首页|麦田不同高度日极端气温预测技术研究

麦田不同高度日极端气温预测技术研究

扫码查看
原文链接
  • 国家科技期刊平台
  • NETL
  • NSTL
  • 万方数据
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.
Study on Prediction Technology of Daily Extreme Temperature at Different Heights in Wheat Field
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.

wheat fieldsmultiple linear regressionBP neural networkforecast of temperatureHRCLDASintelligent grid prediction

朱保美、李密、张继波、黄丙玲

展开 >

山东省气象防灾减灾重点实验室,山东 济南 250031

齐河县气象局,山东 齐河 251100

博山区气象局,山东淄博 255213

山东省气候中心,山东 济南 250031

齐河县农业农村局,山东 齐河 251100

展开 >

麦田 多元线性回归 BP神经网络 气温预测 HRCLDAS 智能网格预报

山东省气象局气象科研引导类项目山东省气象局气象科研重点项目科技创新2030—重大项目德州市气象局科研项目

2021SDYD252023sdqxz122022ZD01195032023dzqxyb09

2024

10.12057/j.issn.1002-0799.2024.04.021

沙漠与绿洲气象
新疆维吾尔自治区气象学会 中国气象局乌鲁木齐沙漠气象研究所

沙漠与绿洲气象

CSTPCD
影响因子：1.007
ISSN：1002-0799
年,卷(期)：2024.18(4)
  • 25