Data-driven artificial intelligence (AI)-based weather prediction (AIWP) models have demonstrated significant potential in weather forecasts, facilitating paradigm shift of prediction from a deductive to an inductive inference. However, this shift raises concerns regarding the performance of the AIWP models in severe weather forecasting. Tropical cyclones (TCs) are one of the most typical cases of severe weather forecasting. In this study, we compare forecasts of Western Pacific TCs in 2023 produced by the AIWP model, Pangu-Weather, with those generated by numerical weather prediction (NWP) models, specifically the European Centre for Medium-Range Weather Forecasts (ECMWF) and the National Centers for Environmental Prediction (NCEP), in the operational context. We analyze the impact of different initial conditions on the AIWP model Pangu-Weather, in TC forecasting. Our analysis includes statistical evaluation of forecast skills related to TC activity, track, intensity, and a case study on the physical structure of TCs. The Pangu-Weather model exhibits superior forecast skills compared to the NWP model regarding TC tracks and environmental variables within TC activity domains, particularly at longer leading times. However, the overly smooth forecasts from Pangu-Weather lead to significant underestimations of intensity and a weakened dynamic-thermodynamic structure of TCs. Additionally, Pangu-Weather shows reduced sensitivity to initial conditions concerning TC structure and intensity, potentially attributable to the limitations of the training dataset and deep learning model employed. Enhancing the application of higher-quality initial conditions and the exploring hybrid models that integrate physical processes with data-driven methods could significantly improve the effectiveness of AIWP models in severe weather forecasting.

Processes at the air-sea interface govern the climate mean state and climate variability by determining the exchange of momentum, heat, and water between the atmosphere and ocean. Traditional climate models compute those exchanges across the air-sea interface by assuming an ocean surface with roughness determined by wind and stability conditions, essentially assuming ocean surface waves are in equilibrium states. In reality, that is rarely the case. Such effects have been emphasized in numerical weather predictions for weather systems like tropical cyclones. An accurate representation of ocean surface waves requires a prognostic ocean surface wave model. The addition of WAVEWATCH III (WW3) to the Community Earth System Model 2 (CESM2) makes it possible to parameterize the impacts of ocean surface waves on momentum and energy exchange. This study documents our implementation of a wave-state-dependent surface flux scheme in CEMS2. Our scheme considers the effects of waves on ocean surface roughness and those of sea spray on surface sensible and latent heat. We found that the new scheme significantly impacts the mean atmospheric circulation and the upper ocean. The errors in mean atmospheric circulation and surface temperature patterns are reduced. The modified surface flux lowers the eddy-driven jet speed and weakens the Hadley circulation. Global mean sea surface temperature (SST) warm bias is reduced due to the cooling of the Southern Ocean and eastern boundary currents. In particular, the eastern Pacific exhibited a weak cooling trend in the historical simulation for the recent decades, reducing the existing SST trend bias in CESM2.

On December 18, 2022, Hawaiian Airlines flight HA35 encountered severe turbulence in a cloud-free region without warning. We simulated this incident using the Model for Prediction Across Scales (MPAS) with a convective permitting grid. We found that the turbulence formed due to the Kelvin-Helmholtz instability (KHI) generated by strong vertical wind shear. At low altitudes, deep convection caused a decrease in wind speed in both upstream and downstream regions. At upper levels, the jet descended and accelerated after flowing over the convection, which acted like a barrier and produced a situation similar to a downslope windstorm. The low Scorer parameter above the jet and the self-induced critical level created the locally enhanced descending jet stream, which destabilized the flow through KHI.

Previous studies on South China’s convective precipitation forecast focused on the effects of multi-scale dynamics and microphysics parameterizations. However, how the uncertainty in aerosol data might cause errors in quantitative precipitation forecast (QPF) has yet to be investigated. In this case study, we estimate the impact of aerosol uncertainties on the QPF for South China’s severe convection using convection-permitting simulations. The variability range of aerosol concentrations is estimated with past observation for the pre-summer months. Simulation results suggest that the rainfall pattern and intensity change notably when aerosol concentrations are varied. The simulation with low aerosol concentrations produces the most intense precipitation, approximately 50\% stronger than the high-concentration simulation. Decreasing aerosol hygroscopicity also increases precipitation intensity, especially in pristine clouds. The aerosol uncertainty changes alter the number of cloud condensation and ice nuclei, which modifies the altitude and amount of latent heating and thereby modulates convection.

The projection of extreme convective precipitation by global climate models (GCM) exhibits significant uncertainty due to coarse resolutions. Direct dynamical downscaling (DDD) of regional climate at kilometer-scale resolutions provides valuable insight into extreme precipitation changes, but its computational expense is formidable. Here we document the effectiveness of machine learning to enable smart dynamical downscaling (SDD), which selects a small subset of GCM data to conduct downscaling. Trained with data for three subtropical/tropical regions, convolutional neural networks (CNNs) retained 92% to 98% of extreme precipitation events (rain intensity higher than the 99th percentile) while filtering out 88% to 95% of circulation data. When applied to reanalysis data sets differing from training data, the CNNs’ skill in retaining extremes decreases modestly in subtropical regions but sharply in the deep tropics. Nonetheless, one of the CNNs can still retain 62% of all extreme events in the deep tropical region in the worst case.

This study investigates the mechanisms by which small-scale turbulence and cloud physics determine the organization of large-scale convection in radiative-convective equilibrium (RCE), an idealization of the tropical atmosphere. Under uniform forcings similar to typical tropical conditions, the atmosphere in RCE might spontaneously separate into dry and moist regions on scales of 100-1000 km, with convective clouds aggregating into a cluster in the latter. This phenomenon is known as convective self-aggregation. Herein, we demonstrate that subtle changes in assumptions related to cloud physics and turbulence on scales of ~1 km can dictate the emergence or suppression of convective self-aggregation, resulting from a bifurcation of the dynamical system. The bifurcation occurs when a small dry patch forms in the domain and is sustained because it contributes to negative effective diffusivity of the circulation. Cloud-radiation feedbacks and turbulence circulation interactions govern the formation of such dry patches, thereby modulating the bifurcation. This sensitive dependence on subgrid process models might be a fundamental barrier to climate predictability in light of inherent uncertainties in microscale processes. Because without the capability to include exact representations of those processes in climate models, slight differences in the different approximations used by modelers can lead to qualitative changes in climate predictions, at least for some processes.