Super local weather forecast in Longyearbyen

Top image: Students with weather station in Svalbard. Photo: Lukas Frank/UNIS.

UNIS students set up measuring stations everywhere. Suddenly, the researchers received much more weather data from a small area in Svalbard.

29 June 2021
Text: Anna Kathinka Dalland Evans, The Norwegian Meteorological Institute. Translation to English by the University Centre in Svalbard (UNIS).

This happened when researchers from the Norwegian Meteorological Institute collaborated with UNIS students on fieldwork in Svalbard.

The aim was to investigate the possibilities of forecasting the weather at an even more detailed level than meteorologists usually manage today.

The Arctic is an area of the world where more accurate weather forecasts will be very useful. The terrain is complicated, and the weather can change abruptly from place to place.

The quiet revolution

Weather forecasts are getting better and better. Statistics show that the forecast for three days ahead today is approximately as good as the forecast two days ahead was 10 years ago.

Most of this development has taken place as a steady stream of improvements and technological innovations, and not through major revolutionary upheavals.

In an article in the journal Nature in 2015, this steady improvement in weather forecasting was called “The quiet revolution of numerical forecasting”. This is a name that has settled among researchers.

Started with UNIS students on fieldwork

Last autumn, researchers at the Norwegian Meteorological Institute published a scientific article about how weather forecasts in Arctic areas can become even better in the future. The work was done as part of the Alertness research project.

”It all really started with a course in meteorology at the University Centre in Svalbard”, says Teresa Valkonen. She is a researcher at the Development Centre for Weather Forecasting at the Norwegian Meteorological Institute.

“The students built their own weather stations. They used weather balloons to collect observations and they even installed temperature sensors on snowmobiles”, says Valkonen.

As a result, in an area around Longyearbyen in Svalbard, the researchers gained access to many more and closer observations, both in space and time, than they usually have through the network of official measuring stations.

Varied terrain provides changing weather

You have probably noticed that valleys or tall buildings can make wind tunnels where there is a lot of wind. The terrain affects what kind of weather we get, in a constant interaction between the ground and the air.

The terrain on Svalbard is varied, with for example many fjords, and there are large local variations in the weather.

A valley can be narrow. The weather inside a fjord can be very different from the weather just outside.

Norway’s grid size

Behind all weather forecasts there is a numerical weather model. In such a model, Norway, and the surrounding sea areas, are divided into a grid.

With the help of large computers, the weather in the future will be calculated for each of these grids. This means that when you receive a forecast for exactly where you are, it is actually a forecast that applies to the entire grid you are in. This is one of the reasons why the weather you experience may be different from what was forecasted.

The weather forecast you get on Yr for Svalbard today is based on grids that have 2.5 kilometres long sides in each direction. This is called the grid size. But the weather can vary a lot in this distance, and the model does not get these very local variations.

Need more and closer observations

The grid into which the researchers divide Norway has steadily become more fine-grained, and the size has become smaller. For example, the grid size in the models of the Meteorological Institute was as big as 10 kilometres when Yr was launched in 2007. Now it is 2.5 kilometres.

If we want more accurate weather forecasts, the grid can be made even smaller. Then the forecast for the area where you are, will be based on a smaller grid and therefore be more precise. This is especially useful when the terrain is very variable.

“Historically, it has been important to constantly increase the resolution, i.e. make the grid size smaller, to get better forecasts. But with more detailed forecasts comes the challenge of how we can assess how good the forecasts are”, says Valkonen.

In order to check how good the forecasts are when using a smaller grid size, more observations are needed of how the weather actually turned out. The observations of, for example, temperature, wind, pressure, and humidity, with which the researchers would like to compare the forecast, are usually not densely enough placed to assess how well the model has forecasted very local differences.

The collaboration with the UNIS students in Svalbard enabled the researchers to investigate further how accurate extra detailed forecasts can be, as they gained access to a much more fine-meshed network of observations of temperature and wind. They studied a grid size that was only 500 meters, i.e. much smaller than what is used for the operational forecasts today.

Why can we not just increase the resolution?

If the grid size in the model is to be made even smaller, many more observations are needed to compare to the model. It is not enough to know what the weather was like at the measuring station in Longyearbyen, we also need to know how the conditions were around the entire Adventfjorden.

Still, it’s not just about making the grid size smaller and smaller. Firstly, it becomes very demanding in terms of resources for computers to run through the models when the grid size become very small. This cost both time and money. For example, halving the grid size requires about eight times as much computing power.

Secondly, changes in one part of the weather model can lead to surprises in another part of the model, and it can even give incorrect results.

“It is a tempting way of thinking to constantly increase the resolution, but it is not always a given that this yields better results”, Valkonen explains.

A weather model is constantly kept in place by observations of how the weather is right now. If the grid size becomes smaller than what is justifiable based on available observations, the model may be given too much freedom, and then the result may be incorrect.

More research provides ever better weather forecast

The researchers in the project in Svalbard found that many natural processes were better described when they used an even smaller grid size than usual. On average, the forecasts for temperature and wind were improved. But they also experienced that they had to focus on other parts of the model, parts that describe snow and the ground, for example, in order not to get wrong results.

“In a model used for operational forecasts, it is not currently possible in terms of resources to use such a high resolution as we have done in this research project”, says Teresa Valkonen.

“But for research purposes, it is very useful to see what works best and which parts of the model we should continue to work on. The overall goal is always to make the forecasts as good as possible”, she says.

The possibility of making the grid size smaller in the operational weather forecast that is sent out to the public is there, but it requires more research on the effects on all parts of the model.

The quiet revolution continues into the future.

Reference:
Teresa Valkonen et al.: Evaluation of a sub-kilometre NWP system in an Arctic fjord-valley system in winter. Tellus A: Dynamic Meteorology and Oceanography, 2020. Doi.org/10.1080/16000870.2020.1838181

FACT BOX
Alertness will provide more accurate Arctic weather forecasts.

A four-year research project which will improve weather forecasts in the Arctic. The Alertness project, Advanced models and weather prediction in the Arctic, is led by the Norwegian Meteorological Institute and funded by the Research Council of Norway.

Researchers from these institutions participate: the Norwegian Meteorological Institute (MET Norway), the University of Bergen (UiB), Norwegian Research Centre (NORCE), the University of Tromsø (UiT), The Royal Netherlands Meteorological Institute (KNMI), the Nansen Environmental and Remote Sensing Center (NERSC) and the University Center at Svalbard (UNIS).

This article was first published in Norwegian on Forskning.no.