基于RF-BiLSTM神经网络的多时次土壤水分动态预测

Multi-time soil moisture dynamic prediction based on RF-BiLSTM neural network

为探究土壤水分变化特征,提高土壤水分的预测准确率,提出基于随机森林与双向长短期记忆神经网络结合的土壤水分预测方法(RF-BiLSTM),采用三亚国家气候观象台2016-2021年10 cm深度的土壤体积含水量小时数据和同期7个气象要素(气温、地面温度、10 cm地温、日照时数、相对湿度、降水量、蒸发量)资料,开展多时次土壤水分预测。结果表明,RF-BiLSTM模型对6、12、24、48 h后的土壤体积含水量预测平均绝对误差(MAE)分别为0.462%、0.702%、0.889%、1.282%,决定系数(R2)分别为0.983、0.967、0.951、0.913,准确率均高于长短期记忆神经网络模型、BP神经网络模型。

In order to explore the change characteristics of soil moisture and improve the prediction accuracy of soil moisture,a soil moisture prediction method based on the combination of random forest and two-way long-term and short-term memory network(RFBiLSTM)was proposed.Using the hourly data of soil volume moisture at the depth of 10 cm from 2016 to 2021 of Sanya National Cli⁃mate Observatory and the data of 7 meteorological elements(air temperature,ground temperature,10 cm ground temperature,sun⁃shine hours,relative humidity,precipitation and evaporation)in the same period,the multi-time soil moisture prediction was carried out.The results showed that the average absolute errors(MAE)of RF-BiLSTM model for predicting soil volume water content after 6,12,24 and 48 hours were 0.462%,0.702%,0.889%and 1.282%respectively,and the determination coefficients(R2)were 0.983,0.967,0.951 and 0.913 respectively.The accuracy was higher than that of the long short-term memory neural network model,and BP neural network model.