I have a times series for temperature and I am trying to decompose it into a

seasonal + trend + stochastic process

to look at what model would fit for the stochastic process.

However, I only have data for 333 days, and I assume my data should include a periodic phenomena, with the period being one year, or 365 days. How can I eliminate the trend and seasonal component when I don't have data for the entire assumed period? Is there a way to implement the method in R?

Here is the series http://www.wikiupload.com/XD5FHJFC7S857MG

EDIT: Plot of this data by @Penguin_Knight

enter image description here

  • $\begingroup$ Why is 333 days "one and a half year?" $\endgroup$ – Penguin_Knight Mar 20 '13 at 20:04
  • $\begingroup$ Edited, my mistake, sorry about that $\endgroup$ – l3win Mar 20 '13 at 20:32
  • 1
    $\begingroup$ I can't believe that it's possible to calculate a seasonal effect for a data set that does not include each season more than once. Much less a data set that doesn't have any data from roughly three months of the year. $\endgroup$ – Wayne Mar 20 '13 at 21:00
  • 1
    $\begingroup$ You are correct, @Wayne: it is impossible, because season and "trend" are completely confounded; there is no information to separate one from the other. $\endgroup$ – whuber Mar 20 '13 at 21:48
  • 1
    $\begingroup$ I fear that without some heroic assumptions (unverifiable from this data and possibly difficult to justify), meaningful progress will be impossible. $\endgroup$ – Glen_b -Reinstate Monica Mar 20 '13 at 22:14

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.