Bayesian analysis of jujube canopy transpiration models: Does embedding the key environmental factor in Jarvis canopy resistance sub-model always associate with improving transpiration modeling?

Abstract

The Jarvis canopy resistance sub-model is commonly used in transpiration modelling, although its optimal structure is rarely studied. It is still not fully clear that to what extent transpiration modeling may differ if using different constraint function forms for key environmental variable in Jarvis sub-model. In this study, various Jarvis canopy resistance sub-model configurations were embedded in the Penman-Monteith model to compare their ability to model daily transpiration of rain-fed jujube (Ziziphus jujuba Mill.) plantations where soil moisture is a key factor of tree water use. Parameters were calibrated using the Bayesian Markov Chain Monte Carlo (MCMC) simulation technique and model comparison was quantified using Deviance Information Criterion (DIC). The results showed significant differences in model performance between the constraint function forms of soil water content. The difference in DIC between the model with the best constraint function form and the other two forms reached 37.66–50.94, much higher than the evaluation criteria for significance (larger than 7). When the best constraint function form was used, the performance of the transpiration model improved. The model performance worsened when the other constraint function forms were used, even worse than those without consideration for soil water content. However, only slight differences in model performance were detected for the constraint function forms of temperature, vapor pressure deficit and photosynthetically active radiation. Using the best configuration of Jarvis canopy resistance sub-model, daily transpiration of jujube plantation was well estimated with overall good accuracy and acceptable uncertainty. The predictions and observations were highly correlated (R2 = 0.87 for calibration and R2 = 0.80 for validation). The results suggested that different constraint function forms of an environmental factor contributed differently to transpiration model performance, and the situation was different for different environmental factors. Including the key environmental factor in the Jarvis canopy resistance sub-model will not always improve the performance of transpiration models.