基于SARIMA模型的近岸海表温度短期预报研究  

A study of short-term forecast of nearshore SST based on SARIMA model

作　　者：赵强 王擎宇 舒志光 ZHAO Qiang;WANG Qingyu;SHU Zhiguang(Ningbo Marine Center,Ministry of Natural Resources,Ningbo 315012,China)

机构地区：[1]自然资源部宁波海洋中心,浙江宁波315012

出　　处：《海洋预报》2024年第1期42-49,共8页Marine Forecasts

基　　金：国家重点研发计划(2018YFC1407000)。

摘　　要：基于石浦海洋站实测数据,采用周期性自回归积分滑动平均方法(SARIMA)构建了逐时海表温度短期预报模型,根据观测数据的周期特征和模型预报误差比选确定了模型参数。结果表明:与采用逐时观测数据作为输入的模型相比,采用逐0.5 h内插数据构建的SARIMA模型的预报结果与实测数据间的相位更为一致,预报误差更小,但进一步将输入数据的时间分辨率提高,72 h逐时预报精度提升不明显;研究还发现模型预报误差总体随输入数据时长的减小而增大;采用366 d逐0.5 h数据构建的SARIMA(2,0,2)(2,1,0)25模型的预报结果较优,0~24 h、24~48 h、48~72 h预报的平均绝对误差分别为0.176℃、0.350℃、0.520℃,相应的均方根误差分别为0.217℃、0.396℃、0.567℃。Based on the Sea Surface Temperature(SST)data of Shipu Station,time-series model of Seasonal Auto Regressive Integrated Moving Average(SARIMA)was used to construct a short-term forecasting model for hourly SST.Model parameters were determined according to the periodic of the data and the model forecasting errors.Compared to the model with original hourly input data,the model with interpolated half-hourly input data shows better performance,and the phases of the forecasts have a better consistent with the observations.Using higher temporal resolution of the input data shows no obvious improvement of the accuracy of the 72 h hourly SST forecasts.The results also show that the forecasting error increases with the reduction of the training data length.SARIMA(2,0,2)(2,1,0)25 model with 366-day interpolated half-hourly SST data shows the best forecasting accuracy.The mean absolute errors of 0~24 h,24~48 h and 48~72 h forecasts are 0.176℃,0.350℃and 0.520℃,the corresponding root mean square error are 0.217℃,0.396℃and 0.567℃,respectively.

关 键 词：周期性自回归积分滑动平均方法 统计预报 海表温度 预报 

分 类 号：P731.31[天文地球—海洋科学]