《海洋预报》｜基于海温因子的传递函数模型在黄海绿潮规模预测中的应用

　　摘要：为寻求基于海温生态因子的绿潮规模预测方法，采用2010—2019年国家卫星海洋应用中心黄海绿潮卫星遥感资料和日本气象厅融合海表温度数据，利用协整检验和Granger因果检验方法对绿潮覆盖面积和海温之间的长期均衡和因果关系进行分析。结果显示：两者间存在长期协整性，海温是绿潮规模变化的Granger原因。通过建立协整模型和传递函数模型，开展了海温对绿潮规模影响的定量计算实例研究。结果表明：两种模型均可有效地刻画2010—2017年绿潮规模的变化过程，2018—2019年绿潮规模的预测结果与遥感实况较吻合，体现了模型的可靠性和可移植性。传递函数模型预测结果的RMSE为160.94，MAE为109.70，整体略优于协整模型的RMSE（171.40）和MAE （122.48），说明数据经预白化处理后可提高预测精准度，海温与绿潮覆盖面积具有动态相关性。

　　关键词：绿潮；海温；协整模型；传递函数模型；Granger因果检验；遥感

　　Abstract：In order to find a method to predict the green tide scale based on sea temperature, the Yellow Sea green tide satellite remote sensing data of the National Satellite Ocean Application Service and the merged sea surface temperature (SST) data of the Japan Meteorological Agency (JMA) from 2010 to 2019 is used to analyze the longterm equilibrium and causal relationship between green tide coverage area and SST based on co-integration test and Granger causality test. The results show that there is long-term co-integration between green tide coverage area and SST, and SST is the Granger cause of the green tide scale variation. The quantitative calculation case study of the influence of sea temperature on green tide scale is carried out by establishing a cointegration model and transfer function model It is shown that both models could effectively depict the variation process of green tide scale from 2010 to 2017, and the prediction results of green tide scale from 2018 to 2019 agree well with the remote sensing monitoring data, which reflects the reliability and portability of the models. The root mean square error (RMSE) and mean average error (MAE) of the predicted results of the transfer function model is 160.94 and 109.70, respectively, which is slightly better than the co-integration model with the RMSE of 171.40 and the MAE of 122.48, indicating that the prediction accuracy could be improved by the pre-whitening process of the SST. Moreover, SST and green tide area has a dynamic correlation.

　　Key words：green tide; sea temperature; cointegration model; transfer function model; Granger causality test;

　　remote sensing