基于气温预报和神经网络的参考作物腾发量预报  

Reference Crop Evapotranspiration Forecasting Based on Temperature Forecasts and Neural Network

作　　者：谢梅香[1,2] 罗玉峰[1,2] 钱子嘉 闫欣[1] 彭世彰[1] 李江安 常晓敏[1] 

机构地区：[1]河海大学水文水资源与水利工程科学国家重点实验室,南京210098 [2]河海大学南方地区高效灌排与农业水土环境教育部重点实验室,南京210098 [3]江苏省高邮市水务局,江苏高邮225600

出　　处：《节水灌溉》2015年第2期33-36,共4页Water Saving Irrigation

基　　金：国家自然科学基金项目(51179048);水利部公益性行业科研专项经费项目(201301014);江苏高校优秀科技创新团队项目

摘　　要：采用反向传播人工神经网络（BP-ANN）逼近气象因子-参考作物腾发量ET0函数关系,以天气预报中的最高和最低气温为输入进行短期ET0预报。收集了南京站实测的2010年7月1日至2013年7月7日逐日气象数据和2012年7月1日至2013年6月30日逐日对未来7d的气象预报数据,以最高、最低气温及相应的日序数为3个输入因子,ET0为输出建立一个包含一个隐含层的3层BP网络,以2010年7月1日至2012年6月30日实测气象数据及通过FAO-56PM公式计算的ET0进行网络,以2012年7月1日至2013年6月30日实测气象数据及通过FAO-56PM公式计算的ET0进行网络验证。将2012年7月1日至2013年6月30日逐日对未来7d的气象预报中的最高、最低气温输入训练及验证后的网络,得到2012年7月1日至2013年6月30日逐日对未来7d的ET0预报值,并与FAO-56PM公式计算的ET0值进行比较以验证预报精度。结果表明,预见期1-7d内,预报的ET0和计算的ET0变化趋势基本一致,预报精度随着预见期的增加而降低;平均准确率（±1.5mm/d以内）达88.08%,相关系数为0.77,均方根误差为1.28mm/d,显示出了较高的预报精度。在局部时间段内出现的ET0,PM和预报ET0的较大差别的原因是该时段内的ET0更多地受到除了日最高和最低气温之外的其他因素的影响。提出的方法 ET0预报,随着气象预报准确度的提高,可实现较为精确的ET0预报。A Back Propagation Artificial Neural Network（BP-ANN）was developed to describe the climate variables-reference crop evapotranspiration（ET0）and weather forecasts were input to the network model to forecast ET0.Observed daily meteorological data from July 1,2010 to July 7,2013 and weather forecast data for 7days ahead from July 1,2012 to June 30,2013 for the Nanjing Station were collected.A three-layer BP-ANN with maximum temperature,minimum temperature and Julian day as input,ET0 as output,and a hidden layer was developed.The observed daily meteorological data from July 1,2010 to July 1,2012 and calculated ET0 by the FAO-56 PM equation were used to train the network,then the observed daily meteorological data from July 1,2012 to July 1,2013 and calculated ET0 by the FAO-56 PM equation were used network validation.The forecasted temperatures 7days ahead for July 1,2012 to June 30,2013 were input to the trained and validated network and ET0 forecasts 7days ahead for July 1,2012 to June 30,2013 were obtained,and the ET0 forecasts were lastly compared with the calculated ET0 by the FAO-56 PM equation and the observed data.The results showed that within the forecast horizon of 1to 7days,ET0 forecasting and the calculated ET0 from the PM equation with observed meteorological data shared the same trend,and the forecast accuracy decreased with the increase in forecast horizon;the high average accuracy rate（88.08%）and correlation coefficient（0.77）and low root mean square error（1.28mm/d）indicated higher forecast accuracy.Some large differences between ET0 forecasts and ET0,PM were caused by the influences of other climate variables other than maximum and minimum temperatures.The proposed method can be used for ET0 forecast and higher accuracy can be achieved with more accurate weather forecasts.

关 键 词：参考作物腾发量 灌溉预报 天气预报 BP人工神经网络 

分 类 号：S161.2[农业科学—农业气象学]