A synthesis of bias and uncertainty in sap flow methods

Sap flow measurements with thermometric methods are widely used to measure transpiration in plants. Different method families exist depending on how they apply heat and track sapwood temperature (heat pulse, heat dissipation, heat field deformation or heat balance). These methods have been calibrated for many species, but a global assessment of their uncertainty and reliability has not yet been conducted. Here we perform a meta-analysis of 290 individual calibration experiments assembled from the literature to assess calibration performance and how this varies across methods, experimental conditions and wood properties (density and porosity types). We used different metrics to characterize mean accuracy (closeness of the measurements to the true, reference value), proportional bias (resulting from an effect of measured flow on the magnitude of the error), linearity in the relationship between measurements and reference values, and precision (reproducibility and repeatability). We found a large intra- and inter-method variability in calibration performance, with a low proportion of this variability explained by species. Calibration performance was best when using stem segments. We did not find evidence of strong effects of wood density or porosity type in calibration performance. Dissipation methods showed lower accuracy and higher proportional bias than the other methods but they showed relatively high linearity and precision. Pulse methods also showed significant proportional bias, driven by their overestimation of low flows. These results suggest that Dissipation methods may be more appropriate to assess relative sap flow (e.g., treatment effects within a study) and Pulse methods may be more suitable to quantify absolute flows. Nevertheless, all sap flow methods showed high precision, allowing potential correction of the measurements when a study-specific calibration is performed. Our understanding of how sap flow methods perform across species would be greatly improved if experimental conditions and wood properties, including changes in wood moisture, were better reported