Modeling water consumption of Populus bolleana by artificial neural network based on fuzzy rules

Abstract

To explore the water consumption characteristics of trees, the thermal dissipation probe technology was used to monitor sap flow of Populus bolleana in east sandy land of Yellow River, from July to November in 2017. Microclimate variables were monitored. We analyzed the diurnal and seasonal variations of water consumption, and proposed the models for water consumption with back propagation neural network (BPNN) and Elman neural network (ENN) based on fuzzy rules. Results showed that the average sap flow rate of P. bolleana was 4.98 g·cm-2·h-1 in growing season (July to October), with solar radiation (Rs), temperature (T), vapor pressure deficit (VPD) and relative humidity (RH) as the main factors affecting sap flow. Due to the influence of meteorological factors, water consumption was characterized by obvious seasonal variation, with that in summer (July-August) being 1.4 times of that in autumn (September-October). BPNN and ENN models based on fuzzy rules were used to simulate water consumption of P. euphratica. The optimal parameter calibration of two models explained more than 80% of the total variation, which indicated that these two models could more accurately simulate water consumption. Compared with the BP neural network model, the simulated results of ENN model showed that the relative error was reduced by 27.0%, RMSE was reduced by 24.3%, Nash-Sutclife efficiency coefficient increased by 67.9%, R2 was higher than 0.80. The ENN model performed better than BPNN model with a higher efficiency and goodness of fitness. ENN model effectively improved the simulating accuracy of water consumption. Therefore, it could be used as an optimal model to estimate water consumption of P. bolleana in east sandy land of Yellow River.