January 2012 - The Met Office runs a 1.5 km grid length model over the UK in order to produce improved forecasts of small scale weather such as thunderstorm and winter snow.

The Met Office runs a global weather forecasting model. Given available computer resources, to cover the whole globe it is necessary to use a grid length of approximately 25 km. This is able to represent the large scale weather patterns, for example the high and low pressure areas one sees on a weather map. A 25 km model can also resolve significantly more detail than this - for example areas of frontal rain and regions of shower activity are also able to be picked out by a model with this grid length. However, there are still many important weather phenomena which take place on more local scales and require a more detailed model.

For this reason the Met Office runs a version of the Unified Model over the UK with a grid length of 1.5 km usually referred to as the UKV. Examples of weather phenomena that require such a model include fog filling valleys, enhanced rain over mountains and higher temperatures in cities. However, the main benefit is to better resolve convective showers or storms which, in extreme cases, can give rise to major flooding events or disruptive snow in winter. These convective clouds are typically less than 10 km horizontal extent and have to be represented as subgrid processes at global model resolutions. In contrast, a 1.5 km model can often produce the convection explicitly on the model grid which can lead to much better forecasts in many situations. An important caveat is that even with a 1.5 km grid length convection is often still under-resolved and a subject of research is how best to address this problem.

The area of the UKV model is shown in figure below. The boundary data for any limited area (i.e. not global) model are provided from a model covering a larger area (either a global model or a larger limited area model). In the case of the UKV the boundaries are currently supplied from the global model. Since the large jump in grid length at the boundaries from 25 km to 1.5 km could cause problems the UKV is a variable resolution model (hence the name). Within the green rectangle in this figure the grid length is 1.5 km, between the green and the red rectangle the grid length smoothly increases to 4 km. For the rest of the area shown outside the red rectangle the grid length is 4 km. These variations in grid length take place independently in the x and y directions so the grid cells are 4x1.5 km along the edges and 4x4 km in the corners as shown.

A second advantage of using a UKV domain annotated with sizes of cells variable resolution model is to push boundary effects further away. The 4 km grid length around the edges is still short enough to allow convection to be resolved in many cases. An issue as air enters a high resolution model from a lower resolution one is that the transition from subgrid convection to resolved convection takes a finite length of time. This can lead to artefacts in the representation of showers. By having an area of 4 km grid around the edge of the domain the area where these artefacts affect the forecast can be pushed away from the area of interest at much lower cost that by simply extending the 1.5 km area.

The model obtains its initial conditions using 3-dimensional variational (3D-Var) data assimilation with additional nudging of radar rainfall data. Eight three-hour assimilation cycles are run each day but only half of these are run out as 36 hour forecasts (i.e. four forecasts per day).

In addition, in order to provide, longer high resolution forecasts a different model with a 4 km grid length is run out as five-day forecasts. This model takes its boundary data directly from the 25 km global model

The top figure shows an example of the advantages of a high resolution model, a comparison with radar can been seen by selecting Enlarge. The charts show total snowfall over 24 hours on 25 November 2010. This was a case where numerous heavy snow-showers being carried inland from the sea in a NE wind caused significant disruption in the north east of England. The figure shows that for the coarser 12 km model with convection not represented explicitly, showers stall over the coast causing a major underestimate of snow inland. This is a well known problem with models of this grid length. In contrast, the UKV is able to represent the showers more realistically and brings the showers inland, producing a much better forecast.

Models with this grid length raise significant issues of interpretation. Chaos theory tells us that smaller scales in the atmosphere are inherently unpredictable for longer times. So although the UKV model might well reproduce an area and general character of showers, the detailed location and timing of each shower will not be correct. The model still gives very useful information, for example a forecast that an afternoon's weather will consist of a number of heavy showers is very different from saying there will be continuous lighter rain even though the total amount of rain might be the same. However, it is important to avoid presenting the information in such a way that the unpredictable smaller scales will be taken literally. For example, it is not possible to infer that in a particular town a shower will start at 3pm and end at 3:20pm simply because that is what was predicted by the model. An important tool in addressing this problem is to move towards probabilistic forecasts. These forecasts can be generated with ensemble techniques where more than one version of the same forecast is run in order to get an estimate of the uncertainty. The Met Office intends to move to an ensemble of convection permitting models.

Last updated: 14 April 2014

- Share this page
- Tweet