Leveraging explainable AI to predict soil respiration sensitivity and its drivers for climate change mitigation

Area of Science:

• Environmental Science
• Soil Science
• Artificial Intelligence

Background:

• Global warming necessitates understanding soil carbon release.
• Soil respiration sensitivity (Q10) is a critical indicator of temperature response.
• Predicting Q10 across diverse soils is challenging due to complex interactions.

Purpose of the Study:

• To predict soil respiration sensitivity (Q10) using explainable artificial intelligence (XAI).
• To identify key factors influencing Q10.
• To segment soils based on Q10 and transition probabilities.

Main Methods:

• Applied machine learning models with XAI (SHAP values).
• Utilized t-SNE and clustering for soil segmentation.
• Validated model performance with accuracy, precision, AUC-ROC, and AUC-PRC metrics.

Main Results:

• Identified glucose-induced soil respiration and bacterial proportion as key Q10 predictors.
• Achieved high model accuracy ([Formula: see text]), precision ([Formula: see text]), AUC-ROC ([Formula: see text]), and AUC-PRC ([Formula: see text]).
• Segmented low Q10 soils, identifying those prone to transitioning to high Q10 states.

Conclusions:

• XAI enhances transparency and interpretability in environmental modeling.
• Findings provide actionable insights for managing soil carbon release under climate change.
• This research bridges AI, environmental modeling, and agricultural applications for climate resilience.