A backward propagation neural network for predicting daily transpiration of poplar

Abstract In this study, a supervised artificial neural network (ANN) trained by back propagation (BP) algorithm s was developed to predict the transpiration of poplar based on six input variables. Based on the transpiration characteristics of trees, a three‐layer BP network was constructed with six input units and one output unit. Daily average temperature, relative humidity, photosynthetic affective radiation, wind speed, sunlight duration and the soil water content 50 cm below the soil surface were considered as the six input variables of the network, which primarily affected the transpiration of poplar. The prediction of daily transpiration of poplar in Heilonggang region, Hebei province was conducted. The research results indicated that R 2 equalled 0.9534 between measured values and predicted values. The maximum relative error, the minimum relative error and the average relative error were 16.85, 1.49, and 4.2%, respectively. The proposed model could describe the relationship between the daily transpiration of poplar, the meteorological factors and soil moisture conditions with a relatively high accuracy. The research results had potential values for the production and management in this polar stand.