Uncertainties in ecophysiological responses to environment, such as the impact of atmospheric and soil moisture conditions on plant water regulation, limit our ability to estimate key inputs for ecosystem models. Advanced statistical frameworks provide coherent methodologies for relating observed data, such as stem sap flux density, to unobserved processes, such as canopy conductance and transpiration. To address this need, we developed a hierarchical Bayesian State-Space Canopy Conductance (StaCC) model linking canopy conductance and transpiration to tree sap flux density from a 4-year experiment in the North Carolina Piedmont, USA. Our model builds on existing ecophysiological knowledge, but explicitly incorporates uncertainty in canopy conductance, internal tree hydraulics and observation error to improve estimation of canopy conductance responses to atmospheric drought (i.e., vapor pressure deficit), soil drought (i.e., soil moisture) and above canopy light. Our statistical framework not only predicted sap flux observations well, but it also allowed us to simultaneously gap-fill missing data as we made inference on canopy processes, marking a substantial advance over traditional methods. The predicted and observed sap flux data were highly correlated (mean sensor-level Pearson correlation coefficient = 0.88). Variations in canopy conductance and transpiration associated with environmental variation across days to years were many times greater than the variation associated with model uncertainties. Because some variables, such as vapor pressure deficit and soil moisture, were correlated at the scale of days to weeks, canopy conductance responses to individual environmental variables were difficult to interpret in isolation. Still, our results highlight the importance of accounting for uncertainty in models of ecophysiological and ecosystem function where the process of interest, canopy conductance in this case, is not observed directly. The StaCC modeling framework provides a statistically coherent approach to estimating canopy conductance and transpiration and propagating estimation uncertainty into ecosystem models, paving the way for improved prediction of water and carbon uptake responses to environmental change.
hierarchical Bayesian model
Bell, David M.; Ward, Eric J.; Oishi, A. Christopher; Oren, Ram; Flikkema, Paul G.; Clark, James S.; Whitehead, David. 2015. A state-space modeling approach to estimating canopy conductance and associated uncertainties from sap flux density data. Tree Physiology. 35(7): 792-802.