I'd add 6th: Your underlying hypothesis of homoskedasticity is wrong - we expect that due to increased positive feedbacks and reduced memory effect (less old ice there...) the variability (and probably also the autocorrelation) should go up. Hence, the statistical significance of these events is overestimated by your eye.
The PIOMAS anomaly chart
is quite striking with anomalies now well above the long term trend, and above the range of the darker shading (1 standard deviation?). This fits in comfortably with a long term linear trend. However Tamino in the past had
shown a statistically significant acceleration and that the trend was not linear. Some possibilities:
1. Tamino's analysis is wrong.
3. My gut feel assessment that a return to the upper boundaries of the typical range for a linear trend must be a significant departure from the non linear trend is wrong.
3. A statistical fluke either caused the appearance of acceleration, or the return to above linear trend.
4. A significant factor (i.e. something not covered by the relevant statistical test for noise) caused a temporary increase in melt and this factor has now changed.
5. Something to do with Tamino's analysis being extent and this being volume.
6. Slow decline - the acceleration was real, but the trend is a gompertz curve that is now entering the deceleration phase.
[/quote]