基于人工神经网络的涡度相关仪观测蒸散量的数据插补方法

Gap filling for evapotranspiration based on BP artificial neural networks

立足于绿洲农田生态系统,采用人工神经网络方法对潜热通量数据进行模拟和插补.根据作物生长季分阶段模拟,对比整体模拟结果,发现分阶段模拟效果(R2=0.91～0.95,RMSE=28.9～41.3 W/m2,MAE=21.3～28.8 W/m2)优于整体模拟效果(R2=0.87～0.92,RMSE=39.6～50.7 W/m2,MAE=27.6～34.9 W/m2).通过模型网络连接权值对各阶段环境因子的相对贡献率作了定量分析,并从数学统计的角度对研究区蒸散发环境因子影响机理进行了分析.结果表明为了提高潜热通量的插补精度,合理地根据作物生长季分阶段建模插补是有必要的.

Based on the oasis farmland ecosystem, an artificial neural network approach was used to simulate the latent heat flux data and gap filling. According to crop growth season stages and comparing the overall simulation results, it was found that the stage simulation results were better than the overall simulation results. In addition, a quantitative analysis was made of the relative contribution of environmental factors in the various stages through the connection weights, and also were analyzed the environmental factors affecting the evapo-transpiration in the study area from mathematical and statistical perspectives. The research showed that crop growing season stages are necessary to improve the accuracy of gap filling of latent heat flux.