随着全球气候变暖加剧,北极地区的大气海洋环境剧烈变化,导致海冰变化更加不稳定,使得海冰预测的难度增大.本研究选择海表温度、2m平均气温、二氧化碳浓度为大气海洋变量,海冰范围距平为时序特征参数,将上述参量作为北极海冰范围(Sea Ice Extent,SIE)的预测要素,建立了面向SIE的多变量长短期记忆(Long Short Term Memory,LSTM)神经网络模型,对比分析了2015-2021年不同时间序列预测模型的预测结果.结果显示:本研究所构建模型的RMSE、MAE、MAPE分别为0.353×106 km2、0.261×106 km2和3.191%.相比于其他预测模型,结合大气海洋变量和时序特征参数后的LSTM模型预测结果误差更小,拟合效果更好,可以消除夏季海冰剧烈变化对预测效果的影响,提高海冰范围的预测精度,对北极航道的通航安全保障工作具有重要的研究与应用价值.
Prediction of Arctic Sea Ice Extent based on atmospheric ocean multivariate and sea ice temporal characteristics
With the intensification of global climate warming,the Arctic region is experiencing drastic changes in its atmospheric and oceanic environment,leading to increased instability in sea ice variations and making sea ice prediction more challenging.In this study,the atmospheric and oceanic variables(sea surface temperature,mean 2m air temperature,carbon dioxide concentration)and the time series feature parameter(sea ice extent anomaly)were selected as predictive elements.Then we constructed a multivariate Long Short-Term Memory(LSTM)neural network model through the predictive elements for the Arctic Sea Ice Extent(SIE)prediction.Finally,the model constructed in this study was evaluated and compared with different time series forecasting models for the period from 2015 to 2021.The results showed that the Root Mean Square Error(RMSE),Mean Absolute Error(MAE),and Mean Absolute Percentage Error(MAPE)for the multivariate LSTM model were 0.353×106 km2,0.261×106 km2 and 3.191%,respectively.Compared to other time series models,the LSTM model,which incorporates atmospheric and oceanic variables as well as time series feature parameters,exhibited reduced prediction errors and superior fitting performance.This capability helps mitigate the impact of abrupt changes in summer sea ice on prediction outcomes and enhances the accuracy of sea ice extent forecasts.Consequently,it holds significant research and practical value in ensuring the safety and security of navigation on Arctic routes.
ArcticSea Ice ExtentLSTM neural network modeltime series predictiontime series characteristics
NETL
NSTL
万方数据
