Use of statistical and neural network techniques to detect how stomatal conductance responds to changes in the local environment

The bulk stomatal conductance term, gs, in the Penman-Monteith equation is modelled as a function of local environmental variables. The stomatal dependencies are frequently described as a product of individual functions, fj, each depending on only one variable Xj, where the functional forms have been estimated through laboratory experiments. Other descriptions for gs are being developed (notably through photosynthesis models) and so methods of comparing model performance are needed. A notion of the ‘best possible fit’ of gs to the Xj is required, thereby providing a benchmark for any model. This paper introduces regression and neural network methods to analyze the stomatal conductance of pine forest, although the techniques are applicable to any vegetation type. In this paper the importance of a strong nonlinear dependence of gs on the Xj is illustrated and further the frequently used ‘Jarvis’ type nonlinear functions, fj, are shown to be nearly optimal.