Avoiding AI Autopilot

The concept of meteorological cancer, that is, the atrophy of human weather forecaster skills, was first discussed within the forecasting community shortly after the introduction of Model Output Statistics (MOS). With the recent, rapid growth in the types and performance capabilities of modern artificial intelligence (AI) and machine learning approaches, this ongoing concern should and can be effectively factored into modernization plans.

In this talk, I will review several applications of AI techniques to weather forecasting problems. The focus will be on how these techniques can be used to enhance decision support efforts while simultaneously building forecaster confidence through physical interpretation of the model output and the running of forecast scenarios. Specific examples will include multiple logistic regression, multi-layer perceptron artificial neural networks, and evolutionary programming applications to the forecast problems of 2-m temperature, heavy rainfall, convective occurrence, and precipitation phase.