Search for: All records

Accurate weather forecasting is critical for science and society. However, existing methods have not achieved the combination of high accuracy, low uncertainty, and high computational efficiency simultaneously. On one hand, traditional numerical weather prediction (NWP) models are computationally intensive because of their complexity. On the other hand, most machine learning-based weather prediction (MLWP) approaches offer efficiency and accuracy but remain deterministic, lacking the ability to capture forecast uncertainty. To tackle these challenges, we propose a conditional diffusion model, CoDiCast, to generate global weather prediction, integrating accuracy and uncertainty quantification at a modest computational cost. The key idea behind the prediction task is to generate realistic weather scenarios at a future time point, conditioned on observations from the recent past. Due to the probabilistic nature of diffusion models, they can be properly applied to capture the uncertainty of weather predictions. Therefore, we accomplish uncertainty quantifications by repeatedly sampling from stochastic Gaussian noise for each initial weather state and running the denoising process multiple times. Experimental results demonstrate that CoDiCast outperforms several existing MLWP methods in accuracy, and is faster than NWP models in inference speed. Our model can generate 6-day global weather forecasts, at 6-hour steps and 5.625-degree latitude-longitude resolutions, for over 5 variables, in about 12 minutes on a commodity A100 GPU machine with 80GB memory. The source code is available at https://github.com/JimengShi/CoDiCast. 

Abstract. The fundamental mechanism underlying tropical cyclone (TC) intensification may be understood from the conservation of absolute angular momentum, where the primary circulation of a TC is driven by the torque acting on air parcels resulting from asymmetric eddy processes, including turbulence. While turbulence is commonly regarded as a flow feature pertaining to the planetary boundary layer (PBL), intense turbulent mixing generated by cloud processes also exists above the PBL in the eyewall and rainbands. Unlike the eddy forcing within the PBL that is negative definite, the sign of eyewall/rainband eddy forcing above the PBL is indefinite and thus provides a possible mechanism to spin up a TC vortex. In this study, we show that the Hurricane Weather Research & forecasting (HWRF) model, one of the operational models used for TC prediction, is unable to generate appropriate sub-grid-scale (SGS) eddy forcing above the PBL due to lack of consideration of intense turbulent mixing generated by the eyewall and rainband clouds. Incorporating an in-cloud turbulent mixing parameterization in the PBL scheme notably improves HWRF's skills on predicting rapid changes in intensity for several past major hurricanes. While the analyses show that the SGS eddy forcing above the PBL is only about one-fifth of the model-resolved eddy forcing, the simulated TC vortex inner-core structure and the associated model-resolved eddy forcing exhibit a substantial dependence on the parameterized SGS eddy processes. The results highlight the importance of eyewall/rainband SGS eddy forcing to numerical prediction of TC intensification, including rapid intensification at the current resolution of operational models. 

The FV3GFS is the current operational Global Forecast System (GFS) at the National Centers for Environmental Prediction (NCEP), which combines a finite-volume cubed sphere dynamical core (FV3) and GFS physics. In this study, FV3GFS is used to gain understanding of rapid intensification (RI) of tropical cyclones (TCs) in shear. The analysis demonstrates the importance of TC structure in a complex system like Hurricane Michael, which intensified to a category 5 hurricane over the Gulf of Mexico despite over 20 kt (10 m s⁻¹) of vertical wind shear. Michael’s RI is examined using a global-nest FV3GFS ensemble with the nest at 3-km resolution. The ensemble shows a range of peak intensities from 77 to 159 kt (40–82 m s⁻¹). Precipitation symmetry, vortex tilt, moisture, and other aspects of Michael’s evolution are compared through composites of stronger and weaker members. The 850–200-hPa vertical shear is 22 kt (11 m s⁻¹) in the mean of both strong and weak members during the early stage. Tilt and moisture are two distinguishing factors between strong and weak members. The relationship between vortex tilt and humidification is complex, and other studies have shown both are important for sheared intensification. Here, it is shown that tilt reduction leads to upshear humidification and is thus a driving factor for intensification. A stronger initial vortex and early evolution of the vortex also appear to be the key to members that are able to resist the sheared environment.