Scaling terrestrial ecosystem water fluxes at the interface of in-situ measurements and machine learning

Over a century of study of ecosystem water fluxes has resulted an abundance of in-situ measurement techniques causing the availability of robust and continuous measurements to quietly grown by orders of magnitude in the last few years. For example, the ten years since the release of the FLUXNET 2015 synthesis dataset (which contained records dating back 25 years) has more than doubled the amount of eddy covariance measurements publicly released, with now over a million total days of measurements taken from over 450 sites globally. Furthermore, other dataset synthesis efforts for sap flux, soil moisture, stream flow, etc., as well as combinations with proximal and remote sensing, quickly result in datasets much larger than can be tackled by an individual. The advancement of machine learning and computational power to digest and utilize this deluge of environmental data hold promise to be able to understanding global water cycles in an unprecedented detail. However, limitations to applying machine learning methods often comes not from computational power, but rather in understanding the particular uncertainties and nuances, as well as unique information on ecosystem functioning, that each dataset brings.

Here, I briefly outline the current state of the art of scaling ecosystem water fluxes from in-situ to regional and global scales through the example of eddy covariance and the FLUXCOM-X framework [1]. Particularly, I highlight the current sources of uncertainties, such as measurement corrections and spatial extrapolation, as well as the potential limitations of machine learning and artificial intelligence in tackling these issues. Furthermore, comparing up-scaled eddy covariance evapotranspiration and transpiration products to terrestrial land surface models demonstrates the discrepancy in the global ratio of transpiration to ET between process based and data driven methods, demonstrating how machine learning from in-situ scales can inform our understanding of global cycle. Finally, I explore how integration of multiple data sources holds promise in isolating individual ecosystem water fluxes and to link the local measurements of individual plants and ecosystems to the regional and global scales.

1 - Nelson and Walther et al., 2024. X-BASE: the first terrestrial carbon and water flux products from an extended data-driven scaling framework, FLUXCOM-X. Biogeosciences 21, 5079–5115. https://doi.org/10.5194/bg-21-5079-2024