Abstract:

Climate change is an important factor affecting wheat yield， so scientific classification of climate year type is very important for accurate prediction of wheat yield. This paper divided the temperature data of 15 meteorological stations （cities or counties） in Henan Province from 1984 to 2018 to warm year， normal year， and cold year； according to the rainfall data， wet year， normal water year， and dry year were divided， and then 9 climatic year types were formed to analyze their laws. The relationship between the climatic year type and the output year type was analyzed combined with the wheat yield data to divide the high year， the average year， and the low year， based on topography， soil， weather and other factors， 5 wheat areas were diveded including northern Henan， central and eastern Henan， western Henan， southern Henan， and Nanyang basin. On this basis， the HP filter method was used to separate the meteorological output and the trend output. Under the driving of meteorological factor， the BP neural network was used to construct a model to predict the meteorological output， and the linear regression model was used to obtain the trend output， and then the two output were superimposed to obtain the actual output to realize the forecast of wheat output. The results showed that the accumulated temperature years in Henan Province were mainly warm year and normal year， and rainfall year was more evenly distributed. The climate year types were mainly normal year and dry year. The frequency of high wheat yield was the highest in warm and wet years with 76.9%. In cold and wet years， the frequency of low wheat yield was the highest with 67.9%； accumulated temperature in meteorological data was the main factor affecting wheat yield fluctuations. In warm years， wheat was easily high yields， and the probability of low yields in cold years was higher； compared the wheat output obtained by the superimposition of the output modeling using meteorological yields and trends with the actual output， the average relative error of each model of the 5 wheat regions in northern Henan， middle eastern Henan， western Henan， southern Henan and Nanyang basin was 0.31%， 0.36%， 0.58%， 0.48% and 0.38%， respectively， which indicated that it was feasible to predict wheat yield using HP filter and BP neural network technology.

Key words: wheat, weather, climate year type, yield, HP filter, BP neural network

摘要：

气候变化是影响小麦产量的重要因素，因此科学划分气候年型，对于准确预测小麦产量至关重要。利用河南省15个气象站点（市、县）1984—2018年的温度数据划分暖温年、平温年、冷温年，依据降雨数据划分湿年、平水年、干年，组合成9种气候年型并分析其规律；结合小麦产量数据划分出丰产年、平产年、低产年，分析气候年型和产量年型的关系。然后结合地形土壤、气象等因素把河南划分为豫北麦区、豫中东部麦区、豫西麦区、豫南麦区和南阳盆地5个麦区。在此基础上，利用HP滤波法分离出气象产量和趋势产量，以气象因子驱动并利用BP神经网络构建模型预测气象产量，利用一元线性回归模型建模得到趋势产量，把两者产量叠加得出实际产量，从而实现小麦产量的预测。结果表明：河南省积温年型以暖温年、正常年型为主，降雨年型分布比较均匀，气候年型中以正常年和干年为主，暖湿年小麦高产频率最高，为76.9%，冷湿年小麦低产频率最高，为67.9%；积温是影响小麦产量波动的主要因素，暖年年型下小麦更容易丰产，冷年时小麦低产概率较高；利用气象产量和趋势产量分别建模叠加得出的小麦产量和实际产量相比，豫北、豫中东、豫西、豫南和南阳盆地五大麦区各模型的平均相对误差分别为0.31%、0.36%、0.58%、0.48%、0.38%，说明利用HP滤波和BP神经网络技术预测小麦产量是可行的。

关键词: 小麦, 气象, 气候年型, 产量, HP滤波, BP神经网络