Analysis of Influencing Factors of Terrestrial Carbon Sinks in China Based on LightGBM Model and Bayesian Optimization Algorithm

Yana Zou and Xiangrong Wang ()
Additional contact information
Yana Zou: Department of Environmental Science & Engineering, Fudan University, Shanghai 200438, China
Xiangrong Wang: Department of Environmental Science & Engineering, Fudan University, Shanghai 200438, China

Sustainability, 2025, vol. 17, issue 11, 1-34

Abstract: With accelerating climate change and urbanization, regional carbon balance faces increasing uncertainty. Terrestrial carbon sinks play a crucial role in advancing China’s sustainable development under the dual-carbon strategy. This study quantitatively modeled China’s terrestrial carbon sink capacity and analyzed the multidimensional relationships between impact factors and carbon sinks. After preprocessing multi-source raster data, we introduced kernel normalized the difference vegetation index (kNDVI) to the Carnegie–Ames–Stanford approach (CASA) model, together with a heterotrophic respiration (R h ) empirical equation, to simulate pixel-level net ecosystem productivity (NEP) across China. A light gradient-boosting machine (LightGBM) model, optimized via Bayesian algorithms, was trained to regress NEP drivers, categorized into atmospheric components (O 3 , NO 2 , and SO 2 ) and subsurface properties (a digital elevation model (DEM), enhanced vegetation index (EVI), soil moisture (SM)), and human activities (land use/cover change (LUCC), POP, gross domestic product (GDP)). Shapley Additive Explanation (SHAP) values were used for model interpretation. The results reveal significant spatial heterogeneity in NEP across geographic and climatic contexts. The pixel-level mean and total NEP in China were 268.588 gC/m 2 /yr and 2.541 PgC/yr, respectively. The north tropical zone (NRZ) exhibited the highest average NEP (828.631 gC/m 2 /yr), while the middle subtropical zone (MSZ) and south subtropical zone (SSZ) demonstrated the most stable NEP distributions. LightGBM achieved high simulation accuracy, further enhanced by Bayesian optimization. SHAP analysis identified EVI as the most influential factor, followed by SM, NO 2 , DEM, and POP. Additionally, LightGBM effectively captured nonlinear relationships and variable interactions.

Keywords: terrestrial carbon sink; net ecosystem productivity; NEP; LightGBM; machine learning; Bayesian optimization; SHAP value (search for similar items in EconPapers)
JEL-codes: O13 Q Q0 Q2 Q3 Q5 Q56 (search for similar items in EconPapers)
Date: 2025
References: Add references at CitEc
Citations:

Downloads: (external link)
https://www.mdpi.com/2071-1050/17/11/4836/pdf (application/pdf)
https://www.mdpi.com/2071-1050/17/11/4836/ (text/html)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:gam:jsusta:v:17:y:2025:i:11:p:4836-:d:1663642

Access Statistics for this article

Sustainability is currently edited by Ms. Alexandra Wu

More articles in Sustainability from MDPI
Bibliographic data for series maintained by MDPI Indexing Manager ().