用于高温次季节预报的深度学习方法

A deep Learning Method for Sub-seasonal Forecast of High Temperature

高温是一种比较常见的气象灾害,对人们日常生活和健康以及国民经济都有一定的影响。准确地预报未来10-30天(次季节)的每日高温天气状况,有助于人们及早地做出对应的决策和降低财产损失。本文设计了一种深度学习模型,通过对两个Encoder提取的大气环流和地表高温的特征进行延迟融合,并结合Decoder逐日输出高温预报结果。实验结果表明,本文的方法有较好的预测效果,相较于单个Encoder的深度学习方法,均方根误差降低了2.12%,异常相关系数提高了2.72%,决定系数提高了0.04。

High temperature is a common meteorological disaster that has a certain impact on people’s daily life and economy.Accurate forecasting of daily high temperature weather for the next 10-30 days(sub-seasonal)can help people make early decisions and reduce damage.In the paper,a deep learning model was design by late fusion the futures of atmospheric circulation data and surface high temperature extracted by two Encoders,and combining with Decoder to output day-by-day high temperature forecasting results.The experimental results show that the proposed method has better prediction effect,with a 2.12%reduction in root mean square error,a 2.72%improvement in anomaly correlation coefficient and a 0.04%improvement in coefficient of determination compared to the deep learning method with a single Encoder.