The release of the Intergovernmental Panel on Climate Change (IPCC) report sent shockwaves throughout the general public. The report, which is the first major review of its kind since 2013, indicates a troubling rise in international temperatures. UN scientists found that temperatures on Earth will likely exceed the 1.5C limit established in the Paris climate agreement in around two decades, adding that a near-2m rise in sea levels by the end of this century "cannot be ruled out". 
All of these projections were made possible thanks to Earth Systems Models (ESMs). ESMs integrate the interactions of atmosphere, ocean, land, ice, and biosphere to estimate the state of regional and global climate under a wide variety of conditions. Typically, such projections take immense super-computing power — until now.
Christopher Fletcher, an associate professor in Geography and Environmental Management (GEM) is an ESM expert in the Faculty of Environment at Waterloo.
Thanks to a partnership with Microsoft AI for Earth via Waterloo AI Institute and graduate students John Virgin (GEM) and William McNally (ENG) they’ve developed a machine learning technique to effectively, and more efficiently identify optimal parameters for the calibration of Earth System Models (ESMs).
These are quantitative projections of future global climate changes, which are critical to inform policy and decision-making as they deliver a more detailed view of earth systems using a fraction of the computing power such modelling took in the past. Thanks to Fletcher and his team we can calibrate the ESMs more efficiently, that will leave more supercomputing cycles free to perform more actual "science," according to Fletcher. “We would hope that this work helps to enable more and better scientific discovery.”
The process used by Fletcher and his team is known as emulation where a statistical model is trained to learn the behaviors of a more complex model.
Fletcher’s team has been able to replicate results that look very similar to the original, more expensive model. Fletcher’s recent paper, Toward efficient calibration of higher-resolution Earth System Models published in the journal Climate Change AI, uses precipitation as an example earth system.
“Precipitation tends to be a challenging target for ESMs, due to its variability in space and time,” says Fletcher. As we understand the science better, we know exactly what the problem is, and the magnitude of the problem that we're dealing with. Then we can know where to find a solution."
This new process works with a convolutional neural network (CNN), trained to run simulations from two lower-resolution (and thus much less expensive) versions of the same ESM, and a reduced number of higher resolution simulations. The idea was developed by a research assistant and engineering student, William McNally. The goal is to invert the regular process, which progressively simplifies any data input (an image) to highlight the most prominent features given.
The parameter tuning, which normally demands a significant amount of CPU energy to output such complex renders of global precipitation and other physical events, can now be reconfigured based on the cross-validation of this CNN. A reduction of 20-40 per cent in CPU processing time can now be achieved.
“If we're able to make this process of calibration, both more efficient and more robust for these modeling centers meanwhile saving a bunch of CPU time,” says Fletcher. “We can have as much confidence as possible in the future projections that emerge from them. We don't have crystal balls to look forward in time, we just have these computer models. And so we need those models to be as good as possible."
For the last several decades, most of the large supercomputing centers around the world are being monopolized by weather and climate simulations, running models that are incredibly computationally, economically, and environmentally intensive.
All the major modeling centers around the world, including in the Canadian Center for Climate Modeling and Analysis, have all been in non-stop preparation, running these models, developing the systems necessary to predict our future climate in preparation for IPCC AR6 through the project CMIP Phase 6 (CMIP6) - World Climate Research Programme.
“Ultimately, these results have a profound impact on how decisions will be made to adapt to our rapidly-changing climate crisis,” says Fletcher.
