Beyond optimality: Dryland ecosystems infrequently use water efficiently for carbon gain

Optimality theory assumes plants maximize carbon gain per unit water lost and is often implemented to scale leaf-level carbon gain and water use to regional and global scales. Optimality theory is often mathematically represented by assuming plant water-use efficiency (WUE) scales with VPD^k, where k = ½ represents expected optimal behavior. It is unclear, however, if this relationship holds in arid and semi-arid ecosystems that are strongly impacted by soil and atmospheric moisture status. We used data from seven flux tower sites along an aridity gradient in New Mexico to answer: how does the relationship between WUE and VPD compare to expectations based on optimality theory? To address this question, we integrated the Dynamic Evapotranspiration Partitioning Approach for Rapid Timescales with a stochastic antecedent model to estimate ecosystem-level WUE (GPP/T) and the net sensitivity of WUE to VPD, or k^Dynamic, which we compare to the theoretical optimal sensitivity of k = ½. Our results show that optimality theory is not always appropriate, and k^Dynamic often deviates from ½, especially at some of the more arid sites or during periods of low soil moisture. At less arid, higher elevation sites, k^Dynamic is most consistent with optimality theory at moderate VPD levels, but not at high VPD. In general, the sensitivity of WUE to VPD is highly variable such that k^Dynamic exhibits notable daily and seasonal variability, suggesting highly dynamic stomatal behavior. These results emphasize that representing plant water-use strategies as dynamic in time and space is critical to improving large-scale estimates of plant water use.

This research was supported by NSF Hydrologic Sciences award EAR1834699 to K.S‐C., K.O., and M.L., a DOE Ameriflux Management Project award to M.L., a USGS‐NSF Internship Program supplement to award EAR1834699, and NASA FINESST award 80NSSC22K1443 to E.R. and K.O. We would also like to thank Michael Fell for his R functions for summarizing Bayesian output. Any use of trade, firm, or product names is for descriptive purposes only and does not imply endorsement by the U.S. Government.