Evaluating soil CO₂ efflux variability across diverse land uses in semi-arid region of Southern India and its prediction through machine learning

Soil CO₂ efflux is a major component of ecosystem respiration and plays a vital role in regulating the global carbon cycle. Variations in land use, soil properties, and seasonal dynamics strongly influence soil CO₂ emissions, particularly in semi-arid regions where climatic constraints interact with land management practices. Despite its importance, integrated field-based assessments combined with advanced modeling approaches remain limited for semi-arid region of India. This study aimed to develop a comprehensive understanding of the variability of soil CO₂ efflux across different land uses and seasons in a semi-arid region of southern India, and to identify suitable machine learning algorithms to model relationships between soil properties and CO₂ efflux for prediction and validation. Soil CO₂ efflux was assessed across six major land uses forest, agriculture, horticulture, plantation, wasteland, and fallow land in the semi-arid region of southern India. Field measurements were conducted during representative dry and wet seasons using an in situ Vaisala GMP343 CO₂ diffusion probe. Soil temperature and moisture were recorded concurrently, and surface soil samples were analysed for physicochemical properties. Statistical analyses, including correlation, ANOVA, and principal component analysis, examined relationships between soil properties and CO₂ efflux. Machine learning models Partial Least Squares Regression (PLSR), Random Forest (RF), Gradient Boosting Regression (GBR), and k-Nearest Neighbors (KNN) were developed to predict soil CO₂ efflux. Model performance was evaluated using R², RMSE, and MAE. Soil CO₂ efflux showed significant seasonal and land-use variability, with higher emissions during the wet season. Forest soils recorded the highest efflux, while fallow and wasteland soils showed the lowest values. Soil moisture, temperature, and organic carbon (OC) were dominant drivers, showing strong correlations with fluxes. Among models, RF performed best (R² = 0.97; RMSE = 3.36), followed by GBR, KNN, and PLSR. RF variable importance indicated soil moisture as the most influential predictor, followed by temperature and OC. Land use, seasonality, and key soil properties strongly regulate soil CO₂ efflux in semi-arid ecosystems. Machine learning, particularly RF, effectively captured complex relationships. However, limitations include daytime-only measurements and exclusion of biological parameters. Future studies should incorporate continuous and year-round monitoring.