The accuracy improvement of sap flow prediction in Picea crassifolia Kom. based on the back-propagation neural network model

Tree transpiration is an important water movement process in forest ecosystems, and it plays a decisive role in the coupling feedback of hydrological and ecological processes. Therefore, identifying the impact of different factors on sap flow can promote efficient water management and improve assessment of the climate change impacts. However, the interaction between sap flow and control factors is not clear, and there is no accurate model to predict sap flow change of Picea crassifolia Kom. This study explored the correlation between sap flow and environmental factors and compared the performances of the back-propagation neural network (BPANN) and multiple regression (MLR) models, and the importance of the time-lag effect and different soil moisture conditions in the response mechanism of sap flow was also explained. A higher fitting performance was found in the BPANN model (R-2 > 0.90) based on the relationships between air temperature (Ta), air relative humidity (RH), solar radiation (Rn) and vapour pressure deficit (VPD) and sap flux density (Q) than that MLR (R-2 = 0.8915). The correlation was improved due to the consideration of time-lag effect. Other variables, such as maximum temperature (Tmax), wind speed (WS) and precipitation (P), explained smaller portions of the variance in sap flow, while minimum temperature (Tmin) and leaf area index (LAI) had almost no effect. Moreover, the R-2 of the water stress condition (REW <0.38) was lower than that of the non-water stress condition (REW >= 0.38), and even lower than the R-2 of the whole experimental period. Therefore, the sap flow prediction model based on BPANN could more reasonably explain the nonlinear relationship between transpiration and control factors, which provided a basis for the estimation of plant water-use and the construction and management of ecological vegetation in alpine arid and semi-arid areas, especially in response to the continuous enhancement of aridification and climate warming.