Disentangling species-specific controls on urban tree transpiration: A Bayesian framework

Urban trees regulate water-energy exchanges via transpiration. Although the Jarvis framework is widely used to describe plant physiological responses to environmental drivers, including vapor pressure deficit ( V P D ), solar radiation ( R s ), air temperature ( T a ), and soil moisture content ( θ ), key parameter and subfunction structural uncertainties remain insufficiently assessed at the species-specific level. Another challenge is modeling the transpiration variability across multiple sites and seasons simultaneously, particularly in climatically heterogeneous urban settings. Using hourly sap flux density ( S F D ) observations from nine urban sites in Munich and Würzburg, Germany (2015–2021), we fitted 81 subfunction configurations using Markov Chain Monte Carlo (MCMC) Bayesian inference under uniform priors for two species with contrasting wood anatomy: diffuse-porous Tilia cordata and ring-porous Robinia pseudoacacia . Widely applicable information criterion (WAIC) based Sobol indices quantified structural sensitivity, and species-specific driver sensitivities were reconstructed under the best-performing structure. We then refitted the best Jarvis structure in a hierarchical Bayesian model, treating sites as random effects and adding a seasonal term. Results suggest that (i) at both the parameter and structure levels, the two species exhibit markedly different transpiration responses to environmental drivers, which can be interpreted in relation to known differences in hydraulic architecture and water-use strategy; (ii) model performance is sensitive to subfunction choice rather than increasing monotonically with structural complexity; and (iii) the hierarchical model captures cross-site variation in the transpiration upper bound and improves fit across sites and seasons, with the intraclass correlation coefficient (ICC) exceeding 80% for both species. Collectively, the framework provides uncertainty-aware, species-level Jarvis parameterizations at the single-tree scale, clarifies how species identity and model structure jointly shape sap flux dynamics, and offers an adaptable basis for interpreting and representing species-specific transpiration in urban microclimate models and urban forestry management.