What Is Numerical Weather Prediction?

Numerical Weather Prediction (commonly known as NWP) is a method used by meteorologists to forecast the weather by solving mathematical equations that describe the behavior of the atmosphere. These equations, known as the fundamental equations of motion, conservation of mass, and thermodynamic principles, are based on physical laws governing the atmosphere's behavior.

What Is a Numerical Weather Model?

NWP encompasses the entire process of using numerical weather models to make weather forecasts. This includes data collection, data assimilation (integrating observed data into the model), model initialization (setting the initial conditions), model integration (running the model forward in time), and post-processing (interpreting and presenting the forecast data).

A numerical weather model calculates the current and future states of the atmosphere with the help of computers and on the basis of measured data, both locally measured and remotely sensed. Complex physical equations are used, which are calculated by high-performance computers with enormous power and processed into high-resolution weather forecasts.

The Three-Dimensional Grid

To obtain weather forecasts, a numerical weather model divides the Earth's atmosphere into a three-dimensional grid. For each grid point, relevant atmospheric parameters (such as temperature, humidity, wind speed, and pressure) are calculated at various altitudes and at fixed time intervals. 

What Is Weather Model Resolution?

Spatial Resolution

The distance between the grid points determines the model's spatial resolution; larger grid spacings result in coarser resolutions. Typically, when referring to the resolution of a weather model, we consider the spacing of grid points at the Equator, where the Earth's circumference is largest. As depicted in the illustration above, the grid size varies with latitude, generally decreasing towards the poles due to the convergence of meridians.

Currently, a 1 km spatial resolution is considered very high. With a high spatial resolution of 1 km or less, many local and dynamic effects which are not captured by models with larger grid cells can be mapped, greatly refining and thus improving weather forecasts at the local level. As illustrated below, when we refer to "1 km," "6 km," or "10 km" resolution, it actually means km² for the grid cell area. When transitioning from a model with a 1 km² grid, such as EURO1k, to a model with a 6 km² grid, such as ICON EU, or a 10 km² grid, such as ECMWF, the resolution decreases exponentially by factors of 36 and 100, respectively.

Temporal Resolution

A high temporal resolution, on the other hand, means that changes in weather are detected by the model over short time intervals – such as every 20 minutes for the high-resolution EURO1k model. For comparison, the global weather model ECMWF has a temporal resolution of 6 hours. Again, higher resolution allows us to physically model the changes to the atmospheric circulation at shorter timescales, producing more reliable forecasts.

Global Weather Models

In global weather models, a mesh size of 10 to 50 kilometers is common. This is usually sufficient to optimally predict conditions in the higher atmosphere, but the topography below the horizontal resolution (10-50 kilometers) cannot be captured by models, and finer-scale events, such as thunderstorms, cannot be modeled as a result.

Thus, forecasts of parameters near the earth’s surface, such as wind, maximum or minimum temperature, can be inaccurate at a local level. Higher resolution global models are required to close these data gaps.

Examples of global weather models are, among many others: the American Global Forecast System (GFS) with about 25 km resolution and the forecast model of the European Centre for Medium-Range Weather Forecasts (ECMWF) with about 10 km resolution.

Local Weather Models

In contrast to the global weather models, the grid points of local weather models are much closer together. This requires a lot more computational capacity and is therefore only calculated for a period of one to three days.

A local weather model only covers a limited domain. In order to include relevant influences coming from outside a certain region, the values at the edges of the domain are taken from a global model. This principle is called "nesting" and is intended to provide better model predictions at the edges.

The EURO1k model from Meteomatics, with 1 km resolution, is an example of a local weather model for Europe. Similarly, the High-Resolution Rapid Refresh (HRRR) model developed by NOAA, with a 3 km resolution, is used for local weather prediction in the United States.

The Meteomatics Weather API has a comprehensive portfolio of weather models, including:

• Global numerical weather models (ECMWF, GFS, DWD, etc)
• Global AI-based weather models (FourCastNet, GraphCast)
• Local numerical weather models (EURO1k, Arome, DWD, HRRR, etc)

Sign Up for a Free Trial of the Weather API

With a Weather API Free Trial you gain access to 1800+ parameters and 110+ weather sources and models.

Explore the power of the API for 14 days for free!