Data-Driven Predictions of the Time Remaining Until Critical Global Warming Thresholds are Reached
Collaborator Prof. Elizabeth Barnes (Colorado State University) and co-author Prof. Noah Diffenbaugh (Stanford University) trained neural networks to predict the number of years until particular global temperature thresholds (e.g. 2 degrees warming above pre-industrial) are reached. Their approach included a quantification of uncertainty with each prediction as well. They trained the neural networks on climate model data alone, and then applied the trained network to observational data. In this way, the trained neural networks act in a way to merge a range of disparate climate model simulations with the real world to provide estimates (with their uncertainty) of 21st Century climate change.
The United Nations Paris Agreement articulates the goal of “holding the increase in the global average temperature to well below 2 °C above preindustrial levels and pursuing efforts to limit the temperature increase to 1.5 °C above preindustrial levels”. This provides a central estimate for the 1.5°C global warming threshold between 2033 and 2035, including a ±1σ range of 2028 to 2039 in the Intermediate (SSP2-4.5) climate forcing scenario, consistent with previous assessments. However, the observational results suggest a higher likelihood of reaching 2°C in the Low (SSP1-2.6) scenario than indicated in some previous assessments—although the possibility it could be avoided is not ruled out. The data-driven framework presented here thus provides a unique, data-driven approach for quantifying the signal of climate change in historical observations and for constraining the uncertainty in climate model projections.
Given their policy relevance, the time remaining until global warming thresholds (e.g. 2C above preindustrial, 1.5C above preindustrial) are reached is of great interest. However, climate models largely disagree on these timings and it is unclear how much time the real world has until reaching these critical thresholds. Here, Diffenbaugh and Barnes train data-driven, machine learning models to predict (with uncertainties) the time remaining given a suite of climate model data as input. After training is complete, they then put observed data into the trained model and predict the time remaining based on observed warming patterns. By training over a range of climate model data, the machine learning algorithms must learn reliable patterns that are consistent across climate models for predicting the time remaining. In this way, the approach uses climate model projections to constrain observed warming. Furthermore, using explainable machine learning techniques, the study is able to visualize which regions of the globe are most important for accurately predicting the time remaining. Given the substantial existing evidence of accelerating risks to natural and human systems at 1.5 °C and 2 °C, the results provide further evidence for high-impact climate change over the next three decades.