Exploring the predictive skill of computer models when predicting Arctic weather

By Marvin Kähnert, ph.d. student, Bjerknessenteret

Over the last decades, climate change has led to greater temperature changes in the Arctic than anywhere else on this planet. This process, known as Arctic amplification, causes unprecedented and disruptive changes in the Arctic region.

Polar low. Photo: NASA

Polar low. Photo: NASA

We can see the Arctic amplification manifesting itself amongst other things in terms of a retreating sea ice, changes in the ecosystems and in weather patterns. In turn, these changes create profound socio-economic challenges as well as opportunities, connected to areas like enhanced tourism, transportation or exploitation of natural resources. All of this points to the fact that there is a growing demand for accurate weather predictions in the Arctic regions.

Marvin Kähnert, ph.d. student Photo: Anna Kathinka Dalland Evans

Marvin Kähnert, ph.d. student
Photo: Anna Kathinka Dalland Evans

There are several weather regimes in the Arctic that pose large difficulties for computer models. Among these are also so-called high impact weather (HIW) events. Examples of high impact weather are polar lows and persistent summertime fog, prevalent in the Arctic.

When studying complicated processes like weather systems in the natural world, we try to catch the underlying physical behaviours in specific equations and then use computers to solve them. Variables like wind speed, temperature or pressure are represented by a letter in our equations that describe the system. We refer to these letters as parameters, and they can have different numerical values. These values change according to our understanding of the atmosphere.

However, our computer models all have the drawback of having a finite vertical and horizontal resolution. The ‘model world’ is not an exact portrayal of our natural world, but more a pixelated version of it.

The problem now is that we can just compute the atmospheric state per pixel so to speak, leaving us blind for every process that happens on a smaller scale. They fall through the grid.

However, these missed processes highly influence the weather and its development and are often important for the end user. Thus, these smaller scale phenomena must somehow be represented in our models. In order to do that we employ so called parameterization schemes. A parameterization uses resolvable variables (e.g. wind speed in a model grid cell) to calculate these small scale phenomena (e.g. height of sea waves/swell due to wind speed).

It is important to recognise that parameterizations are simplified and idealised representations of complex physical processes. As a consequence, each of these schemes inevitable contains some degree of uncertainty. In my ph.d. project, I am working with understanding the behaviour of these schemes and how their uncertainties affect numerical weather prediction models, with special attention given to high impact weather events. I will have a detailed look into the representation of turbulence, the interactions between different parameterization schemes and the susceptibility of designated schemes to small changes to their formulas.

The study of parameterizations is a vital component when dealing with numerical weather prediction and will improve the overall utility of the model. I will give particular attention to the representation of turbulence in the model.

Read more:

WWRP Polar Prediction Project Science Plan. No. 1. World Meteorological Organization Tech. Rep., page 69 pp. 1.