Regional machine learning weather prediction (MLWP) models based on graph neural networks have recently demonstrated remarkable predictive accuracy, outperforming numerical weather prediction models at lower computational costs. In particular, limited-area model (LAM) and stretched-grid model (SGM) approaches have emerged for generating high-resolution regional forecasts. While LAM uses lateral boundaries from an external global model, SGM incorporates a global domain at lower resolution. This study aims to understand how differences in model design impact relative performance and potential applications. Specifically, in a near-identical setup, the strengths and weaknesses of these two approaches are identified for generating deterministic regional forecasts over Europe. Results show that both LAM and SGM are competitive deterministic MLWP models with generally accurate and comparable forecasting performance over the regional domain. Various differences were identified in the performance of the models across applications. LAM is able to successfully exploit high-quality boundary forcings to make predictions within the regional domain and is more suitable when training data is only available in a limited region. SGM is fully self-contained for easier operationalisation, can take advantage of more training data and shows signs of increased (temporal) generalisability. Our paper can serve as a starting point for meteorological institutes to guide their choice between LAM and SGM in developing an operational data-driven forecasting system. • Different architectures have emerged for data-driven regional weather forecasting. • Stretched-grid has global coverage, while limited-area models use boundary forcings. • Strengths and weaknesses of both approaches were explored in a near-identical setup. • Stretched-grid has operationalisation, scalability and generalisability advantages. • Limited-area can exploit high-quality forcings and deal with limited global data.
Wijnands et al. (Sun,) studied this question.