Data from the first research flight (RF01) of the second Dynamics and Chemistry of Marine Stratocumulus (DYCOMS-II) field study are used to evaluate the fidelity with which large-eddy simulations (LESs) can represent the turbulent structure of stratocumulus-topped boundary layers. The initial data and forcings for this case placed it in an interesting part of parameter space, near the boundary where cloud-top mixing is thought to render the cloud layer unstable on the one hand, or tending toward a decoupled structure on the other hand. The basis of this evaluation consists of sixteen 4-h simulations from 10 modeling centers over grids whose vertical spacing was 5 m at the cloud-top interface and whose horizontal spacing was 35 m. Extensive sensitivity studies of both the configuration of the case and the numerical setup also enhanced the analysis. Overall it was found that (i) if efforts are made to reduce spurious mixing at cloud top, either by refining the vertical grid or limiting the effects of the subgrid model in this region, then the observed turbulent and thermodynamic structure of the layer can be reproduced with some fidelity; (ii) the base, or native configuration of most simulations greatly overestimated mixing at cloud top, tending toward a decoupled layer in which cloud liquid water path and turbulent intensities were grossly underestimated; (iii) the sensitivity of the simulations to the representation of mixing at cloud top is, to a certain extent, amplified by particulars of this case. Overall the results suggest that the use of LESs to map out the behavior of the stratocumulus-topped boundary layer in this interesting region of parameter space requires a more compelling representation of processes at cloud top. In the absence of significant leaps in the understanding of subgrid-scale (SGS) physics, such a representation can only be achieved by a significant refinement in resolution-a refinement that, while conceivable given existing resources, is probably still beyond the reach of most centers.

Ten 3-D cloud-resolving model simulations and four 3-D limited area model simulations of an intense mesoscale convective system observed on 23-24 January 2006 during the Tropical Warm Pool-International Cloud Experiment (TWP-ICE) are compared with each other and with observed radar reflectivity fields and dual-Doppler retrievals of vertical wind speeds in an attempt to explain published results showing a high bias in simulated convective radar reflectivity aloft. This high-bias results from ice water content being large, which is a product of large, strong convective updrafts, although hydrometeor size distribution assumptions modulate the size of this bias. Making snow mass more realistically proportional to D 2 rather than D 3 eliminates unrealistically large snow reflectivities over 40 dBZ in some simulations. Graupel, unlike snow, produces high biased reflectivity in all simulations, which is partly a result of parameterized microphysics but also partly a result of overly intense simulated updrafts. Peak vertical velocities in deep convective updrafts are greater than dual-Doppler-retrieved values, especially in the upper troposphere. Freezing of liquid condensate, often rain, lofted above the freezing level in simulated updraft cores greatly contributes to these excessive upper tropospheric vertical velocities. The strongest simulated updraft cores are nearly undiluted, with some of the strongest showing supercell characteristics during the multicellular (presquall) stage of the event. Decreasing horizontal grid spacing from 900 to 100 m slightly weakens deep updraft vertical velocity and moderately decreases the amount of condensate aloft but not enough to match observational retrievals. Therefore, overly intense simulated updrafts may additionally be a product of unrealistic interactions between convective dynamics, parameterized microphysics, and large-scale model forcing that promote different convective strengths than observed.

[1] Hurricane boundary layer (HBL) processes, especially the structure of the coherent large eddy circulations (LECs) and their induced vertical transport, are not well understood. This paper introduces a large eddy simulation (LES) framework in a weather hindcasting mode developed from a multiple scale nested Weather Research and Forecasting (WRF) model. Using the WRF-LES, this study investigated the structure of the HBL LECs and the associated vertical transport during the landfall of Hurricane Ivan (2004). The simulation shows that the HBL LECs exist in a mean stable environment and consist of well-defined updraft and downdraft. Statistically, the HBL LECs are only slightly skewed with the updrafts and downdrafts relatively evenly distributed spatially. The inversion base basically envelopes the upper boundary of LECs. The trough in between two adjacent LECs is where most entrainment takes place, whereas the crest of the LECs is where boundary layer air detrains out of the HBL. In such a way, LECs directly connect the surface, the HBL, and the main body of a hurricane vortex and enhance the exchange of energy, moisture, and momentum between them. It is found that the current boundary layer schemes significantly underestimate the resolved turbulent fluxes due to the fact that the effects of LECs have not been included in the parameterizations. On the basis of the statistical structure of LECs simulated by the WRF-LES, this paper proposes a conceptual updraft-downdraft model that can potentially be implemented in weather forecasting models to parameterize the fluxes induced by the HBL LEC transport.

No abstract

Ten single-column models (SCMs) from eight groups are used to simulate a nocturnal nonprecipitating marine stratocumulus-topped mixed layer as part of an intercomparison organized by the Global Energy and Water Cycle Experiment Cloud System Study, Working Group 1. The case is idealized from observations from the Dynamics and Chemistry of Marine Stratocumulus II, Research Flight 1. SCM simulations with operational resolution are supplemented by high-resolution simulations and compared with observations and large-eddy simulations. All participating SCMs are able to maintain a sharp inversion and a mixed cloud-topped layer, although the moisture profiles show a slight gradient in the mixed layer and produce entrainment rates broadly consistent with observations, but the liquid water paths vary by a factor of 10 after only 1 h of simulation at both high and operational resolution. Sensitivity tests show insensitivity to activation of precipitation and shallow convection schemes in most models, as one would observationally expect for this case.

Observations of precipitating trade wind cumuli show convective invigoration on the downwind side of their cold pools. The authors study convection and cold pools using a nested-Weather Research and Forecasting Model simulation of 19 January 2005-a day from the Rain in Cumulus over the Ocean experiment. The temperature and water vapor mixing ratio drops in simulated cold pools fall within the envelope of observed cases, and the wind enhancement matches observations more closely. Subcloud updrafts downwind and near the cold pool boundary are statistically compared to updrafts further from cold pools. Updrafts near cold pool outflows are moister than the other updrafts and are more likely to originate from overall moister regions. Cold pool-influenced updrafts tend to exceed the other updrafts in vertical velocity and are associated with more cloud liquid water. The strength of circulation within the cold pool boundary is unable to match that because of the low-level environmental wind shear, and the lifted updrafts advect faster than the environmental wind, thereby accessing the ambient environmental moisture converged by cold pool expansion. Cases with higher rain rates correspond to larger cloud cover through the shearing off of the upperlevel cloud, consistent with observations. This study suggests that it is the ability of cold pools to lift thermodynamically favorable air that is critical for secondary convection of trade wind cumuli.

Coastal currents generally flow downshelf with land on the right side (Northern Hemisphere) under the geostrophic balance, and are often strengthened by downwelling‐favorable winds. However, the recent mooring observation in the inner southwestern Yellow Sea showed that coastal transport direction can be substantially changed by tidal forcing. In the survey, the tidal‐averaged transports at two out of three sites remained northward (i.e., in the upshelf direction) and opposite the downwelling‐favorable northerly wind, except during a brief neap tide period. Numerical experiments showed that the incoming Poincaré wave tide from the East China Sea plays a key role in forming this counter‐wind transport system. This tidal wave produces a shoreward tidal stress south of 33.5°N in the inner southwestern Yellow Sea, driving an upshelf transport under the Earth's rotation. Counterpropagating tidal waves from the East China Sea and the northern Yellow Sea collide in coastal water in 32.5–34°N, which produce a standing tidal wave and therefore a mean sea‐surface setup with alongshore and cross‐shelf scales of both >100 km. This sea‐surface setup causes an alongshore sea surface gradient, which veers the upshelf transport to the offshore direction under geostrophic balance. The strong tidal current increases the tidal‐mean bottom resistance in the SCW, thus reduces the wind‐driven current to a magnitude smaller than the tide‐induced residual transport velocity. Therefore, upshelf transport persists in the inner southwestern Yellow Sea, and the Changjiang River Estuary becomes a major source area for the inner southwestern Yellow Sea.

Medical dialogue systems are promising in assisting in telemedicine to increase access to healthcare services, improve the quality of patient care, and reduce medical costs. To facilitate the research and development of medical dialogue systems, we build large-scale medical dialogue datasets -MedDialog, which contain 1) a Chinese dataset with 3.4 million conversations between patients and doctors, 11.3 million utterances, 660.2 million tokens, covering 172 specialties of diseases, and 2) an English dataset with 0.26 million conversations, 0.51 million utterances, 44.53 million tokens, covering 96 specialties of diseases. To our best knowledge, MedDialog is the largest medical dialogue dataset to date. We pretrain several dialogue generation models on the Chinese MedDialog dataset, including Transformer, GPT, BERT-GPT, and compare their performance. It is shown that models trained on MedDialog are able to generate clinically correct and human-like medical dialogues. We also study the transferability of models trained on MedDialog to lowresource medical dialogue generation tasks. It is shown that via transfer learning which finetunes the models pretrained on MedDialog, the performance on medical dialogue generation tasks with small datasets can be greatly improved, as shown in human evaluation and automatic evaluation.