Combined insights from 10 models of the worldwide climate suggest that temperatures are rising faster than previously expected.
The alarming finding, published today in Environmental Research Letters, indicates that most land regions assessed by the Intergovernmental Panel on Climate Change (IPCC) will surpass the 2.7 degrees Fahrenheit (1.5 degrees Celsius) threshold above pre-industrial temperatures.
The researchers worked with an artificially intelligent system known as a convolutional neural network. Neural networks process and interpret information in a way inspired by the human brain. Convolutional neural networks are different from artificial neural networks in that they preserve spatial and temporal relationships in the data, and are very good at solving problems related to image recognition.
The research team trained a convolutional neural network on each of the 43 regions defined by the IPCC. The models were thus trained to predict future temperature changes based on region, rather than on a global scale, providing a more localized and equally weighted vision of shifts in the climate. The team also added a step of transfer learning, which fine-tuned the trained neural network with observational data that made the model’s predictions more realistic. Transfer learning was possible with data from 34 of the IPCC regions.
“It is important to focus not only on global temperature increases but also on specific changes happening in local and regional areas,” said Noah Diffenaugh, a climate scientist at Stanford University and co-author of the research, in a university release. “By constraining when regional warming thresholds will be reached, we can more clearly anticipate the timing of specific impacts on society and ecosystems.”
“The challenge is that regional climate change can be more uncertain,” Diffenbaugh added, “both because the climate system is inherently more noisy at smaller spatial scales and because processes in the atmosphere, ocean and land surface create uncertainty about exactly how a given region will respond to global-scale warming.”
The team predicted temperature increases at multiple temperature thresholds: the 2.7 degrees Fahrenheit (1.5 degrees Celsius) mark, 3.6 degrees Fahrenheit (2 degrees C), and 5.4 degrees Fahrenheit (3 degrees C). The group found that 34 regions are likely to exceed the first threshold by 2040—which sounds relatively far away until you realize how short 16 years can be. Of those 34 regions, the team found that 31 are expected to reach the second warming threshold by 2040, and 26 of them will surpass the third threshold by 2060.
“Our research underscores the importance of incorporating innovative AI techniques like transfer learning into climate modelling to potentially improve and constrain regional forecasts and provide actionable insights for policymakers, scientists, and communities worldwide,” said Elizabeth Barnes, a climate scientist at Colorado State University and lead author of the study, in the same release.
Diffenbaugh and Barnes also published results in Geophysical Research Letters today indicating a 50/50 chance that global warming will still exceed 3.6° Fahrenheit (2 degrees Celsius) even if humankind meets its goals of curtailing greenhouse gas emissions to net-zero in the next 30-odd years.
Unfortunately (shocking the first time that word is used in this article), when Earth’s climate is bent too far out of whack, certain changes are irreversible. Global temperature rises can cause “dangerous and cascading effects,” according to NASA, including heat stress to Earth’s denizens, including humans. The recent team’s study uses a new approach to confirm the hazards of climate change, courtesy of new, AI-fueled projections.
