As regular readers know, we have been using daily gridded sea ice data from satellite microwave sounders published by the NSIDC to discover what we can learn both from and about the data. With Layman Lurker’s help we recently discovered that nearly 100% of the sea ice to the South of 72 degrees North latitude and the equator has melted every year since satellite records started. Since the Antarctic sea ice is known to melt annually as well, we can say that nearly all of the sea ice south outside of 18 degrees from the North pole vanishes every single year. Figure 1 below takes a moment to interpret, but is interesting in that it shows the loss and recovery of Arctic sea ice over a typical year (2003 was chosen at random). This post has a video of all years for your confirmation.
The near complete melt of Arctic ice can be seen at Day of the Year ~250 and latitude 72. The red dot below indicates the visually approximated time and latitude of maximum melt.
So I masked off and ignored all of the data north of 72 degrees representing basically the only multi-annual sea ice on Earth. This assumption corresponds to all of the ice outside of the bright green circle in Figure 3 below. Notice that there is a lot of Northern hemisphere sea ice to the South of this circle.
Figure 4 depicts the approximate latitude for maximum and minimum sea ice formation in our comfortable interglacial period. Since the Antarctic ice also melts nearly completely every year, the red line at the North and upward represents the entire area of multi-year sea ice on Earth. The lines in the southern hemisphere give an interesting reference but you can confirm from video that the Antarctic sea ice melts almost completely every year.
So with all of this groundwork in place, we can now look at the behavior of single year sea ice trends from a reasonably knowledgeable position. The Northern hemisphere trend of single-year ice (Figure 5) is strongly significant – meaning we can measure it easily outside of its normal variance. A linear least squares fit reveals a downward trend of about 800,000 km^2 over the length of the satellite record but remember, the ice in this entire post basically reaches zero area every summer. For reference, there is an average of about 4 million km^2 of this ice in the Northern hemisphere averaged over the entire 34 year record. In a future post, we will look at how this ice trend can be compared to the perennial ice further to the north but from past work, this decline represents a bit under half of the total northern hemisphere trend.
Below is the Southern hemisphere sea ice. I did not mask the Antarctic data because Antarctic sea ice melts almost completely in the summer. The Antarctic has shown a widely un-publicized, statistically significant increase in ice over the years of satellite data. One critique I have heard in past analyses (typically from the extremist types) is that the Arctic ice is most important because it is multi-year ice. We will look at that in the future but that criticism is what has driven this investigation. I would expect that in an extreme global warming event, the multi-year ice would show a heavy downward trend because it wouldn’t form as completely in the winter. The Antarctic has other ideas though.
Below, I have combined what amounts to nearly all of the annually formed sea ice into a single global ice anomaly (Figure 7). We have a negative trend which to my surprise, fell just short of the 95% significance estimate. We know the planet is warming so this was a little puzzling.
Now I’m sure that people will call this exercise “cherry picking” but I don’t think that is a fair critique having picked all of the annual sea ice. Perhaps I could spend even more time to mask the North pole by something other than latitude but I doubt it would change anything. The post shows the trend is nearly significant so other statistical tests would undoubtedly cross the 95% threshold and still other tests would expand the confidence interval even further from the trend. My intent though was to see just how much of the ice melt, was the result of reduced annual refreeze. So with that said, isn’t it interesting that global single-year sea ice has not shown a trend which is easily differentiable from noise in the past 34 years?
28 thoughts on “Single Year Global Sea Ice Shows Minimal Trend”
Garsh, that word Anomaly seems so wrong, even though il cognoscenti love it.
What engineer would use that to describe normalized trend data?
To me, the South Atlantic Anomaly truly deserves the name; someon left the detector on by accident once on a transit, and it was the beginning of the end for HEAO-A2 soft x-ray detector.
Anyway, does any of the popular sources of ‘natural variability’ happen to correlate with Figure 7? This is a question for the peanut gallery, not the hardest-workin’ man in the blog bidness…
Also, how sensitive is the slope to the start/end year? Perhaps UC will bless us with one of his dynamic rolling thingies…
Nice Work, Boys!
Even more to the point, you have at least 30 years of data and can make a climate science claim. 😉
Thanks Jeff. Here is graph of the ice extent maximums south of 72 degrees. The maximums occur annually in March (vs April north of 72 degrees).
Jeff, any chance of running a smoothing filter over it (length approx. 12 months)?
I’m happy to do it if you don’t mind sending the data.
Jonathon, I have been using the sea ice concentration data from KNMI, but it is very close to what Jeff is using. Here is the south 72 lat data with 12 month MA smooth applied.
Thanks Layman Lurker, but it was Jeff’s figure 7 that I was referring to.
It is easy but what are you hoping to see?
This is Jeff’s figure 7 (different y scale) with the 12 mon MA overlaid. Hmmm… I should have used a two sided filter to properly center the smoothed series.
Howdja do it??
coef<-c(rep(.0833333,times=12))# vector of 12 equal weighting coefficients for a 12 month MA smooth
gl.si.ma<-filter(glo.si, coef, method=c("convolution"), sides=1)#MA filter applied to original time series to create smoothed series (use "2 sided" when overlaying original time series to center)
ts.plot(ts(glo.si[12:400],start=1979.75,frequency=12),ts(gl.si.ma[12:400],start=1979.75,frequency=12),col=1:2)#ts plot of smoothed and unsmoothed time series.
Oh I see why the smoothed series appears to lag – I used the wrong start date for the unsmoothed series. It should have been 1978.75
How did you get the timeseries data for filtering?
With the different patterns in the arctic and antarctic I have always wondered if there was a slight change in the “slope” of the earth — so that the north pole got a bit more of the sunlight and the south pole got a bit less.
It seems implausible given it is never mentioned, but on the other hand I have never seen anyone discuss and discount it — and I know that there are wobbles over time. So can anyone point me to a place where it is discussed (and presumably dismissed.)
Jeff, mostly I was wondering where most of the weight causing the trend came from. It looks like the trend can be hugely influenced by the start and end points. Also, there might be cyclic components of interest being masked by the noise.
The party line
I did a little R work this morning wanting to look at possible structural breaks in the north and south 72 latitude sea ice time series. Using the “breakpoint” function in R’s strucchange package, I generated the following graphs: north of 72 lat and south of 72 lat. While doing some google searching to help prepare I came upon this post at Kelly O’day’s blog with some code posted that came in handy for these graphs. The confidence intervals for the date of the structural breaks are shown along the x axis.
Right now I am reading up on the data documentation at NSIDCand RSS V6 (Wentz) to study the circumstances of combining the SSM/I microwave emission data from the F8, F11 and F13 satellites. There is some interesting stuff which seems to relate to these breakpoint dates, but I want to read and ponder a bit more before posting.
We are working on similar things. I wonder if you would mind pasting your breakpoint code for calc/plotting here?
north<-ts(north72,start=1978.75,frequency=12)#create ts object from "north72" data vector
plot(north,ylab="anomaly",main="Sea Ice Breakpoint Analysis North of 72 Latitude")
bp.north<-breakpoints(north ~ 1)#apply breakpoint function to create vector of breakpoints.
breakdates(bp.north) #breakpoint dates are: 1989.750 1999.417 2005.417
lines(bp.north,breaks=NULL,lty=2)#dashed lines for breakpoints
segments.north <- lm(north ~ breakfactor(bp.north))
ci.north <- confint(bp.north)
lines(ts(fitted(segments.north), start = 1978.75,frequency=12), col = "red")#shows magnitude of offsets
lines(ci.north, col = "red")#plot ci's
big h/t Kelly O’Day
It might be worth considering when they switched between sensors/satellites. I’ve not gone back to check the dataset you are using but the Bootstrap details are:
Four sets of satellite data are used to create the Bootstrap sea ice data stream:
Nimbus-7 SMMR, data range: 26 October 1978 through 20 August 1987
DMSP-F08 SSM/I, data range: 9 July 1987 through 18 December 1991
DMSP-F11 SSM/I, data range: 3 December 1991 through 31 December 1996
DMSP-F13 SSM/I, data range: 5 May 1995 through 31 December 2007
DMSP-F17 SSMIS, data range: 1 January 2008 through most current processing
The overlap will likely make it difficult to spot the switch-overs.
Right now I am spending a lot of time studying “The Version 6 Calibration of SSM/I” by Frank Wentz (2010). It explains the data processing methods and presents raw inter-satellite difference time series data for all overlapping periods. Each satellite is assumed to differ from other satellites by an offset due to unique orbit and instrument angles. In addition to the offset, there are other factors which play into a model of inter-satellite differences including an adjustment for time varying differences if deemed necessary. These calculations and calibrations have been much more complicated by problems with the DMSP-F10 satellite. Because of these problems, the last 1.5 years of data was truncated, and the satellite was excluded from the initial calibration mix. F10 was calibrated after the calibration of F11 through F15 was completed. F8 calibration was based on it’s overlap with the adjusted F10.
And this doesn’t even get into the time varying adjustments, or the “intra-scan” adjustment applied to all data from 2000 and on (not before).
#25, And we both know that the sat temp data suffers similar problems. Deeper and deeper we go into the muck.
“So I masked off and ignored all of the data north of 72 degrees representing basically the only multi-annual sea ice on Earth.”
What is the point of doing this? Why not just use the whole Arctic anomaly?
“the red line at the North and upward represents the entire area of multi-year sea ice on Earth”
No it doesn’t. It represents the are where multi-year ice can survive (or form). The area isn’t multi-year sea ice.
“The Northern hemisphere trend of single-year ice…”
Surely the loss of MULTI-YEAR ice is far more significant.
This is a really weird ‘analysis’ Jeff. It has no meaning,.
“One critique I have heard in past analyses (typically from the extremist types) is that the Arctic ice is most important because it is multi-year ice.”
It’s not JUST that, it’s that the loss of ice in the Arctic leads to greater warming. It’s likely to happen a lot faster than the Antarctic, and will happen in our lifetime. The reduction in Albedo is what counts, losing multiyear ice results in lower albedo, as new ice melts faster.
“We have a negative trend which to my surprise, fell just short of the 95% significance estimate. We know the planet is warming so this was a little puzzling.”
Why is this puzzling?
“So with that said, isn’t it interesting that global single-year sea ice has not shown a trend which is easily differentiable from noise in the past 34 years?”
No, because as you said earlier in the post all of this sea ice melts every year. Every year the area you picked sums to zero, so it can’t show anything other than noise. If you wanted to do something really interesting you would do an analysis of WHEN the melt happens. So is there a change in the date when the last ice has melted south of 72N?