A while back Stephan calculated the R values for Volume and Extent at the end of each month to the following September minimum. Accepting a few assumptions (that seem reasonable), squaring those values gives the R2, a measure of the percent of interannual variation that can be explained by knowing those values, and thus reducing the width of the confidence interval for estimating the subsequent September minimum.
https://forum.arctic-sea-ice.net/index.php/topic,2348.msg257955.html#msg257955 Improved Volume prediction accuracy as melt season observations become available. R-square reduction in variability derived from R values posted by Stephan at
https://imgur.com/a/O82kzZZ Linear model estimate for 2020 September average Volume before any 2020 observations: 3.9M (1.2 – 6.6) km3, (95% of cases expected to fall within 1.2 – 6.6M km3, i.e +/- 2.7M km3).
With MARCH observation as predictor, confidence interval (CI) reduction for September average Volume estimate:
4%. Width of 95% CI with March observation: +/- 2.6M km3.
With APRIL Volume observation, confidence interval reduction:
7%. Width of 95% CI with April observation: +/- 2.5M km3.
With MAY Volume observation, confidence interval reduction:
32%. Width of 95% CI with May observation: +/- 1.9M km3.
With JUNE Volume observation, confidence interval reduction:
63%. Width of 95% CI with June observation: +/- 1.0M km3.
With JULY Volume observation, confidence interval reduction:
83%. Width of 95% CI with July observation: +/- 0.5M km3.
With AUGUST Volume observation, conf. interval reduction:
94%. Width of 95% CI with August observation: +/- 0.2M km3.
-----------------------------
Those values indicate that until we have the end of May, and really the end of June, volume observations, prior observations don't give us much foresight about the September volume minimum. Which in turn suggests that melt season conditions (temperature, cloudiness, and storms) that take effect in June, July, and August are the primary determinants for the September volume minimum. (Some of those conditions, such as melt pond formation, may have been established earlier, but do not manifest as changes in volume until after May 31). Thus, about 68% of the variation in September minimum Volume becomes apparent after May 31 (100% minus 32% = 68%).
The same approach for Extent shows the R2 at the end of March, April and May at insignificant level, only reaching 22% by the end of June, and 56% at the end of July. Thus, changes that manifest in July and August account for 78% of the variation in September minimum Extent (100% minus 22% = 78%).
An expert analysis by Walt Meier and NSIDC concluded “Plotting the de-trended maximum versus minimum extent (Figure 2) shows a near-random distribution.” “The seasonal maximum extent and the September minimum extent are not correlated...“ "because summer weather conditions strongly shape the September minimum.”
Figure 2. This plot compares de-trended maximum extent (x-axis) with minimum extent (y-axis). The yearly values shown are calculated by subtracting the linear trend value for that year from the total extent. Credit: W. Meier, NSIDC. From “Maximum extent is not predictive of minimum extent”
https://nsidc.org/arcticseaicenews/2020/03/