Home > GIStemp, LSCF > Lines, Sines, and Curve Fitting 6 – backcast and forecast

Lines, Sines, and Curve Fitting 6 – backcast and forecast

2011 January 14

This is a long post, so I am going to place it behind the “fold.” Click on the read more link to see the curve fits from the previous posts backcast and forecast. This might be a good post to practice your browser’s zoom features.

Line with Sine on Residuals

line sine 1900 2000

R2[1900-2000]= 0.89
R2[1880-2000]= 0.90
R2[1880-2010]= 0.91

line sine 1880 2010


Sine with Line on Residuals

sine line 1900 2000

R2[1900-2000]= 0.86
R2[1880-2000]= 0.85
R2[1880-2010]= 0.81

sine line 1880 2010


Sine with Sine on Residuals

sine sine 1900 2000

R2[1900-2000]= 0.86
R2[1880-2000]= 0.64
R2[1880-2010]= 0.55

sine sine 1880 2010


Line and Sine Simultaneous

both line sine 1900 2000

R2[1900-2000]= 0.89
R2[1880-2000]= 0.90
R2[1880-2010]= 0.91

both line sine 1880 2010


Sine and Sine Simultaneous

both sine sine 1900 2000

R2[1900-2000]= 0.89
R2[1880-2000]= 0.80
R2[1880-2010]= 0.86

both sine sine 1880 2010


Exponential with Sine on Residuals

both sine exp 1900 2000

R2[1900-2000]= 0.89
R2[1880-2000]= 0.90
R2[1880-2010]= 0.92

both sine exp  1880 2010


Sine with Exponential on Residuals

both exp sine 1900 2000

R2[1900-2000]= 0.86
R2[1880-2000]= 0.84
R2[1880-2010]= 0.81

both exp sine 1880 2010


Exponential and Sine Simultaneous

both exp sine 1900 2000

R2[1900-2000]= 0.89
R2[1880-2000]= 0.90
R2[1880-2010]= 0.92

both exp sine 1880 2010


We have some obvious “winners”:
[89,90,92] Exponential and Sine Simultaneous
[89,90,92] Exponential with Sine on Residuals
[89,90,91] Line and Sine Simultaneous
[89,90,91] Line with Sine on Residuals

And some obvious “losers”:
[86,64,55] Sine with Sine on Residuals
[86,84,81] Sine with Exp on Residuals
[86,85,81] Sine with Line on Residuals
[89,80,86] Sine and Sine Simultaneous

About these ads
  1. Ned
    2011 January 14 at 5:43 am | #1

    The lesson here seems to be “Don’t try to include a sine function in models of recent temperature trends.” Am I right?

    You can get what looks like a good fit … if you use all the available data and don’t do any outside-of-sample testing. Your exp + sine model looked good for the 20th century, but underpredicted the past decade’s temperatures. The other ones are even worse.

    One can make a nice-looking exp + sine model using all the data 1880-2010. Will it still look good in 2020? Who knows?

    Perhaps somebody should run this on the internet:

    1,$s/climate cycle/climate oscillation/

  2. 2011 January 14 at 11:36 am | #2

    I don’t know, Ned. I think the lesson might be “things changed in the late60s/early70s.

    There was a flat/cooling period of about 25-30 years from the early 40s to the late 60s-early 70s. Tisdale lately called it a climate shift. Tamino notes it as the beginning of something new.

    Was the 25-30 year cooling natural? If so, why? If so, is it periodic? For me, those questions are still on the table.

Comments are closed.
Follow

Get every new post delivered to your Inbox.

Join 27 other followers