skip to main content


Title: Machine learning classification of significant tornadoes and hail in the U.S. using ERA5 proximity soundings
Abstract Previous studies have identified environmental characteristics that skillfully discriminate between severe and significant-severe weather events, but they have largely been limited by sample size and/or population of predictor variables. Given the heightened societal impacts of significant-severe weather, this topic was revisited using over 150 000 ERA5 reanalysis-derived vertical profiles extracted at the grid-point nearest—and just prior to—tornado and hail reports during the period 1996–2019. Profiles were quality-controlled and used to calculate 84 variables. Several machine learning classification algorithms were trained, tested, and cross-validated on these data to assess skill in predicting severe or significant-severe reports for tornadoes and hail. Random forest classification outperformed all tested methods as measured by cross-validated critical success index scores and area under the receiver operating characteristic curve values. In addition, random forest classification was found to be more reliable than other methods and exhibited negligible frequency bias. The top three most important random forest classification variables for tornadoes were wind speed at 500 hPa, wind speed at 850 hPa, and 0–500-m storm-relative helicity. For hail, storm-relative helicity in the 3–6 km and -10 to -30 °C layers, along with 0–6-km bulk wind shear, were found to be most important. A game theoretic approach was used to help explain the output of the random forest classifiers and establish critical feature thresholds for operational nowcasting and forecasting. A use case of spatial applicability of the random forest model is also presented, demonstrating the potential utility for operational forecasting. Overall, this research supports a growing number of weather and climate studies finding admirable skill in random forest classification applications.  more » « less
Award ID(s):
2048770
NSF-PAR ID:
10318032
Author(s) / Creator(s):
; ; ;
Date Published:
Journal Name:
Weather and Forecasting
ISSN:
0882-8156
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Hodographs are valuable sources of pattern recognition in severe convective storm forecasting. Certain shapes are known to discriminate between single cell, multicell, and supercell storm organization. Various derived quantities such as storm-relative helicity (SRH) have been found to predict tornado potential and intensity. Over the years, collective research has established a conceptual model for tornadic hodographs (large and “looping”, with high SRH). However, considerably less attention has been given to constructing a similar conceptual model for hodographs of severe hail. This study explores how hodograph shape may differentiate between the environments of severe hail and tornadoes. While supercells are routinely assumed to carry the potential to produce all hazards, this is not always the case, and we explore why. The Storm Prediction Center (SPC) storm mode dataset is used to assess the environments of 8,958 tornadoes and 7,256 severe hail reports, produced by right- and left-moving supercells. Composite hodographs and indices to quantify wind shear are assessed for each hazard, and clear differences are found between the kinematic environments of hail-producing and tornadic supercells. The sensitivity of the hodograph to common thermodynamic variables was also examined, with buoyancy and moisture found to influence the shape associated with the hazards. The results suggest that differentiating between tornadic and hail-producing storms may be possible using properties of the hodograph alone. While anticipating hail size does not appear possible using only the hodograph, anticipating tornado intensity appears readily so. When coupled with buoyancy profiles, the hodograph may assist in differentiating between both hail size and tornado intensity. 
    more » « less
  2. Abstract A primary goal of the National Oceanic and Atmospheric Administration Warn-on-Forecast (WoF) project is to provide rapidly updating probabilistic guidance to human forecasters for short-term (e.g., 0–3 h) severe weather forecasts. Postprocessing is required to maximize the usefulness of probabilistic guidance from an ensemble of convection-allowing model forecasts. Machine learning (ML) models have become popular methods for postprocessing severe weather guidance since they can leverage numerous variables to discover useful patterns in complex datasets. In this study, we develop and evaluate a series of ML models to produce calibrated, probabilistic severe weather guidance from WoF System (WoFS) output. Our dataset includes WoFS ensemble forecasts available every 5 min out to 150 min of lead time from the 2017–19 NOAA Hazardous Weather Testbed Spring Forecasting Experiments (81 dates). Using a novel ensemble storm-track identification method, we extracted three sets of predictors from the WoFS forecasts: intrastorm state variables, near-storm environment variables, and morphological attributes of the ensemble storm tracks. We then trained random forests, gradient-boosted trees, and logistic regression algorithms to predict which WoFS 30-min ensemble storm tracks will overlap a tornado, severe hail, and/or severe wind report. To provide rigorous baselines against which to evaluate the skill of the ML models, we extracted the ensemble probabilities of hazard-relevant WoFS variables exceeding tuned thresholds from each ensemble storm track. The three ML algorithms discriminated well for all three hazards and produced more reliable probabilities than the baseline predictions. Overall, the results suggest that ML-based postprocessing of dynamical ensemble output can improve short-term, storm-scale severe weather probabilistic guidance. 
    more » « less
  3. Abstract The magnitude of water vapor content within the near-storm inflow can either support or deter the storm’s upscale growth and maintenance. However, the heterogeneity of the moisture field near storms remains poorly understood because the operational observation network lacks detail. This observational study illustrates that near-storm inflow water vapor environments are both significantly heterogeneous and different than the far-inflow storm environment. This study also depicts the importance of temporal variation of water vapor mixing ratio (WVMR) to instability during the peak tornadic seasons in the U.S. Southeast and Great Plains regions during the Verification of the Origins of Rotation in Tornadoes Experiment Southeast 2018 (VSE18) campaign and the Targeted Observation by Radar and UAS of Supercells (TORUS) campaign, respectively. VSE18 results suggest that the surface processes control WVMR variation significantly in lower levels, with the highest WVMR mainly located near the surface in inflows in the southeast region. In contrast, TORUS results show more vertically homogeneous WVMR profiles and rather uniform water vapor distribution variation occurring in deep, moist stratified inflows in the Great Plains region. Temporal water vapor variations within 5-min periods could lead to over 1000 J kg −1 CAPE changes in both VSE18 and TORUS, which represent significant potential buoyancy perturbations for storms to intensify or decay. These temporal water vapor and instability evolutions of moving storms remain difficult to capture via radiosondes and fixed in situ or profiling instrumentation, yet may exert a strong impact on storm evolution. This study suggests that improving observations of the variability of near-storm inflow moisture can accurately refine a potential severe weather threat. Significance Statement It has long been recognized that better observations of the planetary boundary layer (PBL) inflow near convective storms are needed to improve severe weather forecasting. The current operational networks essentially do not provide profile measurements of the PBL, except for the sparsely spaced 12-hourly sounding network. More frequent geostationary satellite observations do not provide adequately high vertical resolution in the PBL. This study uses airborne lidar profiler measurements to examine moisture in the inflow region of convective storms in the Great Plains and the southeastern United States during their respective tornadic seasons. Rapid PBL water vapor variations on a ∼5 min time scale can lead to CAPE perturbations exceeding 1000 J kg −1 , representing significant perturbations that could promote storm intensification or decay. Severe thunderstorms may generate high-impact weather phenomena, such as tornadoes, high winds, hail, and heavy rainfall, which have substantial socioeconomic impacts. Ultimately, by contrasting characteristics of the convective storm inflow in the two regions, this study may lead to a more accurate assessment of severe weather threats. 
    more » « less
  4. Abstract

    Severe storms produce hazardous weather phenomena, such as large hail, damaging winds, and tornadoes. However, relationships between convective parameters and confirmed severe weather occurrences are poorly quantified in south-central Brazil. This study explores severe weather reports and measurements from newly available datasets. Hail, damaging wind, and tornado reports are sourced from the PREVOTS project from June 2018 to December 2021, while measurements of convectively induced wind gusts from 1996 to 2019 are obtained from METAR reports and from Brazil’s operational network of automated weather stations. Proximal convective parameters were computed from ERA5 reanalysis for these reports and used to perform a discriminant analysis using mixed-layer CAPE and deep-layer shear (DLS). Compared to other regions, thermodynamic parameters associated with severe weather episodes exhibit lower magnitudes in south-central Brazil. DLS displays better performance in distinguishing different types of hazardous weather, but does not discriminate well between distinct severity levels. To address the sensitivity of the discriminant analysis to distinct environmental regimes and hazard types, five different discriminants are assessed. These include discriminants for any severe storm, severe hail only, severe wind gust only, and all environments but broken into “high” and “low” CAPE regimes. The best performance of the discriminant analysis is found for the “high” CAPE regime, followed by the severe wind regime. All discriminants demonstrate that DLS plays a more important role in conditioning Brazilian severe storm environments than other regions, confirming the need to ensure that parameters and discriminants are tuned to local severe weather conditions.

     
    more » « less
  5. Abstract

    Supercell thunderstorms develop low-level rotation via tilting of environmental horizontal vorticity (ωh) by the updraft. This rotation induces dynamic lifting that can stretch near-surface vertical vorticity into a tornado. Low-level updraft rotation is generally thought to scale with 0–500 m storm-relative helicity (SRH): the combination of storm-relative flow, |SRF|, |ωh|, and cosϕ(whereϕis the angle betweenSRFandωh). It is unclear how much influence each component of SRH has in intensifying the low-level mesocyclone. This study surveys these three components using self-organizing maps (SOMs) to distill 15 906 proximity soundings for observed right-moving supercells. Statistical analyses reveal the component most highly correlated to SRH and to streamwise vorticity (ωs) in the observed profiles is |ωh|. Furthermore, |ωh| and |SRF| are themselves highly correlated due to their shared dependence on the hodograph length. The representative profiles produced by the SOMs were combined with a common thermodynamic profile to initialize quasi-realistic supercells in a cloud model. The simulations reveal that, across a range of real-world profiles, intense low-level mesocyclones are most closely linked toωhandSRF, while the angle between them appears to be mostly inconsequential.

    Significance Statement

    About three-fourths of all tornadoes are produced by rotating thunderstorms (supercells). When the part of the storm near cloud base (approximately 1 km above the ground) rotates more strongly, the chance of a tornado dramatically increases. The goal of this study is to identify the simplest characteristic(s) of the environmental wind profile that can be used to forecast the likelihood of strong cloud-base rotation. This study concludes that the most important ingredients for storm rotation are the magnitudes of the horizontal vertical wind shear between the surface and 500 m and the storm inflow wind, irrespective of their relative directions. This finding may lead to improved operational identification of environments favoring tornado formation.

     
    more » « less