麦田不同高度日极端气温预测技术研究  

Study on Prediction Technology of Daily Extreme Temperature at Different Heights in Wheat Field

作　　者：朱保美 李密 张继波 黄丙玲[5] ZHU Baomei;LI Mi;ZHANG Jibo;HUANG Bingling(Key Laboratory for Meteorological Disaster Prevention and Mitigation of Shandong,Jinan 250031,China;Qihe Meteorological Bureau,Qihe 251100,China;Boshan Meteorological Bureau,Zibo 255213,China;Shandong Climate Center,Jinan 250031,China;Qihe Agriculture and Rural Affairs Bureau of Shandong Province,Qihe 251100,China)

机构地区：[1]山东省气象防灾减灾重点实验室,山东济南250031 [2]齐河县气象局,山东齐河251100 [3]博山区气象局,山东淄博255213 [4]山东省气候中心,山东济南250031 [5]齐河县农业农村局,山东齐河251100

出　　处：《沙漠与绿洲气象》2024年第4期159-166,共8页Desert and Oasis Meteorology

基　　金：山东省气象局气象科研引导类项目(2021SDYD25);山东省气象局气象科研重点项目(2023sdqxz12);科技创新2030—重大项目(2022ZD0119503);德州市气象局科研项目(2023dzqxyb09)。

摘　　要：利用2019年1月—2022年6月麦田小气候自动观测站和齐河县国家基本气象观测站资料,分别采用多元线性回归和BP神经网络方法,建立不同天气类型下麦田30、60、150 cm日最高和日最低气温预测模型,分别选取麦田小气候观测站实测值、中国气象局高分辨率陆面数据同化系统(HRCLDAS)逐小时温度实时产品和智能网格省级气温格点预报产品对模型预测值进行对比分析。结果表明:两种模型预测效果均能满足麦田气温预测需求,BP神经网络模型的预测精度略高于多元线性回归模型;两模型针对晴天时150 cm日最高气温预测效果最好,平均绝对误差均为0.5℃,均方根误差均为0.6℃,准确率均为100%;晴天时30 cm日最高气温预测效果最差。两模型针对不同天气类型不同高度日最高、日最低气温预测值与HRCLDAS气温格点实况值的一致性均较好,均方根误差范围分别为2.0~3.9和1.9~4.1℃;与气温格点预报值的误差较大,均方根误差范围分别为2.4~5.1和2.4~5.3℃。多元线性回归模型预测值与HRCLDAS气温格点实况值、气温格点预报值的一致性优于BP神经网络模型。Using the observation data from January 2019 to June 2022 from the wheat field microclimate automatic observatory and the national basic meteorological observatory in Qihe county,the models of multiple linear regression and BP neural network are established in order to predict daily maximum and daily minimum temperature at various heights(30,60,and 150 cm)in the wheat field under different weather conditions.The accuracy and performance of these models are evaluated by comparing the predicted values with microclimate data,HRCLDAS grid temperature data,and grid temperature forecast data provided by the China Meteorological Administration.The results indicate that both models are capable of meeting the temperature forecasting requirements for the winter wheat field.The BP neural network model demonstrates higher prediction accuracy compared to the multiple linear regression model.Specifically,both models perform best in predicting the daily maximum temperature at a depth of 150 cm on sunny days,with mean absolute errors of 0.5℃,root mean square errors of 0.6℃,and 100%accuracy rate for each model.However,both models perform worst in predicting the daily maximum temperature at a depth of 30 cm on sunny days.The predicted values of daily maximum and daily minimum temperatures,varying across weather types and altitudes,align well with the HRCLDAS grid temperature data.The root mean square errors range from 2.0℃to 3.9℃and 1.9℃to 4.1℃,respectively.Conversely,there is a larger error when compared with the grid temperature forecast data,with root mean square errors ranging from 2.4℃to 5.1℃and 2.4℃to 5.3℃for these two models.The multiple linear regression model demonstrates better consistency with both the HRCLDAS grid temperature data and the grid temperature forecast data compared to the BP neural network model.

关 键 词：麦田 多元线性回归 BP神经网络 气温预测 HRCLDAS 智能网格预报 

分 类 号：P463.2[天文地球—大气科学及气象学]