Global soil respiration estimation based on ecological big data and machine learning model.

Area of Science:

• Environmental Science
• Ecology
• Data Science

Background:

• Soil respiration (Rs) is a major CO2 flux from terrestrial ecosystems, but its drivers and models have uncertainties.
• Accurate estimation of global Rs is crucial for understanding climate change and carbon cycling.

Purpose of the Study:

• To estimate the spatial distribution and driving mechanisms of global soil respiration using machine learning and ecological big data.
• To develop and evaluate machine learning models for predicting Rs and identify key environmental factors influencing it.

Main Methods:

• Constructed a 27-dimensional ecological big data set with five categories of environmental factors.
• Applied four machine learning methods under four training strategies to model Rs.
• Utilized the Random Forest (RF) algorithm to estimate global Rs patterns and identify influential environmental drivers.

Main Results:

• The RF model demonstrated superior performance (R² = 0.782, RMSE = 285.896, MAE = 180.419 gCm⁻²y⁻¹), suitable for large-scale Rs estimation.
• Geographical location was found to be a more significant driver of Rs than Mean Annual Precipitation (MAP).
• Enhanced Vegetation Index 2 (EVI2) showed a greater contribution to Rs estimation than EVI and NDVI.

Conclusions:

• Ecological big data and machine learning offer a powerful approach for large-scale Rs estimation and driver analysis.
• The study provides improved insights into the spatial patterns and environmental controls of global soil respiration.
• Findings highlight the importance of geographical factors and advanced vegetation indices in modeling soil respiration.