Measuring and modeling stomatal conductance of cucumber crop in solar greenhouse in Northeast China

Stomatal conductance (Gs), net photosynthesis (An) and environmental variables were measured from October 2006 to January 2007 under stable soil water content in solar greenhouse in Northeast China. Seventy percent of the measurements were used to parameterize the Gs models and others were used to validate these models. The diurnal variations of Gs were bimodal curves in early period and unimodal curves in latter period. The Ball–Berry model was run by using either observed An (BAobs-model) or predicted An (BApred-model), which was simulated from environmental variables. The BApred-model performed less successful than statistical models because of the systematic error in predicting An. The BAobs-model performed well (explained 89% of observed Gs) in the present study. Several statistical methodologies (Partial Least Square, PLS; Neural Net Analysis, NNA) were used to build models to predict Gs. PLS analysis found a high number of components, and the overall components explained 87% of the observed Gs variability. We tested four neural net models and found radial basis function network (RBF) performed the best. In most situations, measurements of photosynthesis are not available and Gs must be predicted from environmental data. In such case, we conclude that the PLS and NNA are more successful in predicting Gs of cucumber crop in solar greenhouse than the Ball–Berry model.