The increase in global temperatures as a consequence of climate change has been a growing concern for a long time. The United Nations Paris Agreement has highlighted the urgency to maintain the increase in global average temperatures well below 2 °C and to ideally pursue the goal to limit temperature increase by 1.5 °C in order to mitigate the impact of climate change. Having an estimate of how much time it would take to reach these threshold temperatures is thus critical to better implement strategies to slow down global warming. While there has been research to identify the timescale to reach these threshold temperatures (‘time-to-threshold’), they are mostly based on linear extrapolations of the observed trends of global temperatures, which assume that temperature changes will follow the same trends as previously observed. In a recent study, Noah S. Diffenbaugh and Elizabeth A. Barnes leveraged artificial neural networks (ANNs) to develop a data-driven approach to more efficiently estimate the time-to-threshold, including the identification of regions that will likely see the increase in temperature in a shorter amount of time.
In order to train, test and validate the ANNs, the authors made use of previously established global climate models: for each of these models, they used flattened maps of temperature trends as input data for the ANNs. The climate models used in their approach have a wide variety of global warming rates classified as ‘Low’, ‘Intermediate’ and ‘High’. Based on the temperature trends observed until 2021, the authors predicted the time taken to reach a global temperature increase of 1.1 °C by 2022 (a known fact) in all of the aforementioned three scenarios for validation purposes. In the Low scenario, the time taken to reach this increase in temperature was 0 to 1 years, and in the High scenario, the time taken was 1 year, which implies that an overall increase of 1.1 °C would have been reached by the year of 2022. Despite varying warming rates, the time taken to reach a threshold temperature was similar in both cases, although there was a larger degree of uncertainty associated with the Low scenario.
This is a preview of subscription content, access via your institution