基于Gash修正模型与神经网络优化模型的刺槐冠层截留模拟

Simulation of Robinia pseudoacacia Canopy Interception Based on Modified Gash Model and Neural Network Model

[目的]对比分析Gash修正模型和神经网络模型在模拟和预测人工林冠层截留的适用性,揭示干旱区刺槐冠层截留及其响应过程,为深入了解森林生态水文过程及其调控机制提供科学依据。[方法]于2019年5—10月,以宁夏河东地区刺槐(Robinia pseudoacacia)人工林为研究对象,定位观测了树干茎流和穿透雨并计算得到冠层截留,采用修正后的Gash模型与神经网络模型对刺槐林林冠截留量进行了模拟。[结果](1)研究区刺槐人工林的穿透雨量、树干茎流量、林冠截留量分别为154.19,5.61,16.5 mm,产生穿透雨和树干茎流的阈值分别为1.37,2.17 mm。(2)Gash修正模型和优化后的神经网络模型均能较好地模拟刺槐冠层截留量,Gash修正模型的绝对误差、均方误差、均方根误差、平均绝对百分比误差分别为0.20%,0.06%,0.24%,52.43%,模拟结果拟合精度达到83%;与Gash修正模型相比,采用麻雀搜索算法优化后的BP神经网络算法模型(SSA-BP),均方误差降低了61.48%,平均绝对误差降低了40.39%,均方根误差降低了37.93%,平均绝对百分比误差降低了50.52%,决定系数提高了1.2%。[结论]在林木冠层截留模拟研究方面,加入麻雀搜索算法后的BP神经网络模型具有较好的可靠性,可以有效降低模拟误差,提高模型的预测精度。

[Objective]The aims of this study are to compare and analyze the applicability of modified Gash model and neural network model in simulating and predicting canopy interception of artificial forest to reveal the canopy interception and its response process of Robinia pseudoacacia in arid area,and to provide scientific basis for further understanding of forest eco-hydrological process and its regulation mechanism.[Method]Robinia pseudoacacia plantation in the east of the Yellow River of Ningxia was taken as the research object,the stemflow and throughfall were observed and the canopy interception was calculated.The modified Gash model and neural network model were used to simulate the canopy interception of Robinia pseudoacacia forest.[Result](1)The throughfall,stemflow and canopy interception of Robinia pseudoacacia plantation in the study area were 154.19 mm,5.61 mm and 16.5 mm,respectively,and the thresholds for throughfall and stemflow were 1.37 mm and 2.17 mm,respectively.(2)Both the Gash modified model and the optimized neural network model could better simulate the canopy interception of Robinia pseudoacacia.The absolute error,mean square error,root mean square error and mean absolute percentage error of the Gash modified model were 0.20%,0.06%,0.24%and 52.43%,respectively.The fitting accuracy of the simulation results reached 83%.Compared with the Gash modified model,the mean square error of the BP neural network algorithm model(SSA-BP)optimized by the sparrow search algorithm was reduced by 61.48%,the mean absolute error was reduced by 40.39%,the root mean square error was reduced by 37.93%,the mean absolute percentage error was reduced by 50.52%,and the coefficient of determination was increased by 1.2%.[Conclusion]In the simulation study of canopy interception,the BP neural network model with sparrow search algorithm has a good reliability,which can effectively reduce the simulation error and improve the prediction accuracy of the model.