The detailed surface rainfall processes associated with landfalling typhoon Kaemi（2006） are investigated based on hourly data from a two-dimensional cloud-resolving model simulation. The model is integrated for 6 da...The detailed surface rainfall processes associated with landfalling typhoon Kaemi（2006） are investigated based on hourly data from a two-dimensional cloud-resolving model simulation. The model is integrated for 6 days with imposed large-scale vertical velocity, zonal wind, horizontal temperature and vapor advection from National Center for Environmental Prediction （NCEP） / Global Data Assimilation System （GDAS） data. The simulation data are validated with observations in terms of surface rain rate. The Root-Mean-Squared （RMS） difference in surface rain rate between the simulation and the gauge observations is 0.660 mm h^-1, which is smaller than the standard deviations of both the simulated rain rate （0.753 mm h^-1） and the observed rain rate （0.833 mm h^-1）. The simulation data are then used to study the physical causes associated with the detailed surface rainfall processes during the landfall. The results show that time averaged and model domain-mean Ps mainly comes from large-scale convergence （QWVF） and local vapor loss （positive QWVT）. Large underestimation （about 15%） of Ps will occur if QWVT and QCM （cloud source/sink） are not considered as contributors to Ps ,QWVF accounts for the variation of P during most of the integration time, while it is not always a contributor to Ps,Sometimes surface rainfall could occur when divergence is dominant with local vapor loss to be a contributor to Ps - Surface rainfall is a result ofmulti-timescale interactions. QWVE possesses the longest time scale and the lowest frequeney the second and QCM of variation with time and may exert impact on P on longer time scales. QWVF possesses longest time scale and lowest frequency and can explain most of the variation of Ps. QWVT possess shorter time scales and higher frequencies, which can explain more detailed variations in Ps. Partitioning analysis shows that stratiform rainfall is dominant from the morning of 26 July till the late night of 27 July. After that, convective rainfall dominates till about 1000 LST 28 July. Before 28 July, the variations of QWVT in rainfall-free regions contribute less to that of the domain-mean QWVT while after that they contribute much, which is consistent to the corresponding variations in their fractional coverage. The variations of QWVF in rainfall regions are the main contributors to that of the domain-mean QWVF, then the main contributors to the surface rain rate before the afternoon of 28 July.展开更多
The self-attention networks and Transformer have dominated machine translation and natural language processing fields,and shown great potential in image vision tasks such as image classification and object detection.I...The self-attention networks and Transformer have dominated machine translation and natural language processing fields,and shown great potential in image vision tasks such as image classification and object detection.Inspired by the great progress of Transformer,we propose a novel general and robust voxel feature encoder for 3D object detection based on the traditional Transformer.We first investigate the permutation invariance of sequence data of the self-attention and apply it to point cloud processing.Then we construct a voxel feature layer based on the self-attention to adaptively learn local and robust context of a voxel according to the spatial relationship and context information exchanging between all points within the voxel.Lastly,we construct a general voxel feature learning framework with the voxel feature layer as the core for 3D object detection.The voxel feature with Transformer(VFT)can be plugged into any other voxel-based 3D object detection framework easily,and serves as the backbone for voxel feature extractor.Experiments results on the KITTI dataset demonstrate that our method achieves the state-of-the-art performance on 3D object detection.展开更多
基金National Basic Research Program of China (973 Program) (2009CB421505)National Natural Science Foundation of China (40775036)Knowledge Innovation Program of Chinese Academy of Sciences (IAP07214)
文摘The detailed surface rainfall processes associated with landfalling typhoon Kaemi（2006） are investigated based on hourly data from a two-dimensional cloud-resolving model simulation. The model is integrated for 6 days with imposed large-scale vertical velocity, zonal wind, horizontal temperature and vapor advection from National Center for Environmental Prediction （NCEP） / Global Data Assimilation System （GDAS） data. The simulation data are validated with observations in terms of surface rain rate. The Root-Mean-Squared （RMS） difference in surface rain rate between the simulation and the gauge observations is 0.660 mm h^-1, which is smaller than the standard deviations of both the simulated rain rate （0.753 mm h^-1） and the observed rain rate （0.833 mm h^-1）. The simulation data are then used to study the physical causes associated with the detailed surface rainfall processes during the landfall. The results show that time averaged and model domain-mean Ps mainly comes from large-scale convergence （QWVF） and local vapor loss （positive QWVT）. Large underestimation （about 15%） of Ps will occur if QWVT and QCM （cloud source/sink） are not considered as contributors to Ps ,QWVF accounts for the variation of P during most of the integration time, while it is not always a contributor to Ps,Sometimes surface rainfall could occur when divergence is dominant with local vapor loss to be a contributor to Ps - Surface rainfall is a result ofmulti-timescale interactions. QWVE possesses the longest time scale and the lowest frequeney the second and QCM of variation with time and may exert impact on P on longer time scales. QWVF possesses longest time scale and lowest frequency and can explain most of the variation of Ps. QWVT possess shorter time scales and higher frequencies, which can explain more detailed variations in Ps. Partitioning analysis shows that stratiform rainfall is dominant from the morning of 26 July till the late night of 27 July. After that, convective rainfall dominates till about 1000 LST 28 July. Before 28 July, the variations of QWVT in rainfall-free regions contribute less to that of the domain-mean QWVT while after that they contribute much, which is consistent to the corresponding variations in their fractional coverage. The variations of QWVF in rainfall regions are the main contributors to that of the domain-mean QWVF, then the main contributors to the surface rain rate before the afternoon of 28 July.
基金National Natural Science Foundation of China(No.61806006)Innovation Program for Graduate of Jiangsu Province(No.KYLX160-781)University Superior Discipline Construction Project of Jiangsu Province。
文摘The self-attention networks and Transformer have dominated machine translation and natural language processing fields,and shown great potential in image vision tasks such as image classification and object detection.Inspired by the great progress of Transformer,we propose a novel general and robust voxel feature encoder for 3D object detection based on the traditional Transformer.We first investigate the permutation invariance of sequence data of the self-attention and apply it to point cloud processing.Then we construct a voxel feature layer based on the self-attention to adaptively learn local and robust context of a voxel according to the spatial relationship and context information exchanging between all points within the voxel.Lastly,we construct a general voxel feature learning framework with the voxel feature layer as the core for 3D object detection.The voxel feature with Transformer(VFT)can be plugged into any other voxel-based 3D object detection framework easily,and serves as the backbone for voxel feature extractor.Experiments results on the KITTI dataset demonstrate that our method achieves the state-of-the-art performance on 3D object detection.
