Optimized Deployment of Atmospheric Models on High-Performance Computing Platforms

Abstract

This study evaluates the performance of two atmospheric reanalysis datasets: ERA5 and NARR (NCEP / NCAR) when used as boundary conditions for the WRF model to simulate meteorological conditions over Mexicali, a rapidly urbanized city located in a semi-arid region of northwestern Mexico. Simulations were carried out over a dry, synoptically stable period to assess the sensitivity of the model to different initial and boundary conditions, while maintaining physical parameterizations fixed. Validation against in situ surface observations focused on key atmospheric variables, including 2-meter air temperature, surface heat fluxes, and 10-meter wind components. The results reveal systematic differences between the two reanalysis products: ERA5 more accurately reproduces near-surface temperature patterns and ground heat fluxes, while NARR tends to overestimate sensible and latent heat fluxes, as well as wind variability.
Both datasets exhibit biases in representing the diurnal cycle and surface energy partitioning, which have significant implications for simulating boundary layer development and urban climate processes in arid environments.