Weather Research and Forecasting model

The WRF model was conceived in the late 1990s as a collaborative effort among the National Center for Atmospheric Research (NCAR), the National Oceanic and Atmospheric Administration (NOAA), the U.S. Air Force, the Naval Research Laboratory, the University of Oklahoma, and the Federal Aviation Administration.¹ It was designed to replace older mesoscale models like MM5 and to serve both research and operational communities. The first public release, WRF 1.0, appeared in 2000, and the model has since undergone continuous development, with major updates every few years. WRF is written in Fortran and is highly modular, allowing users to configure it for diverse applications, from idealized simulations to real-data forecasts. Its open-source nature has fostered a large user community, with regular workshops and a dedicated support forum.

WRF uses a terrain-following hydrostatic-pressure coordinate system and offers a choice of dynamical cores: the Advanced Research WRF (ARW) and the Nonhydrostatic Mesoscale Model (NMM).² The ARW core is the most widely used, employing a Runge–Kutta time integration scheme and a staggered Arakawa C grid. The model supports multiple nesting levels, allowing high-resolution domains to be embedded within coarser ones. Physical parameterizations include options for microphysics, cumulus convection, planetary boundary layer, land surface, and radiation. WRF also includes a data assimilation system (WRFDA) that can ingest observations from radar, satellite, and surface stations. The model is parallelized with MPI and OpenMP, enabling efficient runs on supercomputers.

WRF is used for a wide range of applications, including operational weather forecasting by national meteorological services, such as the National Weather Service in the U.S., and by private companies. It is also a key tool in atmospheric research, supporting studies of severe thunderstorms, hurricanes, and regional climate. WRF has been adapted for air quality modeling (WRF-Chem), where it simulates chemical transport and interactions with meteorology. Additionally, it is used for wind energy resource assessment, agricultural planning, and wildfire smoke dispersion. The model's flexibility allows it to be run at resolutions from kilometers down to tens of meters for local-scale studies.

Beyond its mainstream use, WRF has several niche applications. For instance, it has been used to simulate the atmospheric effects of volcanic eruptions and to study the dispersion of radioactive materials after nuclear accidents. In the polar regions, WRF has been adapted with specialized sea-ice and snow schemes to improve forecasts in the Arctic and Antarctic. The model also supports idealized simulations, such as the classic 'bubble' test, which are used to test numerical schemes. A lesser-known fact is that WRF has a 'single-column' mode that isolates vertical processes, useful for testing parameterizations. Additionally, the WRF-Hydro extension couples the atmosphere with terrestrial hydrology, enabling flood forecasting. The model's community has produced numerous regional configurations, such as the Polar WRF and the Tropical WRF, each tuned for specific environments.

One edge case is the treatment of the 'gray zone' of convection, where cumulus parameterizations become invalid at resolutions around 1–10 km; WRF users often must choose between explicit microphysics and parameterized convection, leading to sensitivity in precipitation forecasts. Another detail is the model's ability to run in 'forecast mode' with real-time data, but it also supports 'climate mode' for long-term simulations, though this requires careful tuning. WRF has been used in educational settings, with universities incorporating it into coursework. The model's code is continuously updated, and users must keep track of version-specific behavior. A notable historical fact is that WRF was used to support the 2008 Beijing Olympics, providing high-resolution forecasts for the event.

Mesoscale

Atmospheric phenomena ranging from a few kilometers to several hundred kilometers in size.

Dynamical core

The set of equations and numerical methods that solve the atmospheric motion.

Parameterization

A simplified representation of a physical process that is too small-scale to be resolved directly.

Nesting

Embedding a higher-resolution domain within a lower-resolution one to focus on a region of interest.

Data assimilation

The process of incorporating observations into a model to improve initial conditions.