基于人工神经网络的柑橘树日间蒸腾速率精细模拟与预测

Precise simulation and prediction of citrus tree daytime transpiration rate based on artificial neural network

蒸腾作用是植物的重要生理过程,精细模拟与预测蒸腾速率有助于植株需水量的确定。本文使用茎流计和气象站监测柑橘树蒸腾速率及周边气象因子,基于人工神经网络,构建10 min尺度的柑橘树日间蒸腾速率预测模型。以环境温度、环境湿度、太阳净辐射、风速四种气象因子组合,构建的4-7-1网络结构的柑橘树蒸腾速率预测模型精度最高,与实测数据的Pearson相关系数高于0.8;与FAO作物系数模型和经验公式模型相比,神经网络模型对10 min间隔的柑橘树蒸腾速率预测更加准确,建模所需数据量更少,在"午休"现象的预测上符合实际规律。结果表明以气象因子作为输入的神经网络模型,能够对10 min间隔的柑橘树蒸腾速率进行更加精细的模拟与预测。

Transpiration is an important physiological process of plants. Accurate simulation and prediction of transpiration rate are helpful to determine the water demand of plants. The packaged stem sap flow gauge and the meteorological station were used to measure the stem sap flow rate and the peripheral meteorological factors. Based on the artificial neural network, a 10-minute daytime transpiration rate prediction model of citrus was established by using different combinations of meteorological factors. The results showed the 4-7-1 network model which was established by combining four meteorological factors, including environmental temperature, environmental humidity, solar net radiation and wind speed had the optimal result, and the Pearson correlation coefficient between prediction results and measured data was higher than 0.8. Compared with FAO coefficient approach model and empirical formula model, neural network model was more accurate in predicting transpiration rate of citrus trees with 10 min interval, less data was required for modeling, and more in line with practical rules in predicting the phenomenon of "lunch break". It shows that the artificial neural network model can simulate and predict the transpiration rate of citrus trees more precisely.