生态学报2026，Vol.46Issue(16)：8942-8956,15.DOI:10.20103/j.stxb.202511092980
基于可解释机器学习的土壤呼吸驱动因素及空间异质性分析
The analysis of soil respiration driving factors and their spatial heterogeneity based on lnterpretable machine learning
摘要
Abstract
Soil respiration(Soil Respiration,Rs)represents the largest carbon flux from terrestrial ecosystems to the atmosphere and plays a crucial role in assessing the terrestrial carbon cycle and understanding climate change.However,due to its significant spatial heterogeneity,accurately estimating regional soil carbon fluxes is highly challenging.The soil carbon flux in different regions is influenced by the complex interactions of various factors,making it difficult to precisely estimate regional soil carbon fluxes.To better understand the dominant drivers of global soil respiration and improve the estimation of regional soil carbon fluxes,this study uses the Global Soil Respiration Database in combination with multiple environmental factors and applies explainable machine learning models to systematically analyze the dominant drivers of global Rs.The aim is to provide insights into accurately estimating regional carbon fluxes and to inform the development of global carbon cycle models.The study begins by identifying the primary controlling factors at the global scale,including mean annual temperature,precipitation,soil bulk density,and the normalized difference vegetation index(normalized difference vegetation index,NDVI).These factors are identified as the key drivers of soil respiration.Further modeling is carried out by stratifying the data according to ecosystem types and climate zones,and exploring the overall model uncertainty.The results reveal significant differentiation across different types of ecosystems.For example,croplands and grasslands are mainly controlled by precipitation,forests are primarily driven by temperature,and shrublands and wetlands are mainly constrained by soil texture.At the climate scale,water stress in arid regions makes soil structure the key controlling factor.In temperate regions,soil bulk density and vegetation activity play critical roles,while in tropical regions,soil temperature and elevation exert dominant regulation.In boreal and polar regions,both temperature stress and nutrient limitation have dual influences on soil respiration.By utilizing the SHAP method within a dual-partition modeling framework of ecosystems and climate zones,this study reveals the dominant factors and their interaction mechanisms that govern global Rs.Shapley Additive Explanations(Shapley Additive Explanations,SHAP)provides a way to interpret the contributions of each factor to the model's output,allowing for a deeper understanding of how various factors and their interactions influence global soil respiration.These findings not only enhance the understanding of the spatial heterogeneity of soil carbon cycling but also provide a theoretical basis for improving global carbon cycle models and refining terrestrial carbon cycle accuracy.By understanding how ecological and climatic factors interact in different regions,this study contributes to more accurate estimates of soil carbon fluxes and supports efforts to address climate change more effectively.关键词
土壤呼吸/空间异质性/机器学习/驱动因素Key words
soil respiration/spatial heterogeneity/machine learning/driving factors引用本文复制引用
王楠,姜俊杰,赵婷,陈哲昊,胡军国..基于可解释机器学习的土壤呼吸驱动因素及空间异质性分析[J].生态学报,2026,46(16):8942-8956,15.基金项目
国家自然科学基金资助项目(32371668,31971493) （32371668,31971493）
