Machine Learning Driven Sensitivity Analysis of E3SM Land Model Parameters for Wetland Methane Emissions

Sandeep Chinta,Xiang Gao,Qing Zhu

2023-12-05

Abstract:Methane (CH4) is the second most critical greenhouse gas after carbon dioxide, contributing to 16-25% of the observed atmospheric warming. Wetlands are the primary natural source of methane emissions globally. However, wetland methane emission estimates from biogeochemistry models contain considerable uncertainty. One of the main sources of this uncertainty arises from the numerous uncertain model parameters within various physical, biological, and chemical processes that influence methane production, oxidation, and transport. Sensitivity Analysis (SA) can help identify critical parameters for methane emission and achieve reduced biases and uncertainties in future projections. This study performs SA for 19 selected parameters responsible for critical biogeochemical processes in the methane module of the Energy Exascale Earth System Model (E3SM) land model (ELM). The impact of these parameters on various CH4 fluxes is examined at 14 FLUXNET- CH4 sites with diverse vegetation types. Given the extensive number of model simulations needed for global variance-based SA, we employ a machine learning (ML) algorithm to emulate the complex behavior of ELM methane biogeochemistry. ML enables the computational time to be shortened significantly from 6 CPU hours to 0.72 milliseconds, achieving reduced computational costs. We found that parameters linked to CH4 production and diffusion generally present the highest sensitivities despite apparent seasonal variation. Comparing simulated emissions from perturbed parameter sets against FLUXNET-CH4 observations revealed that better performances can be achieved at each site compared to the default parameter values. This presents a scope for further improving simulated emissions using parameter calibration with advanced optimization techniques like Bayesian optimization.

Atmospheric and Oceanic Physics,Machine Learning,Data Analysis, Statistics and Probability

What problem does this paper attempt to address?

The paper aims to address several key issues related to wetland methane emissions, primarily focusing on reducing the impact of model parameter uncertainty on prediction accuracy. The main issues the paper attempts to address are as follows:

### Key Issues Addressed by the Paper

1. **Identifying Key Parameters**: The study aims to determine which model parameters are most sensitive to simulated wetland methane emissions through sensitivity analysis (SA). This helps to understand which parameters are crucial for methane emissions and may become the focus of future model improvements.

2. **Temporal and Spatial Characteristics of Parameter Sensitivity**: The study explores the seasonal and annual variations in these parameter sensitivities and the spatial differences between different vegetation types. This helps to understand which parameters have the greatest impact on methane emissions at different times and locations.

3. **Improving Simulated Emissions**: By comparing the methane emissions simulated with perturbed parameter sets to FLUXNET-CH4 observational data, the study evaluates model performance and explores whether it is possible to further improve simulation accuracy through parameter calibration and other methods.

### Research Background and Methods

- **Research Background**: Methane, as the second most important greenhouse gas after carbon dioxide, has a significant impact on global warming. Wetlands are one of the largest natural sources of methane emissions, but their emission amounts are highly uncertain, partly due to uncertain parameters in the multiple physical, biological, and chemical processes included in the models.

- **Research Methods**:
- Using machine learning algorithms (specifically Gaussian Process Regression, GPR) to construct simulation experiments to reduce computational costs.
- Employing variance-based sensitivity analysis methods (such as the Sobol method) to quantify the impact of different input parameters on model outputs.
- Selecting 19 key parameters for sensitivity analysis and using Latin Hypercube Sampling (LHS) to generate different parameter combinations.
- Conducting single-point simulation experiments at 14 FLUXNET-CH4 sites with different vegetation types.

### Main Findings

- Model parameters related to methane production and diffusion generally show the highest sensitivity, despite significant seasonal variations.
- By comparing the emissions simulated with perturbed parameter sets to FLUXNET-CH4 observational data, it was found that better performance could be achieved at each site compared to the default parameter values.
- This indicates that simulation accuracy can be further improved through parameter calibration and advanced optimization techniques (such as Bayesian optimization).

In summary, this study aims to improve the accuracy of wetland methane emission simulations through the comprehensive application of machine learning and sensitivity analysis methods, providing valuable insights and potential directions for future research.