Upscaling eddy covariance measurements of carbon and water fluxes to the continental scale by incorporating GEDI-derived canopy structural complexity metrics

Abstract

Upscaling carbon and water fluxes measured from eddy covariance (EC) sites to regional and global scales with machine learning (ML) methods allows us to assess land-atmosphere carbon and water exchange over these broad scales. Although canopy structure and diversity are crucial in regulating carbon and water fluxes by affecting photosynthetic capacity, turbulence, and seasonal dynamics, ML-based upscaling of these fluxes has typically relied on climate forcing data and satellite-derived vegetation indices, and overlooked structural diversity. We used canopy height (RH) and foliage height diversity (FHD) data derived from NASA's Global Ecosystem Dynamics Investigation (GEDI) instrument to investigate how ecosystem structure and diversity influence the upscaling of EC carbon and water fluxes. We combined canopy structural diversity metrics derived from GEDI, flux tower data of over 90 sites from AmeriFlux and National Ecological Observatory Network (NEON), the Near-Infrared Reflectance of Vegetation (NIRv) derived from the Moderate Resolution Imaging Spectroradiometer (MODIS), and meteorological data from Daymet. ML methods were used to develop predictive models for both gross primary production (GPP) and evapotranspiration (ET) and to generate gridded carbon and water fluxes across the conterminous United States (CONUS). The incorporation of GEDI-derived RH and FHD improved the estimation of GPP by increasing the coefficient of determination (R²) from 0.79 to 0.91 and reducing the root-mean-square error (RMSE) from 1.77 to 1.14 gC m⁻² d⁻¹. Similarly, including RH and FHD increased R² from 0.79 to 0.85 and decreased RMSE from 0.82 to 0.68 mm d⁻¹ for the estimation of daily ET. Using the trained ML models, we generated gridded GPP and ET datasets with 1 km resolution and daily timestep across the CONUS for 2019–2023 (i.e., the GEDI era). Additionally, we explored effects of canopy structural complexity on ecosystem GPP and ET based on our gridded GPP and ET estimates. Annual GPP and ET showed positive logarithmic relationships with FHD, increasing with greater canopy structural complexity, though the responses weakened as FHD continued to rise. Greater canopy complexity was associated with a reduction in the seasonal variability of GPP and ET. Under severe drought events, greater canopy complexity enhanced drought resilience by reducing GPP and ET loss. Incorporating canopy structural diversity can improve the upscaling of EC carbon and water fluxes and our understanding of ecosystem responses to environmental changes.