Quantify the spatio-temporal variability and changes of terrestrial gross primary production across the globe during 2000-2024

Abstract

Gross Primary Production (GPP) is the largest carbon flux in terrestrial ecosystems and plays a critical role in the global carbon cycle, climate regulation, ecosystem integrity, biodiversity maintenance, and food security. The ability to continuously and accurately observe and simulate GPP is fundamental for improving our understanding of biogeochemical feedback, assessing ecosystem responses to environmental change, and informing land management and sustainability strategies. Currently, GPP is primarily derived from eddy covariance flux tower observations and remote sensing-driven model products. However, existing global GPP datasets show considerable disagreement in both magnitude and seasonal dynamics, underscoring the need to improve GPP modeling frameworks for more robust global estimates. Additionally, the impact of natural disturbances (e.g., wildfire) on GPP and its post-disturbance recovery patterns remains poorly understood due to a lack of systematic assessments. This dissertation presents the development, improvement, and application of the Vegetation Photosynthesis Model (VPM), a light use efficiency (LUE) model, as a tool for global GPP estimation. This dissertation addresses two central research questions: (1) Can the newly developed improved versions of the VPM model (v3.0 and v4.0) accurately simulate both the magnitude and seasonal variation of GPP across site and global scales (Chapters 2–5)? (2) How do wildfire disturbances alter ecosystem carbon uptake (GPP), and what are the environmental drivers of post-fire GPP recovery trajectories (Chapters 6–7)? Key findings show that VPM v3.0 can effectively capture observed GPP seasonality at four long-term deciduous broadleaf forest flux tower sites (2000–2020) and across 205 globally distributed sites (spanning 1658 site-years and 11 biomes), with high agreement to flux data (slope = 0.97, R2 = 0.78, RMSE = 1.46 g C m-2 day-1). The model-derived global GPP was estimated at 143 ± 4 Pg C yr-1, and its 8-day time-series product showed strong agreement with other benchmark GPP datasets (e.g., VPM v2.0, BEPS, MOD17, GOSIF, BESS, and FLUXCOM), with correlation coefficients ranging from 0.88 to 0.95. Further extending the model, I incorporated atmospheric CO2 effects into VPM to develop VPM v4.0, which produced a higher global GPP estimate of 155 ± 6 Pg C yr-1, aligning with recent global estimations (150-175 Pg C yr-1). The model also captured interannual GPP variability (1.18 Pg C yr-1) and long-term increasing trends (0.78 Pg C yr-2). Using the GPP dataset estimated by VPM v3.0, I investigated the impacts of fire in two contrasting fire-prone regions: Australia and high-latitude forests. In Australia, I found that ecosystems show strong and rapid post-fire GPP recovery, primarily driven by post-fire precipitation. As a result, interannual variations in continental-scale GPP were more influenced by climate and land-use change than by fire, due to effective compensatory recovery. In high-latitude forests, the first-year post-fire GPP recovery rate has declined since the early 2000s, mainly due to intensifying fire severity, increasing temperature, and vapor pressure deficit (VPD), and decreasing soil moisture. Overall, this dissertation highlights the importance of estimating long-term, high-accuracy GPP products and improves our understanding of ecosystem responses to disturbance through a refined modeling framework. The insights gained offer valuable contributions to carbon cycle research and provide actionable tools for evaluating biosphere–atmosphere feedback under ongoing global change.