基于BP神经网络的河北省降水量空间分布预测研究

Spatial distribution prediction of precipitation in Hebei Province based on BP neural network

基于河北省近年来降水的观测资料,在使用地形、海陆位置、区域植被净初级生产力(Net Primary Productivity,NPP)等环境变量的基础上,构建基于BP神经网络(Back Propagation Neural Network)的降水量预测回归模型,预测生成河北省500 m分辨率的降水量分布图。结果显示:就河北省全省来看,区域年平均降水量为506.72 mm,属半湿润区;BP神经网络驱动的降水量预测模型具有较高的可信度,其验证精度R^(2)达0.82,平均绝对误差(MAE)和均方根误差(RMSE)为55.27、69.47 mm;预测结果显示区域降水量呈现自东部向西北减少趋势,降水量中心位于秦皇岛、唐山,次中心为沧州、衡水;该模型产生的空间误差为随机分布,表明BP模型可作为预测降水量空间分布的有效方法之一。

Based on the observational data of precipitation in Hebei Province in recent years,and on the basis of using environmental variables such as topography,sea-land location,the regression model based on BP neural network is established to forecast and generate the precipitation distribution map with 500m resolution in Hebei province.The results show that the annual average precipitation is 506.72 mm,belonging to the semi-humid region,and the precipitation prediction model driven by BP neural network has high reliability,and its verification accuracy is 0.82,the Mae and RMSE are 55.27 mm and 69.47 mm,respectively,and the results show that the regional precipitation decreases from the east to the northwest,with the precipitation centers located in Qinhuangdao and Tangshan,and the secondary centers located in Cangzhou and Hengshui The spatial error produced by the model is random distribution,which indicates that the BP model can be used as one of the effective strategies to predict the spatial distribution of precipitation.