Datensatz

Zusammenfassung

The estimation of plant transpiration is an important factor to comprehend the effects of environmental variables on global water balance. The land surface models (LSMs) reflect plant response to the environmental variables changes by simulation of stomatal conductance (g_s). However, the plant responses are not well understood and vary by the climate and vegetation types. In this study, we reviewed the simulation of g_s within different LSMs. These g_s simulation models were calibrated and tested by Bayesian Markov chain Monte Carlo and 10^th fold cross-validation. The uncertainty of each g_s simulation model to input variables was determined by global sensitivity analysis. The results show the challenging issues in generalization in these models due to high sensitivity and dependency on parameterization in semi-empirical models and complex environmental stress functions in empirical models. The machine learning (ML) model was used to optimally discover the functional relationship between key environmental variables in the g_s simulation by following the physical constraints. The high accuracy and low uncertainty to input variables in g­_s simulation highlighted the ML model's importance for future studies. The new approach could be used to improve the integrity of the fundamental scientific processes in LSMs for transpiration simulation and water balance prediction.