Convectively induced turbulence (CIT) is a severe aviation hazard. It is challenging to forecast CIT because low-resolution models cannot explicitly resolve convective motions at kilometer scales. In this study, we used the Model for Prediction Across Scales (MPAS) to simulate CIT cases with convection-permitting resolution ( $$\sim $$ ∼ 1 km) in the region of the CIT events and coarse resolution in other parts of the globe. We developed a method to estimate the eddy dissipation rate (EDR) using the resolved wind field of the MPAS simulations. The method is based on explicit filtering and reconstruction in the turbulence modeling for large-eddy simulations (LES). It estimates turbulence kinetic energy (TKE), which is then used to derive EDR. The new method produces different turbulence distribution and intensity than previous methods based on second-order structure functions and convective gravity wave drag, with higher accuracy and better correlation with observations for CIT cases tested in this study. The 1-km resolution simulation generates more accurate EDR and improves spatial patterns, but it is computationally demanding. The 3-km resolution can get benefits from reasonable accuracy and affordable computational cost. Because convection-permitting resolutions are in the gray zone for simulating convection, we evaluated the sensitivity of the prediction to the variations in physical and numerical schemes. Varying cumulus convection parameterization and monotonicity of numerical schemes are identified as practical approaches to generate beneficial ensemble spread. However, the physical perturbation-based ensemble has limitations, and initial condition perturbations are still necessary to encompass uncertainties in the development of convection.

This paper documents a windshear case for an airport in the Qinghai-Tibet Plateau and explores, for the first time, the capability for high-resolution numerical weather simulation of the wind shear features. The windshear appears to be associated with pulses of the wind speed in a low-level easterly jet. The features are basically reproduced quite well with the high-resolution numerical model, though some discrepancies are identified, such as the maximum wind speed of the easterly jet and the magnitude of the eddy dissipation rate as compared with the actual Doppler LIDAR observations. Statistical analysis has been performed between the observation and the simulation results. The sensitivity of the modeling result to the choice of turbulence parameterization scheme has also been studied. The study result shows that it is possible to forecast the windshear feature using a high-resolution numerical weather prediction model for an airport in the complex terrain of the Plateau.

Accurate nowcasting of convective clouds from satellite imagery is essential for mitigating the impacts of meteorological disasters, especially in developing countries and remote regions with limited ground-based observations. Recent advances in deep learning have shown promise in video prediction; however, existing models frequently produce blurry results and exhibit reduced accuracy when forecasting physical fields. Here, we introduce SATcast, a diffusion model that leverages a cascade architecture and multimodal inputs for nowcasting cloud fields in satellite imagery. SATcast incorporates physical fields predicted by FuXi, a deep-learning weather model, alongside past satellite observations as conditional inputs to generate high-quality future cloud fields. Through comprehensive evaluation, SATcast outperforms conventional methods on multiple metrics, demonstrating its superior accuracy and robustness. Ablation studies underscore the importance of its multimodal design and the cascade architecture in achieving reliable predictions. Notably, SATcast maintains predictive skill for up to 24 hours, underscoring its potential for operational nowcasting applications.

Convectively induced turbulence (CIT) is a serious aviation hazard and it is challenging to forecast the CIT in the region near convection. Previous studies used reginal model with high resolution or global model with low resolution and selected empirical indices to diagnose the turbulence. In this study, we used The Model for Prediction Across Scales (MPAS) to simulate some cases of CIT reported near Hong Kong. MPAS allows us to use convection-permitting resolution in the interested area while including the global-large scale circulation with coarser resolutions in other regions. The eddy dissipation rate (EDR) is computed to diagnose the potential occurrence of CIT. We compared three methods for calculating EDR from the resolved flow in the MPAS, the first one based on second order structure function, the second one based on Scale-Similarity in Large Eddy Simulation (LES) and the third is Near Cloud Turbulence (NCT) diagnostics by using Convective Gravity Wave Drag. Comparing with the NOAA Graphical Turbulence Guidance (GTG) product and flight data suggests that computing EDR with Scale-Similarity is more effective and accurate than second order structure function and NCT diagnostics. Resolution is also an important factor in forecast, we tested the method in mesh with different resolutions but similar distributions, the results from low resolution simulations can generate a useful turbulence pattern forecast, but the intensity is weak, highlighting the value of high resolution simulations that can resolve convection. We evaluated the sensitivity to several model physics and numeric options in simulations. Those variations can change the EDR prediction by influencing the intensity and the life cycle of the convection. No particular scheme produces systematically more intense turbulence than others, suggesting varying model physics captures some stochasticity of convection. Compared with flight records of EDR along the flight routes, MPAS could produce in three out of five cases showing maximum EDR is close to the observed intensity of turbulence (EDR>0.4). However, in the other two cases, the results are not satisfactory mainly because of significant location biases of the predicted convection. We also add initial condition perturbation-based large ensemble in one case and find it possible to improve the prediction of the failed cases by influencing the position of the convection. Further work should be conducted to prioritize the ensemble members since only a few members can capture the turbulence and doing the average will erase them easily.