CLIMATE PT.3
A scorching 3.5°C-degree rise or a moderate 0.5? Scientists are unable to predict how high temperatures will rise, but artificial intelligence and hidden satellite data patterns should give us a more accurate idea of future climate.
Climate models are good at predicting the past, but the variations are still too great for reliable predictions of the future. We urgently need more accurate and more localised models.
Is a disaster imminent, or do we have little to fear? Scientists have been processing huge quantities of data on the world’s biggest supercomputers over recent decades. Their climate models have improved markedly, but the future is still uncertain, because the results of the models vary. They predict a temperature rise of between 0.5 and 3.5 degrees by 2100 – and the two outcomes would result in radically different scenarios.
If the world becomes 0.5 degree warmer, we will see some increase in the number and intensity of extreme weather phenomena such as torrential rain and heavy storms, but the situation is probably manageable. If it becomes 3.5 degrees warmer, we face a real climate disaster in which hundreds of millions of people will need to escape from draught, heatwaves, forest fires, flooding and storms.
Either scenario requires action, so the question becomes what action to take. Yet in order to determine the action required, we depend on climate models telling us exactly how the climate will change and where the changes will have the most severe effect.
New applications of artificial intelligence and the ability to find previously-hidden satellite data patterns are now helping scientists to simulate cloud formations and to reveal the drivers of Earth's melting ice. The results should provide a more accurate impression of the future world.
Supercomputers determine climate
The most sophisticated modern climate models are known as global circulation models; these include atmosphere, oceans, ice sheets, and ecosystems. The models are ‘linked’, in that they calculate how different parts of Earth’s climate system affect each other. Rising air temperatures increase evaporation from the ocean, so increasing the water content of the atmosphere, so
increasing the quantity of precipitation, and so on. The interaction between the different factors is described in extremely complex computer code, and the calculations require groups of supercomputers.
Specifically, the models describe Earth’s climate system as a 3D network in which the atmosphere, the oceans and the land are all divided into cubes. In each cube, all the relevant climate variables – temperature, pressure, precipitation, clouds – are considered as the same throughout that cube. The cubes typically measure 100 100km horizontally and one layer of the atmosphere or the ocean vertically. The vertical layers are about 11km thick, and the models typically include about 30 layers of atmosphere and 20 layers of ocean. So each grid is an approximation, whereas in the real world one end of a cube might be covered in clouds while the sun shines in the rest of it.
Computers also divide time into steps which mark the flow from cube to cube of wind and ocean currents; these are typically calculated with intervals of 30 minutes.
Ideally, both the time steps and the cubes would be infinitely small, or at least smaller, perhaps one cubic kilometre with steps of a few seconds. But such calculations would be too great for existing computer power. So with the main model, the results delivered by the application of the laws of physics are supplemented with a kind of rule of thumb, to allow for the uncertainty as best as possible. However, the rules are based on estimates, and different weightings from model to model are an important reason why different final results are delivered.
Testing the past
One important step on the way towards improved climate models is to test how accurate the models are, and then adjust them towards their best possible accuracy. We can’t wait 50 years to see if the models get it right, so instead the models are tested by making them simulate the past climate.
Scientists have data about annual air temperatures back to the end of the ice age 11,700 years ago. The data comes from Greenlandic ice cores and from detailed data on tree ring thickness. Throughout this period there is reasonable agreement between the historic development and the results from today’s climate models.
Scientists also have global air temperature data from 1850 onwards. The climate models also simulate those accurately – but only when they include our emissions of greenhouse gases. If our emissions are left out, the models fail, as they do not reproduce the global temperature rise of 1.1°C that has occurred since 1880.
But on several counts, the models have so far under-estimated the consequences of global warming. The mass loss from the ice sheets of Greenland and Antarctica, for example, has been greater than predicted since the year 2000. In this case the models have since been improved by scientists incorporating the effect of warmer ocean water melting glaciers from underneath.
Ice-free sea intensifies warming
The Arctic sea ice has also receded faster than expected. Again an improvement has been made, in this case using the discovery of previously unrecognised patterns among decades of satellite data, which are helping scientists spot the errors in the models.
The retreating sea ice has global consequences because the white ice reflects sunlight into space, whereas dark, open sea absorbs most of the energy from sunlight, retaining large quantities of heat within the atmosphere. Calculations indicate that
the melting has increased the absorption of solar energy so much that it has contributed to global warming an amount corresponding to 25% of the contribution from increased carbon dioxide in the atmosphere. It is vital that scientists can predict how quickly the sea ice is melting.
All climate models agree that the Arctic Ocean will be ice-free during summers if emissions of carbon dioxide continue at the present level. In some models this will happen within one or two decades; others predict the end of the century. To narrow down the uncertainty, American scientists have taken a close look back at satellite data collected since 1979.
Ice cover in early September has halved since 1979, from some 8 million km then to less than 4 million today. At the same time, the ice volume has been reduced by 75% because of the increased quantity of thin first-year ice. American climate researcher Alex Hall and his colleagues have discovered that the change in the relationship between thin first-year ice and the thicker, older ice could explain the rapid melting. The UN climate panel has used this knowledge to narrow its predictions. The results indicate that the Arctic Ocean will become ice-free over summers by around the year 2050.
The cause of receding sea ice is the higher quantities of carbon dioxide in the atmosphere, and yet scientists still disagree over exactly how much carbon dioxide causes what level of global warming. The gas absorbs infrared heat radiation from Earth, which would otherwise disappear into space. But the radiation includes a spectrum of different wavelengths, and carbon dioxide does not absorb all wavelengths equally well.
Modern climate models are based on a weighted average of the absorption across wavelengths, but each model has its own method for calculating the average, and so they reach different results. New models will be able to calculate for individual wavelengths, reducing that uncertainty. On the other hand, such calculations again require immense computing power.
Clouds are tricky
Scientists face even bigger challenges when they try to calculate the future of cloud cover, which greatly affects the results of the models, since clouds can either cool or heat. Dense low-lying clouds cool the world, because they reflect large quantities of sunlight into space. High-lying thin clouds, on the other hand, allow the light to pass, and at the same time they absorb heat from Earth’s surface.
Dense low-lying clouds are the most common, and the net effect of cloud cover in the world today is cooling. But in a warmer atmosphere, both the existence of different types of clouds and their geographical distribution could change. The dense low-lying cloud cover is now primarily located above the tropics, but changing wind systems could push large quantities of low-lying clouds towards the poles. There they would reflect only the faint sunlight over these higher regions instead of the intense sunlight of the tropics. So in the future, cloud cover might heat the world instead of cooling it.
Again climate models have difficulties supplying accurate predictions because clouds form in much smaller spaces than those large atmospheric cubes used by the models. American scientists now hope to develop methods using artificial intelligence which will use satellite data to simulate cloud formation within a more realistic volume of just a few cubic kilometres. These will then be included in the global models.
Models predict the near future
In recent years, models have become ever better at predicting Earth’s climate for decades and even centuries ahead. But with climate change already taking place, increased accuracy is vital so that governments throughout the world can be more sure how their small parts of the world will be affected in the years to come.
So scientists are developing localised models better at predicting local conditions, providing a clearer image of climate development within the next decade.
The UK is a leader in this field. In 2016, scientists drew up a major report about the risk of flooding in different parts of England and Wales within the following few years. The predictions mined decades of detailed weather satellite data, with the scientists predicting that the quantity of precipitation during intense winter downpours will probably increase by about 10% over the next 10 years – but with an increase up to 30% in south-east England. The detail provided a guide to the UK Government, which intends to spend £700m on improving the country’s defences against flooding.