Data-driven insights into Australia's terrestrial carbon cycle dynamics

Abstract

Australia plays a disproportionately important role in the global terrestrial carbon cycle, yet its contribution remains poorly constrained due to large uncertainties in the magnitude, variability, and drivers of carbon fluxes across the continent. This thesis addresses these gaps by developing and applying a suite of regionally optimised, remote sensing-based datasets to improve the monitoring and understanding of Australia's terrestrial carbon dynamics.

Eddy covariance flux tower observations from the OzFlux network were integrated with satellite and climate data to produce 'AusEFlux' a high-resolution (~1 km) machine learning-derived dataset of gross primary productivity (GPP), ecosystem respiration (ER), and net ecosystem exchange (NEE) for 2003-2022. AusEFlux outperforms leading global upscaling products and terrestrial biosphere models in capturing both the magnitude and interannual variability of carbon fluxes, revealing Australia to be a persistent carbon sink with inter-annual variability strongly linked to rainfall anomalies, and seasonal variability linked to respiration dynamics.

To extend carbon cycle analysis prior to the MODIS era, 'AusENDVI' was developed - an Australia-specific NDVI dataset (1982-2022) that harmonises AVHRR and MODIS time series using gradient-boosted machine learning, orbital parameters, and climate covariates. AusENDVI exhibits improved temporal consistency and accuracy relative to global AVHRR products, providing a robust foundation for multi-decadal vegetation monitoring. Applying AusENDVI enabled a reassessment of long-term trends in Australia's land surface phenology, identifying both widespread amplification of vegetation growth cycles, and significant shifts in the timings of land-surface phenology.

Finally, by combining the AusEFlux modelling framework with AusENDVI, a four-decade GPP record was produced to empirically estimate the CO2 fertilisation effect (CFE) across Australia. Preliminary results indicate the strongest absolute CFEs occur in mesic forests, while arid ecosystems show more modest responses than previously hypothesised.

Collectively, this work demonstrates the value of regionally optimised, data-driven approaches for reducing uncertainties in carbon flux estimates and revealing how Australia's ecosystems respond to climate variability and rising atmospheric CO2. The datasets and methods developed here enable improved benchmarking of process-based models, inform climate-carbon feedback assessments, and support operational monitoring of the Australian carbon cycle. Show more