September 2011 Arctic Sea Ice Extent Forecast

In this post, I use a quadratic regression model to forecast the  September, 2011  Arctic Sea Ice Extent. The model was developed with  1980 – 2010 data. Links to the R script, source data and  how-to article on polynomial regression are provided.

Arctic Sea Ice Extent Forecast for September, 2011

First, here is my forecast: (Click image to enlarge)

ASIE_forecast_2011

Based on the 1980 – 2010 downward Arctic Sea Ice trend,  my forecast is that September, 2011 SIE will decline  0.36 below 2010 levels, to 4.54 million km^2, with a confidence band of +- 0.59.

How Did I Develop My Forecast?

I have written a number of posts on Arctic Sea Ice Extent (here, here, here). In this post, I used the NSDIC‘s monthly data file (link)  to construct a quadratic regression model of September sea ice extent for the 1980 – 2010 period. I then used this model to predict the September, 2011  Arctic Sea Ice Extent.

I have 2 main learning curve sources for this model:

  • Tamino‘s post on Arctic Sea Ice decline provided the basic idea of using a quadratic model to fit Arctic SIE decline.
  • John Quick’s tutorial on polynomial regression provided the how-to instructions I needed to implement Tamino’s approach in R.

RClimate Script and Links

Here is the link to my RClimate script.

About these ads

9 responses to “September 2011 Arctic Sea Ice Extent Forecast

  1. Looking at the conditions of the Arctic Ice this winter, I expect that the Arctic Ice Minimum will be lower than the record minimum set in 2007. The Cryosphere Today site showed that much of the Arctic Ice was at less than 100% coverage. As far as I could determine looking at CT’s archived views, this is the first winter that has occured. The color coding of the ice showed colors of light purple, red, and yellow, meaning that the ice was at 85%-97% coverage over large areas. This was extraordinary.

    Of more significance to me, though, was the concept that we could be witnessing a long term decline of the Arctic Ice Winter Maximum. If that trend continues, that would result in a year round ice free Arctic Ocean. WOW! Now what would that imply for the melting of the Greenland ice cap?

  2. Hi Kelly, I’ve written a new blog post called First forecasts. Thanks a lot for letting me link to your forecast.

  3. The number and range you have derived appear reasonable for 2011, but the data set you are using appears to be flawed for predicting the rate of decline for years beyond 2011.

    What would the September line look like if the Arctic Basin, Canadian Archipelago, and the Greenland Sea were the only regions included in the analysis?

    Including the data from other regions creates a false sense of the steepness of the decline as the mathmatical analysis being applied to the line assumes that these regions could contribute negative values. Once a region reaches zero, it can not decline any further. Including data from regions that previously contained ice, but reached a zero level in September of 2010, skews the slope of the line. For that matter, including any region that is more than 50% depleted compared to the 1979 to 2008 mean level will generate an exaggerated rate of decline that can not be maintained.

    For example, the East Siberian Sea was almost 300,000 km2 below the 1979 to 2008 mean in 2010 and had only 60,000 km2 left in 2010. The Laptev Sea was more than 50% depleted in 2010. The following regions only had a negligible amount of ice in 2010 at the September minimum but previously contributed to the decline.

    Beaufort Sea – decline of about 130,000 km2
    Kara Sea – decline of about 100,000 km2

    The Chuckchi Sea has declined by 100,000 km2 from the 1979 to 2008 mean and hit zero prior to September.

    While the loss of ice from these regions will no doubt cause the rate of decline in the Arctic Basin to increase, there does not appear to be any reason to believe that ice above 80 degrees north latitude, which receives less solar and ocean heat input will decline in the future at the same rate shown by ice below 80 degrees north latitude.

    Please use only data from the Arctic Basin, Canadian Archipelago, and the Greenland Sea to perform your calculations.

  4. Sorry, I made a mistake for the IC of Tamino.
    Tamino used PI for the CI, because the CI (PI) for the 2010 forescast goes down 4 (CI for the trend in 2010 is > 4, and no for the PI ) .
    http://tamino.files.wordpress.com/2010/09/arcmin.jpg

  5. I don’t understand why you and Tamino, give only the confidence intervals of the trend in 2011 (2010 for the graph of Tamino) to predict the probable 2011 september anomaly (2010 for Tamino).
    For me, with this quadratic model, the 2011 september anomaly must be in the prediction intervals (for me in this case, IP are the IC for the september value 2011 only, and no the IC for all the line trend in 2011) :
    http://meteo.besse83.free.fr/imfix/SIEprevsep2011IPquad.png

    For example, I make a comparison of the IC and IP with quadratic and linear model :
    IC trend: http://meteo.besse83.free.fr/imfix/SIEprevsep20112.png
    IP value : http://meteo.besse83.free.fr/imfix/SIEprevsep2011IP.png

    In this graph, http://rhinohide.org/gw/trendtester/img/trendtest-GISTEMP-2000-100.png , Ron say it is CI 95%, but in its R script, it is the prediction intervals 95% (for the red and blue line)

    In the different climate blogs, the IC /IP are not always clear without to see in the code. :-)

    Here, we can see some examples of IC and IP, and other help of R scripts :
    http://zoonek2.free.fr/UNIX/48_R/11.html
    http://zoonek2.free.fr/UNIX/48_R/all.html

    Thank you !

  6. Hi Kelly,

    Is this the first 2011 minimum extent forecast or what? :-)

    If it’s okay with you, I think I’ll refer to this as soon as the Arctic sea ice blog comes back to life again.

    Cheers,

    Neven

    • Neven

      Hope your “winter break” is going well. Id be happy to see you discuss this in your 2011 Arctic Sea Ice blog.

      Kelly

  7. Thanks for your post!
    Your RClimate script isn´t running (just plotting the model).
    Object “a” is not defined:
    a <- predict(q_m)
    and then:
    # Add quadratic trend line
    points(sdf$yr, a, type="l", pch=16,col="blue",lwd=2

    • Edi

      Thanks for catching that. I missed in editing my original script for posting.

      You are now part of the script.

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s