基于气温和DC-BP-NN的河西走廊月度ETo估算模型

Monthly Reference Crop Evapotranspiration Estimation Model Based on Air Temperature and DC-BP-NN in Hexi Corridor

针对参考作物蒸散量(Reference crop evapotranspiration,ETo)估算模型中,标准估算模型——FAO PenmanMonteith(FAO-PM)模型需要充分的气象数据,而基于气温的估算模型精度不足的问题,参考FAO-PM模型结构,基于气温和月序数,融合分治法(Divide and conquer,DC)和误差反向传播神经网络(Back propagation neural network,BP-NN),提出了一种采用DC-BP-NN的月度ETo估算模型;以FAO-PM模型计算的ETo为标准,利用河西走廊酒泉气象站1958年1月—2013年9月的月度气象数据,将DC-BP-NN模型与其余6种基于气温的ETo估算模型(Blaney-Criddle模型、Hargreaves-Samani模型、2种改进的Hargreaves-Samani模型、BP-NN模型、BP-NN1模型)进行对比。结果表明,DC-BP-NN模型的估算精度(均方根误差5.99 mm/月,平均偏差0.99 mm/月,平均绝对百分误差7.18%,决定系数0.988 6)优于其余6种ETo估算模型,该模型可以用于河西走廊农田气象数据不充分条件下的月度ETo估算。

As the standard method for estimating reference crop evapotranspiration（ ETo）,FAO PenmanMonteith（ FAO-PM） model incorporates both the thermodynamic aspect and the aerodynamic aspect of evapotranspiration. The model needs complete agricultural meteorological data to estimate ETo,which is considered to be a difficult task in many locations of Hexi Corridor. Meanwhile,the accuracy of the temperature-based models is insufficient. In order to solve these problems,a monthly ETo estimation model（ DC-BP-NN） was proposed,which integrated air-temperature,divide and conquer（ DC） method and back propagation neural network（ BP-NN） with the structure of FAO-PM model. The model consisted of two BP-NN models： the radiation BP-NN model and the aerodynamic BP-NN model. In the experiments,the data was from Jiuquan Weather Station in Hexi Corridor. The reference standard was obtained by FAO-PM model. The results showed that DC-BP-NN model was superior to the other six ETo estimation models, including Blaney-Criddle model, Hargreaves-Samani model, two improved Hargreaves-Samani models, BP-NN model and BP-NN1 model（ BP-NN model was based on air temperature and monthly ordinal number）,with average root mean square error of 5. 99 mm / month,mean bias error of 0. 99 mm / month,mean absolute percentage error of 7. 18% and determination coefficient of0. 988 6. Therefore,the DC-BP-NN model can be used for estimating monthly ETo in Hexi Corridor withinsufficient meteorological data.