黄土区不同土地经营方式径流量神经网络模拟

基于黄土坡面降雨—径流关系的复杂性和非线性,引用3层前馈型BP网络模型,对不同土地经营方式(草灌地、刈割地、翻耕地)径流量进行模拟,以植被盖度、降雨强度、坡度、土壤前期含水率和土壤容重5个因子作为输入层变量,次降雨下径流量作为输出层变量。利用野外人工模拟降雨试验所得到的不同降雨强度下各类土地经营径流小区的径流量实测资料,对网络进行模拟训练并预测,径流量平均误差不超过10%,且径流量较大的翻耕地训练精度及预测结果较草灌地、刈割准确性更高些。与传统回归统计方法进行了误差比较,结果表明,该模型能更好地预测次降雨的径流量。

Runoff Simulated with Neural Network under Different Management Patterns in Loess Region

Based on the complexity and nonlinear characteristics of rainfall and slope runoff,a three-layer feed-forward back-propagation(BP) neural network model was constructed and used to simulate the runoff under different management patterns(grass and shrub land,harvested land,and tillage land).In this network model,vegetation coverage,rainfall intensity,gradient,antecedent soil moisture and soil bulk density were selected as the input variables,and runoff intensity under individual rainfall event was the only output variable.The network model was trained and validated by using the measured data obtained from different runoff plots under different rainfall intensity.The results showed that the mean error was less than 10%.The training accuracy and predictive results of tillage land were better than grass and shrub land and harvested land.The comparison for the results of the BP neural network model mothod and regression statistics method showed that the BP neural network model was better to predict the runoff under individula rainfall event.