From movie proposals to autonomous vehicles, machine learning has revolutionized modern life. Experts are now helping to solve one of mankind’s greatest problems: climate change. With machine learning, we can use our wealth of historical climate data and observations to improve predictions about the future climate of the earth. And these predictions will play an important role in reducing our climate impact in the years to come.
What is machine learning?
Machine learning is a branch of artificial intelligence, and while it has become a buzzword, it’s essentially a process of extracting patterns from data.
Machine learning algorithms use available data sets to develop a model. This model can then make predictions based on new data that was not part of the original data set. To get back to our climate problem, there are two main approaches machine learning can use to help: We improve our understanding of the climate: observations and modeling.
In recent years, the amount of observational and climate modeling data available has grown exponentially. It is impossible for humans to go through all of this. Fortunately, machines can do that for us.
Observations from space
Satellites continuously monitor the ocean surface and provide scientists with useful information on how ocean currents are changing. NASA’s Surface Water and Ocean Topography (SWOT) satellite mission — scheduled to launch late next year — aims to observe the ocean surface in unprecedented detail compared with current satellites. But a satellite cannot observe the entire ocean at once, it can only see the part of the ocean underneath, and it will take the SWOT satellite 21 days to travel the whole world.
Is there a way to add the missing data so that we always have a complete global picture of the sea surface? This is where machine learning comes in. Machine learning algorithms can use the data retrieved from the SWOT satellite to predict the missing data between each SWOT revolution.
Obstacles in climate modelling
Observations tell us about the present, but to predict future climate we have to rely on comprehensive climate models. The latest IPCC climate report is based on climate projections from various research groups around the world. Scenarios that cast predictions hundreds of years into the future.
To model the weather, computers overlay the oceans, the atmosphere and the land with a computational grid. Based on the current climate, they can then solve the equations of motion for liquids and heat in each box of this grid to model how the climate will develop in the future. The size of each box on the grid is what we call the “resolution” of the model. The smaller the frame size, the finer the flow details that the model can capture.
But running climate models that project hundreds of years into the future brings even the most powerful supercomputers to their knees. Therefore we are currently forced to operate these models with approximate resolution. In fact, sometimes it’s so gross that the flow has nothing to do with reality. Ocean models used for climate projections, for example, often look like the one on the left, but in reality the ocean current looks a lot more like the image on the right.
Unfortunately, we currently do not have the computing power to operate realistic, high-resolution climate models for climate projections. Climate researchers are trying to find ways to include the effects of small-scale and fine turbulent movements in the above picture in the coarse-resolution climate model on the Left. If we succeed in doing this, we can create climate projections that are more precise but still mathematically feasible. This is what we call “parameterization” – the holy grail of climate modeling.
Quite simply, then we can achieve a model that does not necessarily contain all of the complex flow characteristics on a smaller scale (and requires a lot of computing power), but can still integrate its effects into the overall model in a simpler and cheaper way.
A clearer picture
Some parameterizations already exist in coarse-resolution models, but they often cannot work well to effectively integrate small-scale flow properties. Machine learning algorithms can use results from realistic high-resolution weather models (like the one on the right) to develop much more precise parameterizations. As our computing capacity grows along with our climate data, we will be able to use increasingly sophisticated machine learning algorithms to examine this information and offer improved climate models and projections.
