Due to various technical issues,existing numerical weather prediction(NWP)models often perform poorly at forecasting rainfall in the first several hours.To correct the bias of an NWP model and improve the accuracy of ...Due to various technical issues,existing numerical weather prediction(NWP)models often perform poorly at forecasting rainfall in the first several hours.To correct the bias of an NWP model and improve the accuracy of short-range precipitation forecasting,we propose a deep learning-based approach called UNet Mask,which combines NWP forecasts with the output of a convolutional neural network called UNet.The UNet Mask involves training the UNet on historical data from the NWP model and gridded rainfall observations for 6-hour precipitation forecasting.The overlap of the UNet output and the NWP forecasts at the same rainfall threshold yields a mask.The UNet Mask blends the UNet output and the NWP forecasts by taking the maximum between them and passing through the mask,which provides the corrected 6-hour rainfall forecasts.We evaluated UNet Mask on a test set and in real-time verification.The results showed that UNet Mask outperforms the NWP model in 6-hour precipitation prediction by reducing the FAR and improving CSI scores.Sensitivity tests also showed that different small rainfall thresholds applied to the UNet and the NWP model have different effects on UNet Mask's forecast performance.This study shows that UNet Mask is a promising approach for improving rainfall forecasting of NWP models.展开更多
The definition of a reference state close to the realistic atmosphere in an atmospheric model is essential for deriving prognostic deviations and improving numerical accuracy.In this study,a new dynamical framework al...The definition of a reference state close to the realistic atmosphere in an atmospheric model is essential for deriving prognostic deviations and improving numerical accuracy.In this study,a new dynamical framework allowing easy switching between a one-dimensional(1D)and a three-dimensional(3D)time-independent reference state is developed for the semi-implicit semi-Lagrangian solver in a global non-hydrostatic atmospheric model on Yin–Yang grids.The 3D reference state is introduced with consideration of additional horizontal gradient terms of referencestate terms,which is different from the 1D reference state.It is characterized by reduced magnitude of deviations,more accurate pressure gradient force,as well as alleviated numerical noise.Four idealized benchmark tests and multiple full-physics real-case forecasts are carried out to assess the impact of the 3D and 1D reference states.The 3D reference state shows significant advantages in the simulation of atmospheric transport and wave propagation in the idealized experiments.In the real-case forecasts,batched forecasts from June to August 2021 show a comprehensive improvement in medium-range prediction by using the 3D reference state.The new scheme achieves an enhanced prediction skill for large-scale circulation and extends the effective forecast period by 0.8 days in the Northern Hemisphere.展开更多
With the advent of the phased array radar(PAR)technology,it is possible to capture the development and evolution of convective systems in a much shorter time interval and with higher spatial resolution than via tradit...With the advent of the phased array radar(PAR)technology,it is possible to capture the development and evolution of convective systems in a much shorter time interval and with higher spatial resolution than via traditional Doppler radar.Research on the assimilation of PAR observations in numerical weather prediction models is still in its infancy in China.In this paper,the impact of assimilating PAR data on model forecasts was investigated by a case study of a local heavy rainfall event that occurred over Foshan city of Guangdong Province on 26 August 2020,via a series of sensitivity experiments.Both the retrieved three-dimensional wind and hydrometeor fields were assimilated through the nudging method with the Tropical Regional Assimilation Model for South China Sea_Rapid Update Cycle_1km(TRAMS_RUC_1km).The temperature and moisture fields were also adjusted accordingly.The results show that significant improvements are made in the experiments with latent heat nudging and adjustment of the water vapor field,which implies the importance of thermodynamic balance in the initialization of the convective system and highlights the need to assimilate PAR radar observations in a continuous manner to maximize the impact of the data.Sensitivity tests also indicate that the relaxation time should be less than 5 min.In general,for this case,the assimilation of PAR data can significantly improve the nowcasting skill of the regional heavy precipitation.This study is the first step towards operational PAR data assimilation in numerical weather prediction in southern China.展开更多
基金jointly supported by the National Natural Science Foundation of China(Grant No.U1811464)the Hydraulic Innovation Project of Science and Technology of Guangdong Province of China(Grant No.2022-01)the Guangzhou Basic and Applied Basic Research Foundation(Grant No.202201011472)。
文摘Due to various technical issues,existing numerical weather prediction(NWP)models often perform poorly at forecasting rainfall in the first several hours.To correct the bias of an NWP model and improve the accuracy of short-range precipitation forecasting,we propose a deep learning-based approach called UNet Mask,which combines NWP forecasts with the output of a convolutional neural network called UNet.The UNet Mask involves training the UNet on historical data from the NWP model and gridded rainfall observations for 6-hour precipitation forecasting.The overlap of the UNet output and the NWP forecasts at the same rainfall threshold yields a mask.The UNet Mask blends the UNet output and the NWP forecasts by taking the maximum between them and passing through the mask,which provides the corrected 6-hour rainfall forecasts.We evaluated UNet Mask on a test set and in real-time verification.The results showed that UNet Mask outperforms the NWP model in 6-hour precipitation prediction by reducing the FAR and improving CSI scores.Sensitivity tests also showed that different small rainfall thresholds applied to the UNet and the NWP model have different effects on UNet Mask's forecast performance.This study shows that UNet Mask is a promising approach for improving rainfall forecasting of NWP models.
基金Supported by the National Natural Science Foundation of China(42375153,42075151,and 42205157).
文摘The definition of a reference state close to the realistic atmosphere in an atmospheric model is essential for deriving prognostic deviations and improving numerical accuracy.In this study,a new dynamical framework allowing easy switching between a one-dimensional(1D)and a three-dimensional(3D)time-independent reference state is developed for the semi-implicit semi-Lagrangian solver in a global non-hydrostatic atmospheric model on Yin–Yang grids.The 3D reference state is introduced with consideration of additional horizontal gradient terms of referencestate terms,which is different from the 1D reference state.It is characterized by reduced magnitude of deviations,more accurate pressure gradient force,as well as alleviated numerical noise.Four idealized benchmark tests and multiple full-physics real-case forecasts are carried out to assess the impact of the 3D and 1D reference states.The 3D reference state shows significant advantages in the simulation of atmospheric transport and wave propagation in the idealized experiments.In the real-case forecasts,batched forecasts from June to August 2021 show a comprehensive improvement in medium-range prediction by using the 3D reference state.The new scheme achieves an enhanced prediction skill for large-scale circulation and extends the effective forecast period by 0.8 days in the Northern Hemisphere.
基金Supported by the National Natural Science Foundation of China(U1811464 and 40675099)National Key Research and Development Program of China(2018YFC1506900)。
文摘With the advent of the phased array radar(PAR)technology,it is possible to capture the development and evolution of convective systems in a much shorter time interval and with higher spatial resolution than via traditional Doppler radar.Research on the assimilation of PAR observations in numerical weather prediction models is still in its infancy in China.In this paper,the impact of assimilating PAR data on model forecasts was investigated by a case study of a local heavy rainfall event that occurred over Foshan city of Guangdong Province on 26 August 2020,via a series of sensitivity experiments.Both the retrieved three-dimensional wind and hydrometeor fields were assimilated through the nudging method with the Tropical Regional Assimilation Model for South China Sea_Rapid Update Cycle_1km(TRAMS_RUC_1km).The temperature and moisture fields were also adjusted accordingly.The results show that significant improvements are made in the experiments with latent heat nudging and adjustment of the water vapor field,which implies the importance of thermodynamic balance in the initialization of the convective system and highlights the need to assimilate PAR radar observations in a continuous manner to maximize the impact of the data.Sensitivity tests also indicate that the relaxation time should be less than 5 min.In general,for this case,the assimilation of PAR data can significantly improve the nowcasting skill of the regional heavy precipitation.This study is the first step towards operational PAR data assimilation in numerical weather prediction in southern China.
