You may have been enjoying the sun these past few days, and noticed that some website and app forecasts have been somewhat more pessimistic with their predictions of the cloud cover. But why is something as fundamental as cloudy or sunny actually so difficult to forecast?

As with most weather forecasting, computer models are relied upon to produce accurate simulations of the atmosphere to guide cloud-cover forecasts. Because computer resources are finite, the model cannot simulate all the small-scale detail present in the real world, but “sees” the world as a pixellated image. For UK forecasting, the pixel sizes are 1.5km in the horizontal and around 100m in the vertical at the typical altitude of low clouds. Therefore clouds at these scales, or smaller, are very difficult to represent. Their formation, evolution and dissipation is controlled by “parametrizations” – the part of the model used to deal with processes acting on a smaller scale than the pixel size.

Low cloud, or stratocumulus, is particularly difficult because it requires many different parametrizations to interact correctly with each other to produce an accurate forecast. Turbulence, which can both enhance the cloud by mixing moist air from below and destroy the cloud by mixing dry air from above, must interact with solar and thermal radiation, which can directly modify the cloud by heating or cooling the atmosphere, and indirectly modify it by altering the turbulence in a feedback process. These processes must also interact correctly with the light winds and slowly subsiding air typically associated with high-pressure systems, which provide a weak, but important forcing of the cloud field.

Another aspect, which may have been relevant this past week, is how the model uses observations to initialise the forecasts – data assimilation. By far the best source of information on cloud cover is satellite imagery, but when (as has happened this week) low cloud is accompanied by high “cirrus” cloud, the satellite only sees the high cloud, making the full extent of the low cloud difficult to determine, particularly over the oceans. Only when the high cloud clears, or the low cloud reaches land, can the extent of any forecast differences be established and corrected within the model.

We continue to work on improving both the computer model, and methods used to initialise the forecasts from observations. As shown below, a current test model (right) including changes to both the parametrizations and data assimilation produced a much better cloud forecast than the operational model (left) when compared to the actual cloud coverage shown by visible satellite imagery. Improvements like these should filter through to forecasts in the near future.

Visible satellite image at 11am on Thursday 6 April 2017 (above), with cloud forecasts from Wednesday evening valid for the same time, using the current operational model (below left) and trial version for future implementation (below right). Clear skies are represented by white with coloured shading representing clouds.

Despite these complications, our forecasts still offer good guidance: we’re consistently ranked as one of the top operational providers in the world for accuracy and we’re trusted by 84% of the public to provide weather and climate services.

As always keep up to date with the weather in your area using our forecast pages, Twitter or Facebook, as well as using our new mobile app which is available for iPhone from the App store and for Android from the Google Play store. Search for “Met Office” in store.