Reconciling Cross-Scale Discrepancies in CO₂ Fluxes. Preliminary Findings from the BenchFlux Project

The BenchFLUX project represents an important advance in evaluating nature-based climate solutions (NbCS) to address the growing climate crisis. The benchmarking of CO₂ fluxes using flux tower measurements and Earth Observation (EO) data is the project's aim, employing multiple approaches to introduce, compare, and integrate temporal and spatial scales. The methods used account for the nonlinear behavior of carbon flux dynamics across scales. Therefore, measurement harmonizations are fundamental for aligning ground and atmospheric measurements. And thus, BenchFLUX provides reliable models and products that accurately track carbon emissions from small local areas to the global scale.

To achieve this goal, the project combines eddy covariance flux tower ground data with multi-source EO data to create harmonized datasets for various advanced machine learning models at different scales. The processes use cloud computing technologies, such as Google Earth Engine and cloud-optimized workflows, to produce spatial CO₂ flux data at multiple spatial resolutions. The proposed methods, including Bayesian and knowledge-guided approaches to achieve accurate and consistent results, and the final products are nested across different temporal and spatial scales among the six international research teams, serving as an integrated element for cross-scale continuity.

The spatial scalability of these methods is analyzed in the project prototype results. Preliminary monthly average CO₂ exchange (GPP) results are provided from the highly standardized NEON sites database for the higher spatial resolution models, revealing discrepancies at multiple scales during both the growing and non-growing seasons. The initial results will also compare coarser spatial resolution models with the eddy covariance ground truth data. All these ongoing comparisons aim to identify the most reliable methods for scaling carbon flux estimates. This will help determine the best combination of techniques to ensure high local precision and global consistency, ultimately supporting continuous cross-scale resource management.