Constraining Projections Using Decadal Predictions

Daniel J. Befort, Christopher H. O’Reilly, and Antje Weisheimer

Geophysical Research Letters
Work Package 5
Link: https://doi.org/10.1029/2020GL087900

Highlights

Initialised decadal predictions, which use data about today’s climate to forecast climate changes in the future, have undergone rapid development recently. However, these initialised predictions are limited to 10-year forecasts, meaning that the increasing demand for climate information out to 50 years in the future can only be met with uninitialised projections. This paper presents a new technique which chooses a subset of uninitialised projections based on which are closest to decadal predictions in their first 10 years. As long as initialised predictions perform better than uninitialised models, this approach might provide significant benefits in the long-term projections, even beyond the 10 years covered by the decadal predictions. This improved information allows policymakers and communities to make better informed decisions around climate adaptation and mitigation, using more skilful projections of how the climate may change in the coming decades.

Background
Recently, significant efforts have gone into producing climate predictions up to 10 years in the future, known as decadal predictions. These are initialised predictions, meaning the simulations are supplied with data about the current state of the climate system at the start of the prediction, which significantly improves their predictive abilities. There is increasing demand for robust climate information that goes beyond these decadal predictions, out to 50 years in the future. Initialised projections cannot be used for this as 10-year forecasts are their maximum; instead uninitialised climate projections are used. These are thought to be less able to predict short-term variations as they don’t include information about the current state of the climate. This study attempts to combine these approaches to improve our long-term projections, which are crucial to society’s efforts to adapt to long-term climate change and mitigate its effects.

Results
The new method presented here involves constraining uninitialised projections using initialised 10-year predictions, choosing those long-term projections which are most similar to the decadal prediction over their first 10 years. For surface temperatures over the North Atlantic Gyre region, the team found that this sub-selected ensemble provides a more skilful constrained projection ensemble than the larger ensemble encompassing every model simulation. Not only are the projections improved over the 10 years they share with the decadal predictions, they also show improvements beyond 10 years in the future. The improvements shown so far are rather modest, and this method relies on initialised predictions showing improvements over uninitialised projections, which is not always the case. Indeed, this method yielded markedly greater improvements in skill in the North Atlantic than it did over Europe. These findings are an important validation of this constraint technique, which stands to become more useful as decadal predictions continue to improve.

Methods
This study made use of 53 CMIP5 climate models, 9 decadal prediction models producing 72 simulations between them, and 44 uninitialised projection models producing 154 simulations. These simulations covered the period from 1961 to 2015, with the decadal predictions running for 10 years and beginning in each year from 1960 to 2005. For each start date until 2001, uninitialised simulations were then chosen from the full ensemble that were closest to the decadal prediction over the following 10 years, and the performance of this subset was compared with the full range of uninitialised projections and with contemporary climate observations. The analysis primarily focused on surface temperatures over the North Atlantic Gyre region (an area of circulating ocean currents), and to a lesser extent on Europe and the NINO3.4 region of the tropical Pacific.

Policy relevance
The results of this study lend confidence to a useful new technique for improving our climate projections potentially many decades into the future. This allows people and policymakers to make better informed decisions around how best to mitigate climate change or adapt to its effects, in light of these improved projections.