基于Bp神经网络和Ball-Berry模型的胡杨气孔导度模拟  

Modeling of stomatal conductance for Populus euphratica using Back Propagation and Ball-Berry model

作　　者：李培都 司建华[1] 冯起[1] 鱼腾飞[1] 赵春彦[1] 

机构地区：[1]中国科学院寒区旱区环境与工程研究所阿拉善荒漠生态水文实验研究站/内陆河流域生态水文重点实验室,兰州730000 [2]中国科学院大学,北京101408

出　　处：《干旱区资源与环境》2016年第11期191-196,共6页Journal of Arid Land Resources and Environment

基　　金：国家自然科学基金(91025024);中国科学院西部之光项目;中国科学院重点部署项目(KZZD-EW-04-05)资助

摘　　要：以2014年5月至9月胡杨气孔导度和环境因子的实测数据,分析了胡杨气孔导度的季节变化特征,并基于Bp神经网络和Ball-Berry模型对胡杨气孔导度进行了模拟研究。结果表明:胡杨气孔导度峰值的出现时间春季要早于夏、秋季节。在不同季节,光合有效辐射均是影响胡杨气孔导度最敏感的环境因子。利用Bp神经网络模型对春季、夏季和秋季胡杨气孔导度的模拟值与实际观测值基于1:1直线的决定系数最高可达0.9078;利用Ball-Berry模型对春季、夏季和秋季胡杨气孔导度预测值的决定系数最高为0.5807。在不同光合有效辐射水平下也对春季、夏季和秋季的胡杨气孔导度进行了模拟,Bp神经网络对气孔导度模拟值与观测值基于1:1直线决定系数最高为0.6739,Ball-Berry模型的决定系数最高为0.5477,在光合有效辐射较弱时,Ball-Berry模型的模拟效果较Bp神经网络效果好,但随着光合有效辐射的增强,Bp神经网络的模拟精度要高于Ball-Berry模型。Based on measured data of Populus euphratica stomatal conductivity and the environmental factors during period of May to September,2014,the seasonal variation characteristics of P. euphratica stomatal conductance were analyzed and the simulation research was carried out based on Bp neural network. The results show that P. euphratica stomatal conductance changed with seasons and peak slightly lagged. Correlation coefficients between P. euphratica stomatal conductance and PAR in different seasons were higher,and correlation with other environmental factors in different season changed. Bp neural network model simulation of P. euphratica stomatal conduction was good,coefficients of determination（ R2） between the model simulation value and the actual observed value of P. euphratica stomatal conductance in spring,summer and autumn based on 1： 1 line with were0. 3818,0. 5392 and 0. 9078,respectively,Root Mean Square Error（ RMSE） respectively were 0. 1265,0.0541 and 0. 0755mol·m-2·s-1. The predicted value R2 of P. euphratica stomatal conductance was 0. 2578,0. 3558 and 0. 5807 in spring,summer and autumn by the model of Ball-berry,RMSE respectively was 0.0517,0. 1665 and 0. 1192mol·m-2·s-1. The accuracy of P. euphratica stomatal conductance simulation was poor in Ball-berry model. Bp neural network model for different seasons of P. euphratica stomatal conductance was simulated under PAR. It was found that the Bp neural network model simulation of stomatal conductance under different levels of PAR had high precision; stomatal conductance simulation value and observation value in spring,summer and autumn based on the 1： 1 line R2 respectively was 0. 5129,0. 5465 0. 6739,RMSE respectively were 0. 0407,0. 0752 and 0. 1406mol·m-2·s-1. While the Ball-berry model of the three seasons of R2 were 0. 1796,0. 2474 and 0. 5477,RMSE was 0. 1656,0. 1409 and 0. 0624mol·m-2·s-1. It showed that the Bp neural network model for P. euphratica stomatal conductance simulation is better.

关 键 词：气孔导度 神经网络 Ball-Berry模型 胡杨 

分 类 号：Q945[生物学—植物学]