Patterns of association between carbon sinks and multidimensional human driving factors in China: a hybrid machine learning framework

Regarding the urgency of carbon neutrality, the necessity to investigate the dynamics of carbon sinks and their human drivers is overwhelming. However, most of existing literature only examines the nexus between carbon sinks and their human drivers by the traditional econometric regressions with predetermined function forms. Such predefined functions may undermine the asymptotical property of estimation and thus yield potentially biased results. This study addresses such flaw by leveraging the merits of Generalized Additive Model (GAM) to investigate the possible nonlinear relationships between carbon sinks and their human driving factors. Elastic Net (EN) is also employed to select valid variables. The above two machine learning algorithms are integrated to construct a methodological framework for identifying complex relationships. Based on the panel dataset in China from 2007 to 2022, this study finds that China’s carbon sinks are highly concentrated in provinces abundant in ecological resources. EN finally selects 7 valid drivers. With respect to their associations with carbon sinks, GAM identifies distinct relationship patterns across multiple drivers. The proportion of primary industry within GDP exhibits a pronounced U-shaped curve. Per capita disposable income follows a W-shaped trajectory. The total afforestation area shows a gentle N-shaped pattern. Both the total sowing area of crops and the number of domestic patents display fluctuating yet generally upward trends. The total power of agricultural machinery is characterized by an inverted U-shaped curve. These diverse nonlinear patterns significantly extend the theory of Environmental Kuznets Curve, demonstrating that the dynamics of carbon sinks follow complex patterns. These findings imply that carbon sink enhancement policies should be regionally differentiated and targeted at the specific threshold intervals of each driving factor, rather than relying on uniform strategies.