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

Abstract

A feedforward 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 km vs. 40 km) and in situations where using UH was detrimental to forecast skill. These included model spin-up, 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 towards integrating machine learning-based forecasts into the operational forecasting process.

A Benchmark to Test Generalization Capabilities of Deep Learning Methods to Classify Severe Convective Storms in a Changing Climate

“…UH values representative of severe convection also vary seasonally and regionally, based on the climatological environments that drive severe convection activity (Sobash & Kain, 2017;Molina, Allen, & Prein, 2020). Recently, Sobash et al (2020) trained a CNN to forecast severe hazard potential using severe thunderstorm parameters derived from WRF, showing that a CNN can learn from diagnostics. The focus herein lies on evaluating a CNN's ability to classify convection and its out-of-sample robustness to a future climate.…”

“…The recent success of convolutional neural networks (CNNs; Fukushima & Miyake, 1982) in Earth science applications is largely due to their ability to capture nonlinear and translation invariant details among input variables. This class of deep learning models (LeCun et al., 2015) has proven skillful in various atmospheric science tasks, including the detection of weather and climate features (Biard & Kunkel, 2019; Lagerquist et al., 2019; Y. Liu et al., 2016; Toms et al., 2019), emulation of complex model processes (Rasp et al., 2018), and prediction of extreme weather and climate phenomena (Gagne II et al., 2019; Ham et al., 2019; Jergensen et al., 2020; Lagerquist et al., 2020; Sobash et al., 2020; Zhou et al., 2019). This study focuses on convection over the central and eastern contiguous United States (CONUS), which at extremes can produce severe hazards (e.g., hail and tornadoes) that pose societal danger.…”

“…This class of deep learning models (LeCun et al, 2015) has proven skillful in various atmospheric science tasks, including detection of weather and climate features (Y. Liu et al, 2016;Lagerquist et al, 2019;Biard & Kunkel, 2019;Toms et al, 2019), emulation of complex model processes (Rasp et al, 2018), and prediction of extreme weather and climate phenomena Zhou et al, 2019;Ham et al, 2019;Jergensen et al, 2020;Sobash et al, 2020;Lagerquist et al, 2020). This study focuses on convection over the central and eastern contiguous United States (CONUS), which at extremes can produce severe hazards (e.g., hail and tornadoes) that pose societal danger.…”

“…Recently, Sobash et al. (2020) trained a CNN to forecast severe hazard potential using severe thunderstorm parameters derived from WRF, showing that a CNN can learn from diagnostics. The focus herein lies in evaluating a CNN's ability to classify convection and its out‐of‐sample robustness to a future climate.…”

Approximation of a Convective-Event-Monitoring System Using GOES-R Data and Ensemble ML Models

“…MLPs have been widely used in several convective-hazard and precipitation forecasting studies (e.g., [58,59]); however, the application of Deep Neural Network architectures has been preponderant recently in this domain, i.e., Deep Neural Networks (DNN; e.g., [4,60,61], Recurrent Neural Networks (RNN; e.g., [62,63]), and CNN (e.g., [29,64,65]) are the most common). The implementation of Deep Learning techniques in deep convective cloud research will be addressed in later sections.…”