Automatic Unstructured Grid Refinement Using Machine Learning for the Analog Ensemble

The AnEn is a statistical technique to generate probabilistic forecasts. This is a computationally efficient solution to ensemble modeling because it does not require multiple NWP simulations, but a single model realization. However, the required computation can grow very large because atmospheric models are routinely run with increasing resolutions. For example, the NAM contains over 262,792 grids to generate a 12 km prediction. NWP models generally use a structured grid to represent the domain, despite the fact that certain physical changes occur non-uniformly across space and time. For example, temperature changes tend to occur more rapidly in mountains than plateaus.

A new machine learning based algorithm is proposed to dynamically and automatically learn the optimal unstructured grid pattern. This iterative algorithm is guided by machine learning rule generation and instantiation to identify grid vertices. Analog computations are performed only at vertices, therefore minimizing the number of vertices. Identifying their locations are paramount to optimize the available computational resources, minimize queue time, and ultimately achieve better results. The optimal unstructured grid is then used to perform probabilistic forecasts for a variety of applications like uncertainty quantification or renewable energy prediction. In this work, the short-term temperature is used as a study case.