Integrating spatiotemporal variation of climate improves predictability of tree growth

Wu, Fang, Jia, Junwen, Li, Cheng, Jiang, Yuan, Savary, Serge and Cui, Xuefeng 2025. Integrating spatiotemporal variation of climate improves predictability of tree growth. Journal of Ecology 10.1111/1365-2745.70073

PDF - Accepted Post-Print Version Download (1MB)

Abstract

Forests are critical components of the global carbon cycle and thus influence climate change. Climate, in turn, strongly affects forest growth. Projecting the growth–climate relationship to unobserved regions or to past and future periods is crucial for climate change mitigation and adaptation. Tree‐ring data, quantifying changes in wood biomass—a key component of forest productivity—provide valuable insights into this relationship. We investigate the generalizability and application domains of three climate‐based modelling approaches for projecting tree growth. These include a time‐fitted model using climate over time at specific sites, a space‐fitted model using climate across regions and a spatiotemporal (ST)‐fitted model integrating both dimensions. All models were developed using Random Forest to predict the growth of Picea mariana under varying climate conditions. Key findings include (1) Both time‐ and ST‐fitted models performed well for temporal projections at a given observation site (R2 = 0.69, 0.68), while the space‐fitted model performed poorly (R2 ≈ 0). (2) Both space‐ and ST‐fitted models performed well for spatial projections based on multiple sites, each with partial observations (R2 = 0.81, 0.80), while the time‐fitted model performed poorly (R2 ≈ 0). (3) Only the ST‐fitted model generated acceptable projections in completely unobserved regions (R2 = 0.21). (4) Model performance declined progressively from training to internal validation and subsequently to external validation. Synthesis: Our results emphasize that integrating spatiotemporal information improves model generalizability and tree growth projection accuracy under climate change, especially for unobserved regions. The study also highlights the necessity of independent external validation in model evaluation. These findings offer actionable guidance for identifying reliable modelling approaches for tree growth projection, thereby informing forest management strategies for climate change mitigation and adaptation.

Item Type:	Article
Date Type:	Published Online
Status:	In Press
Schools:	Schools > Earth and Environmental Sciences
Additional Information:	License information from Publisher: LICENSE 1: URL: http://onlinelibrary.wiley.com/termsAndConditions#vor, Start Date: 2025-06-02
Publisher:	Wiley
ISSN:	0022-0477
Date of First Compliant Deposit:	20 June 2025
Date of Acceptance:	27 April 2025
Last Modified:	26 Jun 2025 17:15
URI:	https://orca.cardiff.ac.uk/id/eprint/179158

Actions (repository staff only)

Edit Item