skip to main content


This content will become publicly available on May 5, 2024

Title: Diagnosing Storm Mode with Deep Learning in Convection-Allowing Models
Abstract While convective storm mode is explicitly depicted in convection-allowing model (CAM) output, subjectively diagnosing mode in large volumes of CAM forecasts can be burdensome. In this work, four machine learning (ML) models were trained to probabilistically classify CAM storms into one of three modes: supercells, quasi-linear convective systems, and disorganized convection. The four ML models included a dense neural network (DNN), logistic regression (LR), a convolutional neural network (CNN) and semi-supervised CNN-Gaussian mixture model (GMM). The DNN, CNN, and LR were trained with a set of hand-labeled CAM storms, while the semi-supervised GMM used updraft helicity and storm size to generate clusters which were then hand labeled. When evaluated using storms withheld from training, the four classifiers had similar ability to discriminate between modes, but the GMM had worse calibration. The DNN and LR had similar objective performance to the CNN, suggesting that CNN-based methods may not be needed for mode classification tasks. The mode classifications from all four classifiers successfully approximated the known climatology of modes in the U.S., including a maximum in supercell occurrence in the U.S. Central Plains. Further, the modes also occurred in environments recognized to support the three different storm morphologies. Finally, storm mode provided useful information about hazard type, e.g., storm reports were most likely with supercells, further supporting the efficacy of the classifiers. Future applications, including the use of objective CAM mode classifications as a novel predictor in ML systems, could potentially lead to improved forecasts of convective hazards.  more » « less
Award ID(s):
2019758
NSF-PAR ID:
10422713
Author(s) / Creator(s):
; ; ; ; ;
Date Published:
Journal Name:
Monthly Weather Review
ISSN:
0027-0644
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Convection intensity and longevity is highly dependent on the surrounding environment. Ensemble sensitivity analysis (ESA), which quantitatively and qualitatively interprets impacts of initial conditions on forecasts, is applied to very short-term (1–2 h) convective-scale forecasts for three cases during the Mesoscale Predictability Experiment (MPEX) in 2013. The ESA technique reveals several dependencies of individual convective storm evolution on their nearby environments. The three MPEX cases are simulated using a previously verified 36-member convection-allowing model (Δ x = 3 km) ensemble created via the Weather Research and Forecasting (WRF) Model. Radar and other conventional observations are assimilated using an ensemble adjustment Kalman filter. The three cases include a mesoscale convective system (MCS) and both nontornadic and tornadic supercells. Of the many ESAs applied in this study, one of the most notable is the positive sensitivity of supercell updraft helicity to increases in both storm inflow region deep and shallow vertical wind shear. This result suggests that larger values of vertical wind shear within the storm inflow yield higher values of storm updraft helicity. Results further show that the supercell storms quickly enhance the environmental vertical wind shear within the storm inflow region. Application of ESA shows that these storm-induced perturbations then affect further storm evolution, suggesting the presence of storm–environment feedback cycles where perturbations affect future mesocyclone strength. Overall, ESA can provide insight into convection dependencies on the near-storm environment.

     
    more » « less
  2. Abstract The importance of discriminating between environments supportive of supercell thunderstorms and those that are not supportive is widely recognized due to significant hazards associated with supercell storms. Previous research has led to forecast indices such as the energy helicity index and the supercell composite parameter to aid supercell forecasts. In this study three machine learning models are developed to identify environments supportive of supercells: a support vector machine, an artificial neural network, and an ensemble of gradient boosted trees. These models are trained and tested using a sample of over 1000 Rapid Update Cycle version 2 (RUC-2) model soundings from near-storm environments of both supercell and nonsupercell storms. Results show that all three machine learning models outperform classifications using either the energy helicity index or supercell composite parameter by a statistically significant margin. Using several model interpretability methods, it is concluded that generally speaking the relationships learned by the machine learning models are physically reasonable. These findings further illustrate the potential utility of machine learning–based forecast tools for severe storm forecasting. Significance Statement Supercell thunderstorms are a type of thunderstorm that are important to forecast because they produce more tornadoes, hail, and wind gusts compared to other types of thunderstorms. This study uses machine learning to create models that predict if a supercell thunderstorm or nonsupercell thunderstorm is favored for a given environment. These models outperform current methods of assessing if a storm that forms will be a supercell. Using these models as guidance forecasters can better understand and predict if atmospheric conditions are favorable for the development of supercell thunderstorms. Improving forecasts of supercell thunderstorms using machine learning methods like those used in this study has the potential to limit the economic and societal impacts of these storms. 
    more » « less
  3. Abstract Observational data collection is extremely hazardous in supercell storm environments, which makes for a scarcity of data used for evaluating the storm-scale guidance from convection allowing models (CAMs) like the National Oceanic and Atmospheric Administration (NOAA) Warn-on-Forecast System (WoFS). The Targeted Observations with UAS and Radar of Supercells (TORUS) 2019 field mission provided a rare opportunity to not only collect these observations, but to do so with advanced technology: vertically pointing Doppler lidar. One standing question for WoFS is how the system forecasts the feedback between supercells and their near-storm environment. The lidar can observe vertical profiles of wind over time, creating unique datasets to compare to WoFS kinematic predictions in rapidly evolving severe weather environments. Mobile radiosonde data are also presented to provide a thermodynamic comparison. The five lidar deployments (three of which observed tornadic supercells) analyzed show WoFS accurately predicted general kinematic trends in the inflow environment; however, the predicted feedback between the supercell and its environment, which resulted in enhanced inflow and larger storm-relative helicity (SRH), were muted relative to observations. The radiosonde observations reveal an overprediction of CAPE in WoFS forecasts, both in the near and far field, with an inverse relationship between the CAPE errors and distance from the storm. Significance Statement It is difficult to evaluate the accuracy of weather prediction model forecasts of severe thunderstorms because observations are rarely available near the storms. However, the TORUS 2019 field experiment collected multiple specialized observations in the near-storm environment of supercells, which are compared to the same near-storm environments predicted by the National Oceanic and Atmospheric Administration (NOAA) Warn-on-Forecast System (WoFS) to gauge its performance. Unique to this study is the use of mobile Doppler lidar observations in the evaluation; lidar can retrieve the horizontal winds in the few kilometers above ground on time scales of a few minutes. Using lidar and radiosonde observations in the near-storm environment of three tornadic supercells, we find that WoFS generally predicts the expected trends in the evolution of the near-storm wind profile, but the response is muted compared to observations. We also find an inverse relationship of errors in instability to distance from the storm. These results can aid model developers in refining model physics to better predict severe storms. 
    more » « less
  4. Abstract

    The severity, duration, and spatial extent of thunderstorm impacts are related to convective storm mode. This study assesses the skill of the High-Resolution Rapid Refresh Ensemble (HRRR-E) and its deterministic counterpart (HRRRv4) at predicting convective mode and storm macrophysical properties using 35 convective events observed during the 2020 warm season across the eastern United States. Seven cases were selected from each of five subjectively determined convective organization modes: tropical cyclones, mesoscale convective systems (MCSs), quasi-linear convective systems, clusters, and cellular convection. These storm events were assessed using an object-based approach to identify convective storms and determine their individual size. Averaged across all 35 cases, both the HRRR-E and HRRRv4 predicted storm areas were generally larger than observed, with this bias being a function of storm lifetime and convective mode. Both modeling systems also underpredicted the rapid increase in storm counts during the initiation period, particularly for the smaller-scale storm modes. Interestingly, performance of the HRRRv4 differed from that of the HRRR-E, with the HRRRv4 generally having a larger bias in total storm area than the HRRR-E due to HRRRv4 predicting up to 66% more storm objects than the HRRR-E. The HRRR-E accurately predicted the convective mode 65% of the time, with complete misses being very rare (<5% of the time overall). However, an evaluation of rank histograms across all 35 cases revealed that the HRRR-E tended to be underdispersive when predicting storm size for all but the MCS mode.

     
    more » « less
  5. null (Ed.)
    Abstract During the Remote Sensing of Electrification, Lightning, and Mesoscale/Microscale Processes with Adaptive Ground Observations-Cloud, Aerosol, and Complex Terrain Interactions (RELAMPAGO-CACTI) field experiments in 2018–19, an unprecedented number of balloon-borne soundings were collected in Argentina. Radiosondes were launched from both fixed and mobile platforms, yielding 2712 soundings during the period 15 October 2018–30 April 2019. Approximately 20% of these soundings were collected by highly mobile platforms, strategically positioned for each intensive observing period, and launching approximately once per hour. The combination of fixed and mobile soundings capture both the overall conditions characterizing the RELAMPAGO-CACTI campaign, as well as the detailed evolution of environments supporting the initiation and upscale growth of deep convective storms, including some that produced hazardous hail and heavy rainfall. Episodes of frequent convection were characterized by sufficient quantities of moisture and instability for deep convection, along with deep-layer vertical wind shear supportive of organized or rotating storms. A total of 11 soundings showed most unstable convective available potential energy (MUCAPE) exceeding 6000 J kg −1 , comparable to the extreme instability observed in other parts of the world with intense deep convection. Parameters used to diagnose severe-storm potential showed that conditions were often favorable for supercells and severe hail, but not for tornadoes, primarily because of insufficient low-level wind shear. High-frequency soundings also revealed the structure and evolution of the boundary layer leading up to convection initiation, convectively generated cold pools, the South American low-level jet (SALLJ), and elevated nocturnal convection. This sounding dataset will enable improved understanding and prediction of convective storms and their surroundings in subtropical South America, as well as comparisons with other heavily studied regions such as the central United States that have not previously been possible. 
    more » « less