High-impact weather events, such as severe thunderstorms, tornadoes, and hurricanes, cause significant disruptions to infrastructure, property loss, and even fatalities. High-impact events can also positively impact society, such as the impact on savings through renewable energy. Prediction of these events has improved substantially with greater observational capabilities, increased computing power, and better model physics, but there is still significant room for improvement. Artificial intelligence (AI) and data science technologies, specifically machine learning and data mining, bridge the gap between numerical model prediction and real-time guidance by improving accuracy. AI techniques also extract otherwise unavailable information from forecast models by fusing model output with observations to provide additional decision support for forecasters and users. In this work, we demonstrate that applying AI techniques along with a physical understanding of the environment can significantly improve the prediction skill for multiple types of high-impact weather. The AI approach is also a contribution to the growing field of computational sustainability. The authors specifically discuss the prediction of storm duration, severe wind, severe hail, precipitation classification, forecasting for renewable energy, and aviation turbulence. They also discuss how AI techniques can process “big data,” provide insights into high-impact weather phenomena, and improve our understanding of high-impact weather.

A GSI-based data assimilation (DA) system, including three-dimensional variational assimilation (3DVar) and ensemble Kalman filter (EnKF), is extended to the multiscale assimilation of both meso- and synoptic-scale observation networks and convective-scale radar reflectivity and velocity observations. EnKF and 3DVar are systematically compared in this multiscale context to better understand the impacts of differences between the DA techniques on the analyses at multiple scales and the subsequent convective-scale precipitation forecasts. Averaged over 10 diverse cases, 8-h precipitation forecasts initialized using GSI-based EnKF are more skillful than those using GSI-based 3DVar, both with and without storm-scale radar DA. The advantage from radar DA persists for ~5 h using EnKF, but only ~1 h using 3DVar. A case study of an upscale growing MCS is also examined. The better EnKF-initialized forecast is attributed to more accurate analyses of both the mesoscale environment and the storm-scale features. The mesoscale location and structure of a warm front is more accurately analyzed using EnKF than 3DVar. Furthermore, storms in the EnKF multiscale analysis are maintained during the subsequent forecast period. However, storms in the 3DVar multiscale analysis are not maintained and generate excessive cold pools. Therefore, while the EnKF forecast with radar DA remains better than the forecast without radar DA throughout the forecast period, the 3DVar forecast quality is degraded by radar DA after the first hour. Diagnostics revealed that the inferior analysis at mesoscales and storm scales for the 3DVar is primarily attributed to the lack of flow dependence and cross-variable correlation, respectively, in the 3DVar static background error covariance.

Led by NOAA’s Storm Prediction Center and National Severe Storms Laboratory, annual spring forecasting experiments (SFEs) in the Hazardous Weather Testbed test and evaluate cutting-edge technologies and concepts for improving severe weather prediction through intensive real-time forecasting and evaluation activities. Experimental forecast guidance is provided through collaborations with several U.S. government and academic institutions, as well as the Met Office. The purpose of this article is to summarize activities, insights, and preliminary findings from recent SFEs, emphasizing SFE 2015. Several innovative aspects of recent experiments are discussed, including the 1) use of convection-allowing model (CAM) ensembles with advanced ensemble data assimilation, 2) generation of severe weather outlooks valid at time periods shorter than those issued operationally (e.g., 1–4 h), 3) use of CAMs to issue outlooks beyond the day 1 period, 4) increased interaction through software allowing participants to create individual severe weather outlooks, and 5) tests of newly developed storm-attribute-based diagnostics for predicting tornadoes and hail size. Additionally, plans for future experiments will be discussed, including the creation of a Community Leveraged Unified Ensemble (CLUE) system, which will test various strategies for CAM ensemble design using carefully designed sets of ensemble members contributed by different agencies to drive evidence-based decision-making for near-future operational systems.