Pangu-Weather(PGW),trained with deep learning–based methods(DL-based model),shows significant potential for global medium-range weather forecasting.However,the interpretability and trustworthiness of global medium-ra...Pangu-Weather(PGW),trained with deep learning–based methods(DL-based model),shows significant potential for global medium-range weather forecasting.However,the interpretability and trustworthiness of global medium-range DLbased models raise many concerns.This study uses the singular vector(SV)initial condition(IC)perturbations of the China Meteorological Administration's Global Ensemble Prediction System(CMA-GEPS)as inputs of PGW for global ensemble prediction(PGW-GEPS)to investigate the ensemble forecast sensitivity of DL-based models to the IC errors.Meanwhile,the CMA-GEPS forecasts serve as benchmarks for comparison and verification.The spatial structures and prediction performance of PGW-GEPS are discussed and compared to CMA-GEPS based on seasonal ensemble experiments.The results show that the ensemble mean and dispersion of PGW-GEPS are similar to those of CMA-GEPS in the medium range but with smoother forecasts.Meanwhile,PGW-GEPS is sensitive to the SV IC perturbations.Specifically,PGWGEPS can generate realistic ensemble spread beyond the sub-synoptic scale(wavenumbers≤64)with SV IC perturbations.However,PGW's kinetic energy is significantly reduced at the sub-synoptic scale,leading to error growth behavior inconsistent with CMA-GEPS at that scale.Thus,this behavior indicates that the effective resolution of PGW-GEPS is beyond the sub-synoptic scale and is limited to predicting mesoscale atmospheric motions.In terms of the global mediumrange ensemble prediction performance,the probability prediction skill of PGW-GEPS is comparable to CMA-GEPS in the extratropic when they use the same IC perturbations.That means that PGW has a general ability to provide skillful global medium-range forecasts with different ICs from numerical weather prediction.展开更多
Data-driven deep-learning models have shown outstanding performance in global weather forecasting.Understanding the dynamic response processes within these models is crucial for comprehending the embedded physical pro...Data-driven deep-learning models have shown outstanding performance in global weather forecasting.Understanding the dynamic response processes within these models is crucial for comprehending the embedded physical processes and sources of predictability.By applying the classic tropical steady heating experiment to the Pangu-Weather deep-learning model during the austral winter background state,we observe a classic Matsuno-Gill response in the tropics and planetary Rossby waves propagating to the polar regions.The results of the Pangu-Weather model are consistent with those of traditional physics-based general circulation models(GCMs):convective heating forcing in the tropical Atlantic and western Indian Ocean,and convective cooling forcing in the Maritime Continent all deepen the Amundsen Sea Low(ASL),while convective heating forcing in the western Pacific weakens the ASL.The Pangu-Weather model has learned that these tropical basins jointly and linearly regulate the atmospheric circulation around West Antarctica through Rossby waves.However,the Pangu-Weather model overestimates(underestimates)atmospheric responses of heating in the tropical Pacific(Indian and Atlantic)Ocean compared with traditional GCMs,with a much larger contribution of Pacific heating forcing than other basins in changes of the ASL.The physics learned from reanalysis data may be the source of these deep-learning models’predictability,and the accuracy of extended-range forecasting and the potential of seasonal forecasting using deep-learning models may be influenced by overestimation or underestimation of the role of the tropical Pacific,Indian,and Atlantic Oceans.展开更多
基金supported by the joint funds of the Chinese National Natural Science Foundation(NSFC)(Grant No.U2242213)the funds of the NSFC(Grant No.42341209)+2 种基金the National Key Research and Development(R&D)Program of the Ministry of Science and Technology of China(Grant No.2021YFC3000902)the National Science Foundation for Young Scholars(Grant No.42205166)the Joint Research Project for Meteorological Capacity Improvement(Grant No.22NLTSQ008)。
文摘Pangu-Weather(PGW),trained with deep learning–based methods(DL-based model),shows significant potential for global medium-range weather forecasting.However,the interpretability and trustworthiness of global medium-range DLbased models raise many concerns.This study uses the singular vector(SV)initial condition(IC)perturbations of the China Meteorological Administration's Global Ensemble Prediction System(CMA-GEPS)as inputs of PGW for global ensemble prediction(PGW-GEPS)to investigate the ensemble forecast sensitivity of DL-based models to the IC errors.Meanwhile,the CMA-GEPS forecasts serve as benchmarks for comparison and verification.The spatial structures and prediction performance of PGW-GEPS are discussed and compared to CMA-GEPS based on seasonal ensemble experiments.The results show that the ensemble mean and dispersion of PGW-GEPS are similar to those of CMA-GEPS in the medium range but with smoother forecasts.Meanwhile,PGW-GEPS is sensitive to the SV IC perturbations.Specifically,PGWGEPS can generate realistic ensemble spread beyond the sub-synoptic scale(wavenumbers≤64)with SV IC perturbations.However,PGW's kinetic energy is significantly reduced at the sub-synoptic scale,leading to error growth behavior inconsistent with CMA-GEPS at that scale.Thus,this behavior indicates that the effective resolution of PGW-GEPS is beyond the sub-synoptic scale and is limited to predicting mesoscale atmospheric motions.In terms of the global mediumrange ensemble prediction performance,the probability prediction skill of PGW-GEPS is comparable to CMA-GEPS in the extratropic when they use the same IC perturbations.That means that PGW has a general ability to provide skillful global medium-range forecasts with different ICs from numerical weather prediction.
基金supported by the National Natural Science Foundation of China (Grant Nos.42288101 and 42175025)the National Key Research and Devel-opment Program of China (Grant No. 2023YFF0806700)
文摘Data-driven deep-learning models have shown outstanding performance in global weather forecasting.Understanding the dynamic response processes within these models is crucial for comprehending the embedded physical processes and sources of predictability.By applying the classic tropical steady heating experiment to the Pangu-Weather deep-learning model during the austral winter background state,we observe a classic Matsuno-Gill response in the tropics and planetary Rossby waves propagating to the polar regions.The results of the Pangu-Weather model are consistent with those of traditional physics-based general circulation models(GCMs):convective heating forcing in the tropical Atlantic and western Indian Ocean,and convective cooling forcing in the Maritime Continent all deepen the Amundsen Sea Low(ASL),while convective heating forcing in the western Pacific weakens the ASL.The Pangu-Weather model has learned that these tropical basins jointly and linearly regulate the atmospheric circulation around West Antarctica through Rossby waves.However,the Pangu-Weather model overestimates(underestimates)atmospheric responses of heating in the tropical Pacific(Indian and Atlantic)Ocean compared with traditional GCMs,with a much larger contribution of Pacific heating forcing than other basins in changes of the ASL.The physics learned from reanalysis data may be the source of these deep-learning models’predictability,and the accuracy of extended-range forecasting and the potential of seasonal forecasting using deep-learning models may be influenced by overestimation or underestimation of the role of the tropical Pacific,Indian,and Atlantic Oceans.
