Meso-Scale Forecasts – how they work

About this page

Meso-scale forecast apply Numerical Weather Prediction to limited areas using high resolution.

Related pages

On this page -


Global Numerical Weather Prediction models run by National Met Services mostly use grid lengths of about 13 – 17 km. It is a simple geometrical fact that any grid can only define shapes of about 4 or 5 grid lengths. It follows that global models can only represent weather on scales of about 50-80 km. Detail smaller this will be artefacts of the mathematics and, so, are filtered out.

Meso-scale NWP (Limited Area) Models are used to predict smaller scale detail. These forecasts come into two broad categories; those run by National Met Services and those run by private sector firms.

Several interlinking factors determine how LAMs are run and what they can produce. The resolution of the global model used to provide boundary conditions for the LAM area is a limiting factor. The size of the area determines how far ahead predictions are useful. Computer power determines grid length and the area size over which the models are run. The quality of detailed data determines what the model “knows” about the initial conditions. The limited lifetime of the small weather detail described by fine scale grids restricts the detail that the model can predict and how far ahead it can be predicted.


National Weather Service LAMs

The Met Office runs a model on a 1.4 km grid over the British Isles. This is "nested" in a model over a larger area run with a 4 km grid itself nested within a global model having a grid length of about 17 km. All are run at 70 levels in the vertical. The 1.4 km model requires as much computing power as their global model with a 17 km grid.

Courtesy of the UK Met Office.

The Danish Met Serv ice, DMI, runs a model with a grid length of 5 km at 40 levels and containing a 2km grid model

Denmark is one of a group of European countries running a meso-scale model known as HIRLAM. Other countries produce meso-scale forecasts relevant to their national areas of interest.

Limiting factors for LAMs

Quality of the initial analysis

  • Shower clouds and similarly sized features will not generally be well represented.
  • Analysis of the mass of (mainly satellite) data is a major computational problem.

Short lifetimes of small weather features.

  • Those small features that are analysed may not exist long enough to be predicted usefully. A large convective cloud, say 10 km across will have a lifetime of less than 6 hours.

Lateral boundary input

  • The models “know” nothing about small detail entering the model area. In mobile weather situations, the benefits of meso-scale will decrease rapidly in time.


  • Small weather detail cannot be explicitly predicted.
  • Models can only take chaos into account through the use of model ensembles.

National Met Service Model Inputs


National Met Service

Commercail firm

Global input resolution

13 to 17 km

25 km

Observational input

From 1 km up to about 40 km. Using all data in a 4-D analysis

25 km data from the US GFS

Topographic data

High resolution

High resolution

Wetness of ground


Probably not


Meso-scale forecasts

These are some of the meso-scale forecasts readily available and of most interest to UK sailors. A more extensive list can be found at Grib-And-Objective-Forecasts-Reviewed

Model[[,& Forecast Period

Computational Grid km

Sample fprecast

Internet Source



Rain area

Android Apps
Met Office site

The presentation is good for rain areas but non-existent for winds.


5 km


 DMI Website

This is a good and useful presentation but cannot show detail smaller than about 25 m size.


2.5 km

YR.NO Website

Zooming in provides moredata. Tapping on a location anywhere on the map gives a meteogram including sea state.


5 km

 AEMet Website

Not such a good presentation as the Danish service. A little clumsy in operation. Show island wind shadows well.

Many other sources of meso-scale and Global models will be found on a GRIB Services list page.

Do they really help?

These are two forecasts from the Met Office 1.5 km model that uses satellite and radar data to get a really fine scale analysis. Forecasts are run at 0300 and 1500 UTC. The outout is valid for 24 hours.

This is a 9 hour forecast and actual for 1200 hours.

Attach:17mar18.png Δ Δ This is a 15 hour forecast and actual for 1800 hours.

Although the forecasts look pretty good, there are some significant errors even on the 9 hour forecast.

These three forecasts all verifying at the same time give some idea about how little real extra information there is in LAMs, even with detailed weather data input, as oppose to the GFS.

The GFS outputs GRIB data on a 0.25 degree grid, about 25 km.

Attach: Weatherwise Ad.png Δ Δ

The Turk-Marine Weatherwise forecast uses the European Centre For Medium Range Forecast ouput on about a 15 km grid. They use the US WRF (Weather Research Forecast) to compute and output on a 12 km grid.

The Croatian Met Service, (DHMZ) uses output on an 8 km grid from a French LAM known as ALADIN . Over the Adriatic, they compute on a 2 km grid. using detailed weather data to initialise their forecast

As far as I can see, there is little extra information in the more detailed forecasts vis a vis the GFS and considerable doubt whether differences are significant. More data does not necessarily mean more information.


Meso-scale models are very useful to National Weather Services for short term warning purposes. Risk can be accessed through use of ensembles.

They should help sailors because of better topographic representation than in global models. However, the complexity of the atmosphere and chaos limits their value in deterministic use.