New modelling system could one day help to predict mega-flood rainstorms

In 2021, Germany experienced one of its greatest natural disasters in recent history. A massive rainstorm dumped more than a month’s worth of rain in just 24 hours in several regions, causing widespread flooding. Thousands were displaced from their homes, more than 200 people died and infrastructure and property worth billions of euros were destroyed.

Meteorologists say that part of the strength and staying power behind the storm was a mesoscale convective system – a raincloud system that can span hundreds of kilometres and lasts longer than a normal thunderstorm. Currently, there is no effective way to predict mesoscale convective systems since how they form remains poorly understood.

But new research, published in the Journal of Geophysical Research: Atmospheres, could one day help to give advance warning for these enormous storms. Atmospheric scientists are developing an algorithm that can recognise cold pools: an atmospheric phenomenon that is thought to be involved in forming mesoscale convective systems.

“When we think about climate change and the extreme events we have observed in Europe over the last few years, it is vital to understand the processes that lead to the organisation of these big systems,” says Jannik Höller, lead author and an atmospheric scientist at the University of Copenhagen, Denmark who studies the organisation of clouds.

Convection and cold pools

A foundational principle of meteorology should be familiar from your oven settings – convection, the fact that warm air rises and cold air sinks.

A cold pool forms when rain evaporates before it reaches the ground, creating a region of cold air. Since cold air is denser than warm air, the cold air sinks, dispersing like ripples on a pond as it reaches the ground. “When it spreads to the side, it can push the warmer and lighter ambient air up and thus trigger new clouds and new convection,” Höller says. Under the right conditions, this new convection can lead to a mesoscale convective system.

Höller and his team set out to develop an algorithm that can scan satellite data for telltale signs of cold pools. They found that using just a few metrics – cloud-top temperature measured by infrared bands on satellites and estimated rainfall – might be enough to identify a nascent cold pool.

The algorithm performed well with simplified models – artificially generated satellite data that reflected atmospheric conditions less complicated than what meteorologists see in the real world.

But Höller and his team knew that the algorithm would need to parse more complex systems to be useful. To put it through its paces, they tested it on a simulated version of a cold pool hotspot – Dakar, Senegal.

A hotspot for cold pools

Senegal is an interesting case study for meteorologists because it is sandwiched between the Sahara Desert to the north and the moist tropics to the south, explains Jan Haerter, co-author and Professor of Atmospheric and Climate Physics at the University of Potsdam, Germany.

“Small shifts in this boundary between the desert and the tropics can affect this area very strongly,” Haerter says. “It is a hotspot for water availability and scarcity and for events of extreme rain.”

Senegal has strong easterly winds that push over the African landmass toward the west coast. This is conducive to forming cold pools, which is easier over a large landmass, and there are very few mountain ranges to disrupt these cloud formations. At the same time, “there are some hills that might lead to forced convection and the sea breeze from the Atlantic Ocean – very complex mechanisms that could confuse the algorithm and lead to false positives,” Höller says.

Höller and his team used real atmospheric data from Senegal on 4 August 2022, a day with plenty of observed convective systems, to establish the starting parameters for a new simulation. The researchers plugged in the temperature, humidity and other measures from that Thursday in Senegal and let the model run for 24 simulated hours.

The algorithm could determine whether pixels from satellite imagery were involved in a cold pool or not with an accuracy higher than 90%, the authors say – although it appeared to get tripped up by certain atmospheric phenomena.

The algorithm occasionally mistook cirrus clouds for cold pools because of their chilly cloud-top temperatures, and organised cloud fronts driven by that easterly wind apparently looked like the ripples of a cold pool’s gust front, causing false positives.

Sometimes the algorithm missed particularly large cold pools if it could not recognise a spreading gust front as one big shape. Fortunately, this may be less of a problem with real satellite data, because lower data resolution forces the algorithm to zoom out and process larger regions at a time – making capturing a recognisable shape more likely.

The future of nowcasting

The researchers say the next step is to retrain the algorithm using real satellite data. This will require atmospheric scientists to meticulously hand label cold pools, pixel by pixel, to show the algorithm what it needs to look for in the noise and context of a real satellite image.

“When the scores are as good as they are now for the simulated clouds, then I think we can use the method to identify cold pools in satellite images [RW1] all over the world,” Höller says.

Haerter is optimistic that the algorithm could one day be used in nowcasting, the shortest-term version of a meteorologist’s forecast. By detecting cold pools as they form, “we might be able to determine, maybe with a two-hour lead time, the likelihood of a devastating event such as Germany’s flood occurring,” Haerter says. That early warning could be broadcasted to residents of an area via their phone’s emergency alert system or by text message.

Although these mesoscale convective systems are big news in Europe, “The City of Dakar is flooded by mesoscale convective systems almost every year,” Haerter adds. “It is important to find out how to reduce the impact of these events and understand them better.”