Time-lagged Causal Inference between Climate and Vegetation Growth

Abstract

Climate change, characterized by increasing temperature, dryness, and extreme weather events, alters terrestrial ecosystem functioning including carbon sequestration capacity of plants. Despite the critical role of vegetation in global biogeochemical cycles and climate impact mitigation, its adaptive, time-lagged responses to environmental stressors and feedback to the climate remain poorly understood. This thesis addressed these gaps by applying advanced causal inference methods to investigate time-lagged and bidirectional causal relationships between climate and vegetation. First, I used Granger causality test to investigate cumulative causal effects of temperature, soil moisture (SM) and vapor pressure deficit (VPD) on satellite observed global vegetation photosynthesis (solar-induced fluorescence, SIF) and greenness (enhanced vegetation index, EVI) records. The findings reveal that terrestrial ecosystems characterized by arid and cold climates show more concurrent climate-vegetation interactions than other ecosystems. Whereas humid ecosystems with higher tree cover show substantial time-lag response of vegetation to climate by up to 6 months. Building on this, I applied the Peter and Clark Momentary Conditional Independence (PCMCI) algorithm to examine bidirectional causal effects between hydroclimate (SM, VPD) and vegetation greenness (EVI, leaf area index, LAI) in dryland ecosystems. Results consistently showed strong climate-to-greenness causality and weaker but significant vegetation feedback to climate, both modulated by aridity and tree cover, emphasizing dryland vulnerability. I also employed PCMCI to investigate tropical and subtropical humid and arid ecosystems, to study how canopy structure complexity (foliage height diversity, FHD; tree-cover fraction) mediates these hydroclimate-vegetation causal effects. The results demonstrated that complex canopies and higher tree-cover mitigate tropical and subtropical vegetation sensitivity to SM and VPD variability. This mitigation is notably stronger in humid ecosystems compared to drylands while vegetation feedback to amelioration of VPD is more pronounced in less complex or sparsely vegetated dryland canopies. Collectively, this thesis enhanced the understanding of the roles of aridity, canopy complexities, tree cover and time lag in bidirectional climate-vegetation interactions. By identifying directional causal effects and quantifying time-lagged responses modulated by climatic conditions and canopy structures, this thesis provides crucial insights into vegetation dynamics under increasing environmental stress towards hotter and drier conditions, thereby improving the predictive capacity of future behaviour of the carbon sink and water cycles.