基于GA-BP网络的西藏高海拔地区ET_0预报

ET_0 forecast on the basis of GA-BP network in high altitude areas of Tibet

选择那曲县(海拔4 450 m)、改则县(海拔4 700 m)作为西藏高原气候典型地区,通过遗传-神经(GA-BP)网络训练,应用1983—2012年30年的数据建立GA-BP网络模型,采用前一年的气象资料预报当年的参考作物腾发量,当2010—2012年连续3年的预报值均满足设定的阈值下限时,输出预测结果,这样使得模型在保证了预报精确度的同时兼具预报稳定性。结果发现:经GA-BP网络确定的2010—2012年3年模型预报值与真实值间的线性关系明显,决定系数R^2分别达到0.8805、0.9363、0.9167,斜率接近于1;多年的模拟预报值与实际值之间的相对误差均处于0.1以下,小于设定的阈值下限。对于易获得气象资料的地区,研究成果可对高海拔地区未来月际间作物需水量的变化进行预判,进而为将来灌溉制度的制定提供依据;对于缺测气象资料的地区,通过本文建立的网络模型,结合气象条件类似的站点,可在大时间尺度下对该地区ET_0变化趋势进行模拟,同时对下年度灌溉制度的拟定提供指导。

As the typical climate area of Tibet plateau,Nagqu County（ 4 450 m above sea level） and Gerze County（ 4 700 m above sea level） were chosen to build a Genetic Algorithm- Back Propagation（ GA- BP） model through the GA- BP network training using data from 1983 to 2012. The ET_0 was obtained by the monthly meteorological data from the previous year. When the forecast values for consecutive years between 2010 and 2012 met the threshold limit set,the forecast values would be exported,which could ensure accuracy and stability of the forecast.The results showed that the there was a great linear relationship between the predicted values by GA- BP model and the real values,reaching R^2 values of 0. 8805,0. 9363 and 0. 9167,respectively. The relative error produced by predicted values and real values were all smaller than 0. 1,which was less than the threshold. In conclusion,the model can be used to predict the ET_0 of different months during crop growth period,and then the inter-monthly water demand of crops can be estimated in the future. It can further provide the basis for future irrigation schedule. For areas lacking meteorological data,based on the network model discussed in this article,the ET_0 variation in these areas can be simulated within a big time frame by referring to other stations with similar meteorological conditions,which can provide guidance for the irrigation schedule of next year.