We are working on a new statistical method for predicting interannual variability in global mean surface air temperature (GMST). The method uses the preceding few years of globally gridded temperature anomalies and Partial Least Squares regression to predict the GMST of the following couple of years. See our recent Nature paper for information on applying Partial Least Squares regression in a different climate context.
The plot below shows our forecast for 2017 (using no data from 2017) compared to the just-released 2017 value for the NASA GISTEMP dataset. We also show our forecast for 2018 and 2019 with 68% confidence intervals. The method suggests that 2018 is likely to be colder than 2017 but record warmth is ‘more-likely-than-not’ in 2019.
We are still in the midsts of sensitivity tests, the method is unpublished and it has not undergone peer review. Thus, these results should be considered to be part of a ‘beta version’ of our method.
Another caveat is that the method cannot possibly predict events like large volcanic eruptions which would drastically alter any annual GMST anomaly and invalidate our forecast.