Sea-surface wind is a vital meteorological element in marine activities and climate research.This study proposed the spectral attention enhanced multidimensional feature fusion convolutional long short-term memory(LST...Sea-surface wind is a vital meteorological element in marine activities and climate research.This study proposed the spectral attention enhanced multidimensional feature fusion convolutional long short-term memory(LSTM)network(SAMFF-Conv-LSTM),a novel approach for sea-surface wind-speed prediction that emphasizes the temporal characteristics of data samples.The model incorporates the Fourier transform to extract time-and frequency-domain features from wave and wind variables.For the 12 h prediction,the SAMFF-ConvLSTM achieved a correlation coefficient of 0.960 and a root mean square error(RMSE)of 1.350 m/s,implying a high prediction accuracy.For the 24 h prediction,the RMSE of the SAMFF-ConvLSTM was reduced by 38.11%,14.26%,and 13.36%compared with those of the convolutional neural network,gated recurrent units,and convolutional LSTM(ConvLSTM),respectively.These results confirm the superior reliability and accuracy of the SAMFF-ConvLSTM over traditional models in theoretical and practical applications.展开更多
In recent years,deep learning methods have gradually been applied to prediction tasks related to Arctic sea ice concentration,but relatively little research has been conducted for larger spatial and temporal scales,ma...In recent years,deep learning methods have gradually been applied to prediction tasks related to Arctic sea ice concentration,but relatively little research has been conducted for larger spatial and temporal scales,mainly due to the limited time coverage of observations and reanalysis data.Meanwhile,deep learning predictions of sea ice thickness(SIT)have yet to receive ample attention.In this study,two data-driven deep learning(DL)models are built based on the ConvLSTM and fully convolutional U-net(FC-Unet)algorithms and trained using CMIP6 historical simulations for transfer learning and fine-tuned using reanalysis/observations.These models enable monthly predictions of Arctic SIT without considering the complex physical processes involved.Through comprehensive assessments of prediction skills by season and region,the results suggest that using a broader set of CMIP6 data for transfer learning,as well as incorporating multiple climate variables as predictors,contribute to better prediction results,although both DL models can effectively predict the spatiotemporal features of SIT anomalies.Regarding the predicted SIT anomalies of the FC-Unet model,the spatial correlations with reanalysis reach an average level of 89%over all months,while the temporal anomaly correlation coefficients are close to unity in most cases.The models also demonstrate robust performances in predicting SIT and SIE during extreme events.The effectiveness and reliability of the proposed deep transfer learning models in predicting Arctic SIT can facilitate more accurate pan-Arctic predictions,aiding climate change research and real-time business applications.展开更多
The prediction of PM_(2.5)concentration has attracted considerable research efforts in recent years.However,due to the lack of open dataset,the data processed by existing intelligent methods are only values at single ...The prediction of PM_(2.5)concentration has attracted considerable research efforts in recent years.However,due to the lack of open dataset,the data processed by existing intelligent methods are only values at single stations or mean value in a small region,while the data in real applications are all gridded values in large regions.This incompatibility in data format makes intelligent methods cannot be integrated into the practical process of PM_(2.5)prediction.In this paper,first we build a large dataset with gridded data obtained from the numerical prediction field,then an intelligent prediction method with gridded data as the basic input and output format is proposed.To capture both the spatial and temporal characteristics in data,the ConvLSTM(convolutional long short term memory)model is applied,which can utilize the advantages of both the CNN(convolutional neural network)and LSTM models.However,ConvLSTM has defects in processing multi-feature data:the more features model uses,the worse the forecasting result will be.To improve the prediction accuracy of ConvLSTM further,the attention mechanism is applied,which can describe more accurately the importance of different features and different regions for the prediction accuracy.On the built large dataset of PM_(2.5)gridded concentrations,when we predict the next hour’s value using the past 6 h,the RMSE(root mean square error)of the conventional MLR(multi-linear regression)and ConvLSTM are respectively 6.44g∕m^(3)and 6.24g∕m^(3),when the attention mechanism is incorporated into ConvLSTM,the RMSE can be decreased to 4.79g∕m^(3).展开更多
基金supported by the National Natural Science Foundation(No.42176020)the Open Research Fund of State Key Laboratory of Target Vulnerability Assessment(No.YSX2024KFYS001)+1 种基金the National Key Research and Development Program(No.2022YFC3105002)the Project from Key Laboratory of Marine Environmental Information Technology(No.2023GFW-1047).
文摘Sea-surface wind is a vital meteorological element in marine activities and climate research.This study proposed the spectral attention enhanced multidimensional feature fusion convolutional long short-term memory(LSTM)network(SAMFF-Conv-LSTM),a novel approach for sea-surface wind-speed prediction that emphasizes the temporal characteristics of data samples.The model incorporates the Fourier transform to extract time-and frequency-domain features from wave and wind variables.For the 12 h prediction,the SAMFF-ConvLSTM achieved a correlation coefficient of 0.960 and a root mean square error(RMSE)of 1.350 m/s,implying a high prediction accuracy.For the 24 h prediction,the RMSE of the SAMFF-ConvLSTM was reduced by 38.11%,14.26%,and 13.36%compared with those of the convolutional neural network,gated recurrent units,and convolutional LSTM(ConvLSTM),respectively.These results confirm the superior reliability and accuracy of the SAMFF-ConvLSTM over traditional models in theoretical and practical applications.
基金supported by the National Natural Science Foundation of China(Grant Nos.41976193 and 42176243).
文摘In recent years,deep learning methods have gradually been applied to prediction tasks related to Arctic sea ice concentration,but relatively little research has been conducted for larger spatial and temporal scales,mainly due to the limited time coverage of observations and reanalysis data.Meanwhile,deep learning predictions of sea ice thickness(SIT)have yet to receive ample attention.In this study,two data-driven deep learning(DL)models are built based on the ConvLSTM and fully convolutional U-net(FC-Unet)algorithms and trained using CMIP6 historical simulations for transfer learning and fine-tuned using reanalysis/observations.These models enable monthly predictions of Arctic SIT without considering the complex physical processes involved.Through comprehensive assessments of prediction skills by season and region,the results suggest that using a broader set of CMIP6 data for transfer learning,as well as incorporating multiple climate variables as predictors,contribute to better prediction results,although both DL models can effectively predict the spatiotemporal features of SIT anomalies.Regarding the predicted SIT anomalies of the FC-Unet model,the spatial correlations with reanalysis reach an average level of 89%over all months,while the temporal anomaly correlation coefficients are close to unity in most cases.The models also demonstrate robust performances in predicting SIT and SIE during extreme events.The effectiveness and reliability of the proposed deep transfer learning models in predicting Arctic SIT can facilitate more accurate pan-Arctic predictions,aiding climate change research and real-time business applications.
文摘The prediction of PM_(2.5)concentration has attracted considerable research efforts in recent years.However,due to the lack of open dataset,the data processed by existing intelligent methods are only values at single stations or mean value in a small region,while the data in real applications are all gridded values in large regions.This incompatibility in data format makes intelligent methods cannot be integrated into the practical process of PM_(2.5)prediction.In this paper,first we build a large dataset with gridded data obtained from the numerical prediction field,then an intelligent prediction method with gridded data as the basic input and output format is proposed.To capture both the spatial and temporal characteristics in data,the ConvLSTM(convolutional long short term memory)model is applied,which can utilize the advantages of both the CNN(convolutional neural network)and LSTM models.However,ConvLSTM has defects in processing multi-feature data:the more features model uses,the worse the forecasting result will be.To improve the prediction accuracy of ConvLSTM further,the attention mechanism is applied,which can describe more accurately the importance of different features and different regions for the prediction accuracy.On the built large dataset of PM_(2.5)gridded concentrations,when we predict the next hour’s value using the past 6 h,the RMSE(root mean square error)of the conventional MLR(multi-linear regression)and ConvLSTM are respectively 6.44g∕m^(3)and 6.24g∕m^(3),when the attention mechanism is incorporated into ConvLSTM,the RMSE can be decreased to 4.79g∕m^(3).
