The Portable Model for multi-scale Atmospheric Prediction (PMAP): Capabilities and development workflows

The Portable Model for multi-scale Atmospheric Prediction (PMAP) is a numerical model currently under development at ECMWF and ETH aimed at large-eddy simulation (LES) of atmospheric flows at the weather-system scale. It builds on a non-hydrostatic, locally conserving, finite-volume, 3D semi-implicit dynamical core coupled to state-of-the-art physics parametrizations. Written entirely in Python, it leverages the GT4Py domain-specific language to achieve high performance and portability – running straightforwardly on laptops and GPU-accelerated HPC systems alike. The systematic separation of concerns between domain science (physics, numerics) and performance engineering (parallelisation, hardware optimisation) provides new avenues for model development, setup, and refinement, which we present in two related contributions.

In this first contribution, we highlight PMAPs strengths as a numerical model framework to flexibly develop and refine numerical algorithms, as well as to implement and extend diagnostics to address research questions in atmospheric sciences. This is illustrated here with an LES tracer transport experiment over complex terrain. We first demonstrate PMAPs capability to perform decametre-scale LES using a height-based terrain-following vertical coordinate in steep terrain with slopes exceeding 80°. This is possible thanks to the locally conservative advection and the 3D semi-implicit integration scheme, which ensures stability of the integration and regularization of the flow. Moreover, we exemplarily show how the Python-based model formulation facilitates evaluating and improving various aspects of flux-form semi-Lagrangian tracer transport schemes in terms of directional splitting approaches and monotonic limiters, and how these impact simulated power spectra of the tracer fields. Lastly, we present how available model diagnostics can easily be extended to perform targeted analyses of sensitivities to implementation details.