Abstract: Tea tree growth, yield and quality are influenced by its canopy structure, therefore simulating tea tree canopy indexes by using mathematical models can instruct the tea cultivation and management. Considering the tea tree canopy each index not only affected each other, but also related to previous years growth data, the difference equations models with multi-variable were established for 18 years data of 15 growth indexes of tea plant canopy. A combination of clustering and neural network was employed to estimate the model parameters. Calculating results showed that the mean absolute percentage error (MAPE) of our model was 0.045828, which was less than the MAPE 0.0618 of one variable difference equation model, also less than the MAPE 0.0842 of a regression equation model. The results indicated that the predictive accuracy of our models was improved comparing with one variable models. This method can be used for modeling of multidimensional time sequence of difference equations.