Weather forecast improvements: the science behind the forecasts

There are two main global numerical weather prediction (NWP) models that are used by most meteorologists: ECMWF (the European model) and GFS (the American model). Every year ECMWF have a meeting for meteorologists who use their model data, for exchanging ideas and experiences on the use of ECMWF data. I was there again last week, presented some of my ideas and as always, walked away with new insights!

The NWP models form the basis of our information for writing a forecast, and these models are typically updated once or twice a year to keep on pushing the bounds and improving the forecasts we have. These improvements fall into two main brackets:

  1. a better understanding of the physical processes and interactions occurring in Earth’s atmosphere and ocean, and
  2. more computer power allowing for higher resolution forecasts to be run.


Improvements to physical processes

An example of the former is that the ECMWF (and GFS!) are known to have too much `drizzle’ in the forecasts when the actual weather is dry. In the ECMWF update (known as a `cycle’) which is to be implemented this Autumn, the way that clouds are modelled has been updated to address this issue.

The challenge for us forecasters is to adjust to the model changes and get a feel for any biases in the new cycle, as well as update any statistical adjustments we use. It won’t change how wet the UK is, but I’m looking forward to less days when the forecast looks too gloomy!

Improvements to model resolution

As to computer power: imagine an old fashioned globe. The latitude and longitude lines give a grid, similar to the grid on an Ordnance Survey map. As we have more computer power available, the distance between the grid lines used in the model are reduced leading to more accurate forecasts. Over the last 25 years the number of grid points used in the ECMWF NWP model has increased exponentially: currently they are about 16km apart for the high resolution run.

Anna Maidens from UK Met Office presented results from the Hadley Centre model showing that increasing the resolution in ocean modelling makes a big difference to capturing the Gulf Stream. In particular the sea surface temperatures off of Nova Scotia and Greenland are more accurately captured.

This strikes me as being particularly relevant for UK weather forecasts since many low pressure systems form in this region, and errors in how/where they form lead to bad forecasts for the UK.  Take earlier this month as an example:

On Tuesday 3rd June 2014 the forecast issued for Friday 6th June has a low forming near Nova Scotia…

The model's idea of what will form near Nova Scotia

Fig 1: On Tues 3rd June the model’s forecast of what will form near Nova Scotia on Fri 6th June

and the forecast for Tues 10th June (again issue on Tues 3rd June) shows that low to be mid-Atlantic and beginning to influence UK weather

Fig 2: and how that low will evolve by Tues 10th June

Fig 2: …and how that low will evolve by Tues 10th June

What actually happens?

The low forms slightly differently to that forecast on Friday 6th June (compare Fig 1 & Fig 3)…

Fig 3: what actually forms on Friday 6th June

Fig 3: what actually forms on Friday 6th June

leading to a very different pattern by Tues 10th (compare Fig 2 & Fig 4)

Fig 4: How the low actually evolves by Tues 10th June

Fig 4: How the low actually evolves by Tues 10th June

The deeper low that actually formed helped the high pressure build over the UK which is currently giving glorious sunshine.  The forecast for today from Tues 3rd June was far less bright.  With improvements to ocean model resolution, UK weather forecasts should improve.  It’s not scheduled for this year, but does give hope that forecasts will continue to improve for some years to come!