Data Comparisons

Wed Ice Update:

21 Comments

Yes. For now it’s all ice, all the time. I admit watching this has distracted me from surface temperatures. This should end soon. Very soon. (Then I can start pondering things like “peer review” being discussed at all the other blogs. 🙂 )

The fascination with ice will have to end: we are approaching the area minimum and even though we aren’t betting on area, I can’t help looking at it daily. Note: We are not betting on area– we are betting on extent. But the extent minimum usually comes after the area minimum. This fact is making lots of us bettors watch to see if the minimum is reached.

Because we are so close, I’ve exploded the graph of daily NH ice areas based on data available at the The Cryosphere Today:

This years trace is shown in black.

Notice that roughly 40 days ago, the NH area was at record low values for this time of year and remained there for roughly 3-4 weeks. Then, ice area loss slowed relative to 2007 and for about a week, the CT area was no longer in record low territory. However, area then declined and area fell back into record low territory falling as low as 2.980 millions square kilometers. (See black dashed line.) When the area hit that value, I thought for sure it would soon fall below the all time record low of 2.919 million square kilometers, but reported ice area decided to increase for a little while, but began to fall again. The CT area now stands at 3.002 million square kilometers (this is noted with the grey dashed line hovering just above the black dashed line). This is a record low for this day of the year, but remains above the all time low of 2.919 millions square kilometers.

Will the CT ice area hit a new daily minimum? I’d say the chance is greater than 50%-50%. Based on average remaining losses for this day of year, we’d anticipate the ice area will hit 2.887 million square kilometers. On the other hand, the record low had already been reached by this time of year in 4 out of 32 years since 1979. (I didn’t pull out data back to 1972.) In comments, Carrick, Dewitt and others are debating whether the date for the area or extent minimums have shifted. Whether it has or it hasn’t, the data are noisy enough that it remains possible we could have an ice minimum this early, but it’s unlikely. If someone offers you an even money bet and you prefer to win rather than lose money, unless you know something about Arctic weather forecasting, it would be wiser to bet the minimum has not yet occurred. (If you know something about Arctic weather forecast, please share!)

Now, on to the bets.
The race for the Quatloos
The Quatloo bet is on 7 day area.

The forecasts are shown in below left (click to enlarge). According to The Blackboard’s forecasting method, the predicted minimum is 4.39 million square kilometers, with ±95% confidence interval of 4.19 t0 4.59. My own bet is now on the top edge of the prediction window– it’s unlikely I’ll win :(. The front runner is BobZ :). To those rooting for a new minimum, my forecast method pegs it’s likelihood at 13.4%; this means it’s possible. For those rooting that this years extent would exceed last years: my forecast now places the probability near “a snowballs chance in hell”. In fact, it’s unlikely the extent will exceed 2008’s value. The extent will almost certainly be lowest or 2nd lowest in the JAXA/GSFC record.

Above left, you can see the histogram. The current extent is shown with a black line. My bet is yellow. Values corresponding to bets are in purple. The forecast value is shown in blue. Bob’s Z’s status as front runner is shown in text in the lower left.

You will also note that I also checked to see how many times since 1972 the 7 day smooth had achieved a minimum by this date: The number is 5 times out of 39 years. So, not many. JAXA extent has been declining consistently. The current area and ice volume are low, and remaining losses tend to be higher when the area when these are low. Though not record breaking, recent extent losses were in the high end for this time of year.

Expect more losses to come next week.

21 thoughts on “Wed Ice Update:”

  9. Yes: I have to admit your mathematics is looking better than my complete guesswork! Fun, isn’t it?

  10. lucia,

    But the extent minimum usually comes after the area minimum.

    That’s been true recently, but if you look at the full record from 1979-2010, the average difference is only 1 day later for the extent minimum.

    When you splice the GSFC data to JAXA, do you transform the GSFC data based on the difference observed in the overlapping period? My calculation is that JAXA = GSFC*0.920186+227497.

  11. Dewitt–
    I don’t transform because there are only a few months of overlap. I don’t think we can be sure that the difference is something that would be consistent each year.

  12. Re: lucia (Sep 8 07:10),

    We must be getting data from different places. I have GSFC data through 2006 so there are over four years of overlap. Unfortunately, I can’t find that link any more. I think it’s buried somewhere in The Air Vent.

  16. Re: lucia (Sep 8 08:30),

    I have that one as well. You can do linear transforms between all of them. Sometimes you have to shift the date a day or two one way or the other. The interpolated data back to 2002 is new to me, though. A simple splice between the gsfc and JAXA data in 2002 is going to result in a large step change.

    gsfc 12/31/2002 13.256 Mm²
    JAXA 12/31/2002 12.721 Mm²

    I’m going to have to recalculate my transform based on the updated datasets. I can see from Carrick’s data that he has transformed the gsfc data to match the JAXA data. He’s also resampled it to a constant 365/year to make it easier to transform to frequency domain.

  17. DeWitt, I don’t rescale the data. All I do is average the two sets together in the overlap region. I ended up using this GSFC set because it has a longer baseline, and in the overlap region, the two sets looked very similar (suggesting to me they use the same or similar underling core algorithms for estimating extent).

    I also don’t know well simply rescaling will work in matching the different sets to each other.

    I like the data set that includes the different geographical regions, I wish there was one that went to current that included this. It’s a shame it ends in 2007.

  18. DeWitt

    The interpolated data back to 2002 is new to me, though. A simple splice between the gsfc and JAXA data in 2002 is going to result in a large step change.

    Sure. If I were trying to tease out changes in seasons etc. I’d worry about that a lot. But for what I’m doing, it’s an uncertainty. Given the short overlap between the sets I had, I didn’t think there was enough information to correctly “correct” one towards the other, so I didn’t “correct”.

    For my purposes, likely the best thing one could do is create a forecast using “corrected” and “uncorrected” data and report both. I haven’t considered it worth it for betting purposes.

  19. Re: Carrick (Sep 8 09:28),

    The GSFC and JAXA data are pretty close for the minimum (see graph). The maximum is quite a bit different. If you’re looking at the variation over the whole year, not rescaling is going to bias the results. There’s a residual seasonal variation that a linear rescaling doesn’t fix. I interpret that as a difference in the sensor and/or algorithm. That’s going to put a crimp in trying to look for differences in actual extent between GSFC pre-2002 and JAXA post-2002.

  20. DeWitt, I recognize that’s a problem with splicing the data sets. Of course this happens internally too, as which satellite you’re using changes, etc.

    If you look closely at the overlap, there is a shift in day of year where the seasonal effects kick in for JAXA relative to GSFC. That means at best, you’d have to use some type of complex valued transform to align the two series, assuming the relationship is even linear (which I actually doubt).

    The problem is JAXA starts just when things get interesting, and GSFC stops just when they really get interesting. It’s almost like you’d have to get the raw files, and process them yourself (ick) if you wanted to get a fully consistent picture.

  21. Re: lucia (Sep 7 16:50),

    I hope it levels off right now; it would surprise me if it did.

    The latest data from the MASIE page shows the Arctic Ocean extent still decreasing. That should bottom out before the total reaches bottom, so it looks like we have a few more days yet. Another sign is some actual increase in extent. Usually there is some fluctuation before there’s a steady upward trend. We haven’t seen that yet (except for CT area) either.

    Carrick’s quartic fit data gives an expected minimum date of 9/13. Using the actual minimum I get an expected minimum on 9/16. Using data that I didn’t have previously, there does appear to be a statistically significant trend in the minimum extent date amounting to an increase of ~7.5 days over 40 years.

Comments are closed.