Search for: All records

Abstract The development of deep learning (DL) weather forecasting models has made rapid progress and achieved comparable or better skill than traditional Numerical Weather prediction (NWP) models, which are generally computationally intensive. However, applications of these DL models have yet to be fully explored, including for severe convective events. We evaluate the DL model Pangu‐Weather in forecasting tornadic environments with one‐day lead times using convective available potential energy (CAPE), 0–6 bulk wind difference (BWD6), and 0–3 km storm‐relative helicity (SRH3). We also compare its performance to the National Centers for Environmental Prediction (NCEP)'s Global Forecast System (GFS), a traditional NWP model. Pangu‐Weather generally outperforms GFS in predicting BWD6 and SRH3 at the closest grid point and hour of the storm report. However, Pangu‐Weather tends to underpredict the maximum values of all convective parameters in the 1–2 hr before the storm across the surrounding grid points compared to the GFS. 

Abstract Improving the skill of medium-range (3–8 day) severe weather prediction is crucial for mitigating societal impacts. This study introduces a novel approach leveraging decoder-only transformer networks to post-process AI-based weather forecasts, specifically from the Pangu-Weather model, for improved severe weather guidance. Unlike traditional post-processing methods that use a dense neural network to predict the probability of severe weather using discrete forecast samples, our method treats forecast lead times as sequential “tokens”, enabling the transformer to learn complex temporal relationships within the evolving atmospheric state. We compare this approach against post-processing of the Global Forecast System (GFS) using both a traditional dense neural network and our transformer, as well as configurations that exclude convective parameters to fairly evaluate the impact of using the Pangu-Weather AI model. Results demonstrate that the transformer-based post-processing significantly enhances forecast skill compared to dense neural networks. Furthermore, AI-driven forecasts, particularly Pangu-Weather initialized from high resolution analysis, exhibit superior performance to GFS in the medium-range, even without explicit convective parameters. Our approach offers improved accuracy, and reliability, which also provides interpretability through feature attribution analysis, advancing medium-range severe weather prediction capabilities. 

Abstract This study explores how tornadic supercell soundings significantly differ from the same‐location and same‐hour baseline environment soundings, sampled from the days prior to or following the event. Permutation testing is used to identify whether sounding‐derived parameters mixed‐layer convective available potential energy and 0–1 km storm‐relative helicity are significantly different between the tornadic and baseline environment. Typically, in an environment with marginal values of certain key environmental parameters, anomalous values of those environmental parameters are more strongly associated with supercell tornadoes. Furthermore, many tornadic events already exhibit environmental conditions favorable for tornadic supercells a day prior to the event itself. Generally, supercell tornadoes that occur during typical peak tornadic activity time frames are easier to distinguish from baseline (non‐tornadic) environments compared to those occurring in other time frames. Spatiotemporal variations of distinguishability between tornadic and baseline environmental parameters add complexity to traditional parameter‐based fixed threshold forecasting. 

Recent research suggests that surface elevation variability may influence tornado activity, though separating this effect from reporting biases is difficult to do in observations. Here we employ Bayes’s law to calculate the empirical joint dependence of tornado probability on population density and elevation roughness in the vicinity of Arkansas for the period 1955–2015. This approach is based purely on data, exploits elevation and population information explicitly in the vicinity of each tornado, and enables an explicit test of the dependence of results on elevation roughness length scale. A simple log-link linear regression fit to this empirical distribution yields an 11% decrease in tornado probability per 10-m increase in elevation roughness at fixed population density for large elevation roughness length scales (15–20 km). This effect increases by at least a factor of 2 moving toward smaller length scales down to 1 km. The elevation effect exhibits no time trend, while the population bias effect decreases systematically in time, consistent with the improvement of reporting practices. Results are robust across time periods and the exclusion of EF1 tornadoes and are consistent with recent county-level and gridded analyses. This work highlights the need for a deeper physical understanding of how elevation heterogeneity affects tornadogenesis and also provides the foundation for a general Bayesian tornado probability model that integrates both meteorological and nonmeteorological parameters.