Six cases of models

The previous post included an attempt to explain land surface temperatures as estimated by the BEST project using a dynamic linear model including regressions on both quarterly CO2 concentrations and ocean heat content. The idea was to check the explanatory power of these independent variables in the context of a strong assault on the viability of doing anything like that, even challenging the very notion of attempting to demonstrate a connection.

But to properly assess how good or bad these predictors are, CO2 concentrations and ocean heat content, it’s necessary to fit and calculate AICc values (see previous post) for six additional cases. The seventh case, with local level, CO2 regression, and ocean heat content regression, was reported in the previous post, and had this result:
best-DLM-land-only-regression-CO2-ocean-heat-20160605-163022
(Click on image to see larger figure. Use your browser Back Button to return to blog.)
This had an AICc of -718.3.

Note the AICc values are comparable for the land only temperature data. The smallest value is judged to be the best model of the bunch compared.

The six cases are:

  1. a local level model only
  2. regression on only the quarterly atmospheric CO2 series
  3. regression on only the ocean heat content series
  4. regression on quarterly atmospheric CO2 and on ocean heat content without local level
  5. local level with regression on quarterly atmospheric CO2 series
  6. local level with regression on quarterly ocean heat content series

The result from the local level only model, AICc = -534.9:
best-DLM-land-only-LL-only-20160606-213938
(Click on image to see larger figure. Use your browser Back Button to return to blog.)

The result from the CO2 series regression only model, AICc = -534.9:
best-DLM-land-only-regression-CO2-noLL-20160606-214833
(Click on image to see larger figure. Use your browser Back Button to return to blog.)
Note that the AICc is no different than the local level model only.

The result from the ocean heat content regression only model, AICc = -534.9, also no better than local level only, or CO2 only:
best-DLM-land-only-regression-ocean-heat-20160606-215239
(Click on image to see larger figure. Use your browser Back Button to return to blog.)

The result from the regression on atmospheric CO2 concentration series and ocean heat content, without local level, AICc = -592.8:
best-DLM-land-only-regression-CO2-ocean-heat-20160606-215627
(Click on image to see larger figure. Use your browser Back Button to return to blog.)

The result from the regression on atmospheric CO2 concentration series with the local level model, AICc = -685.7:
best-DLM-land-only-LL-with-regression-CO2-20160606-220500
(Click on image to see larger figure. Use your browser Back Button to return to blog.)

And, finally, the result from the regression on ocean heat content with the local level model, AICc = -690.0:
best-DLM-land-only-LL-regression-ocean-heat-20160606-221105
(Click on image to see larger figure. Use your browser Back Button to return to blog.)
This is, in fact, better than the CO2 only, or the local level and CO2. But it’s decidedly inferior to the regressions on CO2, ocean heat content, and the local level model, which had its AICc = -718.3.

Code and figures are in this tarball, which, again, contains the R image (workspace) having the data.

About hypergeometric

See http://www.linkedin.com/in/deepdevelopment/ and http://667-per-cm.net
This entry was posted in AMETSOC, anemic data, Anthropocene, astrophysics, Bayesian, Berkeley Earth Surface Temperature project, BEST, carbon dioxide, climate, climate change, climate data, climate disruption, climate models, dlm package, dynamic linear models, dynamical systems, environment, fossil fuels, geophysics, Giovanni Petris, global warming, greenhouse gases, Hyper Anthropocene, information theoretic statistics, maths, maximum likelihood, meteorology, model comparison, numerical software, Patrizia Campagnoli, Rauch-Tung-Striebel, Sonia Petrone, state-space models, stochastic algorithms, stochastic search, SVD, time series. Bookmark the permalink.

One Response to Six cases of models

  1. Pingback: On Munshi mush | Hypergeometric

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