I'm going to hold off on voting until the May monthly data is in, but I can report a bunch of predictions based on the data through April. This is a long post because it includes seven different predictions. If anyone wants to discuss the different prediction techniques, it might be better to do that in the sea ice area and extent data thread. If people think this entire post belongs there instead of here, I can move it.

These are purely statistical predictions based on the available extent data, and do not reflect weather, ice conditions, etc. Note that the behavior of the ice changed in 2007, so the quality of these predictions reflects the ability, or lack thereof, of these predictors to capture that change.

Starting with just the September data for each year, so these predictions could have been made in October and would not have changed, we have: a linear fit predicts 4.8 (million km^2) with 95% confidence interval 3.6-6.0. (Note that I have not attempted to validate any of these confidence intervals.)

If instead we assume there is a constant annual change plus noise (so this September = last September - constant + noise), then the prediction is 5.3 with 95% CI 3.8-6.8. This value is high primarily because last year's extent was high.

If we fit a smoothing spline to the September data, the spline tracks the change in 2007 better than the other predictors, so predicts this year will be 4.4. I have not computed the 95% CI, but it is likely to be substantially larger than the other predictors.

If we build an ARIMA time series model for the September data, we predict 4.6, with 95% CI 3.5-5.7.

If we look at all of the data, the first thing we can try is taking the average change from April to September as a prediction of the change this year. Unfortunately, this does not reflect the changing ice behavior well, and predicts 5.8 with 95% CI 4.1-7.5. On the other hand, this should improve as we get closer to September, unlike the previous predictions.

Computing the monthly anomaly and fitting a smoothing spline fails for the same reason, predicting 5.5. As usual, I have not computed the CI for the spline, but it is likely to be very large. The spline does not track the data since 2007 well, and this is unlikely to improve even as more data comes in for the rest of the year.

Finally, fitting a SARIMA time series model to all of the monthly data predicts 4.7, with 95% CI 4.0-5.4. Frankly, I think CI is overoptimistically small, but the accuracy of the prediction should improve as the year goes on.

Once the May monthly data is available, I will update the SARIMA prediction and use that as my prediction.