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

Abstract

Global warming is one of the most pressing and critical problems facing the world today. It is mainly caused by the increase in greenhouse gases in the atmosphere, such as carbon dioxide (CO₂). Understanding how soils respond to rising temperatures is critical for predicting carbon release and informing climate mitigation strategies. Q₁₀, a measure of soil microbial respiration, quantifies the increase in CO₂ release caused by a [Formula: see text] Celsius rise in temperature, serving as a key indicator of this sensitivity. However, predicting Q₁₀ across diverse soil types remains a challenge, especially when considering the complex interactions between biochemical, microbiome, and environmental factors. In this study, we applied explainable artificial intelligence (XAI) to machine learning models to predict soil respiration sensitivity (Q₁₀) and uncover the key factors driving this process. Using SHAP (SHapley Additive exPlanations) values, we identified glucose-induced soil respiration and the proportion of bacteria positively associated with Q₁₀ as the most influential predictors. Our machine learning models achieved an accuracy of [Formula: see text], precision of [Formula: see text], an AUC-ROC of [Formula: see text], and an AUC-PRC of [Formula: see text], ensuring robust and reliable predictions. By leveraging t-SNE (t-distributed Stochastic Neighbor Embedding) and clustering techniques, we further segmented low Q₁₀ soils into distinct subgroups, identifying soils with a higher probability of transitioning to high Q₁₀ states. Our findings not only highlight the potential of XAI in making model predictions transparent and interpretable, but also provide actionable insights into managing soil carbon release in response to climate change. This research bridges the gap between AI-driven environmental modeling and practical applications in agriculture, offering new directions for targeted soil management and climate resilience strategies.

Keywords: Climate Change; Explainable Artificial Intelligence (XAI); Machine Learning; Q10; Soil Respiration Sensitivity.

Fig. 1

Workflow for predicting Q₁₀ sensitivity in soils. Environmental, biochemical, and microbiome variables are used as input features. The Extra Trees Classifier distinguishes between high and low Q₁₀ soils, with SHAP values providing interpretability. Clustering of low-Q₁₀ soils identifies subgroups at higher risk of transitioning to high Q₁₀. The histogram uses a total of 24 bins.

Fig. 2

ROC (a) and Precision-Recall (b) curves showing the performance of the machine learning classifiers. The bold blue line represents the average curve across 20 repetitions, while the shaded region corresponds to one standard deviation around the mean. The individual-colored curves represent the ROC and Precision-Recall curves from each of the 20 repetitions, illustrating the variability in model performance across iterations. The dashed red line indicates the performance of a random classifier (chance level).

Fig. 3

SHAP summary plot of feature importance in predicting Q₁₀ sensitivity. The x-axis represents each feature’s contribution to the model, where positive values indicate an increased likelihood of high Q₁₀. The color gradient (blue = low, red = high) represents feature values. The most influential predictors include “Bacteria_Positive” and “Microbial_Biomass”, followed by “Alkane” and “SOC” (soil organic carbon).

Fig. 4

t-SNE projections of input features and SHAP values. (a) Input feature-based t-SNE shows overlap between high-Q₁₀ (orange) and low-Q₁₀ (blue) soils. (b) SHAP value-based t-SNE reveals improved separation between high and low Q₁₀ soils. The Euclidean distance between the centroids of high and low soils in the input feature space was 5.22, whereas in the SHAP-based t-SNE embeddings, it increased to 20.59, highlighting improved class separability.

Fig. 5

Clustering of low-Q₁₀ soils based on SHAP-derived t-SNE embeddings. (a) K-means clustering achieved the highest Silhouette score. (b) Visualization of six identified clusters. (c) Predicted probabilities of transitioning to high Q₁₀ across clusters, with Clusters 4, 6, and 2 showing the highest risk.

Fig. 6

Box plots of the four most important variables identified by the SHAP summary plot across the six clusters. (a) Distribution of “Microbial_Biomass” values across clusters, showing that Clusters 4 and 6 have significantly higher values. (b) The distribution of “Bacteria Positive” values shows that Clusters 4 and 6 exhibit higher levels. (c) and (d) display the distributions for “Alkane” and “SOC” (Soil Organic Carbon), respectively, highlighting differences in these key variables across the identified clusters. These higher values in Clusters 4 and 6 are associated with an increased probability of transitioning to high Q₁₀ soils.