/data download metrics and experimental validation.

Leveraging advanced observations, process-based modeling, and cutting-edge machine learning, this project will provide a solution to enhance the global soil CH₄ sink. By deepening our understanding of the mechanisms driving soil CH₄ sinks and pinpointing potential hot moments and spots across future scenarios, we seek to pave the way for methanotrophy enhancement via soil amendments.

With KGML solution, we will develop a user-friendly web-app and Python library (Detailed in Section 2.2.3) to (1) facilitate quick access to KGML pre-simulated regional CH₄ potentials under various climate and soil conditions, and (2) enable out-of-scenario simulations based on customized soil amendment plans.