Intelligent Tornado Prediction Engine   

 We are developing deep learning  models  to analyze tornadic precursors in order to enhance real-time tornado prediction.   
a graphic showing radar-like storm coverage over an google earth image of land, with red/yellow/green zones of predicted tornado likelihood.
The Intelligent Tornado Prediction Engine will utilize a massive open-source dataset to train deep learning models capable of identifying precursors to tornadoes.

Tornadoes in the Southeastern United States pose a unique threat to life and property due to the combination of different storm types (e.g., supercellular, quasi-linear convective systems [QLCSs], tropical) and times (daytime and nocturnal), as well as the density of manufactured homes and other vulnerable residential structures. At the same time, tornadoes continue to pose a significant threat throughout the United States in an ever-changing climate, leading to changes in tornado location, intensity, and frequency. Tornado warning lead times continue to stagnate at levels which can limit the ability for meaningful public response. This is especially the case for QLCS and tropical cyclone tornadoes where, historically, warning lead times have been significantly shorter than for supercellular tornadoes. Thus, new paradigms must be explored for improving lead times, using the plethora of data now available to National Weather Service forecasters with the advent of rapid-update satellite, radar, and numerical weather prediction models.  

One of the most rapidly advancing approaches to many problems in the atmospheric sciences is deep learning (DL), a form of artificial intelligence that is popular in image processing for extracting high-level features from extremely large datasets. DL is capable of combining many very large datasets in order to “learn” trends and features in the data based on history, making it an ideal candidate to search for combined precursors amongst several different data sources.  When a large number of tornadic cases are combined with a large number of null cases (from similar-looking storms), DL has the potential to be able to discern between tornadic and non-tornadic precursors. Discovery of new combinations of precursors could be used to train forecasters, determine new physical mechanisms for tornadogenesis, and even feed future probabilistic prediction techniques. 

We are developing DL models that utilize a massive, open-source dataset for training and validation in order to analyze combinations of precursors, detect ongoing tornadoes, and investigate the potential for real-time probabilistic tornado prediction. These could form the basis for enhanced decision support systems, which increase tornado prediction lead times and lower false alarm rates, allowing more effective public response and lowering the risks to populations in the path of these events.