Convective Hazards Research Articles

Evaluating Machine Learning–Based Probabilistic Convective Hazard Forecasts Using The HRRR: Quantifying Hazard Predictability and Sensitivity to Training Choices

Abstract The High-Resolution Rapid Refresh (HRRR) model provides hourly updating forecasts of convective-scale phenomena, which can be used to infer the potential for convective hazards (e.g., tornadoes, hail, and wind gusts), across the United States. We used deterministic 2019–20 HRRR, version 4 (HRRRv4), forecasts to train neural networks (NNs) to generate 4-hourly probabilistic convective hazard forecasts [neural network probability forecasts (NNPFs)] for HRRRv4 initializations in 2021, using storm reports as ground truth. The NNPFs were compared to the skill of a smoothed updraft helicity (UH) baseline to quantify the benefit of the NNs. NNPF skill varied by initialization time and time of day but was all superior to the UH forecast. NNPFs valid at hours between 1800 and 0000 UTC were most skillful in aggregate, significantly exceeding the baseline forecast skill. Overnight NNPFs (i.e., valid 0600–1200 UTC) were least skillful, indicating a diurnal cycle in hazard predictability that was present across all HRRRv4 initializations. We explored the sensitivity of HRRRv4 NNPF skill to NN training choices. Including an additional year of 2021 HRRRv4 forecasts for training slightly improved skill for 2022 HRRRv4 NNPFs, while reducing the training dataset size by 40% using only forecasts with storm reports was not detrimental to forecast skill. Finally, NNs trained with 2018–20 HRRRv3 forecasts led to a reduction in NNPF skill when applied to 2021 HRRRv4 forecasts. In addition to documenting practical predictability challenges with convective hazard prediction, these findings reinforce the need for a consistent model configuration for optimal results when training NNs and provide best practices when constructing a training dataset with operational convection-allowing model forecasts. Significance Statement Convective hazards, such as hail and tornadoes, are often challenging to predict. To improve hazard predictions, we used machine learning (ML) to generate forecasts of convective hazards across the United States leveraging forecasts of prior events. The ML hazard forecasts were consistently better than a non-ML approach and varied in skill based on the time of day, with nighttime forecasts being particularly challenging. ML forecasts of wind gusts and hail were more skillful than tornadoes. Different strategies of constructing the dataset of prior events led to differences in forecast performance; thus, this work provides recommendations for how to assemble these datasets and train ML models to generate improved forecasts of severe weather events.

Radar Outage Costs and the Value of Alternate Datasets

Abstract Quantifying the costs of radar outages allows value to be attributed to the alternate datasets that help mitigate outages. When radars are offline, forecasters rely more heavily on nearby radars, surface reports, numerical weather prediction models, and satellite observations. Monetized radar benefit models allow value to be attributed to individual radars for mitigating the threat to life from tornadoes, flash floods, and severe winds. Eighteen radars exceed $20 million in annual benefits for mitigating the threat to life from these convective hazards. The Jackson, Mississippi, radar (KDGX) provides the most value ($41.4 million), with the vast majority related to tornado risk mitigation ($29.4 million). During 2020–23, the average radar is offline for 2.57% of minutes or 9.27 days per year and experiences an average of 58.9 outages per year lasting 4.32 h on average. Radar outage cost estimates vary by location and convective hazard. Outage cost estimates concentrate at the top, with 8, 2, 4, and 5 radars exceeding $1 million in outage costs during 2020, 2021, 2022, and 2023, respectively. The KDGX radar experiences outage frequencies of 4.92% and 5.50% during 2020 and 2023, resulting in outage cost estimates > $2 million in both years. Combining outage cost estimates for all radars suggests that approximately $29.1 million in annual radar outage costs may be attributable as value to alternative datasets for helping mitigate radar outage impacts. Significance Statement This study combines information on radar status and monetized radar benefit models to attribute value to individual radars, estimate radar outage costs, and quantify the potential value of alternative datasets during outage-induced gaps in coverage. Eighteen radars exceed $20 million in annual benefits for mitigating the combined threat to life from tornadoes, flash floods, and severe winds. The first and third most valuable radars, both in Mississippi, experience outage frequencies twice the national average, accounting for a disproportionate share of the overall outage costs. Our findings suggest that characterizing and mitigating these outages might provide a near-term solution to better protect these communities from convective hazards. Combining outage cost estimates for all radars suggests that approximately $29.1 million in annual radar outage costs may be attributable as value to alternative datasets for helping mitigate the impacts of radar outages.

Generative Ensemble Deep Learning Severe Weather Prediction from a Deterministic Convection-Allowing Model

Abstract An ensemble postprocessing method is developed for the probabilistic prediction of severe weather (tornadoes, hail, and wind gusts) over the conterminous United States (CONUS). The method combines conditional generative adversarial networks (CGANs), a type of deep generative model, with a convolutional neural network (CNN) to postprocess convection-allowing model (CAM) forecasts. The CGANs are designed to create synthetic ensemble members from deterministic CAM forecasts, and their outputs are processed by the CNN to estimate the probability of severe weather. The method is tested using High-Resolution Rapid Refresh (HRRR) 1–24-h forecasts as inputs and Storm Prediction Center (SPC) severe weather reports as targets. The method produced skillful predictions with up to 20% Brier skill score (BSS) increases compared to other neural-network-based reference methods using a testing dataset of HRRR forecasts in 2021. For the evaluation of uncertainty quantification, the method is overconfident but produces meaningful ensemble spreads that can distinguish good and bad forecasts. The quality of CGAN outputs is also evaluated. Results show that the CGAN outputs behave similarly to a numerical ensemble; they preserved the intervariable correlations and the contribution of influential predictors as in the original HRRR forecasts. This work provides a novel approach to postprocess CAM output using neural networks that can be applied to severe weather prediction. Significance Statement We use a new machine learning (ML) technique to generate probabilistic forecasts of convective weather hazards, such as tornadoes and hailstorms, with the output from high-resolution numerical weather model forecasts. The new ML system generates an ensemble of synthetic forecast fields from a single forecast, which are then used to train ML models for convective hazard prediction. Using this ML-generated ensemble for training leads to improvements of 10%–20% in severe weather forecast skills compared to using other ML algorithms that use only output from the single forecast. This work is unique in that it explores the use of ML methods for producing synthetic forecasts of convective storm events and using these to train ML systems for high-impact convective weather prediction.

Major Role of Marine Heatwave and Anthropogenic Climate Change on a Giant Hail Event in Spain

AbstractA severe hailstorm that occurred in Spain on 30 August 2022, caused material and human damage, including one fatality due to giant hailstones up to 12 cm in diameter. By applying a pseudo‐global warming approach, here we evaluate how a simultaneous marine heatwave (and anthropogenic climate change) affected a unique environment conductive to such giant hailstones. The main results show that the supercell development was influenced by an unprecedented amount of convective available energy, with significant contributions from thermodynamic factors. Numerical simulations where the marine heatwave is not present show a notable reduction in the hail‐favorable environments, related mainly to modifications in thermodynamic environment. Our simulations also indicate that the environment in a preindustrial‐like climate would be less favorable for convective hazards and thus the hailstorm event would likely not have been as severe as the observed one, being possible to perform a novel attribution of such kind.

Atmospheric environments associated with tornadoes in southern Brazil and neighboring areas as compared to other modes of convective hazards

Atmospheric environments associated with tornadoes in southern Brazil and neighboring areas as compared to other modes of convective hazards

The climatology and nature of warm-season convective cells in cold-frontal environments over Germany

Abstract. Cold fronts provide an environment particularly favourable for convective initiation in the mid-latitudes and can also be associated with convective hazards such as flooding, wind, hail and lightning. We build a climatology of cold-frontal convective cells between 2007–2016 for April–September in a cell-front distance framework by combining a radar-based cell detection and tracking dataset and automatic front detection methods applied to reanalysis data. We find that on average around twice as many cells develop on cold-frontal cell days compared to non-cold-frontal cell days. Using the 700 hPa level as a reference point, we show the maximum cell frequency is 350–400 km ahead of the 700 hPa front, which is marginally ahead of the typical surface front location. The 700 hPa front location marks the minimum cell frequency and a clear shift in regime between cells, with a weakened diurnal cycle on the warm side of the 700 hPa cold front and strongly diurnally driven cells on the cold side of the 700 hPa front. High cell frequency is found several hundreds of kilometres ahead of the surface front, and cells in this region are most likely to be associated with mesocyclones, intense convective cores and lightning. Namely, mesocyclones were detected in around 5.0 % of pre-surface-frontal cells compared to only 1.5 % of non-cold-frontal cells. The findings in this study are an important step towards a better understanding of cold-frontal convection climatology and links between cold fronts and convective hazards.

Biases and Skill of Four Two-Moment Bulk Microphysics Schemes in Convection-Allowing Forecasts for the 2018 Hazardous Weather Testbed Spring Forecasting Experiment Period

Abstract A proof-of-concept systematic evaluation of convective hazards is applied to short-term (1–6 h) forecasts using the Morrison, National Severe Storms Laboratory (NSSL), Predicted Particle Properties (P3), and Thompson two-moment microphysics schemes for the 2018 NOAA Hazardous Weather Testbed Spring Forecasting Experiment (HWT SFE) period (hereafter “MORR,” “NSSL,” “P3,” and “THOM” experiments, respectively). Four convective line cases are highlighted to elaborate on relative experiment biases/skill. Composite reflectivity and 1-h accumulated precipitation are examined to determine storm coverage/precipitation biases/skill utilizing point-based verification with a neighborhood. Simulated 1–6-km updraft helicity and observed 3–6-km azimuthal shear and MESH are examined to consider simulated rotation and hail core prediction with object-based scores. Over the full season, MORR displays little overall storm coverage bias relative to NSSL, P3, and THOM underprediction. The equitable threat score (ETS) and fractions skill score (FSS) of P3 are lower than the other experiments. P3 and THOM underpredict convective regions with intense reflectivity relative to MORR and NSSL overprediction. All experiments underpredict precipitation amounts. P3 light precipitation FSS is lower than other experiments. Rotation object verification exhibits sensitivity to microphysics experiments, as microphysics has an indirect influence on storm dynamics. While P3 has the largest hail object underprediction, all experiments grossly overpredict the number of hail objects in convective line cases despite forecast objects defined with the same product (MESH) and threshold as observations. The importance of microphysics ice parameterization and ongoing scheme updates highlight the need to apply this verification framework to optimal/updated schemes before optimizing ensemble design.

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 semisupervised CNN–Gaussian mixture model (GMM). The DNN, CNN, and LR were trained with a set of hand-labeled CAM storms, while the semisupervised 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 United States, 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. Significance Statement Whether a thunderstorm produces hazards such as tornadoes, hail, or intense wind gusts is in part determined by whether the storm takes the form of a single cell or a line. Numerical forecasting models can now provide forecasts that depict this structure. We tested several automated algorithms to extract this information from forecast output using machine learning. All of the automated methods were able to distinguish between a set of three convective types, with the simple techniques providing similarly skilled classifications compared to the complex approaches. The automated classifications also successfully discriminated between thunderstorm hazards, potentially leading to new forecast tools and better forecasts of high-impact convective hazards.

A Review of Machine Learning for Convective Weather

Abstract We present an overview of recent work on using artificial intelligence (AI)/machine learning (ML) techniques for forecasting convective weather and its associated hazards, including tornadoes, hail, wind, and lightning. These high-impact phenomena globally cause both massive property damage and loss of life, yet they are very challenging to forecast. Given the recent explosion in developing ML techniques across the weather spectrum and the fact that the skillful prediction of convective weather has immediate societal benefits, we present a thorough review of the current state of the art in AI and ML techniques for convective hazards. Our review includes both traditional approaches, including support vector machines and decision trees, as well as deep learning approaches. We highlight the challenges in developing ML approaches to forecast these phenomena across a variety of spatial and temporal scales. We end with a discussion of promising areas of future work for ML for convective weather, including a discussion of the need to create trustworthy AI forecasts that can be used for forecasters in real time and the need for active cross-sector collaboration on testbeds to validate ML methods in operational situations. Significance Statement We provide an overview of recent machine learning research in predicting hazards from thunderstorms, specifically looking at lightning, wind, hail, and tornadoes. These hazards kill people worldwide and also destroy property and livestock. Improving the prediction of these events in both the local space as well as globally can save lives and property. By providing this review, we aim to spur additional research into developing machine learning approaches for convective hazard prediction.

Thunderstorm and hailstorm environments in Mexico

AbstractThunderstorm and hailstorm environments from 1981 to 2018 are analysed using the proximity soundings approach based on the European Centre for Medium‐Range Weather Forecasts Reanalysis v5 (ERA5) reanalysis dataset. Given the nature of the severe weather dataset, four proximity soundings 0000, 0600, 1200, and 1800 local time were retrieved per day with hail activity or thunderstorms. After quality control, a total of 30,760 vertical profiles were used and grouped seasonally and spatially to consider the local variation of the convective environments. Diverse instability, kinematic, and composite parameters were calculated to characterize the environments for these two convective hazards. The results indicate differences mainly modulated by orographic features, continentality, regional climate, and general synoptic characteristics. Higher instability environments of thunderstorms and hailstorms are observed in the subregions near moisture sources such as oceans, principally during summer and autumn when tropical activity occurs. High wind shear environments are concentrated during winter and spring, mainly in central and northern regions of the country, associated with the passage of frontal systems. The covariates are reasonably accurate at predicting severe weather, but their prediction skill decreases during the warm season when severe weather is more common in Mexico. Increased knowledge of severe weather and associated convective hazards will improve weather forecasting and may be helpful in disaster risk reduction due to meteorological hazards in Mexico.

Initial Condition Convection-Allowing Ensembles with Large Membership for Probabilistic Prediction of Convective Hazards

Abstract Convection-allowing model (CAM) ensemble forecasts provide quantitative probabilistic guidance of convective hazards that forecasters would otherwise qualitatively assess. Various initial condition (IC) strategies can be used to generate CAM probabilistic forecasts, but it is still unclear how different configurations perform. Schwartz et al. verified five 10-member IC CAM ensembles over one month of 0000 UTC initializations with a focus on precipitation. Here, we initialize four 42-member IC CAM ensembles every 12 h over 6 weeks and verify forecasts of precipitation, column maximum reflectivity, and hourly maximum updraft helicity. The Texas Tech University real-time EnKF ensemble and three additional IC ensemble modeling systems are verified. Holding the model configuration constant, additional ICs are generated by downscaling time-lagged Global Ensemble Forecast System (GEFS) members, applying correlated random noise to Global Forecast System (GFS) analyses, and recentering EnKF perturbations about GFS analyses. We found that ensemble ICs constructed with correlated random noise and EnKF perturbations about GFS analyses both produced higher-quality precipitation forecasts than downscaled GEFS and EnKF strategies. However, downscaled GEFS and EnKF perturbations about GFS analyses frequently initialized more skillful forecasts of reflectivity than ICs with random perturbations, suggesting that flow-dependent perturbations are important for forecasting deep convection. Even with a suboptimal EnKF configuration, our findings still echo those of Schwartz et al. We extend their work by exploring 1) verification of additional convective hazards and 2) empirical scaling of IC perturbations as a computationally inexpensive method for improving CAM ensemble forecasts.

Исследование мезомасштабной конвективной системы в центральных районах ЕТР 7 августа 2021 года

The article provides a case study of the formation of a severe convective storm as part of a quasilinear mesoscale convective system on August 7, 2021 that moved over the territory of the central part of European Russia. The predicted fields of meteorological characteristics and the following data of remote sensing of the atmosphere were analyzed: radar, satellite, lightning detection and atmospheric electric field data. A number of features were revealed in the distribution of meteorological parameters, which are useful for nowcasting of hazards associated with the formation of stable severe convective structures. Keywords: mesoscale convective system, severe convective storm, satellite and lightning detection signatures, instability indices, convective hazards, nowcasting, atmospheric electric field

An eight-year climatology of the warm-season severe thunderstorm environments over North China

Using proximity soundings taken during the warm seasons (May–September) of 2011–2018 over North China, this work investigated the environments of severe thunderstorms that caused the following four types of convective hazards: only hail (H), only thunderstorm high winds (W), only short-duration heavy rainfall (R), and two or three types of the above three hazards (hybrid). The environments for non-severe thunderstorms and no-rain days were also studied for comparison. Soundings in each weather category were further classified into mountain and plain groups. A set of parameters was tested for skill in discriminating between various categories. Results indicate that the environments over the plains, compared with those over the mountains, are generally characterized by stronger instability, more moisture, and higher vertical wind shear. Over both mountains and plains, the precipitable water (PW) shows the greatest skill overall in discriminating between various categories; the instability parameters show considerable skill in discriminating between the severe thunderstorm, non-severe thunderstorm, and no-rain categories but rather limited skill in discriminating between the four severe thunderstorm categories; the wind shear parameters show very limited skill in discriminating between various categories; the lifting condensation level (LCL) distinguishes well the environments of no-rain and R from those of other categories; and the mid-tropospheric pseudo-equivalent potential temperature is the only parameter investigated here that shows utility in discriminating between H and W environments. Probability distributions in two-parameter spaces reveal that great variations exist regarding the environments under which different weather types are most likely to occur.

Using Conditional Generative Adversarial 3-D Convolutional Neural Network for Precise Radar Extrapolation

Radar echo extrapolation is a basic but essential task in meteorological services. It could provide radar echo prediction results with high spatiotemporal resolution in a computationally efficient way, and effectively enhance the operational system's forecasting capability for meteorological hazards. Traditional methods perform extrapolation by estimating echo motions between contiguous radar data. This strategy is difficult to characterize complex nonlinear meteorological processes effectively, and it is difficult to benefit from large historical data. Recently, machine learning (ML) models have been used for radar echo extrapolation. These methods have effectively improved extrapolation quality in a data-driven way and from the statistical perspective. Although the ML-based methods show excellent performance, they usually produce blurry extrapolations. This leads to underestimating radar echo intensity and making echo lack small-scale details. Moreover, it makes models difficult to predict severe convective hazards. To solve this problem, a two-stage extrapolation model based on 3-D convolutional neural network and conditional generative adversarial network is proposed. These two models form the “pre-extrapolation” and “postprocessing” paradigm. The pre-extrapolation model is trained in the traditional way and performs rough extrapolation. The postprocessing model uses the pre-extrapolation result as input and is trained with the adversarial strategy. It could correct the echo intensity and increase the echo's details. In the experiment, our model could provide more precise radar echo extrapolations than other methods, especially for intense echoes and convective systems, in the data of North China from 2015 to 2016.

Statistical Characteristics and Environmental Conditions of the Warm-Season Severe Convective Events over North China

Based on severe weather reports, surface precipitation observations, surface routine observations, and the European Center for Medium-Range Weather Forecasts ERA5 reanalysis dataset during the warm seasons (May–September) of 2011–2018 over North China, this paper analyzes the statistical characteristics and environmental conditions of three types of severe convective events. Results are compared between events with different altitudes (i.e., mountains and plains), severities (i.e., ordinary and significant), and months. Hail and thunderstorm high winds (THWs) are more common over the mountains whereas short-duration heavy rainfall (SDHR) is more frequent over the plains. The occurrence frequency of severe convective events exhibits distinct monthly and diurnal variations. Analyses of the environmental parameters provide reference for the potential forecasting of severe convective events over this region. Specifically, the 850–500 hPa temperature lapse rate (LR85), pseudo-equivalent potential temperature at 500 hPa (thetase500), and precipitable water (PW) are skillful in distinguishing hail and THW environments from SDHR environments, and thetase500 is useful in discriminating between hail and THW environments. The convective environments over the plains are characterized by significantly higher (lower) PW (LR85) compared with mountains. The skill of these parameters in forecasting the severity of the convective hazards is limited. Probability distributions in the two parameters space indicate that the occurrence of significant hail requires both higher most unstable convective available potential energy (MUCAPE) and stronger 0–6 km bulk wind shear (SHR6) compared with ordinary hail. Compared with ordinary THWs, the significant THWs over the mountains depend more on the SHR6 whereas those over the plains rely more on the MUCAPE. The significant SDHR events over the plains tend to occur under a variety of instability conditions. The thermodynamic parameters (i.e., MUCAPE, thetase500, and downdraft convective available potential energy), and PW are significantly higher in July–August, whereas the LR85 and vertical wind shear are apparently higher in May, June and September.

Severe Convective Storms across Europe and the United States. Part I: Climatology of Lightning, Large Hail, Severe Wind, and Tornadoes

AbstractAs lightning-detection records lengthen and the efficiency of severe weather reporting increases, more accurate climatologies of convective hazards can be constructed. In this study we aggregate flashes from the National Lightning Detection Network (NLDN) and Arrival Time Difference long-range lightning detection network (ATDnet) with severe weather reports from the European Severe Weather Database (ESWD) and Storm Prediction Center (SPC) Storm Data on a common grid of 0.25° and 1-h steps. Each year approximately 75–200 thunderstorm hours occur over the southwestern, central, and eastern United States, with a peak over Florida (200–250 h). The activity over the majority of Europe ranges from 15 to 100 h, with peaks over Italy and mountains (Pyrenees, Alps, Carpathians, Dinaric Alps; 100–150 h). The highest convective activity over continental Europe occurs during summer and over the Mediterranean during autumn. The United States peak for tornadoes and large hail reports is in spring, preceding the maximum of lightning and severe wind reports by 1–2 months. Convective hazards occur typically in the late afternoon, with the exception of the Midwest and Great Plains, where mesoscale convective systems shift the peak lightning threat to the night. The severe wind threat is delayed by 1–2 h compared to hail and tornadoes. The fraction of nocturnal lightning over land ranges from 15% to 30% with the lowest values observed over Florida and mountains (~10%). Wintertime lightning shares the highest fraction of severe weather. Compared to Europe, extreme events are considerably more frequent over the United States, with maximum activity over the Great Plains. However, the threat over Europe should not be underestimated, as severe weather outbreaks with damaging winds, very large hail, and significant tornadoes occasionally occur over densely populated areas.

Severe Convective Storms across Europe and the United States. Part II: ERA5 Environments Associated with Lightning, Large Hail, Severe Wind, and Tornadoes

AbstractIn this study we investigate convective environments and their corresponding climatological features over Europe and the United States. For this purpose, National Lightning Detection Network (NLDN) and Arrival Time Difference long-range lightning detection network (ATDnet) data, ERA5 hybrid-sigma levels, and severe weather reports from the European Severe Weather Database (ESWD) and Storm Prediction Center (SPC) Storm Data were combined on a common grid of 0.25° and 1-h steps over the period 1979–2018. The severity of convective hazards increases with increasing instability and wind shear (WMAXSHEAR), but climatological aspects of these features differ over both domains. Environments over the United States are characterized by higher moisture, CAPE, CIN, wind shear, and midtropospheric lapse rates. Conversely, 0–3-km CAPE and low-level lapse rates are higher over Europe. From the climatological perspective severe thunderstorm environments (hours) are around 3–4 times more frequent over the United States with peaks across the Great Plains, Midwest, and Southeast. Over Europe severe environments are the most common over the south with local maxima in northern Italy. Despite having lower CAPE (tail distribution of 3000–4000 J kg−1 compared to 6000–8000 J kg−1 over the United States), thunderstorms over Europe have a higher probability for convective initiation given a favorable environment. Conversely, the lowest probability for initiation is observed over the Great Plains, but, once a thunderstorm develops, the probability that it will become severe is much higher compared to Europe. Prime conditions for severe thunderstorms over the United States are between April and June, typically from 1200 to 2200 central standard time (CST), while across Europe favorable environments are observed from June to August, usually between 1400 and 2100 UTC.

Updraft-Based Adaptive Assimilation of Radial Velocity Observations in a Warn-on-Forecast System

AbstractThe Warn-on-Forecast system (WoFS) provides short-term, probabilistic forecasts of severe convective hazards including tornadoes, hail, and damaging winds. WoFS initial conditions are created through frequent assimilation of radar (reflectivity and radial velocity), satellite, and in situ observations. From 2016 to 2018, 5-km radial velocity Cressman superob analyses were created to reduce the observation counts and subsequent assimilation computational costs. The superobbing procedure smooths the radial velocity and subsequently fails to accurately depict important storm-scale features such as mesocyclones. This study retrospectively assimilates denser, 3-km radial velocity analyses in lieu of the 5-km analyses for eight case studies during the spring of 2018. Although there are forecast improvements during and shortly after convection initiation, 3-km analyses negatively impact forecasts initialized when convection is ongoing, as evidenced by model failure and initiation of spurious convection. Therefore, two additional experiments are performed using adaptive assimilation of 3-km radial velocity observations. Initially, an updraft variance mask is applied that limits radial velocity assimilation to areas where the observations are more likely to be beneficial. This experiment reduces spurious convection as well as the number of observations assimilated, in some cases even below that of the 5-km analysis experiments. The masking, however, eliminates an advantage of 3-km radial velocity assimilation for convection initiation timing. This problem is mitigated by additionally assimilating 3-km radial velocity observations in locations where large differences exist between the observed and ensemble-mean reflectivity fields, which retains the benefits of the denser radial velocity analyses while reducing the number of observations assimilated.

A Comparison of Neural-Network and Surrogate-Severe Probabilistic Convective Hazard Guidance Derived from a Convection-Allowing Model

AbstractA feed-forward neural network (NN) was trained to produce gridded probabilistic convective hazard predictions over the contiguous United States. Input fields to the NN included 174 predictors, derived from 38 variables output by 497 convection-allowing model forecasts, with observed severe storm reports used for training and verification. These NN probability forecasts (NNPFs) were compared to surrogate-severe probability forecasts (SSPFs), generated by smoothing a field of surrogate reports derived with updraft helicity (UH). NNPFs and SSPFs were produced each forecast hour on an 80-km grid, with forecasts valid for the occurrence of any severe weather report within 40 or 120 km, and 2 h, of each 80-km grid box. NNPFs were superior to SSPFs, producing statistically significant improvements in forecast reliability and resolution. Additionally, NNPFs retained more large magnitude probabilities (>50%) compared to SSPFs since NNPFs did not use spatial smoothing, improving forecast sharpness. NNPFs were most skillful relative to SSPFs when predicting hazards on larger scales (e.g., 120 vs 40 km) and in situations where using UH was detrimental to forecast skill. These included model spinup, nocturnal periods, and regions and environments where supercells were less common, such as the western and eastern United States and high-shear, low-CAPE regimes. NNPFs trained with fewer predictors were more skillful than SSPFs, but not as skillful as the full-predictor NNPFs, with predictor importance being a function of forecast lead time. Placing NNPF skill in the context of existing baselines is a first step toward integrating machine learning–based forecasts into the operational forecasting process.

GLM and ABI Characteristics of Severe and Convective Storms

AbstractThe recent deployment of the Geostationary Lightning Mapper (GLM) on board GOES‐16 and GOES‐17 provides a new perspective of total lightning production for the severe convective storms research and operational communities. While the GLM has met its performance targets, further understanding flash characteristics and the physical limitations of the GLM are required to increase the applicability of the data. Derived cloud‐top height and infrared (IR) brightness temperature products from the Advanced Baseline Imager (ABI) are used to assess data quality and characteristics from gridded GLM imagery across 7 weeks of active severe weather: 13 April through 31 May 2019. Areas with cloud tops colder than 240 K typically produced lightning, though this becomes less certain near the edge of the field of view due to algorithm limitations. Increasing flash rates were observed to correlate with decreasing flash areas, increasing cloud‐top heights, and colder cloud‐top temperatures. However, flash rates and size were more strongly tied to convective intensity and proximity to convective hazards at the surface due to the ability to delineate between convective and stratiform precipitation. Results show that merging ABI and GLM data sets could add value to both machine learning and statistical‐based algorithms and also forecast applications with each providing unique details, although parameters such as GOES‐16 viewing angle should be considered. Lastly, two case studies (24 and 27 May 2019) are used to help interpret the results from the 7‐week sampling period and identify GLM and ABI trends related to thunderstorm evolution.