Improving model robustness to drought stress by constraining plant hydraulics with complementary in situ measurements

Droughts threaten ecosystems worldwide and are projected to occur more frequently and with greater intensity in the future. Accurately projecting ecosystem responses to these climate extremes relies on vegetation models. While process-based models are evolving fast and many models now represent plant hydraulic processes, including these mechanisms often comes at the cost of an increase in the number of difficult-to-constraint model parameters. Concurrently, recent experimental advances open a new avenue for parameter constraining, by providing high-temporal resolution data on plant hydraulic variables, for example continuous in situ measurements of water potential and sap flux. However, much of this novel data has not yet been considered by vegetation models.

Here, we utilize a rare, comprehensive time-series of data obtained in the Hainich experimental forest, specifically high-temporal resolution datasets of (1) sap flux, (2) stem water potential, (3) Net Ecosystem Exchange (NEE), and (4) Evapotranspiration (ET). With this data spanning both the water and carbon axes of plant function, we constrain the terrestrial biosphere model QUINCY and the latest development of its plant hydraulic architecture. We find that integrating such complementary experimental data yields three key outcomes. First, it evaluates the physical representation of plant hydraulic theory within the model. Second, it results in tighter constraints on plant-hydraulic parameters. High-temporal resolution water potential and sap flow data are vital here, as they resolve the diurnal lags necessary to identify capacitance parameters that remain unidentifiable under daily or weekly sampling. By capturing these fast-response dynamics, the model not only narrows parameter uncertainty but also reveals critical functional interdependencies and correlations that define plant hydraulic strategy. Third, these constraints yield more robust projections by significantly reducing the variability in simulated stocks and fluxes under future climate scenarios. We conclude that the growing availability of continuous data from novel physiological sensors is essential to constrain and build trust in increasingly complex vegetation models, as demonstrated here for plant hydraulics.