AbstractThe depth‐wise distribution of root water uptake is typically inferred through linear mixing models that utilize knowledge of stable water isotopes in soil and plants. However, these existing models often represent the water uptake profile in discrete segments, potentially introducing significant uncertainty and bias into results. In this study, we introduced a novel root water uptake mixing model that combines a Bayesian linear mixing framework with a continuous root water uptake pattern, named CrisPy. To evaluate the performance of CrisPy, we conducted virtual and field‐based tests under several types of prior information. CrisPy showed accurate and robust reconstruction of the true root water uptake profile under various prior information settings in the virtual test. By contrast, the discrete mixing model, MixSIAR was greatly influenced by the prior information and deviated from the true profile. The root mean squared error of the uptake proportions from CrisPy ranged from 3.6% to 7.4%, while MixSIAR exhibited values of 6.3%–15.2%. Furthermore, posterior predictive checking indicated that CrisPy effectively reconstructed the mean and standard deviations of plant water isotopic compositions in both virtual and field‐based tests. MixSIAR, however, underestimated the mean and overestimated the standard deviation of these compositions. These findings collectively support the enhanced accuracy, greater robustness, and reduced uncertainty of CrisPy in comparison to MixSIAR. Therefore, CrisPy provides a powerful tool for partitioning plant water sources.
Read full abstract