Estimating evapotranspiration from temperature and wind speed data using artificial and wavelet neural networks (WNNs)
Estimating evapotranspiration from temperature and wind speed data using artificial and wavelet neural networks (WNNs)
复制标题
DOI：
10.1016/j.agwat.2014.03.014
复制
发表时间：
2014-07-01
影响因子：
6.7
通讯作者：
Lee, Teang Shui
中科院分区：
文献类型：
--
作者：
Falamarzi, Yashar;Palizdan, Narges;Lee, Teang Shui
蒸散量是水文循环的重要组成部分，其准确预测是所有水资源应用的基础。本研究利用人工神经网络（ANN）与小波神经网络（WNN），从温度与风速资料预测日蒸散量。在这项研究中使用的小波神经网络模型是一个神经网络模型与一个隐藏层和小波函数作为激活函数。利用澳大利亚Redesdale气候站2009-2012年的气候资料进行分析。采用FAO-PM 56方法计算ET日参考值。以最高气温、最低气温和风速作为输入，以蒸散量序列的参考值作为输出，建立了两种模式的神经网络和小波网络模型。为了评估小波变换分解输入数据对模型效率的影响，分别利用原始数据集和分解后的时间序列对模型进行了标定和验证。研究了风速作为第三输入对模型性能的影响。结果表明，ANN和WNN模型预测ET的准确度都在可接受的水平。然而，wavlet-WNN 261（2个输入，隐藏层中的6个神经元和一个输出）在RMSE、APE、N. S. R值分别为1.03 mm/d、22%、0.79和0.89。(C)2014爱思唯尔有限公司版权所有。
Evapotranspiration (ET) is a major component of the hydrologic cycle and its accurate forecasting is essential in all water resources applications. In this study, artificial neural network (ANN) and wavelet neural network (WNN) were utilized to forecast daily ET from temperature and wind speed data. The WNN model used in this study is a neural network model with one hidden layer and a wavelet function as an activation function. The climatic data of Redesdale climatology station, Australia for the period 2009-2012 were utilized for the analysis. The daily reference values of ET were calculated by the FAO-PM56 method. The maximum temperature, minimum temperatures and wind speed data were used as the inputs and the reference values of ET data series was utilized as the output of the ANN and WNN models. In order to assess the effect of decomposing the input data by wavelet transform on the models efficiency, the original dataset and separately the decomposed time series were applied for calibrating and validating the models. The influence of using wind speed data as the third input on the performance of models was also investigated. The results showed that both the ANN and WNN models predicted ET at an acceptable accuracy level. However, the wavlet-WNN261 (2 inputs, 6 neurons in the hidden layer and one output) performed the best with the RMSE, APE, N.S. and R values of 1.03 mm/day, 22%, 0.79 and 0.89, respectively. (C) 2014 Elsevier B.V. All rights reserved.