An AI-Machine Learning Probabilities (AI-MLP) Forecast System for Hazardous Weather Prediction

Data driven global AI models have garnered significant attention of late owing to their competitiveness with state-of-the-art dynamics-based numerical weather prediction (NWP) systems. After training, the AI models can simulate synoptic-scale patterns with relative ease and speed, and recent quantitative evaluations have suggested they can simulate large-scale temperature, pressure, and winds with the same or better accuracy than NWP models. However, these AI models suffer from some of the same deficiencies as NWP models for applications in high-impact weather domains. Namely, their rather coarse resolution doesn’t lend itself well to explicit predictions of convection hazards. To get around this issue with traditional NWP models, postprocessing methods have been used to generate explicit forecasts of hazards. As an example, the Colorado State University Machine Learning Probabilities (CSU-MLP) forecast system leverages random forests (RFs) and inputs from a global NWP ensemble to generate daily probabilistic forecasts at lead times of 1-8 days for excessive rainfall, tornado, severe hail, and severe wind hazards. The CSU-MLP uses simulated environmental parameters like shear, CAPE, and precipitable water to generate forecasts of hazards, with meteorological expertise guiding the model design and a long historical record for model training. Output forecasts mimic operational outlooks from NWS operational centers, making these types of forecasts particularly valuable to operational forecasting.

In a similar way, output fields from the data-driven AI models can be used to generate hazardous weather outlooks. In this work, we apply the previously detailed CSU-MLP framework to global AI model inputs, herein termed the AI-MLP, to generate daily probabilities of hazardous weather. With limited outputs from the models, feature engineering is used to select the most appropriate input variables that drive the postprocessing system. The operations-like outlooks are compared to similar operational CSU-MLP products as well as outlooks generated from human forecasters to examine how data driven AI models can be used for small-scale hazardous weather prediction. Further, explainable AI techniques are leveraged to decipher how MLP systems respond to inputs from different forecast models (e.g., Global Ensemble Forecast System in CSU-MLP, global AI models in AI-MLP) and whether global AI systems could be used in an ingredients-based forecasting paradigm.