Methane Concentration Inversion Based on Multi-Feature Fusion and Stacking Integration.

Area of Science:

• Environmental Science
• Data Science
• Atmospheric Chemistry

Background:

• Traditional methane concentration inversion methods often rely on simplistic features and models, resulting in suboptimal accuracy.
• Accurate methane monitoring is crucial for understanding greenhouse gas dynamics and climate change mitigation efforts.

Purpose of the Study:

• To develop and validate a novel methane concentration inversion method that overcomes the limitations of existing approaches.
• To improve the accuracy and generalization capability of methane concentration inversion using advanced machine learning techniques.

Main Methods:

• Proposed a methane concentration inversion method integrating multi-feature fusion with Stacking ensemble learning.
• Employed a series-parallel cascade structure of base and meta-models to capture complex feature relationships.
• Conducted experimental validation in the eastern Xinjiang region.

Main Results:

• The proposed Stacking ensemble model demonstrated superior inversion performance compared to other typical methods.
• Achieved high performance metrics: R² of 0.9747, Root Mean Square Error (RMSE) of 2.8294, and Mean Absolute Error (MAE) of 1.5299.
• Identified seasonal methane concentration patterns, with lower averages in spring/autumn and higher averages in summer/winter.

Conclusions:

• The multi-feature fusion and Stacking ensemble learning approach significantly enhances methane concentration inversion accuracy.
• The method effectively explores intrinsic relationships between various feature factors and methane concentration.
• The findings provide a more robust tool for methane emission monitoring and analysis.