Machine learning approach for Studying the multifunctionality of soil against global climate changes

Abstract

Soil ecosystem multifunctionality (EMF) represents the soil biodiversity and the soil capacity for sustainable development. Due to the high heterogeneity of climate and land use changes, mapping the patterns of global soil EMF in the past and future is necessary and challenging. EMF data from 790 sampling points worldwide were analyzed using a random forest algorithm with SHAP analysis, partial dependence analysis and structural equation modeling to elucidate driving mechanisms of soil EMF under global change and to forecast the global distribution of soil EMF. This also unveiled the interplay between climate and land use changes on EMF. This work revealed that EMF hotspots are distributed in the Caribbean, Southeast Asia and Eastern Europe and are twice as common in these areas than they are in western Asia, North Africa and South Asia. The interplay of multiple dominant factors has antagonistic or synergistic effects and generates tipping points, which are critical for understanding the change processes of EMFs. From 2007 to 2018, land use changes were the dominant factor leading to fluctuations in EMF. However, climate change will become the dominant factor in the future. Land use optimization can mitigate EMF fluctuations in response to climate change. Changes from deserts to grasslands in Africa and from forests to grasslands in Oceania can combat the decline in EMF induced by climate change by 2100. According to the distribution patterns of EMF and optimization, hotspot regions could be protected, and land use planning could be conducted to prevent the degeneration of soil.

Keywords

• Climate change
• Land use
• Soil
• Multifunctionality
• Machine learning